Latent Space: The AI Engineer Podcast — Practitioners talking LLMs, CodeGen, Agents, Multimodality, AI UX, GPU Infra and all things Software 3.0

The podcast by and for AI Engineers! In 2023, over 1 million visitors came to Latent Space to hear about news, papers and interviews in Software 3.0. We cover Foundation Models changing every domain in Code Generation, Multimodality, AI Agents, GPU Infra and more, directly from the founders, builders, and thinkers involved in pushing the cutting edge. Striving to give you both the definitive take on the Current Thing down to the first introduction to the tech you'll be using in the next 3 months! We break news and exclusive interviews from OpenAI, tiny (George Hotz), Databricks/MosaicML (Jon Frankle), Modular (Chris Lattner), Answer.ai (Jeremy Howard), et al. Full show notes always on https://latent.space www.latent.space

Latest episodes

Aug 4, 2023 • 59min

[AI Breakdown] Summer AI Technical Roundup: a Latent Space x AI Breakdown crossover pod!

Jul 26, 2023 • 55min

FlashAttention 2: making Transformers 800% faster w/o approximation - with Tri Dao of Together AI

FlashAttention was first published by Tri Dao in May 2022 and it had a deep impact in the large language models space. Most open models you’ve heard of (RedPajama, MPT, LLaMA, Falcon, etc) all leverage it for faster inference. Tri came on the podcast to chat about FlashAttention, the newly released FlashAttention-2, the research process at Hazy Lab, and more. This is the first episode of our “Papers Explained” series, which will cover some of the foundational research in this space. Our Discord also hosts a weekly Paper Club, which you can signup for here. How does FlashAttention work?The paper is titled “FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness”. There are a couple keywords to call out:* “Memory Efficient”: standard attention memory usage is quadratic with sequence length (i.e. O(N^2)). FlashAttention is sub-quadratic at O(N). * “Exact”: the opposite of “exact” in this case is “sparse”, as in “sparse networks” (see our episode with Jonathan Frankle for more). This means that you’re not giving up any precision.* The “IO” in “IO-Awareness” stands for “Input/Output” and hints at a write/read related bottleneck. Before we dive in, look at this simple GPU architecture diagram:The GPU has access to three memory stores at runtime:* SRAM: this is on-chip memory co-located with the actual execution core. It’s limited in size (~20MB on an A100 card) but extremely fast (19TB/s total bandwidth)* HBM: this is off-chip but on-card memory, meaning it’s in the GPU but not co-located with the core itself. An A100 has 40GB of HBM, but only a 1.5TB/s bandwidth. * DRAM: this is your traditional CPU RAM. You can have TBs of this, but you can only get ~12.8GB/s bandwidth, which is way too slow.Now that you know what HBM is, look at how the standard Attention algorithm is implemented:As you can see, all 3 steps include a “write X to HBM” step and a “read from HBM” step. The core idea behind FlashAttention boils down to this: instead of storing each intermediate result, why don’t we use kernel fusion and run every operation in a single kernel in order to avoid memory read/write overhead? (We also talked about kernel fusion in our episode with George Hotz and how PyTorch / tinygrad take different approaches here)The result is much faster, but much harder to read:As you can see, FlashAttention is a very meaningful speed improvement on traditional Attention, and it’s easy to understand why it’s becoming the standard for most models.This should be enough of a primer before you dive into our episode! We talked about FlashAttention-2, how Hazy Research Group works, and some of the research being done in Transformer alternatives.Show Notes:* FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness (arXiv)* FlashAttention-2* Together AI* From Deep Learning to Long Learning* The Hardware Lottery by Sara Hooker* Hazy Research* Is Attention All You Need?* Nvidia CUTLASS 3* SRAM scaling slows* Transformer alternatives:* S4* Hyena* Recurrent Neural Networks (RNNs)Timestamps:* Tri's background [00:00:00]* FlashAttention’s deep dive [00:02:18]* How the Hazy Research group collaborates across theory, systems, and applications [00:17:21]* Evaluating models beyond raw performance [00:25:00]* FlashAttention-2 [00:27:00]* CUDA and The Hardware Lottery [00:30:00]* Researching in a fast-changing market [00:35:00]* Promising transformer alternatives like state space models and RNNs [00:37:30]* The spectrum of openness in AI models [00:43:00]* Practical impact of models like LLAMA2 despite restrictions [00:47:12]* Incentives for releasing open training datasets [00:49:43]* Lightning Round [00:53:22]Transcript:Alessio: Hey everyone, welcome to the Latent Space podcast. This is Alessio, Partner and CTO-in-Residence at Decibel Partners. Today we have no Swyx, because he's in Singapore, so it's a one-on-one discussion with Tri Dao. Welcome! [00:00:24]Tri: Hi everyone. I'm Tri Dao, excited to be here. [00:00:27]Alessio: Tri just completed his PhD at Stanford a month ago. You might not remember his name, but he's one of the main authors in the FlashAttention paper, which is one of the seminal work in the Transformers era. He's got a lot of interest from efficient transformer training and inference, long range sequence model, a lot of interesting stuff. And now you're going to be an assistant professor in CS at Princeton next year. [00:00:51]Tri: Yeah, that's right. [00:00:52]Alessio: Yeah. And in the meantime, just to get, you know, a low pressure thing, you're Chief Scientist at Together as well, which is the company behind RedPajama. [00:01:01]Tri: Yeah. So I just joined this week actually, and it's been really exciting. [00:01:04]Alessio: So what's something that is not on the internet that people should know about you? [00:01:09]Tri: Let's see. When I started college, I was going to be an economist, so I was fully on board. I was going to major in economics, but the first week I was at Stanford undergrad, I took a few math classes and I immediately decided that I was going to be a math major. And that kind of changed the course of my career. So now I'm doing math, computer science, AI research. [00:01:32]Alessio: I had a similar thing. I started with physics and then I took like a programming course and I was like, I got to do computer science. I don't want to do physics. So FlashAttention is definitely, everybody's using this. Everybody loves it. You just released FlashAttention 2 last week. [00:01:48]Tri: Yeah. Early this week on Monday. Yeah. [00:01:53]Alessio: You know, AI time. Things move fast. So maybe let's run through some of the FlashAttention highlights, some of the innovation there, and then we can dive into FlashAttention 2. So the core improvement in FlashAttention is that traditional attention is a quadratic sequence length. And to the two, FlashAttention is linear, which obviously helps with scaling some of these models. [00:02:18]Tri: There are two factors there. So of course the goal has been to make attention go faster or more memory efficient. And ever since attention became popular in 2017 with the Transformer paper, lots and lots of folks have been working on this. And a lot of approaches has been focusing on approximating attention. The goal is you want to scale to longer sequences. There are tons of applications where you want to do that. But scaling to longer sequences is difficult because attention scales quadratically in sequence length on both runtime and memory, as you mentioned. So instead of trying to approximate attention, we were trying to figure out, can we do the same computation and maybe be more memory efficient? So in the end, we ended up being the memory is linear in sequence length. In terms of computation, it's still quadratic, but we managed to make it much more hardware friendly. And as a result, we do get wall clock speed up on the order of 2 to 4x, which really helps because that just means that you'll be able to train with 2 to 4x longer sequence length for the same cost without doing any approximations. As a result, lots of folks have been using this. The thing is available in a lot of libraries that do language model training or fine tuning. [00:03:32]Alessio: And the approximation thing is important because this is an exact thing versus a sparse. So maybe explain a little bit the difference there. [00:03:40]Tri: For sure. So in addition, essentially you compute pairwise similarity between every single element in a sequence against each other. So there's been other approaches where instead of doing all that pairwise computation, you only compute similarity for some pairs of elements in the sequence. So you don't do quadratic number of comparison. And this can be seen as some form of sparsity. Essentially you're ignoring some of the elements. When you write down the matrix, you essentially say, OK, I'm going to pretend there's zero. So that has some benefits in terms of runtime and memory. But the trade-off is that it tends to do worse in terms of quality because you're essentially approximating or ignoring some elements. And I personally have worked on this as well for a few years. But when we talk to practitioners who actually train models, especially at large scale, they say, tend not to use these approximate attention methods. Because it turns out, this was surprising to me at the time, was that these approximation methods, even though they perform fewer computation, they tend to not be faster in walk-on time. So this was pretty surprising because back then, I think my background was more on the theoretical side. So I was thinking of, oh, how many flops or floating point operations are you performing? And hopefully that correlates well with walk-on time. But I realized that I was missing a bunch of ideas from the system side where flops or floating point operations don't necessarily correlate with runtime. There are other factors like memory reading and writing, parallelism, and so on. So I learned a ton from just talking to systems people because they kind of figured this stuff out a while ago. So that was really eye-opening. And then we ended up focusing a lot more on memory reading and writing because that turned out to be the majority of the time when you're doing attention is reading and writing memory. [00:05:34]Alessio: Yeah, the I.O. awareness is probably one of the biggest innovations here. And the idea behind it is, like you mentioned, the FLOPS growth of the cards have been going up, but the memory bandwidth, not as much. So I think maybe that was one of the assumptions that the original attention paper had. So talk a bit about how that came to be as an idea. It's one of those things that like in insight, it's like, obviously, why are we like rewriting to like HBM every time, you know, and like once you change it, it's clear. But what was that discovery process? [00:06:08]Tri: Yeah, in hindsight, a lot of the ideas have already been there in the literature. And I would say is it was somehow at the intersection of both machine learning and systems. And you kind of needed ideas from both sides. So on one hand, on the system side, so lots of systems folks have known that, oh, you know, kernel fusion is great. Kernel fusion just means that instead of performing, you know, loading the same element, instead of performing an operation, write it down, load it back up and perform the second operation, you just load it once, perform two operations and then write it down again. So that saves you kind of memory read and write in the middle there. So kernel fusion has been a classic. There's been other techniques from the system side, like tiling, where you perform things in the form of computations in block, again, so that you can load it into a really fast memory. Think of it as a cache. And this is, again, classical computer science ideas, right? You want to use the cache. So the system folks have been thinking about these ideas for a long time, and they apply to attention as well. But there were certain things in attention that made it difficult to do a complete kernel fusion. One of which is there is this softmax operation in the middle, which requires you to essentially sum across the row of the attention matrix. So it makes it difficult to kind of break it, because there's this dependency. So it makes it difficult to break things into a block. So on the system side, people have been thinking about these ideas, but it's been difficult to kind of do kernel fusion for the entire operation. On the machine learning side, people have been thinking more algorithmically. They say, okay, either we can approximate attention, or there's this trick called the online softmax trick, which says that because of softmax, the way it's written mathematically, you can actually break it up into smaller pieces, do some rescaling, and still get the right answer. So this online softmax trick has been around for a while. I think there was a paper from NVIDIA folks back in 2018 about this. And then there was a paper from Google. So Marcus, Rob, and Stats wrote a paper late 2021 on using this online softmax trick to break attention up into smaller pieces. So a lot of the ideas were already there. But it turns out, you kind of need to combine ideas from both sides. So you need to understand that, hey, we want to do kernel fusion to reduce memory written writes. But we also need this online softmax trick to be able to break the softmax into smaller pieces so that a lot of the systems tricks kind of carry through. We saw that, and it was kind of a natural idea that we ended up using ideas from both sides, and it ended up working pretty well. Yeah. [00:08:57]Alessio: Are there any downsides to kernel fusion? If I think about databases and the reasons why we have atomic operations, you know, it's like, you have observability and fallback in between them. How does that work with attention? Is there anything that we lose by fusing the operations? [00:09:13]Tri: Yeah, I think mostly on the practical side is that you lose a little bit of flexibility in the sense that, hey, now you have, for example, faster attention, it's just a subroutine that you would call to do attention. But as a researcher, let's say you don't want that exact thing, right? You don't want just attention, let's say you want some modification to attention. You want to do, hey, I'm going to multiply the query and key, but then I'm going to do this extra thing before I carry on. So kernel fusion just means that, okay, we have a subroutine that does the entire thing. But if you want to experiment with things, you won't be able to use that fused kernel. And the answer is, can we have a compiler that then automatically does a lot of this kernel fusion? Lots of compiler folks are thinking about this, either with a new language or you can embed it in PyTorch. PyTorch folks have been working on this as well. So if you write just your code in PyTorch and they can capture the graph, can they generate code that will fuse everything together? That's still ongoing, and it works for some cases. But for attention, because of this kind of softmax rewriting stuff, it's been a little bit more difficult. So maybe in a year or two, we'll have compilers that are able to do a lot of these optimizations for you. And you don't have to, for example, spend a couple months writing CUDA to get this stuff to work. Awesome. [00:10:41]Alessio: And just to make it clear for listeners, when we say we're not writing it to memory, we are storing it, but just in a faster memory. So instead of the HBM, we're putting it in the SRAM. Yeah. [00:10:53]Tri: Yeah. [00:10:54]Alessio: Maybe explain just a little bit the difference there. [00:10:56]Tri: Yeah, for sure. This is kind of a caricature of how you think about accelerators or GPUs in particular, is that they have a large pool of memory, usually called HBM, or high bandwidth memory. So this is what you think of as GPU memory. So if you're using A100 and you list the GPU memory, it's like 40 gigs or 80 gigs. So that's the HBM. And then when you perform any operation, you need to move data from the HBM to the compute unit. So the actual hardware unit that does the computation. And next to these compute units, there are on-chip memory or SRAM, which are much, much smaller than HBM, but much faster. So the analogy there is if you're familiar with, say, CPU and RAM and so on. So you have a large pool of RAM, and then you have the CPU performing the computation. But next to the CPU, you have L1 cache and L2 cache, which are much smaller than DRAM, but much faster. So you can think of SRAM as the small, fast cache that stays close to the compute unit. Physically, it's closer. There is some kind of asymmetry here. So HBM is much larger, and SRAM is much smaller, but much faster. One way of thinking about it is, how can we design algorithms that take advantage of this asymmetric memory hierarchy? And of course, lots of folks have been thinking about this. These ideas are pretty old. I think back in the 1980s, the primary concerns were sorting. How can we sort numbers as efficiently as possible? And the motivating example was banks were trying to sort their transactions, and that needs to happen overnight so that the next day they can be ready. And so the same idea applies, which is that they have slow memory, which was hard disk, and they have fast memory, which was DRAM. And people had to design sorting algorithms that take advantage of this asymmetry. And it turns out, these same ideas can apply today, which is different kinds of memory. [00:13:00]Alessio: In your paper, you have the pyramid of memory. Just to give people an idea, when he says smaller, it's like HBM is like 40 gig, and then SRAM is like 20 megabytes. So it's not a little smaller, it's much smaller. But the throughput on card is like 1.5 terabytes a second for HBM and like 19 terabytes a second for SRAM, which is a lot larger. How do you think that evolves? So TSMC said they hit the scaling limits for SRAM, they just cannot grow that much more. HBM keeps growing, HBM3 is going to be 2x faster than HBM2, I think the latest NVIDIA thing has HBM3. How do you think about the future of FlashAttention? Do you think HBM is going to get fast enough when maybe it's not as useful to use the SRAM? [00:13:49]Tri: That's right. I think it comes down to physics. When you design hardware, literally SRAM stays very close to compute units. And so you don't have that much area to essentially put the transistors. And you can't shrink these things too much. So just physics, in terms of area, you don't have that much area for the SRAM. HBM is off-chip, so there is some kind of bus that essentially transfers data from HBM to the compute unit. So you have more area to essentially put these memory units. And so yeah, I think in the future SRAM probably won't get that much larger, because you don't have that much area. HBM will get larger and faster. And so I think it becomes more important to design algorithms that take advantage of this memory asymmetry. It's the same thing in CPU, where the cache is really small, the DRAM is growing larger and larger. DRAM could get to, I don't know, two terabytes, six terabytes, or something, whereas the cache stays at, I don't know, 15 megabytes or something like that. I think maybe the algorithm design becomes more and more important. There's still ways to take advantage of this, I think. So in the future, I think flash attention right now is being used. I don't know if in the next couple of years, some new architecture will come in and whatnot, but attention seems to be still important. For the next couple of years, I still expect some of these ideas to be useful. Not necessarily the exact code that's out there, but I think these ideas have kind of stood the test of time. New ideas like IO awareness from back in the 1980s, ideas like kernel fusions, tiling. These are classical ideas that have stood the test of time. So I think in the future, these ideas will become more and more important as we scale models to be larger, as we have more kinds of devices, where performance and efficiency become much, much more important. [00:15:40]Alessio: Yeah, and we had Jonathan Frankle on the podcast, and if you go to issattentionallyouneed.com, he has an outstanding bet, and he does believe that attention will be the state of the art architecture still in a few years. Did you think flash attention would be this popular? I'm always curious on the research side, you publish a paper, and obviously you know it's great work, but sometimes it just kind of falls flat in the industry. Could you see everybody just starting to use this, or was that a surprise to you? [00:16:11]Tri: Certainly, I didn't anticipate the level of popularity. Of course, we were extremely happy to have people using this stuff and giving us feedback and so on, and help us improve things. I think when we were writing the paper, I remember sending an email to one of my advisors, and like, hey, I'm excited about this paper, but I think the most important thing will be the artifact, which is the code. So I knew that the code will be valuable. So we kind of focus a lot on the code and make sure that the code is usable and as fast as can be. Of course, the idea, the paper presents the ideas and explain it and have experiments that validate the idea, but I knew that the artifact or the code was also pretty important. And that turned out to be the right focus, which is, you know, we put out the paper, we release the code and continue working on the code. So it's a team effort with my co-authors as well. [00:17:07]Alessio: We mentioned Hazy Research a bunch of times on the podcast before. I would love for you to spend five minutes just talking about how does the group work? How do people get together? How do you bounce ideas off of each other? Yeah. [00:17:21]Tri: So Hazy Research is a research group at Stanford led by one of my advisors, Chris Re. I love the people there. It was one of the best experiences I had. They've made my PhD so much more enjoyable. And I think there are a couple of ways that the group has been working pretty well. So one is, I think there's a diverse pool of people who either, you know, some of them focus on algorithms and theory, some of them focus on building systems, some of them focus on applications. And as a result, there is this flow of idea. So as an example, some of us were working on like more algorithms and theory, and then we can talk to the folks building systems and say, hey, let's try it out and let's put it in the systems and see how it is. And there you will get feedback from systems folks. They will say, hey, we implemented this, or we tried this and this is where it doesn't work, something like that. And once we put it in the systems, the application folks can use the algorithm or new methods or new models. And we again get great feedback from them because the application folks, for example, some of my good friends, they focus on medical imaging or seizure detection. And that is the problem they care about. And if your method doesn't work on the task they care about, they will tell you. Whereas I think a lot of people in machine learning, they're a little bit more flexible. So they will be like, hey, it doesn't work on seizure detection. Let's try some other task, right? But having that direct feedback of like, hey, it doesn't work there, let's figure out why. I think that that feedback allows us to do better work. And I think that kind of process of exchanging ideas, validating it in a real system so that applications folks can try it out and give you feedback. That cycle has been very, very useful. And so that's one, having a diverse group of people. The other one is, and this is something I really appreciate from advice from Chris was try to understand the fundamental, right? And he's happy letting me go off and read some textbooks and playing with things because I think a lot of research ideas come from understanding the old literature and see how it fits with the new landscape. And so if you just new archive papers every day, that's great, but you also need to read textbooks. And that's one advice I got from Chris, which is understand the fundamentals. And I think that allows us to do more impactful work. [00:19:46]Alessio: How do you think about academia versus industry? I feel like AI / Machine Learning has been an area where up until three, four years ago, most of the cutting edge work was being done in academia. And now there's all these big industry research labs. You're obviously going to Princeton, so you're an academia believer. How should people think about where to go? Say I'm doing my master's, I have to decide between doing a PhD and going into OpenAI Anthropic. How should I decide? [00:20:15]Tri: I think they kind of play a complementary role, in my opinion. Of course, I also was considering different paths as well. So I think right now, scaling matters a lot, especially when you talk about language models and AI and so on. Scaling matters a lot. And that means that you need compute resources and you need infrastructure and you need engineers time. And so industry tends to have an advantage when it comes to scaling things. But a lot of the ideas actually came from academia. So let's take Attention, which got popular with the Transformer in 2017. Attention actually has been around for a while. So I think the first mention was in 2014, a paper from Bernadot and others and Yoshua Bengio, which is coming from academia. A lot of ideas did come from academia. And scaling things up, of course, I think OpenAI has been great at scaling things up. That was the bet that they made after, I think, GPT-2. So they saw that scaling these things up to back then was 1.5 billion parameter seemed to give you amazing capabilities. So they really committed to that. They really committed to scaling things. And that turned out to be, it's been a pretty successful bet. I think for academia, we're still trying to figure out exactly what we're doing in this shifting landscape. And so lots of folks have been focusing on, for example, evaluation. So I know the Stanford Center for Foundation Model led by Percy, they have this benchmark called HELM, which is this holistic benchmark. So trying to figure out, okay, characterizing the landscape of different kinds of models, what people should evaluate, what people should measure, and things like that. So evaluation is one role. The other one is understanding. So this has happened historically where there's been some development in the industry and academia can play a role in explaining, understanding. They have the luxury to slow down trying to understand stuff, right? So lots of paper on understanding what's really going on, probing these models, and so on. I think I'm not as familiar with the NLP literature, but my impression is there's a lot of that going on in the NLP conferences, which is understanding what these models are doing, what capabilities they have, and so on. And the third one I could see is that the academia can take more risky bets in the sense that we can work on stuff that is quite different from industry. I think industry, my impression is you have some objective. You're trying to say, hey, for this quarter, we want to scale the model in this particular way. Next quarter, we want the model to have these capabilities. You're trying to get objectives that maybe, I don't know, 70% that will work out because it's important for the company's direction. I think for academia, the way things work is you have many, many researchers or PhD students, and they're kind of pursuing independent directions. And they have a little bit more flexibility on, hey, I'm going to try out this seemingly crazy idea and see, let's say there's a 30% chance of success or something. And however you define success, for academia, a lot of the time, success just means like, hey, we found something interesting. That could eventually go into industry through collaboration and so on. So I do see academia and industry kind of playing complementary roles. And as for someone choosing a career, I think just more and more generally, industry would be probably better in terms of compensation, in terms of probably work-life balance. But my biased perspective is that maybe academia gives you a little bit more freedom to think and understand things. So it probably comes down to personal choice. I end up choosing to be a professor next year at Princeton. But of course, I want to maintain a relationship with industry folks. I think industry folks can provide very valuable feedback to what we're doing in academia so that we understand where the field is moving because some of the directions are very much influenced by what, for example, OpenAI or Google is doing. So we want to understand where the field is moving. What are some promising applications? And try to anticipate, okay, if the field is moving like this, these applications are going to be popular. What problems will be important in two, three years? And then we try to start thinking about those problems so that hopefully in two, three years, we have some of the answers to some of these problems in two, three years. Sometimes it works out, sometimes it doesn't. But as long as we do interesting things in academia, that's the goal. [00:25:03]Alessio: And you mentioned the eval side. So we did a Benchmarks 101 episode. And one of the things we were seeing is sometimes the benchmarks really influence the model development. Because obviously, if you don't score well on the benchmarks, you're not going to get published and you're not going to get funded. How do you think about that? How do you think that's going to change now that a lot of the applications of these models, again, is in more narrow industry use cases? Do you think the goal of the academia eval system is to be very broad and then industry can do their own evals? Or what's the relationship there? [00:25:40]Tri: Yeah, so I think evaluation is important and often a little bit underrated. So it's not as flashy as, oh, we have a new model that can do such and such. But I think evaluation, what you don't measure, you can't make progress on, essentially. So I think industry folks, of course, they have specific use cases that their models need to do well on. And that's what they care about. Not just academia, but other groups as well. People do understand what are some of the emerging use cases. So for example, now one of the most popular use cases is Chatbot. And then I think folks from Berkeley, some of them are from Berkeley, call them MLCs. They set up this kind of Chatbot arena to essentially benchmark different models. So people do understand what are some of the emerging use cases. People do contribute to evaluation and measurement. And as a whole, I think people try to contribute to the field and move the field forward, albeit that maybe slightly different directions. But we're making progress and definitely evaluation and measurement is one of the ways you make progress. So I think going forward, there's still going to be just more models, more evaluation. We'll just have better understanding of what these models are doing and what capabilities they have. [00:26:56]Alessio: I like that your work has been focused on not making benchmarks better, but it's like, let's just make everything faster. So it's very horizontal. So FlashAttention 2, you just released that on Monday. I read in the blog post that a lot of the work was also related to some of the NVIDIA library updates. Yeah, maybe run us through some of those changes and some of the innovations there. Yeah, for sure. [00:27:19]Tri: So FlashAttention 2 is something I've been working on for the past couple of months. So the story is the NVIDIA CUTLASS team, they released a new version of their library, which contains all these primitives to allow you to do matrix multiply or memory loading on GPU efficiently. So it's a great library and I built on that. So they released their version 3 back in January and I got really excited and I wanted to play with that library. So as an excuse, I was just like, okay, I'm going to refactor my code and use this library. So that was kind of the start of the project. By the end, I just ended up working with the code a whole lot more and I realized that, hey, there are these inefficiencies still in Flash Attention. We could change this way or that way and make it, in the end, twice as fast. But of course, building on the library that the NVIDIA folks released. So that was kind of a really fun exercise. I was starting out, it's just an excuse for myself to play with the new library. What ended up was several months of improvement, improving Flash Attention, discovering new ideas. And in the end, we managed to make it 2x faster and now it's pretty close to probably the efficiency of things like matrix multiply, which is probably the most optimized subroutine on the planet. So we're really happy about it. The NVIDIA Cutlass team has been very supportive and hopefully in the future, we're going to collaborate more. [00:28:46]Alessio: And since it's an NVIDIA library, can you only run this on CUDA runtimes? Or could you use this and then run it on an AMD GPU? [00:28:56]Tri: Yeah, so it's an NVIDIA library. So right now, the code we release runs on NVIDIA GPUs, which is what most people are using to train models. Of course, there are emerging other hardware as well. So the AMD folks did implement a version of Flash Attention, I think last year as well, and that's also available. I think there's some implementation on CPU as well. For example, there's this library, ggml, where they implemented the same idea running on Mac and CPU. So I think that kind of broadly, the idea would apply. The current implementation ended up using NVIDIA's library or primitives, but I expect these ideas to be broadly applicable to different hardware. I think the main idea is you have asymmetry in memory hierarchy, which tends to be everywhere in a lot of accelerators. [00:29:46]Alessio: Yeah, it kind of reminds me of Sara Hooker's post, like the hardware lottery. There could be all these things that are much better, like architectures that are better, but they're not better on NVIDIA. So we're never going to know if they're actually improved. How does that play into some of the research that you all do too? [00:30:04]Tri: Yeah, so absolutely. Yeah, I think Sara Hooker, she wrote this piece on hardware lottery, and I think she captured really well of what a lot of people have been thinking about this. And I certainly think about hardware lottery quite a bit, given that I do some of the work that's kind of really low level at the level of, hey, we're optimizing for GPUs or NVIDIA GPUs and optimizing for attention itself. And at the same time, I also work on algorithms and methods and transformer alternatives. And we do see this effect in play, not just hardware lottery, but also kind of software framework lottery. You know, attention has been popular for six years now. And so many kind of engineer hours has been spent on making it as easy and efficient as possible to run transformer, right? And there's libraries to do all kinds of tensor parallel, pipeline parallel, if you use transformer. Let's say someone else developed alternatives, or let's just take recurrent neural nets, like LSTM, GRU. If we want to do that and run that efficiently on current hardware with current software framework, that's quite a bit harder. So in some sense, there is this feedback loop where somehow the model architectures that take advantage of hardware become popular. And the hardware will also kind of evolve to optimize a little bit for that kind of architecture and software framework will also evolve to optimize for that particular architecture. Right now, transformer is the dominant architecture. So yeah, I'm not sure if there is a good way out of this. Of course, there's a lot of development. Things like, I think compilers will play a role because compilers allow you to maybe still be much more efficient across different kinds of hardware because essentially you write the same code and compiler will be able to make it run efficiently different kinds of hardware. So for example, there's this language Mojo, they're compiler experts, right? And their bet is AI models will be running on different kinds of devices. So let's make sure that we have really good compilers with a good language that then the compiler can do a good job optimizing for all kinds of devices. So that's maybe one way that you can get out of this cycle. But yeah, I'm not sure of a good way. In my own research, I have to think about both the algorithm new model and how it maps to hardware. So there are crazy ideas that seem really good, but will be really, really difficult to run efficiently. And so as a result, for example, we can't really scale some of the architectures up simply because they're not hardware friendly. I have to think about both sides when I'm working on new models. [00:32:50]Alessio: Yeah. Have you spent any time looking at some of the new kind of like AI chips companies, so to speak, like the Cerebras of the world? Like one of their innovations is co-locating everything on the chip. So you remove some of this memory bandwidth issue. How do you think about that? [00:33:07]Tri: Yeah, I think that's an interesting bet. I think Tesla also has this Dojo supercomputer where they try to have essentially as fast on-chip memory as possible and removing some of these data transfer back and forth. I think that's a promising direction. The issues I could see, you know, I'm definitely not a hardware expert. One issue is the on-chip memory tends to be really expensive to manufacture, much more expensive per gigabyte compared to off-chip memory. So I talked to, you know, some of my friends at Cerebros and, you know, they have their own stack and compiler and so on, and they can make it work. The other kind of obstacle is, again, with compiler and software framework and so on. For example, if you can run PyTorch on this stuff, lots of people will be using it. But supporting all the operations in PyTorch will take a long time to implement. Of course, people are working on this. So I think, yeah, we kind of need these different bets on the hardware side as well. Hardware has, my understanding is, has a kind of a longer time scale. So you need to design hardware, you need to manufacture it, you know, maybe on the order of three to five years or something like that. So people are taking different bets, but the AI landscape is changing so fast that it's hard to predict, okay, what kind of models will be dominant in, let's say, three or five years. Or thinking back five years ago, would we have known that Transformer would have been the dominant architecture? Maybe, maybe not, right? And so different people will make different bets on the hardware side. [00:34:39]Alessio: Does the pace of the industry and the research also influence the PhD research itself? For example, in your case, you're working on improving attention. It probably took you quite a while to write the paper and everything, but in the meantime, you could have had a new model architecture come out and then it's like nobody cares about attention anymore. How do people balance that? [00:35:02]Tri: Yeah, so I think it's tough. It's definitely tough for PhD students, for researchers. Given that the field is moving really, really fast, I think it comes down to understanding fundamental. Because that's essentially, for example, what the PhD allows you to do. It's been a couple of years understanding the fundamentals. So for example, when I started my PhD, I was working on understanding matrix vector multiply, which has been a concept that's been around for hundreds of years. We were trying to characterize what kind of matrices would have theoretically fast multiplication algorithm. That seems to have nothing to do with AI or anything. But I think that was a time when I developed mathematical maturity and research taste and research skill. The research topic at that point didn't have to be super trendy or anything, as long as I'm developing skills as a researcher, I'm making progress. And eventually, I've gotten quite a bit better in terms of research skills. And that allows, for example, PhD students later in their career to quickly develop solutions to whatever problems they're facing. So I think that's just the natural arc of how you're being trained as a researcher. For a lot of PhD students, I think given the pace is so fast, maybe it's harder to justify spending a lot of time on the fundamental. And it's tough. What is this kind of explore, exploit kind of dilemma? And I don't think there's a universal answer. So I personally spend some time doing this kind of exploration, reading random textbooks or lecture notes. And I spend some time keeping up with the latest architecture or methods and so on. I don't know if there's a right balance. It varies from person to person. But if you only spend 100% on one, either you only do exploration or only do exploitation, I think it probably won't work in the long term. It's probably going to have to be a mix and you have to just experiment and kind of be introspective and say, hey, I tried this kind of mixture of, I don't know, one exploration paper and one exploitation paper. How did that work out for me? Should I, you know, having conversation with, for example, my advisor about like, hey, did that work out? You know, should I shift? I focus more on one or the other. I think quickly adjusting and focusing on the process. I think that's probably the right way. I don't have like a specific recommendation that, hey, you focus, I don't know, 60% on lecture notes and 40% on archive papers or anything like that. [00:37:35]Alessio: Let's talk about some Transformer alternatives. You know, say Jonathan Franco loses his bet and Transformer is not the state of the art architecture. What are some of the candidates to take over? [00:37:49]Tri: Yeah, so this bet is quite fun. So my understanding is this bet between Jonathan Franco and Sasha Rush, right? I've talked to Sasha a bunch and I think he recently gave an excellent tutorial on Transformer alternatives as well. So I would recommend that. So just to quickly recap, I think there's been quite a bit of development more recently about Transformer alternatives. So architectures that are not Transformer, right? And the question is, can they do well on, for example, language modeling, which is kind of the application that a lot of people care about these days. So there are methods based on state space methods that came out in 2021 from Albert Gu and Curran and Chris Re that presumably could do much better in terms of capturing long range information while not scaling quadratically. They scale sub-quadratically in terms of sequence length. So potentially you could have a much more efficient architecture when sequence length gets really long. The other ones have been focusing more on recurrent neural nets, which is, again, an old idea, but adapting to the new landscape. So things like RWKV, I've also personally worked in this space as well. So there's been some promising results. So there's been some results here and there that show that, hey, these alternatives, either RNN or state space methods, can match the performance of Transformer on language modeling. So that's really exciting. And we're starting to understand on the academic research side, we want to understand, do we really need attention? I think that's a valuable kind of intellectual thing to understand. And maybe we do, maybe we don't. If we want to know, we need to spend serious effort on trying the alternatives. And there's been folks pushing on this direction. I think RWKV scale up to, they have a model at 14 billion that seems pretty competitive with Transformer. So that's really exciting. That's kind of an intellectual thing. We want to figure out if attention is necessary. So that's one motivation. The other motivation is Transformer Alternative could have an advantage in practice in some of the use cases. So one use case is really long sequences. The other is really high throughput of generation. So for really long sequences, when you train with Transformer, with flash attention and so on, the computation is still quadratic in the sequence length. So if your sequence length is on the order of, I don't know, 16K, 32K, 100K or something, which some of these models have sequence length 100K, then you do get significantly slower in terms of training, also in terms of inference. So maybe these alternative architectures could scale better in terms of sequence length. I haven't seen actual validation on this. Let's say an RNN model release with context length, I don't know, 100K or something. I haven't really seen that. But the hope could be that as we scale to long sequences, these alternative architectures could be more well-suited. Not just text, but things like high resolution images, audio, video, and so on, which are emerging applications. So that's one, long sequences. Number two is a high throughput generation, where I can imagine scenarios where the application isn't like an interactive chatbot, but let's say a company wants to batch as many requests as possible on their server, or they're doing offline processing, they're generating stuff based on their internal documents, that you need to process in batch. And the issue with Transformer is that during generation, it essentially needs to keep around all the previous history. It's called the KV cache. And that could take a significant amount of memory, so you can't really batch too much because you run out of memory. I am personally bullish on RNNs. I think RNNs, they essentially summarize the past into a state vector that has fixed size, so the size doesn't grow with the history. So that means that you don't need as much memory to keep around all the previous tokens. And as a result, I think you can scale to much higher batch sizes. And as a result, you can make much more efficient use of the GPUs or the accelerator, and you could have much higher generation throughput. Now, this, I don't think, has been validated at scale. So as a researcher, I'm bullish on this stuff because I think in the next couple of years, these are use cases where these alternatives could have an advantage. We'll just kind of have to wait and see to see if these things will happen. I am personally bullish on this stuff. At the same time, I also spend a bunch of time making attention as fast as possible. So maybe hatching and playing both sides. Ultimately, we want to understand, as researchers, we want to understand what works, why do the models have these capabilities? And one way is, let's push attention to be as efficient as possible. On the other hand, let's push other alternatives to be as efficient at scale, as big as possible, and so that we can kind of compare them and understand. Yeah, awesome. [00:43:01]Alessio: And I think as long as all of this work happens and open, it's a net positive for everybody to explore all the paths. Yeah, let's talk about open-source AI. Obviously, together, when Red Pajama came out, which was an open clone of the LLAMA1 pre-training dataset, it was a big thing in the industry. LLAMA2 came out on Tuesday, I forget. And this week, there's been a lot of things going on, which they call open-source, but it's not really open-source. Actually, we wrote a post about it that was on the front page of Hacker News before this podcast, so I was frantically responding. How do you think about what open-source AI really is? In my mind, in open-source software, we have different levels of open. So there's free software, that's like the GPL license. There's open-source, which is Apache, MIT. And then there's kind of restricted open-source, which is the SSPL and some of these other licenses. In AI, you have the open models. So Red Pajama is an open model because you have the pre-training dataset, you have the training runs and everything. And then there's obviously RandomLens that doesn't make it one-to-one if you retrain it. Then you have the open-weights model that's kind of like StableLM, where the weights are open, but the dataset is not open. And then you have LLAMA2, which is the dataset is not open, the weights are restricted. It's kind of like not really open-source, but open enough. I think it's net positive because it's like $3 million of flops donated to the public. [00:44:32]Tri: How do you think about that? [00:44:34]Alessio: And also, as you work together, what is your philosophy with open-source AI? Right, right. [00:44:40]Tri: Yeah, I think that's a great question. And I think about it on maybe more practical terms. So of course, Meta has done an amazing job training LLAMA1, LLAMA2. And for LLAMA2, they make it much less restrictive compared to LLAMA1. Now you can use it for businesses, unless you are a monthly active user or something like that. I think just this change will have a very significant impact in the kind of landscape of open-source AI, where now lots of businesses, lots of companies will be using, I expect will be using things like LLAMA2. They will fine-tune on their own dataset. They will be serving variants or derivatives of LLAMA2. Whereas before, with LLAMA1, it was also a really good model, but your business companies weren't allowed to do that. So I think on a more practical term, it's kind of shifting the balance between a closed-source model like OpenAI and Anthropic and Google, where you're making API calls, right? And maybe you don't understand as much of what the model is doing, how the model is changing, and so on. Versus now, we have a model with open weight that is pretty competitive from what I've seen in terms of benchmarks, pretty competitive with GPT 3.5, right? And if you fine-tune it on your own data, maybe it's more well-suited for your own data. And I do see that's going to shift the balance of it. More and more folks are going to be using, let's say, derivatives of LLAMA2. More and more folks are going to fine-tune and serve their own model instead of calling an API. So that shifting of balance is important because in one way, we don't want just a concentration of decision-making power in the hands of a few companies. So I think that's a really positive development from Meta. Of course, training the model takes a couple of millions of dollars, but engineers have and I'm sure they spend tons of time trying many, many different things. So the actual cost is probably way more than that. And they make the weights available and they allow probably a lot of companies are going to be using this. So I think that's a really positive development. And we've also seen amazing progress on the open source community where they would take these models and they either fine-tune on different kinds of data sets or even make changes to the model. So as an example, I think for LLAMA1, the context lane was limited to 2K. Like a bunch of folks figured out some really simple methods to scale up to like 8K. [00:47:12]Alessio: Like the RoPE. [00:47:13]Tri: Yes. I think the open source community is very creative, right? And lots of people. LLAMA2 will, again, kind of accelerate this where more people will try it out. More people will make tweaks to it and make a contribution and then so on. So overall, I think I see that as still a very positive development for the field. And there's been lots of libraries that will allow you to host or fine-tune these models, like even with quantization and so on. Just a couple of hours after LLAMA2 was released, tons of companies announcing that, hey, it's on our API or hosting and so on and together did the same. So it's a very fast-paced development and just kind of a model with available weights that businesses are allowed to use. I think that alone is already a very positive development. At the same time, yeah, we can do much better in terms of releasing data sets. Data sets tend to be... Somehow people are not incentivized to release data sets. So philosophically, yeah, you want to be as open as possible. But on a practical term, I think it's a little bit harder for companies to release data sets. Legal issues. The data sets released tend to be not as eye-catchy as the model release. So maybe people are less incentivized to do that. We've seen quite a few companies releasing data sets together. Released a red pajama data set. I think Cerebus then worked on that and deduplicate and clean it up and release slim pajama and so on. So we're also seeing positive development on that front, kind of on the pre-training data set. So I do expect that to continue. And then on the fine-tuning data set or instruction tuning data set, I think we now have quite a few open data sets on instruction tuning and fine-tuning. But these companies do pay for human labelers to annotate these instruction tuning data set. And that is expensive. And maybe they will see that as their competitive advantage. And so it's harder to incentivize these companies to release these data sets. So I think on a practical term, we're still going to make a lot of progress on open source AI, on both the model development, on both model hosting, on pre-training data set and fine-tuning data set. Right now, maybe we don't have the perfect open source model since all the data sets are available. Maybe we don't have such a thing yet, but we've seen very fast development on the open source side. I think just maybe this time last year, there weren't as many models that are competitive with, let's say, ChatGPT. [00:49:43]Alessio: Yeah, I think the open data sets have so much more impact than open models. If you think about Elusive and the work that they've done, GPT-J was great, and the Pythia models are great, but the Pyle and the Stack, everybody uses them. So hopefully we get more people to contribute time to work on data sets instead of doing the 100th open model that performs worse than all the other ones, but they want to say they released the model. [00:50:14]Tri: Yeah, maybe the question is, how do we figure out an incentive structure so that companies are willing to release open data sets? And for example, it could be like, I think some of the organizations are now doing this where they are asking volunteers to annotate and so on. And maybe the Wikipedia model of data set, especially for instruction tuning, could be interesting where people actually volunteer their time and instead of editing Wikipedia, add annotation. And somehow they acknowledge and feel incentivized to do so. Hopefully we get to that kind of level of, in terms of data, it would be kind of like Wikipedia. And in terms of model development, it's kind of like Linux where people are contributing patches and improving the model in some way. I don't know exactly how that's going to happen, but based on history, I think there is a way to get there. [00:51:05]Alessio: Yeah, I think the Dolly-15K data set is a good example of a company saying, let's do this smaller thing, just make sure we make it open. We had Mike Conover from Databricks on the podcast, and he was like, people just bought into it and leadership was bought into it. You have companies out there with 200,000, 300,000 employees. It's like, just put some of them to label some data. It's going to be helpful. So I'm curious to see how that evolves. What made you decide to join Together? [00:51:35]Tri: For Together, the focus has been focusing a lot on open source model. And I think that aligns quite well with what I care about, of course. I also know a bunch of people there that I know and trust, and I'm excited to work with them. Philosophically, the way they've been really open with data set and model release, I like that a lot. Personally, for the stuff, for example, the research that I've developed, like we also try to make code available, free to use and modify and so on, contributing to the community. That has given us really valuable feedback from the community and improving our work. So philosophically, I like the way Together has been focusing on open source model. And the nice thing is we're also going to be at the forefront of research and the kind of research areas that I'm really excited about, things like efficient training and inference, aligns quite well with what the company is doing. We'll try our best to make things open and available to everyone. Yeah, but it's going to be fun being at the company, leading a team, doing research on the topic that I really care about, and hopefully we'll make things open to benefit the community. [00:52:45]Alessio: Awesome. Let's jump into the lightning round. Usually, I have two questions. So one is on acceleration, one on exploration, and then a takeaway. So the first one is, what's something that already happened in AI machine learning that you thought would take much longer than it has? [00:53:01]Tri: I think understanding jokes. I didn't expect that to happen, but it turns out scaling model up and training lots of data, the model can now understand jokes. Maybe it's a small thing, but that was amazing to me. [00:53:16]Alessio: What about the exploration side? What are some of the most interesting unsolved questions in the space? [00:53:22]Tri: I would say reasoning in the broad term. We don't really know how these models do. Essentially, they do something that looks like reasoning. We don't know how they're doing it. We have some ideas. And in the future, I think we will need to design architecture that explicitly has some kind of reasoning module in it if we want to have much more capable models. [00:53:43]Alessio: What's one message you want everyone to remember today? [00:53:47]Tri: I would say try to understand both the algorithm and the systems that these algorithms run on. I think at the intersection of machine learning system has been really exciting, and there's been a lot of amazing results at this intersection. And then when you scale models to large scale, both the machine learning side and the system side really matter. [00:54:06]Alessio: Awesome. Well, thank you so much for coming on 3. [00:54:09]Tri: This was great. Yeah, this has been really fun. [00:54:11] Get full access to Latent Space at www.latent.space/subscribe

Jul 19, 2023 • 1h 20min

Llama 2: The New Open LLM SOTA (ft. Nathan Lambert, Matt Bornstein, Anton Troynikov, Russell Kaplan, Whole Mars Catalog et al.)

Jul 17, 2023 • 1h 1min

AI Fundamentals: Datasets 101

In April, we released our first AI Fundamentals episode: Benchmarks 101. We covered the history of benchmarks, why they exist, how they are structured, and how they influence the development of artificial intelligence.Today we are (finally!) releasing Datasets 101! We’re really enjoying doing this series despite the work it takes - please let us know what else you want us to cover!Stop me if you’ve heard this before: “GPT3 was trained on the entire Internet”.Blatantly, demonstrably untrue: the GPT3 dataset is a little over 600GB, primarily on Wikipedia, Books corpuses, WebText and 2016-2019 CommonCrawl. The Macbook Air I am typing this on has more free disk space than that. In contrast, the “entire internet” is estimated to be 64 zetabytes, or 64 trillion GB. So it’s more accurate to say that GPT3 is trained on 0.0000000001% of the Internet.Why spend $5m on GPU time training on $50 worth of data?Simple: Garbage in, garbage out. No matter how good your algorithms, no matter how much money/compute you have, your model quality is strongly determined by the data you train it on and research scientists think we just don’t need or have that much high quality data. We spend an enormous amount of effort throwing out data to keep the quality high, and recently Web 2.0-era UGC platforms like StackOverflow, Reddit, and Twitter clamped down on APIs as they realize the goldmines they sit on.Data is the new new oil. Time for a primer!Show Notes* Our 2 months worth of podcast prep notes!* The Token Crisis paper* Ilya Sutskever on datasets * OpenAI Tokenizer* Kaplan Scaling Laws Lecture* Chinchilla Paper* Sasha Rush’s Tweet* Karpathy’s Build Conference Presentation* LIMA Paper* Phi-1 by Microsoft* Washington Post Article on datasets* Our episode with Jonathan Frankle* Our episode with Mike Conover* BloombergGPT* Datasets* HuggingFace Hub* CommonCrawl, Overview* C4* List of Dirty, Naughty, Obscene, and Otherwise Bad Words* OpenWebText* books3* OpenAssistant * The Stack* The Pile* LAION* Audio:* LibriSpeech: A dataset of audio recordings of audiobooks* CommonVoice: A dataset of audio recordings of people speaking different languages* Voxforge: A dataset of audio recordings of people speaking different languages* Switchboard: A dataset of audio recordings of telephone conversations* Fisher Corpus: A dataset of audio recordings of news broadcasts* Chinese:* CMRC (Chinese Machine Reading Comprehension 2018)* DuReader* ChID* Copyright & Privacy:* https://stablediffusionlitigation.com/* https://haveibeentrained.com/* https://githubcopilotlitigation.com/* https://twitter.com/moyix/status/1662131770463072257* OpenAI Opt Out Process* Check if you’re in The Stack* Deduplication* Deduplicating Training Data Makes Language Models Better* Deduplicating Training Data Mitigates Privacy Risks in Language Models* Contamination* CodeForces example Get full access to Latent Space at www.latent.space/subscribe

Jul 10, 2023 • 2h 4min

Code Interpreter == GPT 4.5 (w/ Simon Willison, Alex Volkov, Aravind Srinivas, Alex Graveley, et al.)

Jul 2, 2023 • 1h

[Practical AI] AI Trends: a Latent Space x Practical AI crossover pod!

Part 2 of our podcast feed swap weekend! Check out Cognitive Revolution as well."Data" Dan Whitenack has been co-host of the Practical AI podcast for the past 5 years, covering full journey of the modern AI wave post Transformers. He joined us in studio to talk about their origin story and highlight key learnings from past episodes, riff on the AI trends we are all seeing as AI practitioner-podcasters, and his passion for low-resource-everything!Subscribe on the Changelog, RSS, Apple Podcasts, Twitter, Mastodon, and wherever fine podcasts are sold!Show notes* Daniel Whitenack – Twitter, GitHub, Website* Featured Latent Space episodes:* Benchmarks* Reza Shabani* MosaicML and MPT* Segment Anything* Mike Conover* Featured Practical AI episodes:* From notebooks to Netflix scale with Metaflow* Capabilities of LLMs 🤯* ML at small organizations* Prediction Guard* Data DanTimestamps* 00:00 Welcome to Practical AI* 01:16 Latent Space Podcast* 04:00 Practical AI Podcast* 06:20 Prediction Guard* 08:05 Daniel's favorite episodes* 10:21 Alessio's favorite episode* 10:54 Swyx's favorite episode* 12:44 Listener favorites* 15:14 LLMOps* 17:06 Reza Shabani* 19:06 Benchmarks 101* 20:06 Roboflow* 21:38 Mode collapse* 26:21 Rajiv Shah* 28:01 Staying on top of things* 33:11 Kirsten Lum* 34:31 datadan.io* 38:48 Prompt engineering* 40:38 Unique challenges engineers face* 42:51 AI-UX* 45:31 NLP data sets* 50:49 Unlabeled data sets* 55:07 Lightning round!* 55:20 What's already happened in AI?* 56:27 Unsolved questions in AI* 58:01 Get hands on* 58:53 OutroTranscriptFull transcript is over at the Changelog site! Get full access to Latent Space at www.latent.space/subscribe

Jul 1, 2023 • 2h 5min

[Cognitive Revolution] The Tiny Model Revolution with Ronen Eldan and Yuanzhi Li of Microsoft Research

Thanks to the over 1m people that have checked out the Rise of the AI Engineer. It’s a long July 4 weekend in the US, and we’re celebrating with a podcast feed swap!We’ve been big fans of Nathan Labenz and Erik Torenberg’s work at the Cognitive Revolution podcast for a while, which started around the same time as we did and has done an incredible job of hosting discussions with top researchers and thinkers in the field, with a wide range of topics across computer vision (a special focus thanks to Nathan’s work at Waymark), GPT-4 (with exceptional insight due to Nathan’s time on the GPT-4 “red team”), healthcare/medicine/biotech (Harvard Medical School, Med-PaLM, Tanishq Abraham, Neal Khosla), investing and tech strategy (Sarah Guo, Elad Gil, Emad Mostaque, Sam Lessin), safety and policy, curators and influencers and exceptional AI founders (Josh Browder, Eugenia Kuyda, Flo Crivello, Suhail Doshi, Jungwon Byun, Raza Habib, Mahmoud Felfel, Andrew Feldman, Matt Welsh, Anton Troynikov, Aravind Srinivas). If Latent Space is for AI Engineers, then Cognitive Revolution covers the much broader field of AI in tech, business and society at large, with a longer runtime to go deep on research papers like TinyStories. We hope you love this episode as much as we do, and check out CogRev wherever fine podcasts are sold!Subscribe to the Cognitive Revolution on:* Website* Apple Podcasts* Spotify* YoutubeGood Data is All You NeedThe work of Ronen and Yuanzhi echoes a broader theme emerging in the midgame of 2023: * Falcon-40B (trained on 1T tokens) outperformed LLaMA-65B (trained on 1.4T tokens), primarily due to the RefinedWeb Dataset that runs CommonCrawl through extensive preprocessing and cleaning in their MacroData Refinement pipeline. * UC Berkeley LMSYS’s Vicuna-13B is near GPT-3.5/Bard quality at a tenth of their size, thanks to fine-tuning from 70k user-highlighted ChatGPT conversations (indicating some amount of quality). * Replit’s finetuned 2.7B model outperforms the 12B OpenAI Codex model based on HumanEval, thanks to high quality data from Replit usersThe path to smaller models leans on better data (and tokenization!), whether from cleaning, from user feedback, or from synthetic data generation, i.e. finetuning high quality on outputs from larger models. TinyStories and Phi-1 are the strongest new entries in that line of work, and we hope you’ll pick through the show notes to read up further.Show Notes* TinyStories (Apr 2023)* Paper: TinyStories: How Small Can Language Models Be and Still Speak Coherent English?* Internal presentation with Sebastien Bubeck at MSR* Twitter thread from Ronen Eldan* Will future LLMs be based almost entirely on synthetic training data? In a new paper, we introduce TinyStories, a dataset of short stories generated by GPT-3.5&4. We use it to train tiny LMs (< 10M params) that produce fluent stories and exhibit reasoning.* Phi-1 (Jun 2023)* Paper: Textbooks are all you need (HN discussion)* Twitter announcement from Sebastien Bubeck:* phi-1 achieves 51% on HumanEval w. only 1.3B parameters & 7B tokens training dataset and 8 A100s x 4 days = 800 A100-hours. Any other >50% HumanEval model is >1000x bigger (e.g., WizardCoder from last week is 10x in model size and 100x in dataset size). Get full access to Latent Space at www.latent.space/subscribe

Jun 20, 2023 • 1h 13min

Commoditizing the Petaflop — with George Hotz of the tiny corp

We are now launching our dedicated new YouTube and Twitter! Any help in amplifying our podcast would be greatly appreciated, and of course, tell your friends! Notable followon discussions collected on Twitter, Reddit, Reddit, Reddit, HN, and HN. Please don’t obsess too much over the GPT4 discussion as it is mostly rumor; we spent much more time on tinybox/tinygrad on which George is the foremost authority!We are excited to share the world’s first interview with George Hotz on the tiny corp!If you don’t know George, he was the first person to unlock the iPhone, jailbreak the PS3, went on to start Comma.ai, and briefly “interned” at the Elon Musk-run Twitter. Tinycorp is the company behind the deep learning framework tinygrad, as well as the recently announced tinybox, a new $15,000 “luxury AI computer” aimed at local model training and inference, aka your “personal compute cluster”:* 738 FP16 TFLOPS* 144 GB GPU RAM* 5.76 TB/s RAM bandwidth* 30 GB/s model load bandwidth (big llama loads in around 4 seconds)* AMD EPYC CPU* 1600W (one 120V outlet)* Runs 65B FP16 LLaMA out of the box (using tinygrad, subject to software development risks)(In the episode, we also talked about the future of the tinybox as the intelligence center of every home that will help run models, at-home robots, and more. Make sure to check the timestamps 👀 )The tiny corp manifestoThere are three main theses to tinycorp:* If XLA/PrimTorch are CISC, tinygrad is RISC: CISC (Complex Instruction Set Computing) are more complex instruction sets where a single instruction can execute many low-level operations. RISC (Reduced Instruction Set Computing) are smaller, and only let you execute a single low-level operation per instruction, leading to faster and more efficient instruction execution. If you’ve used the Apple Silicon M1/M2, AMD Ryzen, or Raspberry Pi, you’ve used a RISC computer.* If you can’t write a fast ML framework for GPU, you can’t write one for your own chip: there are many “AI chips” companies out there, and they all started from taping the chip. Some of them like Cerebras are still building, while others like Graphcore seem to be struggling. But building chips with higher TFLOPS isn’t enough: “There’s a great chip already on the market. For $999, you get a 123 TFLOP card with 24 GB of 960 GB/s RAM. This is the best FLOPS per dollar today, and yet…nobody in ML uses it.”, referring to the AMD RX 7900 XTX. NVIDIA’s lead is not only thanks to high-performing cards, but also thanks to a great developer platform in CUDA. Starting with the chip development rather than the dev toolkit is much more cost-intensive, so tinycorp is starting by writing a framework for off-the-shelf hardware rather than taping their own chip. * Turing completeness considered harmful: Once you call in to Turing complete kernels, you can no longer reason about their behavior. Since they have to be able to execute any instruction, they are much more complex. To optimize Turing kernels performance, you fall back to caching, warp scheduling, and branch prediction. Since neural networks only need ADD/MUL operations and only rely on static memory accesses, there’s no need to have Turing completeness. This design decision allows tinygrad to optimize instructions at a much lower level. As you might have guessed, CUDA is Turing-complete; this is one of the main differences that tinycorp wants to leverage to be competitive. All that — covered in the first 10 minutes of our discussion. George came ready to go deep, so we went for it. Some of the other technical questions we went through:* Laziness: why laziness is important and how operation fusing can help with memory efficiency* Debugging & CI: Why great developer experience is a priority in tinygrad* Quantization: what’s the right level of quantization, how lossless are these transformations, his quick takes on Mojo and ggml, and why fp16 is the target for their out-of-the-box LLaMA. * Building rigs for individual use: we talked a bit about the design tradeoffs of building these machines with low noise and a single power plug, the difference that PCIe 4 vs 3 makes, and more.The “personal compute cluster” is $15,000, but for businesses interested in local training and inference, George also estimates that he will be able to build you a H100-class GPU that is 5-10x faster (than a H100) for the same price.Misc: Bitter Lessons, Core Insights, Remote WorkOutside of tiny, we also talked about one of George’s favorite units of measure “a person of compute”. Much of the AGI talk has been benchmark-driven, but looking at it from a compute throughput can also be interesting. One person of compute is roughly 20 PFLOPS (64 A100s, or a single dense 42U A100 rack); one A100 is ~$10-15,000, so the GPUs by themselves will come out at $640,000-$1,000,000. We also covered a wide range of topics, including his self analysis on GPT-4, Elon Musk, Remote Work, Computer Vision and the Comma Body, and life above/below the API (and above/below the Kanban board). See show notes and timestamps for more!Show Notes* “Unlocked iPhone Traded for Nissan 350Z”* “Unlocked iPhone” on YouTube (August 21st, 2007)* “The Light It Up Contest” on YouTube (February 13th, 2011)* Comma.ai* NHTSA cease and desist* The Hero’s Journey* The Portal Story* A Person of Compute* Above / Below the API Line (swyx take)* The Bitter Lesson* The Goddess of Everything Else (listen to George read it)* Meditations on Moloch* George’s email to Lisa Su, AMD’s CEO:Timestamps* [00:00:00] Intros & tinygrad’s “Portal Story”* [00:03:00] Thesis #1* [00:03:50] Thesis #2* [00:05:00] Thesis #3 + Turing completeness discussion* [00:10:00] tinygrad’s creation and core ideas* [00:16:00] Operation fusing in tinygrad* [00:17:00] Debugging & profiling in tinygrad* [00:18:30] Tinygrad vs Pytorch competitiveness* [00:20:30] geohot vs AMD* [00:25:00] On ggml* [00:26:00] Tinygrad’s CI philosophy* [00:26:30] On Mojo* [00:28:00] ggml quantization is made up* [00:31:00] Work for tiny: benchmark int8 vs fp16* [00:33:00] Why you can’t build tinybox - Design constraints* [00:35:00] The Personal Compute Cluster* [00:37:00] Shoutout to our MosaicML podcast* [00:39:00] FLOPcoin and other use cases for the tinybox* [00:43:00] Rumors on GPT-4 architecture* [00:46:00] The Bitter Lesson* [00:48:00] Hiring and Changing mind on remote work* [00:52:00] Above/Below The API* [00:55:40] Comma Bodies & Computer Vision* [00:58:40] Merging with the machine and AI girlfriends* [01:02:00] Is AI gonna kill us all?* [01:09:00] Why Avatar 2 was badTranscriptSwyx: Hey everyone, welcome to the Latent Space podcast. This is Swyx, writer and editor of Latent Space. And Alessio is taking over with the intros, Alessio is Partner and CTO in residence at Decibel Partners. [00:00:20]Alessio: Hey everyone, today we have Geohot on the podcast, aka George Hotz. Everybody knows George, so I'm not going to do a big intro. A couple of things that people might have missed: you traded the first ever unlocked iPhone for a Nissan 350Z and three new iPhones. You were then one of the first people to break into the PS3 to run arbitrary code. You got sued by Sony, you wrote a rap song to fight against that, which is still live on YouTube, which we're going to have on the show notes. Did not go to Tesla to build vision, and instead you started Comma.ai, which was an amazing engineering feat in itself until you got a cease and desist from the government to not put these things on the street and turned that into a research only project. [00:01:00]George: You know they're out there. [00:01:01]Alessio: Yeah, yeah. [00:01:03]Swyx: They're out there. [00:01:04]Alessio: But like in a, you know, you market them as a research kind of like no warranty. [00:01:06]George: Because I use the word dev kit, that's not about the government, that's nothing to do with the government. We offer a great one-year warranty. The truth about that is it's gatekeeping. What's the difference between a dev kit and not a dev kit? Nothing. Just the question of do you think it's for you? And if you think it's for you, buy it. It's a consumer product. We call it a dev kit. If you have a problem with that, it's not for you. [00:01:28]Swyx: That's great insight. [00:01:30]Alessio: I was going through your blog posts to get ready. You've wrote this post about The Hero's Journey. And you linked this thing called the portal story, which is kind of the set of stories in movies and books about people living this arbitrary life. And then the run to this magic portals kind of takes them into a new, very exciting life and dimension. When you wrote that post, you talked about TinyGrad, which is one of the projects we're working on today. You mentioned this is more of a hobby, something that is not going to change the course of history. Obviously, you're now going full speed into it. So we would love to learn more about what was the portal that you ran into to get here. [00:02:03]George: Well, what you realize is... You know what made me realize that I absolutely had to do the company? Seeing Sam Altman go in front of Congress. Why? What are the odds they nationalize NVIDIA? What are the odds that large organizations in the government, but of course I repeat myself, decide to try to clamp down on accessibility of ML compute? I want to make sure that can't happen structurally. So that's why I realized that it's really important that I do this. And actually, from a more practical perspective, I'm working with NVIDIA and Qualcomm to buy chips. NVIDIA has the best training chips. Qualcomm has the best inference chips. Working with these companies is really difficult. So I'd like to start another organization that eventually in the limit, either works with people to make chips or makes chips itself and makes them available to anybody. [00:02:48]Alessio: Can you share three core pieces to TinyCorp? Maybe we can dive into each of them. So XLA, PrimTorch, those are the complex instruction system. TinyGrad is the restricted instruction system. So you're kind of focused on, again, TinyGrad being small, not being overcomplicated and trying to get as close to the DSP as possible in a way where it's at more. [00:03:08]George: Well, it's a very clear analogy from how processes are developed. So a lot of processes back in the day were CISC, complex instruction set, system 360, and then x86. This isn't how things stayed. They went to now the most common processor is ARM, and people are excited about RISC-V. No one's excited about it. RISC-V is even less complex than ARM. No one is excited about CISC processors anymore. They're excited about reduced instruction set processors. So TinyGrad is, we are going to make a RISC offset for all ML models. And yeah, it can run all ML models with basically 25 instead of the 250 of XLA or PrimeTorch. So about 10x less complex. [00:03:47]Swyx: Yep. [00:03:48]Alessio: You talk a lot about existing AI chips. You said if you can’t write a fast ML framework for GPUs, you just cannot write one for your own chip. So that's another one of your core insights. I don't know if you want to expand on that. [00:03:59]George: Yeah. I mean, your chip is worse, right? There's no way the chip that you're going to tape out, especially on the first try, is going to be easier to use than an AMD GPU, right? And yet there's no good stack for AMD GPUs. So why do you think you can make one for your chip? You can't, right? There's one other company, aside from NVIDIA, who's succeeded at all at making training chips. What company? [00:04:20]Swyx: AMD? Intel? [00:04:22]George: No, no, no. I've never trained. Who's trained a model on AMD or Intel? Cerebras. [00:04:26]Swyx: Cerebras! [00:04:27]George: I'm talking about, you might know some startups who trained models on these chips. [00:04:31]Alessio: Oh, TPU. [00:04:32]George: Exactly. Right? So Midjourney is trained on TPU, right? Like a lot of startups do actually train on TPUs. And they're the only other successful training chip, aside from NVIDIA. But what's unique about Google is that they also wrote their own ML framework, right? And if you can't write your own ML framework that is performant on NVIDIA, there's no way you're going to make it performant on your stuff. [00:04:53]Alessio: And they started from TensorFlow and then they made the chip after. [00:04:56]Swyx: Yeah, exactly. Exactly. [00:04:58]George: And you have to do it in that direction. Otherwise, you're going to end up, you know, Cerebras, one of those things, a million... Has anyone ever seen a Cerebras? No one's ever like, oh, I trained my model on a Cerebras. Most people are like, I trained my model on GPUs. Some people, 20%, are like, I trained my model on TPUs. [00:05:14]Alessio: And then the third one, which is the one that surprised me the most, is Turing completeness is harmful. It should be avoided. It made sense once I read it, but maybe tell us a bit more about how you got there. [00:05:25]George: Okay. So CPUs devote tons of their silicon and power to things like reorder buffers and speculative execution and branch predictors. And the reason that you need all these things is because at compile time, you can't understand how the code's going to run. This is Rice’s theorem. This is the halting problem and its limit. And this is not like, oh, the halting problem is theoretical. No, no, no, no. It's actually very real. Does this branch get taken or not? Well, it depends on X. Where does X come from? Yeah, forget it, right? But no branches depend on X in a neural net. Every branch is a static loop. Like if you're doing a matrix multiply, it's a static loop over the inner dimension. And neural networks are even better. No loads even depend on X, right? So with a GPU shader, right, your load might depend on which texture you're actually loading into RAM. But with a neural network, your load is, well, I load that way. Why? Well, because I load that way the other million times I ran the same net. Every single time you run the net, you do the exact same set of loads, stores, and arithmetic. The only thing that changes is the data. And this gives you a very powerful ability to optimize that you can't do with CPU-style things, which have branches, and even GPU-style things, which have loads and stores. Well, GPUs, if you want GPU-style stuff, you have like load based on X, you now need a cache hierarchy, and not an explicit cache hierarchy, an implicit cache hierarchy with eviction policies that are hard-coded into the CPU. You start doing all this stuff, and you're never going to get theoretically good performance. Again, I don't think there's 100X. Some startups will talk about 100X, and they'll talk about absolutely ridiculous things like clockless computing or analog computing. Okay, here, analog computing just won't work. And clockless computing, sure, it might work in theory, but your EDA tools are... Maybe AIs will be able to design clockless chips, but not humans. But what actually is practical is changing cache hierarchies and removing branch predictors and removing warp schedulers, right? GPUs spend tons of power on warp scheduling because we have to hide the latency from the memory. We'll have to hide the latency if everything's statically scheduled. [00:07:25]Alessio: Why do you think people are still hanging on to Turing completeness? [00:07:27]Swyx: Well, because it's really easy. [00:07:29]George: Turing Complete is just really easy to just, oh, you know, it would just be so nice if I could do like an if statement here and actually branch the code, right? So it requires a lot more thought to do it without Turing Completeness. [00:07:41]Swyx: And would this be qualitatively different than TPUs? [00:07:44]George: So TPUs are a lot closer. Yeah. TPUs are a lot closer to what I'm talking about than like CUDA. Okay, so what is CUDA? Well, CUDA is a C-like language, which compiles to an LLVM-like IR, which compiles to PTX, which compiles to SAS, which are all Turing Complete. TPUs are much more like this. Yeah. Their memory is pretty statically managed. They have a V—I did some reverse engineering on the TPU. It's published in TinyGrad. It has like a VLIW instruction, and it runs them. So it's similar. I think the TPUs have a few problems. I think systolic arrays are the wrong choice. I think they have systolic arrays because that was the guy's PhD, and then of course Amazon makes— [00:08:20]Swyx: Could you summarize systolic arrays for us? [00:08:21]George: Systolic arrays are just—okay, so basically you have like—it's a way to do matrix multiplication. Think of a grid of mollax, and then the grid can multiply, and then shift, multiply, then shift, multiply, then shift. And they are very power efficient, but it becomes hard to schedule a lot of stuff on them if you're not doing like perfectly sized dense matrix multiplies, which you can argue, well, design your models to use perfectly sized dense matrix multiplies, sure. [00:08:47]Swyx: Thanks for indulging on these explanations. I think we need to keep our audience along with us by pausing every now and then to explain key terms. [00:08:56]George: When I say explain a systolic array, I just immediately get a picture in my head of like tilting a matrix and shifting it. It's hard to kind of explain. Yeah. [00:09:04]Swyx: Yeah. We'll do something. We'll do something. We'll have show notes. [00:09:08]George: And we edit in visuals. Yeah, yeah, yeah. There's some great graphics that just show you, oh, so that's what a systolic array is. But it's a mollax shift machine that looks kind of different from the typical ALU sort of machine. I think the right answer is something that looks more like queues that feed into ALUs, and then you can prefetch the loads from the memory, put in a bunch of queues, and then the queue is just like, and feeds into another queue over here. But yeah, but that's not even the main problem with TPUs. The main problem with TPUs is that they're closed source. Not only is the chip closed source, but all of XLA is open source. But the XLA to TPU compiler is a 32 megabyte binary blob called libTPU on Google's cloud instances. It's all closed source. It's all hidden stuff. And you know, well, there's a reason Google made it closed source. Amazon made a clone of the TPU. It's called Inferentia. Or they have some other name for it, a training. Tranium. Yeah, yeah, yeah. And look, it's a clone of the TPU. But Google's software at least kind of works. [00:09:58]Alessio: So those are kind of like the three core pieces. The first thing you're working on, that you've been working on, is TinyGrad. And one of your Twitch streams, you said, is the best thing you've ever written. [00:10:07]Swyx: Yeah. [00:10:08]Alessio: Tell us a bit more about that creation. [00:10:10]George: For a long time, TinyGrad had a hard limit at a thousand lines of code. And what this would force you to do is really make sure you were not wasting lines. I got rid of the restriction because it became a little code golfy at the end. But once like the core framework of TinyGrad was there in those thousand lines, but like the core framework, the ideas are expressed with no boilerplate. If you go read PyTorch, you know, PyTorch I think is actually pretty good code. I think Facebook's pretty good, but there's so much boilerplate. Go in PyTorch and try to track down how an LGU actually works. [00:10:44]Swyx: Just a lot of instructions. [00:10:45]George: Oh, you're going to be diving down a long stack from Python to C to custom libraries to dispatchers to, and then I don't even know how to read TensorFlow. I don't even know where's an LU in TensorFlow. [00:10:55]Swyx: Nobody knows. [00:10:56]George: Someone at Google knows maybe. Google as an organism knows. I don't know if anyone individual at Google knows. [00:11:02]Alessio: What are like the important ergonomics like for a developer as you think about designing the TinyGrad API? [00:11:07]George: So the TinyGrad front end looks very similar to PyTorch. There's an even higher level front end you can use for TinyGrad, which is just ONNX. We have better support for ONNX than Core ML does. And we're going to have, I think we're going to pass ONNX Runtime soon, too. People think ONNX Runtime, that's the gold standard for ONNX. No, you can do better. [00:11:23]Swyx: Pass them in what, specifically? Test compliance tests. [00:11:26]George: So ONNX has a big set of compliance tests that you can check out. And we have them running in TinyGrad, and there's some failures. We're below ONNX Runtime, but we're beyond Core ML. So that's where we are in ONNX support now. But we will pass ONNX Runtime soon because it becomes very easy to add ops because you don't need to do anything at the lower levels. You just do it at this very high level, and TinyGrad compiles it to something that's fast using these minimal ops. You can write, most concretely, what TinyGrad can do that PyTorch can't really do, is if you have something like A times B plus C. If you write that in NaivePyTorch, what it's going to do on the GPU is read A, read B in a kernel, and then store A times B in memory, and then launch another kernel to do A times B plus C. Okay, got to do those loads from memory. It's a whole extra round trip to memory that I just didn't have to do. And you're like, yeah, but you can use the Torch JIT, and it corrects this. Yeah, for that one example, for that one example of MUL/ACC, but, oh, now you did three multiplies? Six multiplies? It won't compile arbitrary code. [00:12:26]Swyx: And have you looked into the other approaches like PyTorch Lightning to accelerate PyTorch itself? [00:12:32]George: Well, PyTorch Lightning, my understanding is, it's mostly a framework around PyTorch, right? PyTorch Lightning is not going to fix this fundamental problem of I multiply six tensors together. It's not going to fix it going to memory any more than a single read from each and a single write to the output. There are lower level things in PyTorch that are, I'm not exactly sure what Dynamo does, but I know they're generating some Triton stuff, which is going to generate the kernels on the fly. But, you know, PyTorch Lightning is at a higher level of abstraction. So TinyGrad's front-end stuff looks like PyTorch. I made a few tweaks. There's a few things I don't like about PyTorch. Why is Relu a class? Really, what's the state? You make a class, and there's a state. Everything should just be Torch functional and then Relu, but just dot Relu on the tensor. There's things in Torch where you have to do tensor dot and not a tensor dot. It just shows an API that's not perfectly refined. But when you're doing stuff TinyGrad style where you don't have lines, well, it has to work this way. Because even the lines to express the, well, you can't use the where operator in PyTorch. Why is it true case, condition, false case? Ugh, that's how Python expresses ifs. It's disgusting. Turner operators are much nicer. It should be, I can do my like a less than zero dot where a comma one, right? [00:13:46]Swyx: The very pandas-like API? [00:13:50]George: It looks like Torch, NumPy, pandas. They're all very similar. I tried to take the cleanest subset of them and express them. But like I said, you can also interact with it using ONNX. I have a rewrite of StableDiffusion, I have a rewrite of Llama, I have a rewrite of Whisper. You can look at them. They're shorter than the Torch versions, and I think they're cleaner. And you stream them all? [00:14:05]Swyx: Yeah. Very nice. [00:14:07]Alessio: So what's the other important concept that you're leveraging to do operation fusing? [00:14:11]George: Yeah, you have basically like a few different like models for the simplest one is eager is as soon as the interpreter sees A times B, it actually dispatches A times B, right? Then you have graph like TensorFlow, which will put A times B into a graph, and then we'll do absolutely nothing until you actually compile the graph at the end. I like this third choice, which is somewhere in the middle, laziness. Laziness is you don't know when the ops are going to dispatch, and don't worry about that. You don't have to worry about this as a programmer, you just write out all your stuff. And then when you actually type `.numpy`, it'll be ready by the time you copy the thing back to CPU. Or you can do `.realize`, and it will actually like force that tensor to be allocated in RAM. And if you think about it, PyTorch is kind of lazy in a way, but they didn't extend the paradigm far enough, right? When I do A times B in PyTorch, it's going to launch a CUDA kernel to do A times B. But it's not going to wait for that CUDA kernel to complete. So you're getting the worst possible worlds. You're getting the same laziness, but you also can't get fusion, because PyTorch doesn't know that I'm then going to do plus C. There's no way for it to be like, whoa, whoa, whoa, don't launch that CUDA kernel. Whoa, just do this one too. Right? Again, PyTorch is working on this, and it's a little bit harder. In Kama, I felt like I was competing against a lot of idiots. Here, I'm competing against smart, very smart people who've made some, I think, different trade-offs. Whereas, if you're trying to build something that is just straight up good on NVIDIA, and we have a lot of people and complexity to throw at it, yeah, PyTorch made a lot of the right choices. I'm trying to build something that manages complexity. You can always make your software do more. The magic is when you can make your software do more without adding complexity, right? Because complex things eventually collapse under their own weight, so it's kind of... [00:15:58]Alessio: How does fusing actually work? [00:16:00]George: There's this thing called lazy.py, and when you do A times B, that's... It's put into a graph, but it's a very local graph. There's no global graph optimizations. And even this can change, right? Again, the programming model for TinyGrad does not preclude eagerness, right? Laziness is not guaranteed laziness. It's just going to try its best. So you put in A times B, and that's a binary op, right? And then you put in A times B, that's a node in the graph. It's a virtual node because it's not realized yet, plus C. Okay, here's a new node, which takes the C tensor in here and takes the output of A times B. It's like, whoa, there's two binary ops. Okay, we'll just fuse those together. Okay, here I have a kernel. This kernel has A, B, and C as inputs. It does A times B plus C in the local registers, and then outputs that to memory. And you can graph.one in TinyGrad. Another amazing thing that TinyGrad has that I've not seen in any other framework is two things. Graph equals one, which is an environment variable. It will output a complete graph of all the operations. Other people are like, oh, you can use PyTorch, export it to ONNX, and use Netron. Yeah, you can. Like, what? That's not what's real. Graph equals one will show you the actual kernels that were dispatched to the GPU. You can also type debug equals two, which will print those kernels out in your command line, and it will tell you the exact number of flops and the exact number of memory accesses in each kernel. So you can immediately see, wait a second, okay, this kernel used this many flops. This was the gigaflops. This is how many bytes it read, and this was the gigabyte per second. And then you can profile without having to like, okay, I mean, in theory, in PyTorch, Sure, use the NVIDIA Insight Profiler. No one does that. No one does, of course, because it's so difficult, right? Like, actually, NVIDIA used to, I think CUDA 9 was the last one that had it. They had a command line one, but now it's like, okay, I'm going to generate this blob, use this NVIDIA GUI tool to convert it into a Chrome trace, and then load it. Yeah, no one does that, right? Just type debug equals two in any TinyGrad model, and it will show you all the kernels that it launches and the efficiency of each kernel, basically. [00:17:58]Swyx: Yeah, this is something that John Carmack has often commented about, is that when you code, you need to build in your instrumentation or observability right into that. I wonder if whatever John is working on, he's adopting this style, and maybe we can sort of encourage it by, I don't know, naming it and coining a certain kind of debugging style? [00:18:16]George: If he would like to start contributing to TinyGrad, I'd be so happy. [00:18:19]Swyx: You should hook up with them. [00:18:22]George: I've chatted with them a few times. I'm not really sure what his company's doing, but no, I mean, hopefully we get TinyGrad to a point where people actually want to start using it. So TinyGrad right now is uncompetitive on NVIDIA, and it's uncompetitive on x86. [00:18:36]Swyx: And specifically, what do you care about when you say uncompetitive? Speed. [00:18:39]George: Share of speed. It's correct. The correctness is there. The correctness for both forwards and backwards passes is there. But on NVIDIA, it's about 5x slower than PyTorch right now. Like 5x, wow, this is unsurmountable. No, there's reasons it's 5x slower, and I can go through how we're going to make it faster. It could be 100x slower, so we're making progress. But there's one place where it actually is competitive, and that's Qualcomm GPUs. So TinyGrad is used to run the model in OpenPilot. Like right now, it's been live in production now for six months. And TinyGrad is about 2x faster on the GPU than Qualcomm's library. [00:19:10]Swyx: What about Qualcomm architecture? [00:19:12]George: What makes it doable? Well, because the world has spent how many millions of man hours to make NVIDIA fast? And Qualcomm has a team of 10 Qualcomm engineers? Okay, well, who can I beat here? What I propose with TinyGrad is that developer efficiency is much higher. But even if I have 10x higher developer efficiency, I still lose on NVIDIA, right? You know, okay, I didn't put 100,000 man hours into it, right? If they put a million, like, that's what I'm saying. But that's what I'm saying we can get. And we are going to close this speed gap a lot. Like I don't support TensorCourse yet. That's a big one that's just going to, okay, massively close the gap. And then AMD. I don't even have a benchmark for AMD because I couldn't get it compiled. Oh, and I tried. Oh, I tried. I spent a day. Like, I spent actually a day trying to get PyTorch. And I got it built. I got it kind of working, then I tried to run a model, like, there's all kinds of weird errors and the rabbit holes are so deep on this. I'm like, you know, you can compare the speed. Right now, you can run LLAMA, you can run anything you want on AMD. It already all works. Any OpenCL backend works, and it's not terribly slow. I mean, it's a lot faster than crashing. So it's infinitely times faster than PyTorch on AMD. But pretty soon, we're going to start getting close to theoretical maximums on AMD. That's really where I'm pushing. And I want to get AMD on MLPerf in a couple months, hopefully. [00:20:26]Swyx: Now that you bring up AMD. [00:20:27]Alessio: Yeah, let's dive into that. Because when you announced the Semicore fundraise, you mentioned one of your first goals is like build the framework, runtime and driver for AMD. And then on June 3rd on Twitch, you weren't as excited about AMD anymore. Maybe let's talk a bit about that. You compared the quality of commit messages from the AMD kernel to the Intel work that people are doing there. What's important to know? [00:20:51]George: When I said I want to write a framework, I never intended on writing a kernel driver. I mean, I flirted with that idea briefly, but realistically, there's three parts to it, right? There's the ML framework, there's the driver, and then there's the user space runtime. I was even down to rewrite the user space runtime. I have a GitHub repo called CUDA IOControlSniffer. It's terribly called. But you can actually launch a CUDA kernel without CUDA. So you don't need CUDA installed. Just the NVIDIA open source driver and this open source repo can launch a CUDA kernel. So rewriting the user space runtime is doable. Rewriting the kernel driver? [00:21:26]Swyx: I don't even have docs. [00:21:27]George: I don't have any docs for the GPU. Like it would just be a massive reverse engineering project. I wasn't complaining about it being slow. I wasn't complaining about PyTorch not compiling. I was complaining about the thing crashing my entire computer. It panics my kernel. And I have to wait five minutes while it reboots because it's a server motherboard and they take five minutes to reboot. So I was like, look, if you guys do not care enough to get me a decent kernel driver, there's no way I'm wasting my time on this, especially when I can use Intel GPUs. Intel GPUs have a stable kernel driver and they have all their hardware documented. You can go and you can find all the register docs on Intel GPUs. So I'm like, why don't I just use these? Now, there's a downside to them. Their GPU is $350. You're like, what a deal. [00:22:03]Swyx: It's $350. [00:22:04]George: You know, you get about $350 worth of performance. And if you're paying about $400 for the PCIe slot to put it in, right, like between the power and all the other stuff, you're like, okay, nevermind. You got to use NVIDIA or AMD from that perspective. But I sent an email to Lisa Su. She responded. [00:22:19]Swyx: Oh. [00:22:20]George: And I've had a few calls since. And like, what I tried to do, first off, like, thank you for responding. It shows me that like, if you don't care about your kernel panicking, I can't, like, this is just a huge waste of my time, right? I'll find someone who will care. I'm not asking for your seven by seven Winograd convolution when transposed to be fast. Like, I'm not asking for that. I'm asking literally for- The basics of getting it running. Oh, and this isn't TinyGrad. This is your demo apps. I ran their demo apps in loops, and I got kernel panics. I'm like, no, okay. No, Lisa Su reached out, connected with a whole bunch of different people. They sent me a pre-release version of RockM 5.6. They told me you can't release it, which I'm like, guys, why do you care? But they say they're going to release it by the end of the month, and it fixed the kernel panic. The guy managed to reproduce it with the two GPUs and the computer, and yeah, sent me a driver, and it works. I had that experience, and then I had another experience where I had two calls with, like, AMD's, like, communication people. I was just like, I tried to explain to these people, like, open source culture. Like, it's not open source if you dump the source code on a GitHub repo and then forget about it until the next release. It's not open source if all your issues are from 2022. Like, it's just no one's going to contribute to that project, right? Sure, it's open source in a very, like, technical sense. To be fair, it's better than nothing. It's better than nothing, but I fixed a bug in Nickel that I fixed. There's a fun fact, by the way. If you have a consumer AMD GPU, they don't support peer-to-peer, and their all-reduce bandwidth is horrendously slow because it's using CUDA kernels to do the copy between the GPUs, and it's putting so many transactions on the PCIe bus that it's really slow. But you can use CUDA memcpy, and there's a flag to use CUDA memcpy, but that flag had a bug. I posted the issue on Nickel. I expected nothing to happen. The NVIDIA guy replied to me within an hour. He's like, try this other flag. I'm like, okay, I tried the other flag. It still doesn't work, but here's a clean repro. And I spent, like, three hours writing a very clean repro. I ended up tracking the issue down myself, but just the fact that somebody responded to me within an hour and cared about fixing the issue? Okay, you've shown that it's worth my time, and I will put my time in because, like, let's make this better. Like, I'm here to help. But if you show me that, you know, you're like, you're the kernel panics. That's just, like, expected. Okay. [00:24:36]Swyx: Well, it sounds like AMD is getting the message. [00:24:38]George: They are. And I just, I don't really think they've had someone explain to them, like, like, I was like, you can, like, build in public. And they're like, what's an example of building in public? I'm like, go look at PyTorch. Go look at PyTorch. I have two minor things merged into PyTorch because it's very responsive, you know? [00:24:53]Alessio: So that's kind of like the lowest level of the stack. And then at a slightly higher level, obviously, there's TinyGrad, there's Mojo, there's ggml. How are you thinking about breadth versus, like, depth? Like, where you decided to focus early on? [00:25:06]George: So ggml is very much like a, okay, everyone has M1s, right? Actually, I was thinking, in the beginning, I was thinking of something more like ggml, focused on the M1s. But ggml showed up and was just like, we're actually just focusing on the M1s. And actually, M1 PyTorch is considerably better than AMD PyTorch. M1 PyTorch works, it only gives wrong answers sometimes, and it only crashes sometimes. But, like, some models kind of run. When I was writing the metal backend, I was comparing to MPS PyTorch, and I had, like, a discrepancy. TinyGrad checks all its outputs compared to Torch, and I had one where it didn't match. I'm like, I checked the matrix by hand, it matches TinyGrad, I don't understand. And then I switched PyTorch back to CPU, and it matched. I'm like, oh. Well, there's, like, bugs, like, if you, like, transpose the matrix, because, like, I think it has to do with, like, multi-views in PyTorch, and, like, weird under-the-hood stuff that's not exposed to you, like, there's bugs. And maybe they fixed them, but, like, you know, it seems like there was a lot of momentum. Again, because you're getting how many engineers care about making PyTorch work on M1, right? Thousands, tens of thousands. And you have an open development process, and guess what? It's going to be good. How many engineers care about AMD working, PyTorch AMD working? Well, you got 10 guys that work for AMD, and then, like, a couple hobbyists. [00:26:15]Swyx: You revealed an interesting detail about how you debug. You hand-check the matrix math? No, I don't hand-check it. [00:26:20]George: One of the best tests in TinyGrad is a file called testops.py. And it's just a hundred small examples written in TinyGrad and PyTorch, and it checks both the forwards and backwards to make sure they match. [00:26:34]Swyx: Good test suite. Yeah. Very important. [00:26:35]George: That's, I mean, that's one of them where, like, I really, I put a lot of effort into CI for TinyGrad. I think CI is super important. Like, I want that green check to mean I can merge this, right? Like, I don't want my tests to, and if the green check, if you somehow manage to introduce a bug and get the green check, okay, we're fixing the test, top priority. [00:26:51]Swyx: Mojo? [00:26:52]George: It's closed source. No, I'm not that interested. Do you know what I mean? Like, look, I like Chris Lattner. I think he's going to do great things, and I understand the, like, kind of the wisdom, even, in keeping it closed source. But, you know, I'm interested when it's open. [00:27:05]Swyx: Yeah. You have an interesting design deviation from him, because he's decided to be a, well, promised to be a superset of Python, and you have decided to break with PyTorch APIs. And I think that affects learnability and transportability of code. [00:27:18]George: You know, if the PyTorch thing ends up being, like, a stumbling block, I could write a perfect PyTorch instead of import PyTorch. Instead of, like, yeah, import torch, you type import tinytorchestorch. And if that really becomes the stumbling block, I will do that. No, Chris Lattner went much further than PyTorch. Replicating the PyTorch API is something I can do with a couple, you know, like an engineer monitor. [00:27:44]Swyx: A shim. [00:27:44]George: Right, like a shim, yeah. Replicating Python? [00:27:47]Swyx: Hoo-hoo-hoo! [00:27:48]George: There's a big graveyard of those projects. How's Piston going? How's Jython? [00:27:57]Swyx: PyPy? Oh, you can go way back. [00:27:59]Alessio: So your core mission is commoditizing the petaflop. And then your business goal is to sell computers for more than the cost to make, which seems super reasonable. And you're going to have three tiny boxes? [00:28:11]Swyx: Red, green, blue? No, no, no, no, no, no, no. [00:28:13]George: That was my... Look, you know, a lot of people, like, I love, you know, leaning into, like, saying I'm giving up, right? It's great to give up, right? Giving up is this wonderful thing. It's so liberating. And then, like, you can decide afterward if you really give up or not. There's very little harm in saying you give up, except, like, you know, great, Twitter haters have something to talk about, and all press is good press, kids, so... Just red, only red. [00:28:32]Swyx: Tiny box, red. Tiny box, red. [00:28:34]George: Unless AMD, you know, upsets me again, and then we're back to other colors. We have other colors to choose from. [00:28:41]Alessio: When you think about hardware design, what are some of the numbers you look for? So, teraflops per second is one, but, like, memory bandwidth is another big limiter. Like, how do you make those trade-offs? [00:28:52]George: Well, I mean, fundamentally, I'm limited to what GPUs I can buy. But, yeah, for something that I think a lot of people are going to want to reasonably do, with, um... A coworker of mine described them as luxury AI computers. Right? Like, luxury AI computers for people. And that's, like, what we're building. And I think a common thing people are going to want to do is run, like, Large Llama. Right? Or Large, like, Falcon or whatever. [00:29:13]Swyx: FB-16 Llama. [00:29:14]George: FB-16, exactly. Exactly. Um, you know, Int8, I think, can work. I think that, like, what GGML is doing to go to, like, N4. Like, this doesn't work. Like, have you done... I mean, maybe they have. But, like, I read what it was, and I was like, this isn't from any paper. This is just some... Squeezing as much as possible. Yeah, you made up some quantization standards to make it run fast. And, like, maybe it works. But, okay, where's, like, the Hellaswag number? Right? Where's your, uh... [00:29:38]Swyx: The thesis is right. That, like, if you have hundreds of billions of parameters, that the individual quantization doesn't actually matter that much. [00:29:44]George: Well, the real way to look at all of that is to just say you want to compress the weights, right? It's a form of weight compression. Quantization is a form of weight compression, right? Now, this is obviously not lossless. It's not a lossless compressor, right? If it's a lossless compressor, and you can show that it's correct, then, okay, we don't have to have any other conversation. But it's a lossy compressor. And how do you know that your loss isn't actually losing the power of the model? Maybe int4 65B llama is actually the same as FB16 7B llama, right? We don't know. Maybe someone has done this yet, but I looked for it when it, like, first came out and people were talking about it. And I'm like, it's not from a paper, right? The indate stuff is from a paper where they... Like, some of the indate stuff is from a paper. There's one paper, I think it's, like, indate... LLM.indate, where they actually do all the tests. And they didn't go fully indate. They made, like, 90% of it indate and kept, like, 10% of it in FB16 for what they called, like, the outliers or whatever. So I think that this is not quite so easy. [00:30:37]Swyx: And I think being able... [00:30:38]George: Well, so first off, if you're training, no one's gotten training to work with indate yet. There's a few papers that vaguely show it. But if you're training, you're going to need BF16 or float16. So this is why I target that. Now, the thing that you're going to want to do is run these large language models out of the box on your hardware in FB16, and that's memory bandwidth. So you need large amounts of memory bandwidth, too. So ask how I trade off memory bandwidth in Flop, so what GPUs can I buy? [00:31:02]Alessio: So first of all, you have this hiring process, which is you've got to solve one of the bounties that are open on TinyGrad. There's no technical interview. One of them is indate support. Do you already have some things you want to test on? [00:31:14]Swyx: We have indate support. What I'd like to see somebody do [00:31:16]George: is just load the ggml indate llama into TinyGrad and then benchmark it against the FB16 one. Indate already works in TinyGrad. It doesn't actually do the math in indate. It does all the math still in FB32. So indate can mean you just have your weights in indate, or indate can mean you actually do your math in indate. And doing your math in indate, the big gain that people care about is actually having your weights in indate, because weights in indate mean less memory and less memory bandwidth, whereas the math, keep it in FB32. With on M1s, it doesn't matter what data type you're doing in the GPU. I'm not even sure it can do indate, but FB16 and FB32 is the same tariff ops. So yeah, no, that's one of the bounties. One of the bounties is get indate llama running [00:31:58]Swyx: with the indate weights. [00:32:00]George: And then actually, what you could even do, if you really want to test this, just take the FB16 weights, convert them to indate, then convert them back to FB16, then compare the unconverted and converted. [00:32:10]Swyx: Oh, that's a nice hack. Oh, yeah. Right, like- This should be lossless in the other direction. Yeah, I think FB16, [00:32:17]George: it should be lossless in the other direction. I'm actually not 100% about that. Why not? Oh, because like, you ever try to like, like if you want to represent, if it was like int16, it's not lossless. [00:32:25]Swyx: Sure. [00:32:26]George: All of indate can be represented in FB16, but I'm not 100% about that. [00:32:29]Swyx: Just drop the bytes. We just have to do it, right? [00:32:32]George: Just literally do it. There's only 256 to check, like. But yeah, either way, or I mean, int4, definitely. So do your int4, convert it back, and now see, even with int4 weights and FB32 math, like, okay, how much has your performance degraded this model? [00:32:47]Alessio: I think like the, you're planning to release the first tiny box, ship them in like two to six, eight months, something like that. What's top of mind for you in terms of building a team? Who should, who are you calling for? [00:32:59]George: So as the GPU is picked out and you're like, well, I could make that computer with the GPUs. And my answer is, can you? Do you know how hard it is to put six GPUs in a computer? And people think it's really easy. And it's really easy to put one GPU in a computer. It's really easy to put two GPUs in a computer, but now you want to put in eight. Okay, so I'll tell you a few things about these GPUs. They take up four slots. You can buy the nicest super micro. You can't put eight of those in there. You need two slot blowers. [00:33:25]Swyx: If you want to use one of those, [00:33:25]George: those for you super micros, you need two slot blowers or water cooling, right? If you're trying to get the four slot cards in there, you're going to need some form of water cooling. There are some like Chinese 40 nineties that are blowers, right? You have any blowers or water cooling if you're trying to get it in those things, right? [00:33:37]Swyx: So are you doing water? [00:33:39]George: No, I'm not using that chassis. Okay, so now you want to get six GPUs in a computer. So that's a big challenge. You're like, oh, I'll just use a PCIe extenders. I saw it online as tech tips. It works great. No, it doesn't. Try PCIe extenders that work at PCIe 4.0 and interconnect bandwidth, super important. They don't work at 3.0. No PCIe extender I've tested, and I've bought 20 of them, works at PCIe 4.0. So you're going to need PCIe re-drivers. Now, okay, how much is that adding cost, right? Like these things all get really hard. And then tiny boxes, I've even had another constraint to it. I want this thing to be silent, not totally silent, but my limit is like 45, maybe 50 DB, but not super micro machine, 60 DB. We have a small, we have a compute cluster at comma. You gotta wear ear protection to go in there. Like- [00:34:24]Swyx: Yeah, I've seen some videos where you give a tour. Oh yeah. It's noisy. It's super loud. [00:34:28]George: You got all these machines just screaming. All those, like if you have a blower, what is that thing? 10,000 RPM, just screaming. Like I want to be able to use the normal big GPU fans and make this thing so it can sit under your desk, plug into one outlet of power, right? Six GPUs, your GPUs are 350 Watts each. Can't plug that into a wall outlet. Okay, so how are you going to deal with that? Good questions, right? [00:34:51]Swyx: And you're not sharing them. [00:34:52]George: Well, that one, I mean, that one is pretty obvious. You have to limit the power on the GPUs, right? You have to limit the power on the GPUs. Now you can limit power on GPUs and still get, you can use like half the power and get 80% of the performance. This is a known fact about GPUs, but like that's one of my design constraints. So when you start to add all these design constraints, good luck building a tiny box yourself. Obviously it can be done, but you need something that has actually quite a bit of scale and resources to do it. [00:35:15]Alessio: And you see like the, under the desk, it's like one of the main use cases, kind of like individual developer use or. [00:35:21]George: Yeah, what I also see is more of a, like an AI hub for your home, right? As we start to get like home robotics kind of stuff, you don't want to put the inference on the robot, but you also don't want to put the inference on the cloud. Well, you don't want to put it on the robot because, okay, it's 1500 Watts, tiny box. You'll put batteries and charge them, bad idea. Just wireless. Wireless is 0.5 milliseconds, right? This is super fast. You don't want to go to the cloud for two reasons. One, cloud's far away. Okay, it's not that far away. You can kind of address this. But two, cloud's also mad expensive. Like cloud GPUs are way more expensive than running that GPU at your house. At least any rates you're going to get, right? Maybe if you commit to buy, well, yeah, I'm going to buy 10,000 GPUs for three years, then maybe the cloud will give you a good rate. But like, you want to buy one GPU in the cloud? I mean, okay, you can go to like vast, but like if you're going on Azure AWS, so that's expensive. [00:36:12]Swyx: This is like a personal data center instead of a cloud data center. [00:36:16]George: We like the term compute cluster. So we can use NVIDIA GPUs. [00:36:20]Swyx: Yeah, data centers may be a little bit dated. It's a compute cluster, [00:36:23]George: which is totally legal under the CUDA license agreement. [00:36:26]Swyx: You talk a lot about the PCIe connection. Do you think there's any fat there to trim? What do you mean? You're limited by bandwidth. [00:36:32]George: Okay, for some things, yes. So bandwidth is roughly 10x less than what you can get with NB-linked A100s, right? NB-linked A100s are going to have, and then you can even get like full fabric and NVIDIA really pushes on that stuff, 600 gigabytes per second, right? And PCIe, four, you're going to get 60, right? So you're getting 10x less. That said, why do you need the bandwidth, right? And the answer is you need it for training huge models. If you're training on a tiny box, your limit's going to be about 7 billion. If you're training on big stuff, your limit's going to be like 70 billion, right? Okay, you can hack it to get a bit higher. You can hack it, like GPT hacked it to get a bit higher, but like that 65 billion in LLAMA, like there's a reason they chose 65 billion, right? And that's what can reasonably fit model parallel on a GPU, right? So yes, you are going to end up training models. The cap's going to be like 7 billion, but I actually heard this on your podcast. I don't think that the best chatbot models are going to be the big ones. I think the best chatbot models are going to be the ones where you had a thousand training runs instead of one. And I don't think that the interconnect bandwidth is going to matter that much. [00:37:33]Swyx: So what are we optimizing for instead of compute optimal? What do you mean compute optimal? You're talking about this, the LLAMA style models where you train for like 200x. You train longer, yeah. [00:37:41]George: Yeah, yeah, yeah. You can always make your model better by doing one of two things, right? And a comma, we just have a strict limit on it. You can always make your model better by training longer, and you can always make your model better by making it bigger. But these aren't the interesting ones, right? Particularly the making it bigger because training it longer, fine. You're getting a better set of weights. The inference is the same. The inference is the same whether I trained it for a day or a week. Okay, if it's 1 billion versus 10 billion, well, I 10x my inference too, right? So I think that these big models are kind of, sure, they're great if you're research labs and you're trying to like max out this hypothetical thing. [00:38:13]Swyx: Which you can talk about later. Yeah, yeah, yeah. [00:38:15]George: But if you're like a startup or you're like an individual or you're trying to deploy this to the edge anywhere, you don't need that many weights. [00:38:22]Swyx: Yeah, yeah. You actually don't want that many weights. Optimizing for inference rather than capabilities doing benchmarks. Yes. [00:38:29]George: And I think the inference thing, right? There's gonna be so much more. Right now, the ratio between like training and inference on clouds, I think it's only still, I think it's like two or three X, right? It's two or three X more inference, which doesn't make any sense. It's way more inference. [00:38:41]Swyx: Yeah. [00:38:42]George: There should be 10 to 100 X more inference in the world than training. But then also like, what is training, right? You start to see these things like LoRa, like it's kind of blurring the lines between inference and training. And I think that that blurred line is actually really good. I'd like to see much more like on-device training or on-device fine tuning of the final layer. We're pushing toward this stuff at Comma, right? Like why am I shipping a fixed model? I totally want this model to fine tune based on like how your left tire is flat, right? Every time you cut the same turn because your left tire is flat, well, it should learn that, right? [00:39:11]Swyx: So would Comma pursue parameter efficient fine tuning? Yeah. [00:39:16]George: We're looking into stuff like that. I mean, Comma is already very parameter efficient because we have to like run this thing in a car and you have to like cool it and power it. [00:39:22]Alessio: And so this kind of like intelligence cluster you have in your home, you see when the person is using third-party model, they load them locally and kind of do the final fine tuning. It kind of stays within the box. [00:39:33]George: I think that that's one version of it for the privacy conscious. I also see a world where you can have your tiny box in its down cycles, mine flop coin, right? You know, it turns out not all crypto is a scam. [00:39:45]Swyx: There's one way to tell if crypto is a scam. [00:39:46]George: If they're selling the coin before they make the product, [00:39:49]Swyx: it's a scam. [00:39:49]George: If they have the product and then they sell the coin, it's maybe not a scam, right? So yeah, my thought is like each tiny box would let you, would have a private key on it. And you have to do it this way. You can't just let anyone join because of Sybil attacks, right? [00:40:01]Swyx: There's a real problem of like, [00:40:01]George: how do I ensure your data is correct? And the way that I ensure your data is correct on the tiny net is if you ever send wrong data, you're banned from the network for life. [00:40:08]Swyx: Yeah. [00:40:09]George: Your $15,000 hardware box is banned. [00:40:11]Swyx: So, you know, don't cheat. [00:40:11]George: Obviously if it messes up, we'll forgive you. [00:40:14]Swyx: Somebody's going to try to jailbreak your devices. There's no jailbreak. [00:40:17]George: There's no jailbreak. [00:40:18]Swyx: It's just a different network. [00:40:19]George: Well, there's just a private key on ea ch device, right? Like if you buy a tiny box from the tiny corp, [00:40:23]Swyx: I give you a private key. [00:40:23]George: It's in my backend server, right? You want to hack my server, that's illegal. Anything you want to do on the device, the device is yours. My server's mine, right? [00:40:29]Swyx: Yeah. Have you looked into like a federated training at all? [00:40:33]George: Okay. There's orders of magnitude of federated training. You mean like over the cloud and stuff? [00:40:37]Swyx: Over the internet? Yeah. Over the internet, but also distributed on a bunch of devices, right? [00:40:41]George: Yeah, I'm very bearish on this stuff. Because your interconnect bandwidth, right? So, okay. At the high end, you have your interconnect bandwidth of NVLink, which is 600 gigabytes per second, right? The tiny box has 60 gigabytes per second. And then your internet has 125 megabytes per second, right? Not gigabits, 125 megabytes, right? So, okay. That's how many orders of magnitude we're talking here? Like from 60 down to 125? Like, all right, that's over a hundred X. That's 400 X, right? So like, what you can do is inference, right? Like there's, for inference, you don't care, right? For inference, there's so little bandwidth at the top and the bottom of the model that like, yeah, you can do federated inference, right? And that's kind of what I'm talking about. There's also interesting things to push into, like you're like, but okay, what if you want to run closed source models? This stuff gets kind of interesting, like using TPMs on the boxes and stuff. But then someone might jailbreak my device. So, you know, maybe we don't try to do that. [00:41:34]Alessio: Yeah, what's like the enterprise use case? Do you see companies buying a bunch of these and like stacking them together? [00:41:39]George: The tiny box is like the first version of what we're building. But what I really want to do is be on the absolute edge of flops per dollar and flops per watt. These are the two numbers that matter. So the enterprise use case is you want to train, like Kama, right? So Kama just built out a new compute cluster. It's about a person and a half. [00:41:56]Swyx: A person being 20 petaflops. [00:41:58]George: A person is 20 petaflops. It's about 30 petaflops. We built out a little compute cluster and, you know, we paid double what you theoretically could per flop, right? You theoretically could pay half per flop if you designed a bunch of custom stuff. And yeah, I mean, I could see that being, you know, a tiny corp. Kama's going to be the first customer. I'm going to build a box for Kama and then I'm going to show off the box I built for Kama and be like, okay, like, do you want to build? I sell $250,000 training computers. Or how much is one H100 box? [00:42:26]Swyx: It's 400 grand? [00:42:27]George: Okay, I'll build you a 400 grand training computer and it'll be 10x better than that H100 box. Again, not for every use case. For some, you need the interconnect bandwidth. But for 90% of most companies' model training use cases, the tiny box will be 5x faster for the same price. [00:42:41]Alessio: You mentioned the person of compute. How do we build a human for $20 million? [00:42:47]George: Well, it's a lot cheaper now. So like I said, Kama spent about half a million on our person and a half, so. [00:42:54]Alessio: What are some of the numbers people should think of when they compare compute to like people? So GPT-4 was 100 person years of training. That's more like on the timescale. 20 petaflops is one person. I think you, right now the math was that for the price of the most expensive thing we build, which is the International Space Station, we could build one Tampa of. Yeah, yeah, one Tampa of compute. [00:43:16]Swyx: Yeah, which is the ultimate currency of measurement. [00:43:20]George: Yeah, yeah, we could build. So like the biggest training clusters today, I know less about how GPT-4 was trained. I know some rough numbers on the weights and stuff, but Lama- [00:43:28]Swyx: A trillion parameters? [00:43:30]George: Well, okay, so GPT-4 is 220 billion in each head, and then it's an eight-way mixture model. So mixture models are what you do when you're out of ideas. So, you know, it's a mixture model. They just train the same model eight times, and then they have some little trick. They actually do 16 inferences, but no, it's not like- [00:43:45]Swyx: So the multimodality is just a vision model kind of glommed on? [00:43:49]George: I mean, the multimodality is like obvious what it is too. You just put the vision model in the same token space as your language model. Oh, did people think it was something else? The mixture has nothing to do with the vision or language aspect of it. It just has to do with, well, okay, we can't really make models bigger than 220 billion parameters. We want it to be better. Well, how can we make it better? Well, we can train it longer, and okay, we've actually already maxed that out. We're getting diminishing returns there. [00:44:13]Swyx: Okay. A mixture of experts. [00:44:14]George: Yeah, a mixture of experts. We'll train eight of them, right? [00:44:16]Swyx: So, all right. [00:44:17]George: So, you know, the real truth is whenever a start, whenever a company is secretive, it's because they're hiding something that's not that cool. And people have this wrong idea over and over again that they think they're hiding it because it's really cool. [00:44:28]Swyx: It must be amazing. [00:44:29]George: It's a trillion parameters. No, it's a little bigger than GPT-3, and they did an eight-way mixture of experts. Like, all right, dude, anyone can spend eight times the money and get that. Coming back to what I think is actually gonna happen is, yeah, people are gonna train smaller models for longer and fine-tune them and find all these tricks. OpenAI used to publish stuff on this, you know, [00:44:47]Swyx: when they would publish stuff [00:44:48]George: about how much better the training has gotten holding compute constant. It's gotten a lot better, right? Think, compare like BatchNorm to NoBatchNorm. [00:45:00]Swyx: Is you're finding algorithms like FlashAttention? [00:45:02]George: Yeah, well, FlashAttention, yeah. And FlashAttention is the same compute. FlashAttention is an interesting fact where it's actually the identical compute. It's just a more efficient way to do the compute. But I'm even talking about like, look at the new embeddings people are using, right? They used to use these like boring old embeddings. Now, like, Lama uses that complex one, and now there's like Alibi. I'm not up-to-date on all the latest stuff, but those tricks give you so much. [00:45:23]Swyx: There's been a whole round trip with positional embeddings. I don't know if you've seen this discussion. I haven't followed exactly. [00:45:29]George: I mean, you quickly run into the obvious problem with positional embeddings, which is you have to invalidate your KV cache if you run off the context. So that's why I think these new ones, [00:45:38]Swyx: they're playing with them, [00:45:38]George: but I'm not an expert on like the latest up-to-date language model stuff. [00:45:43]Alessio: What are some of the things, I mean, that people are getting wrong? So back to autonomous driving, there was like the whole like LiDAR versus vision thing. People don't get into accidents because they cannot see well. They get into accidents because they get distracted and all these things. Do you see similarities today on like the Pathway GI? [00:45:59]George: Nothing I say about this is ever gonna compete with how Rich Sutton stated it. [00:46:03]Swyx: Rich Sutton, the writer of [00:46:04]George: Reinforcement Learning, The Bitter Lesson. Nothing I say is ever gonna compete with, The Bitter Lesson's way better than any way I'm going to phrase this. Just go read that, and then like, I'm sorry it's bitter, but you actually just have to believe it. Like over and over again, people make this mistake. They're like, oh, we're gonna hand engineer this thing. No, like stop wasting time. [00:46:22]Swyx: I mean, OpenAI is not taking The Bitter Lesson. They were leaders in deep learning for a long, long, long time. [00:46:27]George: Well, OpenAI was the absolute leader to the thesis that compute is all you need, right? [00:46:31]Swyx: And there's a question of how long [00:46:32]George: this thesis is going to continue for. It's a cool thesis, and look, I think I would be lying along with everybody else. I was into language models like way back in the day for the Hutter Prize. I got into AI through the Hutter Prize. Like 2014, I'm trying to build compressive models of Wikipedia. And I'm like, okay, why is this so hard? What this is is a language model, right? And I'm playing with these Bayesian things, and I'm just like, oh, but I get it. I have two data points, and they're almost the same, but how do I measure that almost, right? I just wrapped my head around this, and this was around the time Karpathy released the first RNN that generated the Shakespeare stuff. And I'm like, okay, I get it, right? It's neural networks that are compressors. Now, this isn't actually, you can't actually win the Hutter Prize with these things because the Hutter Prize is MDL. It's the model, size of the model plus the size of the encodings, embeddings. So yeah, you can't, I mean, probably now you can because it's gotten so good. But yeah, back in the day, you kind of couldn't. So I was like, okay, cool. [00:47:29]Swyx: This is what it is. [00:47:29]George: I kind of get it. I didn't expect that it would continue to work this well. I thought there'd be real limits to how good autocomplete could get. That's fancy autocomplete. But yeah, it works well. So like, yeah, what is OpenAI getting wrong? Technically, not that much. I don't know. If I was a researcher, why would I go work there? [00:47:48]Swyx: Yes, so why is OpenAI like the Miami Heat? [00:47:51]George: No, look, this is my technical stuff. I don't really want to harp on this, but like, why go work at OpenAI when you could go work at Facebook as a researcher? OpenAI can keep ideologues who, you know, believe ideological stuff and Facebook can keep every researcher who's like, dude, I just want to build AI and publish it. [00:48:08]Alessio: Yeah, any other thoughts, tiny corp, bounties? [00:48:11]George: You know, I've been thinking a lot about like what it means to hire in today's world. Okay, look, I'm a believer that machines are going to replace everything in about 20 years. So, okay, what is that thing that people can still do that computers can't? And this is a narrowing list, but like, you know, back in the day, like imagine I was starting a company in 1960. Oh, and we're going to have to hire a whole bunch of calculators in the basement to do all the, you know, math to support the, dude, have you heard about computers? Why don't we just buy a few of those? Oh, wow, man, you're right. So like, I feel like that's kind of happening again. And I'm thinking about, I will post in my Discord, I'll be like, who wants to like, okay, I just changed my unary ops used to be log and exp in like E. I changed them to be log two and exp two because hardware has log two and exp two accelerators. [00:48:59]Swyx: Yeah, and of course you can just change your base. [00:49:00]George: It's one multiply to get it back to E. But like, I made the primitives log two and exp two, right? I just posted in the Discord. I'm like, could someone put this pull request up? And someone eventually did and I merged it. But I'm like, this is almost to the level [00:49:12]Swyx: where models can do it. [00:49:14]George: We're almost to the point where I can say that to a model and the model can do it. [00:49:17]Swyx: Have you tried? Yeah, I don't know. [00:49:20]George: I think autocomplete went further than I thought it would, but I'm also relatively unimpressed with these chatbots. The problem is if your loss function is categorical cross entropy on the internet, your responses will always be mid. [00:49:32]Swyx: Yes, mode collapse is what I call it, I don't know. [00:49:35]George: Maybe, I'm not even talking about mode collapse. You're actually trying to predict the, like, look, I rap. I'm a hobbyist rapper. When I try to get these things to write rap, the raps sound like the kind of raps you read in the YouTube comments. [00:49:45]Swyx: Nursery school. [00:49:46]George: Yeah, it's like, all right, great. You rhyme box with fox, sick rhyme, bro. You know, and Drake is rhyming give it up for me with napkins and cutlery, right? Like, all right, come on. [00:49:55]Swyx: He's got like this thing about orange. Orange is famous so you can't rhyme it. Yeah, yeah, yeah, yeah, yeah. [00:49:59]George: But now, of course, you know, four-inch screws and orange juice is in GPT's training course. Yeah, so I think it went further than everyone kind of thought it would. But the thing that I really want to see is like somebody put 10 LLMs in a room and have them discuss the answer before they give it to me. Right, like, you can actually do this, right? And I think the coding things have to be the same way. There is no coder alive, no matter how good you are, that sits down, well, I'm going to start at cell A1 and type my program, and then I'm going to press run and it's going to work. No one programs like that. So why do we expect the models to, right? So there's a lot that, like, still needs to be done. But, you know, at the tiny corp, I want to be on the cutting edge of this, too. I want to be, like, program generation. I mean, what is TinyGrad? It's a compiler, it generates programs. Generate the fastest program that meets the spec, right? Why am I not just having ML do that? So, you know, it's kind of a, you have to exist fluidly with the machines. And I've come around on a lot of stuff. I'm like, wait, TinyGrad, TinyCorp should be a remote company. I can't do this in person. [00:50:58]Swyx: Really? [00:50:58]George: Yeah, like, comma makes sense to be in person. Like, comma, sure. Yeah, we're getting off in San Diego. [00:51:04]Swyx: But that was a six-year-old company, right? [00:51:05]George: And it works, and it works for a certain type of people [00:51:08]Swyx: and a certain type of culture. [00:51:08]George: But what's going to be different this time? Okay, remote, but now it's remote. And now I'm getting these, like, people who apply, and I'm like, I literally have a thousand applications. I'm not calling you to do a technical screen. I can't really tell anything from a technical screen. What am I going to do? Make a code on a whiteboard? Like, bring up a shared notebook document, so we could, oh, like, that's not going to work. Okay, so then I'm moved to the next thing. We do this at Comma with good success, programming challenges. [00:51:31]Swyx: I've also found them to be, like, [00:51:32]George: completely non-predictive. I found one thing to actually be predictive, and it's, wait a second, just write code in TinyGrad. It's open source, right? And yeah, so, you know, I'm talking to a few people who've been contributing, and, like, contribute, or, you know, the job's not for you. But you can do it remote, and it's, look, it's a chill job. Like, you're not, you're like, oh, yeah, well, I work for the tiny corp. Like, well, you're writing MIT-licensed software. Like, you see what it's doing, right? Like, we'll just, I think, think of it as maybe more of, like, a stipend than a salary. And then also some equity. Like, if, you know, I get rich, we all get rich. [00:52:01]Alessio: How do you think about agents and kind of, like, thinking of them as people versus, like, job to be done? Sean built this thing called Small Developer. [00:52:09]Swyx: It's in the same vein. Or, like, the human in the loop with the language model and just iterating while you write code. I think that's absolutely where it goes. [00:52:17]Alessio: And there's, like, a, it's not, like, one thing. It's, like, there's Small Interpreter. There's, like, Small Debugger. It's kind of, like, all these different jobs to be done. [00:52:24]Swyx: It's a small world. [00:52:25]Alessio: Yeah, it's a, I know, this is, like, the small box is, like, small AI meets tiny corp. [00:52:29]Swyx: So we're all in the same wavelength. [00:52:30]Alessio: How do you think about that? Do you think people will have a human-like interaction where it's, like, oh, this is, like, the AI developer, or, like, is it I'm the human being supercharged by the AI tools? [00:52:41]George: Oh, I think it's, yeah, much more like I'm the human supercharged by the AI tools. I think that, like, coding is tool-complete. Like, driving's not tool-complete. We hire people to drive who are, like, below the API line. Right, there's an API line in the world, right? [00:52:53]Swyx: Love that. Yes. [00:52:53]George: Yeah, yeah, yeah, there's an API line in the world. And, like, you can think, like, Uber's a really clear example, right? There's the people below the API line and the people above the API line. And the way you can tell if you're below or above, by the way, is is your manager a computer, right? Who's the manager of the Uber driver? [00:53:06]Swyx: Well, a computer, right? Does the machine tell you what to do or do you tell machines what to do? Exactly, exactly. [00:53:09]George: So, coding is tool-complete, right? [00:53:13]Swyx: Coding is tool-complete. [00:53:13]George: Coding is above the API line. So it will always be tools supercharging your coding workflow. And it will never be you performing some, like, task. Like, okay, well, I can do everything except for actually starting a Docker container. Like, it just doesn't make any sense, right? Yeah, so it will always be sort of tools. And, you know, look, we see the same stuff with all the, like, people are like, stable diffusion's gonna replace artists or whatever. It's like, dude, like- [00:53:38]Swyx: It's gonna create new artists. [00:53:39]George: Did Photoshop replace artists? [00:53:41]Swyx: Like, what are you talking about, right? [00:53:42]George: Like, you know, a real artist's finger paint. They can't use brushes. Brushes are, you know, brushes are gonna replace all the, okay, like, I just can't. Like, it's all just tools and the tools are gonna get better and better and better. And then eventually, yes, the tools are going to replace us. But, you know, that's still 20 years away. So, you know, I got a company to run in the meantime. [00:54:02]Swyx: So I've written about the API line before and I think that's from Venkatesh. I don't know if you've got your directive to it. I don't know, I definitely took it from someone. [00:54:07]George: It's definitely not mine. [00:54:08]Swyx: It's VGR. But I also have a speculated, a higher line than that, which is the Kanban board. Like, who tells the programmers what to do, right? So are you above or below the Kanban board? Has that evolved your management thinking? [00:54:21]George: Yeah, like, that's sort of what I mean. Like, it's like, I'm just gonna describe the pull request in two sentences and then like, yeah. [00:54:28]Swyx: So you are running the Kanban board? Or the bounties, you know? [00:54:31]George: Yes, the bounties are the Kanban board, exactly. And that is kind of the high level. And then like, yeah, we'll get AIs to fill in some and we'll get people to fill in others. And that's also what it means to be like, full-time at TinyCorp, right? Would you start, and I wrote this up pretty concretely. I'm like, okay, step one is you do bounties for the company. Step two is you propose bounties for the company, right? You don't obviously pay them, we pay them. [00:54:52]Swyx: But you propose them. [00:54:52]George: And I'm like, yeah, that's a good bounty. That like, helps with the main workflow of the company. And step three is you get hired full-time, you get equity, we all, you know, maybe get rich. [00:55:01]Swyx: What else are you designing differently about the employee experience? [00:55:04]George: You know, some people really like to like, [00:55:06]Swyx: like keep a separation, right? [00:55:07]George: Some people really like to keep a separation between like employees and management or customers and employees. Like a comma, you know, the reason I do the DevKit thing, it's like, dude, you buy a comma thing, you're an employee of the company. Like you're just part of the company. It's all the same thing. There's no like secrets, there's no dividing lines. There's no like, it's all a spectrum for like, you know, down here at the spectrum, like you pay. And then up here at the spectrum, you get paid. You understand this is the same spectrum of college, right? Like for undergrad, you pay, and then you get up here to like, you know, I'm doing a PhD program, you get paid. Okay, well, cool. Welcome to the, you know. [00:55:39]Alessio: What about comma bodies? You mentioned a lot of this stuff is clearly virtual, but then there's below the API line you actually need. [00:55:47]Swyx: Wait, this is a thing that's been announced? Comma bodies? We sell them. You can buy them. [00:55:51]George: They're a thousand bucks on our website. [00:55:53]Swyx: Oh, okay, no, no, no. I'm thinking about like the, what Tesla announced with like the humanoid robots. It's the same thing. [00:55:58]George: Except of course, we made the comma version of it. Tesla uses 20 actuators. We use two, right? Like how do you build the simplest possible thing that can like turn the robotics problem into entirely a software problem? So right now it is literally just a comma three on a pole with two wheels. It balances, keeps the comma three up there. And like, there's so much you could do with that already. [00:56:21]Swyx: Right? [00:56:22]George: Like this should replace, how many security guards could this replace? Right? If this thing could just competently wander around a space and take pictures and, you know, focus in on things, send you a text message when someone's trying to break into your building, you know, like, like this could already do so much, of course, but the software is not there yet. Right? So how do we turn robotics into a thing where it's very clearly a software problem? You know, that people don't accept that self-driving cars are a software problem. Like, I don't, I don't know what to tell you, man. Like literally just watch the video yourself and then drive with a joystick, right? Can you drive? And we've actually done this test. We've actually done this test where you've had someone, okay, you just watch this video and here's a joystick and you got to drive the car. And of course they can drive the car. It takes a little bit of practice to get used to the joystick, but the problem is all the model, right? So I can now make the model better. [00:57:07]Swyx: Our second most popular episode ever was about segment anything coming out of Facebook, which as far as I understand the state of the art in computer vision, what are you hoping for there that you need for Karma? [00:57:17]George: I haven't used segment anything. Like they large, large YOLOs or not. I've used like large YOLOs and I'm super impressed by them. [00:57:24]Swyx: Yeah. [00:57:25]George: I got to check out segment anything. I don't think it's a distinct problem, right? Okay, here's something that I'm interested in. All right, we have great LLMs. We have great text to speech models and we have great speech to text models. Okay, so why can I not talk to an LLM? Like I'd have a normal conversation with it. [00:57:39]Swyx: You can with the latency of like two seconds every time. Right? [00:57:42]George: And then it feels so unnatural. It's this like staccato. Like I don't like the RLHF models. I don't like the tuned versions of them. You take on the personality of our customer support agent. Right? [00:57:53]Swyx: Like, oh, come on. [00:57:54]George: I like LLMA more than ChatGPT. ChatGPT's personality just graded on me. Whereas LLMA, like, cool. I read a little bit of pretext paragraph. I can put you in any scenario I want, right? Like, that's interesting to me. So yeah, I think there is really no like distinction between computer vision and language and any of this stuff. It's all eventually going to be fused into one massive. So to say computer vision is solved, well, it doesn't make any sense because what's the output of a computer vision model? Segmentation? Like, what a weird task, right? [00:58:26]Swyx: Who cares? OCR? [00:58:28]George: Who cares? [00:58:29]Swyx: I don't care if you can segment [00:58:29]George: which pixels make up that laptop. I care if you can pick it up. [00:58:32]Alessio: And you're going to have the local cluster. You're going to have the body. [00:58:36]Swyx: Yeah. [00:58:37]George: Yeah, I think that's kind of where that goes. [00:58:39]Swyx: Maybe we can paint the future of like, the year is 2050. You've achieved all you wanted at TinyCorp. What is the AI enabled future like? [00:58:48]George: Well, TinyCorp's the second company. Comma was the first. Comma builds the hardware infrastructure. TinyCorp builds the software infrastructure. The third company is the first one that's going to build a real product. And that product is AI Girlfriend. No, like I'm dead serious, right? Like, this is the dream product. This is the absolute dream product. Girlfriend is just the like- [00:59:08]Swyx: Stand-in. [00:59:09]George: Well, no, it's not a stand-in. No, no, no, no. I actually mean it, right? So I've been wanting to merge with a machine ever since I was like, mad little. [00:59:15]Swyx: Like, you know, I was just like, [00:59:16]George: how do I merge with a machine, right? [00:59:18]Swyx: And like, you can look at like, [00:59:19]George: maybe the Elon style way of thinking about it is Neuralink, right? I'm like, I don't think we need any of this, right? You ever, some of your friends maybe, they get into relationships and you start thinking of, you know, them and their partner as the same person. You start thinking of them as like one person. I mean, they are kind of like merged, right? Like, humans can just kind of do this. It's so cool. It's this ability that we already have. Right, so I don't need to put, you know, electrodes in my brain to merge with a machine. I need an AI Girlfriend, right? So that's what I mean. Like, this is the third product. This is the third company. And yeah, in 2050, I mean like, ah, it's so hard. I just like, maybe I can imagine like 2035. I don't even know 2050, but like, yeah, 2035. Like, yeah, that'd be really great. [01:00:03]Swyx: In terms of merging, like, isn't it, shouldn't you work on Brain Upload rather than AI Girlfriend? Brain Upload, right? [01:00:09]George: I don't need Brain Upload either. Like, there's thousands of hours of me on YouTube, right? Yes. How much of my brain's already uploaded? [01:00:17]Swyx: That's only the stuff that you voice. Yeah, it's not that different. [01:00:20]George: It's not that different, right? You really think a model with, you know, an exaflop of compute couldn't extract everything that's really going on in my brain? I'm a pretty open person, right? Like, I'm not running a complex filter. Humans can't run that complex of a filter. Like, humans just can't. Like, this is actually a cool quirk of biology. It's like, well, humans like can't lie that well. [01:00:39]Alessio: So is it good or bad to put all of your stream of consciousness out there? [01:00:43]George: I mean, I think it's good. [01:00:45]Swyx: I mean, he's streaming every day. I want to live forever. We said off mic that we may be the first immortals, right? Yeah, this is how you live forever. [01:00:54]George: It's a question of, okay, how many weights do I have? Right, okay, let's say I have a trillion weights, right? So talking about a terabyte, 100 terabytes here. [01:01:02]Swyx: Okay, but it's not really 100 terabytes, right? [01:01:03]George: Because it's Kolmogorov complexity. How much redundancy is there in those weights? So, like, maximally compressed, how big is the weight file for my brain? Quantize it whatever you want. Quantization is a poor man's compression. I think we're only talking really here about, like, maybe a couple gigabytes, right? And then if you have, like, a couple gigabytes of true information of yourself up there, cool, man. Like, what does it mean for me to live forever? [01:01:27]Swyx: Like, that's me. No, I think that's good. [01:01:29]Alessio: And I think there's a bit of, like, a professionalization of social media, where, like, a lot of people only have what's, like, PC out there, you know? And I feel like you're going to get, going back to the ChatGPT thing, right? You're going to train a model on, like, everything that's public about a lot of people. [01:01:44]Swyx: And it's like- [01:01:45]George: Then no one's going to run their model and they're going to die. Don't put PC on social media. [01:01:49]Swyx: We're moving on to what would normally be called the lightning round, but just general tics, because you're a generally interesting person with many other interests. What does the goddess of everything else mean to you? [01:01:59]George: Oh, it means that AI is not really going to kill us. [01:02:01]Swyx: Really? [01:02:01]George: Of course. [01:02:02]Swyx: Tell us more. [01:02:03]George: Lex asked me this, like, is AI going to kill us all? And I was quick to say yes, but I don't actually really believe it. I think there's a decent chance that AI kills 95% of us. [01:02:11]Swyx: Okay. [01:02:12]Alessio: But they saw on your Twitch streams that you're with them, so they're not going to- [01:02:16]Swyx: No, I don't think, I actually, [01:02:18]George: I don't also think it's AI. Like, I think the AI alignment problem is so misstated. I think it's actually not a question of whether the computer is aligned with the company who owns the computer. It's a question of whether that company's aligned with you or that government's aligned with you. And the answer is no, and that's how you end up dead. [01:02:31]Swyx: So what the goddess of everything else means to me [01:02:32]George: is like, the complexity will continue. Paper clippers don't exist. [01:02:37]Swyx: You know, there are forces. [01:02:38]George: The paper clipper is cancer, right? The paper clipper is really just a perfect form of cancer. And the goddess of everything else says, yeah, but cancer doesn't win, you know? [01:02:48]Swyx: Yeah, it's a beautiful story for those who haven't heard it. And you read it out and I listened to it. Yeah, what are you grateful for today? [01:02:55]George: Oh man, I mean, it's all just like, I haven't, I haven't thinking about this stuff forever. Like, that it's actually like happening and it's happening in an accessible way too. I guess that's what I'm really grateful for. It's not like, AI is not some Manhattan project style. You don't know anything about it. Closed doors. [01:03:12]Swyx: Closed doors. [01:03:13]George: I'll fight really hard to keep it that way. I'm grateful for just how much is released out there and how much I can just learn and stay up to date. And I guess I'm grateful to the true fabric of reality that, you know, I didn't need differential equations to understand it. Like, I don't need some like, there's a limit to my math abilities. I can do most undergrad math, but I took some grad math classes and okay, now we're getting to the end of what I can do. And it's just the actual like, end of what I can do. Like, I'm limited by my brain, but you know, ML stuff, hey, you need high school math. [01:03:45]Swyx: You know what I mean? [01:03:46]George: When I learned to multiply a matrix, seventh grade, [01:03:48]Swyx: like, it's all easy. You need more electrical engineering than you need high school math early. [01:03:52]George: Yeah, well, you need electrical engineering to like, build the machines, but even that, like, these machines are simpler than the machines that have existed before. The compute stack looks really nice. So, you know, yeah, I just, I'm grateful that it's all happening and I get to understand it. [01:04:05]Alessio: John Carmack mentioned there's about six insights we have left. Do you have an intuition for what some of the paths [01:04:11]Swyx: people should be taking? [01:04:12]Alessio: Obviously you're working on one. What are some of the other branches of the tree that people should go under? [01:04:17]George: I don't think I'm working on one of the six insights. I don't think TinyGrid's any one of the six insights. Something I really like that Elon does, and I try to be inspired by it, is look at the boring tunnel machine and ask how you can build a 10X cheaper one. All right, look at the rocket. How can I build a 10X cheaper one? All right, look at the electric car and say, how can I build a 10X cheaper, like, cheaper or, you know, can go further or whatever, whatever, whatever, right? And you just do the straight up physics math, right? I'm trying to do the same thing with ML frameworks, right? And in doing so, making sure that this stuff remains accessible. You could imagine a world where if Google TPUs were actually the ultimate, if Google TPUs were actually the best training things, I mean, actually, you know, I'm kind of grateful for NVIDIA, right? Because if Google TPUs were the ultimate, now you have this huge closed source compiler in between XLA and the hardware, and yeah, that's just a really bad thing. So, I mean, something that is somewhat upsetting about the Tiny Core is that it is trying to prevent downside, but it's not all trying to prevent downside. Like, we're also building computers and we're gonna build some awesome, powerful, cheap computers along the way. So, no, I'm not really working directly on any of the six tricks. I also think the six tricks are kind of gonna be like luck. [01:05:25]Swyx: I think it's just gonna be like, you know, [01:05:26]George: please tell me more about what covariate shift is and how that inspired you to come up with batch normalization. Please tell me more about why it's a transformer and it has a query, a key, and a value, right? Like Schmidt-Huber described it better in fast weights. I mean, my theory about why transformers work have nothing to do with this attention mechanism and just the fact that it's semi-weight sharing, right? Because the weight matrix is being generated on the fly, you can compress the weight matrix, right? Like, this is what that, there's an operation in the transformer, which, and by the way, this is like, Qualcomm's SNPE can't run transformers for this reason. So, most matrix multipliers in neural networks are weight times values, right? Whereas when you get to the outer product in transformers, well, it's weight times weight. It's values times values, right? So, SNPE doesn't even support that operation, right? So, it's like that operation that gives the transformer its power. It has nothing to do with the fact that it's attention, [01:06:20]Swyx: right? [01:06:21]George: And this is a funny, like, but that is one of the six tricks, right? Batch, like these norms are a trick. Transformers are a trick. Okay, six more. [01:06:29]Swyx: So, you talk about attention as weight compression. [01:06:33]George: Compression is not exactly the right word. What I mean is that the weight can change dynamically based on the context. So, there was this thing in PAC-8 in the Hutter Prize that I absolutely loved, and I've never seen it again in neural networks, and it's a really good trick. Okay, imagine you have 256 weight sets for a layer, right? And then you choose which of the weight sets you're loading in based on some context. And that context can come from another neural net, right? So, I have another neural net, which projects 256 wide, one hot, do a softmax, predict it, and then I actually load the weights in. And I can do this operation at both test time and train time. I can do this operation at both training and inference, and I load in the weights given the context. Like, that is what transformers do. But transformers, instead of having 256 discrete ones, it's actually just that, but continuous. Which is funny that that was in language models, and I just like, when I understood that about transformers, I'm like, oh, this is a real trick, and why are they using the word attention? [01:07:23]Alessio: And today is actually the anniversary of attention is all you need. What? [01:07:27]Swyx: Oh, that's so cool. [01:07:28]Alessio: Today, six years ago. [01:07:29]Swyx: Six years. [01:07:30]George: Six years. [01:07:31]Swyx: Changed the world. Wow. [01:07:32]George: Well, there's one of your envelope tricks, right? And you could easily write it on an envelope, think about how you write out that. How many times have you written that? Because it's not in any libraries, because it's all used a little differently each time. Like, you just write out that exact same, you know. [01:07:45]Swyx: You've name checked Elon a few times. I think about both of you as systems thinkers. Input, output, thinking something in between. What's different about your style versus his? [01:07:53]George: Elon's fundamental science for the world is physics, mine is information theory. But you do a lot of physics as well. [01:07:58]Swyx: I mean, like, you base it on- [01:07:59]George: And Elon does a lot of information theory as well, too. But the difference maybe is expressed in what your ambitions are, right? Elon's ambitions may be like- [01:08:08]Swyx: Go to Mars. Go to Mars, right? [01:08:10]George: Go to Mars is the ultimate modernist physics ambition, right? It's a physics problem getting to Mars, right? [01:08:16]Swyx: Well, what are electric cars? [01:08:17]George: It's a physics problem, right? Okay, now he's like pushing on the autonomy stuff, and you push a little on information theory. But fundamentally, his dreams are physics-based dreams. My dreams are information-based dreams. I want to live forever in virtual reality with my AI girlfriend. Those are the aspirations of someone who accepts information theory as a core science. So I think that's the main difference between me and him. He has physics-based aspirations, and I have information-based aspirations. [01:08:39]Swyx: Mark Andreessen, he is a- Hi, Mark. He's a listener. He's a big proponent of effective accelerationism. You've been a bit more critical. Why do you say that IAC is not taken seriously by its adherents? [01:08:50]George: Oh, well, only the left takes ideology seriously. It's just like a fact, right? [01:08:55]Swyx: Is the right more cynical? Is that what it is? [01:08:57]George: I don't know. [01:08:58]Swyx: It's like the left actually manages [01:08:59]George: to get energy around the ideologies, right? [01:09:02]Swyx: Look, here you have- [01:09:03]George: You have two effective altruists named Sam going in front of Congress. Only one of them is in jail. [01:09:08]Swyx: You know, it's interesting. [01:09:09]George: They're both calling for regulation in their respective spaces, right? [01:09:11]Swyx: So SBF is definitely like kind of wolf in sheep's clothing, kind of, right? Like he only adopted IAC or EA to market. [01:09:19]George: Oh, and Sam Altman is a genuinely good guy who is not interested in power-seeking for himself. [01:09:24]Swyx: All right. Okay, okay. We don't have to go there. Fair enough, fair enough. [01:09:27]George: But no, IAC is not like, like you are not serious, right? Mark Andreessen, I like Mark Andreessen, but it's like someone who's like 2019, whose like eyes were opened about like the political world being not exact. You mean all the people on the news were lying to me? [01:09:42]Swyx: Bro, they were lying to you. [01:09:43]George: Like, okay, we all figured this out five years ago. Now, what are you going to do about it? I'm going to complain about it on Twitter. Great, and that's what IAC is. [01:09:50]Alessio: Last and maybe most important, why was Avatar 2 bad? [01:09:55]Swyx: Oh, I have a whole, you can go on my blog. [01:09:56]George: I rewrote the script of Avatar 2. I wrote a script that actually might make you feel something for the characters. I killed Jake Sully in the first scene. Like you had to. Do you really think his second story art topped his first one? No, of course not. You had to kill the guy and make the movie about the brothers, right? And just that alone and realizing that, like you could have kept the Titanic scene. [01:10:16]Swyx: It would have been fine. [01:10:16]George: I didn't even take it out. I left your Titanic scene, James Cameron, but I wrote you a story. So, you know, you're just, just, just. [01:10:23]Swyx: He needs ships to sink in water. [01:10:24]George: Look, it's a great scene, but like the movie was just like, like the Roman, I've never. [01:10:30]Swyx: Great CGI, you know, let down by the writing maybe. It's a beautiful world. [01:10:34]George: And that's why like I care so much, right? Like you don't hear me ranting about Pirates of the Caribbean 2 being a terrible story. Cause come on, what do you expect, man? Like Johnny Depp's like, wow, I had a movie that made me rich. I love this. [01:10:44]Alessio: But this goes back to like the midpoint. You know, I think you wrote like, feels like ChatGPT wrote the movie and that's my worry a little bit. It's like kind of converging towards that. [01:10:53]Swyx: Oh, I. Malik, Malik wrote the movie. Sorry, I didn't want to interrupt you. [01:10:59]George: I closed a pull request two days ago. I was like, was this written by ChatGPT? And I just closed it. [01:11:04]Swyx: Like, you know what? [01:11:05]George: I honestly feel bad if you were a human who wrote this. [01:11:07]Swyx: Incapable of being more perplexed. [01:11:09]George: But if you, if I have a classifier running in my head that asks, you know, is this a AI or is this a human? Like, you know, the only way to deal with all this, like, like, like, oh God, it's like the worst possible. Like, you know, people are like, how are you mad about like these chatbots? You're not mad about like Tesla. I don't want to buy a Tesla. I don't have to buy a Tesla. And it won't really impact my life negatively. But if I don't want to use a chatbot, it's still going to impact my life negatively. All the amount of like personalized spam that now makes me spend more cycles on my classifier to tell if it's spam or not, because you can now use AIs and generate this so cheaply. Like, no, I mean, we have to move to a model where everything's just a dollar, right? Like you want to send me an email, it's a dollar. Like you guys wouldn't care. None of my friends would care. No one would care, except the spammers, right? Like we just got to move to those sort of models. [01:11:54]Swyx: Awesome. [01:11:55]Alessio: One last message you want everyone to remember. [01:11:58]George: Go try TinyGrad. I hope that we're a serious competitor to what's out there. And then I want to take it all the way. We'll start with just building something for GPUs and then we'll start building chips and then we'll start building fabs and then we'll start building silicon mines and then we'll have the first self-reproducing robot using. [01:12:15]Swyx: Yeah, okay. All right, George. [01:12:18]Alessio: Thank you so much for coming on. [01:12:19]Swyx: You did a big inspiration. Thank you. Thanks. [01:12:21]Swyx: Thank you. [01:12:29] Get full access to Latent Space at www.latent.space/subscribe

Jun 14, 2023 • 1h 28min

Emergency Pod: OpenAI's new Functions API, 75% Price Drop, 4x Context Length (w/ Alex Volkov, Simon Willison, Riley Goodside, Joshua Lochner, Stefania Druga, Eric Elliott, Mayo Oshin et al)

Full Transcript and show notes: https://www.latent.space/p/function-agents?sd=pfTimestamps:[00:00:00] Intro[00:01:47] Recapping June 2023 Updates[00:06:24] Known Issues with Long Context[00:08:00] New Functions API[00:10:45] Riley Goodside[00:12:28] Simon Willison[00:14:30] Eric Elliott[00:16:05] Functions API and Agents[00:18:25] Functions API vs Google Vertex JSON[00:21:32] From English back to Code[00:26:14] Embedding Price Drop and Pinecone Perspective[00:30:39] Xenova and Huggingface Perspective[00:34:23] Function Selection[00:39:58] Designing Code Agents with Function API[00:42:16] Models as Routers[00:46:48] Prompt Engineering replaced by Finetuning[00:52:15] The 2 Code x LLM Paradigms[00:56:30] Smol Models for the future[00:58:54] The Evolution of the GPT API[01:03:27] Functions API Security vs Prompt Injection[01:16:18] GPT Model Upgrades[01:17:36] JSONformer[01:21:03] Closing Comments - What We Want Next Get full access to Latent Space at www.latent.space/subscribe

Jun 8, 2023 • 49min

From RLHF to RLHB: The Case for Learning from Human Behavior - with Jeffrey Wang and Joe Reeve of Amplitude

Welcome to the almost 3k latent space explorers that joined us last month! We’re holding our first SF listener meetup with Practical AI next Monday; join us if you want to meet past guests and put faces to voices! All events are in /community.Who among you regularly click the ubiquitous 👍 /👎 buttons in ChatGPT/Bard/etc?Anyone? I don’t see any hands up.OpenAI has told us how important reinforcement learning from human feedback (RLHF) is to creating the magic that is ChatGPT, but we know from our conversation with Databricks’ Mike Conover just how hard it is to get just 15,000 pieces of explicit, high quality human responses. We are shockingly reliant on good human feedback. Andrej Karpathy’s recent keynote at Microsoft Build on the State of GPT demonstrated just how much of the training process relies on contractors to supply the millions of items of human feedback needed to make a ChatGPT-quality LLM (highlighted by us in red):But the collection of good feedback is an incredibly messy problem. First of all, if you have contractors paid by the datapoint, they are incentivized to blast through as many as possible without much thought. So you hire more contractors and double, maybe triple, your costs. Ok, you say, lets recruit missionaries, not mercenaries. People should volunteer their data! Then you run into the same problem we and any consumer review platform run into - the vast majority of people send nothing at all, and those who do are disproportionately representing negative reactions. More subtle problems emerge when you try to capture subjective human responses - the reason that ChatGPT responses tend to be inhumanly verbose, is because humans have a well documented “longer = better” bias when classifying responses in a “laboratory setting”.The fix for this, of course, is to get out of the lab and learn from real human behavior, not artificially constructed human feedback. You don’t see a thumbs up/down button in GitHub Copilot nor Codeium nor Codium. Instead, they work an implicit accept/reject event into the product workflow, such that you cannot help but to give feedback while you use the product. This way you hear from all your users, in their natural environments doing valuable tasks they are familiar with. The prototypal example in this is Midjourney, who unobtrusively collect 1 of 9 types of feedback from every user as part of their workflow, in exchange for much faster first draft image generations:The best known public example of AI product telemetry is in the Copilot-Explorer writeup, which checks for the presence of generated code after 15-600 second intervals, which enables GitHub to claim that 40% of code is generated by Copilot.This is fantastic and “obviously” the future of productized AI. Every AI application should figure out how to learn from all their real users, not some contractors in a foreign country. Most prompt engineers and prompt engineering tooling also tend to focus on pre-production prototyping, but could also benefit from A/B testing their prompts in the real world.In short, AI may need Analytics more than Analytics needs AI.Amplitude’s Month of AIThis is why Amplitude is going hard on AI - and why we recently spent a weekend talking to Jeffrey Wang, cofounder and chief architect at Amplitude, and Joe Reeve, head of AI, recording a live episode at the AI + Product Hackathon where 150+ hackers gathered to compete for over $22.5k in prizes from Amplitude, New Relic, LanceDB, AWS, and more.To put things in perspective, Amplitude is a legendary YC alum with $238M of revenue in 2022 — our first guests representing the AI efforts of a public company!We chatted about how they have been approaching AI in their product (“question to chart” BI, text field autofill, instrumenting Amplitude with Amplitude), some of the issues they’ve had with different models, and the importance of first-party data in the world of LLMs. Another topic that came out of the Q&A was this idea of almost an “AmplitudeGPT”; rather than using language to simply generate a query, you could have these models investigate reasons for why certain behavior is happening in your user base. It was a really good discussion, and hope you all enjoy listening to it! Sections* [00:00:47] Amplitude's founding story and pivot* [00:03:28] Amplitude as an AI company and opportunities* [00:07:14] Limitations and challenges with using AI models* [00:10:56] Using Amplitude's product to build Amplitude - instrumenting AI* [00:12:32] Existing ML models in Amplitude's product and customer use cases* [00:15:50] “A/Z testing” and adaptable products* [00:19:33] The future of analytics and dashboards* [00:21:03] Optimizing for metrics in chatbots and AI products* [00:26:22] Using general models vs. fine-tuned models* [00:30:24] The importance of models vs. data - Amplitude's data set* [00:39:00] Lightning Round + Q&AShow Notes* Amplitude* Sonalight to Amplitude pivot announcement* The Slack origin story* Reverse Engineering Copilot* Simon Willison’s blogTranscriptEditor’s note: all timestamps are 1 minute behind because we hadn’t yet added the intro before making these. Sorry about that!Alessio: Thank you everyone for coming. Hopefully, some of you have listened to the podcast before, if you haven't, we focus on AI research and application. So we don't focus on “AI is going to kill us all”. We don't think about virtual girlfriends. We don't think about all of these more societal things. We're focused on models: how do you build them? How do you train them? How do you use them in production? What are some of the limitations on getting these things from demos to things that millions of users use? And obviously, a lot of you are building things. Otherwise, you wouldn't be here. And some of you have been building things for a long time, and now have a new paradigm that you want to build on top of. So I'm excited to dive in here. And maybe, I mean, I'm sure most people know you, but maybe you want to do intros and give a little background. [00:00:47]Jeffrey: Sure. Yeah, hey, everyone, met you all this morning, but I'm Jeffrey. I'm one of the co-founders and Chief Architect here at Amplitude. Been working on this product analytics thing, helping people understand user behavior data and make great product decisions and build better products for the last decade or so. And obviously, AI is a technology that we've been leveraging for a long time, but the recent trends are particularly exciting. And yeah, we have a lot of thoughts on how to apply that to our space, what we're doing in our product, and what we think the future of AI and product development and product data is. So excited to talk through some of those. [00:01:20]Joe: Yeah, I'm Joe, Joe Reeve. I've got a background in sort of startups and tech, been professional software engineer since I was 16, quit college. And at the moment, I'm running sort of AI R&D efforts here at Amplitude. Super excited about all the new stuff, but also all the stuff that Amplitude's been doing for a long time and how we're sort of getting renewed interest and excitement and abilities to push that even further forwards. [00:01:44]Swyx: So I think it's useful for people listening on the podcast and also some people here. Can you contextualize Amplitude as an AI company? Like what does that mean to you? What unique opportunities do you guys have? [00:02:02]Jeffrey: Sure, yeah, happy to speak to that. So, you know, if we think about the fundamental thing that our customers of Amplitude try to do, it's they want to look at their product data and they want to figure out how do I make my product better? And the really cool thing about product data is that one, it's often like very high fidelity, right? Digital products compared to, you know, let's say physical products before them have way more information about what's going on. And so that's why product data is, you know, even a thing at all, right? You finally have that feedback loop of, hey, I built this thing. This is how people are using it. Now let me learn from that and make my product better. Now, one of the downsides of that is that the data is massive. If you look at any of the internet scale products out there, they generate enormous amounts of data. And the ability of humans to kind of sift through that data is obviously limited. At Amplitude, we try to give people as many tools, whether AI or not, in order to process that. But at the end of the day, if you could get from the data and what user behavior is happening in your product to the insights of how to make your product better without as much manual work, that's kind of the holy grail of product analytics. And so in some sense, Amplitude has always been a company on the path to AI because figuring out how to make your product better from data is ultimately an AI problem. And so we're kind of just solving all the barriers in the way, like getting data in first, building good models for short-term things. And long-term, it's always been about, hey, how can you take product data and automatically make your product better as fast as possible? [00:03:28]Alessio: So that's the future of Amplitude. And a lot of people here probably want to start companies and whatnot. So maybe you want to give a 60 seconds of why you started Amplitude and what the story was like and maybe the first three to six months, what the challenges were. [00:03:42]Jeffrey: Yeah, of course. It's funny that we talk about this because the start of Amplitude is actually almost more AI than the current state. And so actually my two co-founders, Spencer and Curtis, they went through YC originally with not Amplitude, but SonaLite, which was a text-by-voice company. So it was kind of before the era of Siri and those types of technologies where they wanted to build something that would read text messages to them, that's easy, but also do voice recognition so that you could send text messages, say when you're driving, without having to pull out your phone. And so they worked on it and it was really popular back when they were doing it. After they finished YC, they realized the big innovation that they needed to figure out in order to make that successful was being really good at voice recognition, which was a different problem. They're awesome software engineers, but they don't come from an ML background. And so it's like, okay, are we going to spend the next five years solving voice recognition? Not really the thing that they had in mind when they were building product. But one thing that they happened to stumble upon as they were working on that was they spent a lot of time thinking about, hey, what was hard about that product? What made users churn? What made users really love it and engage? And they built a bunch of analytics tools to help them understand that. And they were really kind of shocked that those tools didn't exist out there in the market or they were like much more primitive than they wanted. And it turns out a bunch of other people in their YC batch felt the same. And they were like, hey, that analytics thing you're building, we want that. For you to text by voice, we want your analytics product. And so they're like, okay, fine. We will pivot, natural language and voice recognition isn't really our thing. And so we'll do distributed systems and analytics instead. That's where I came in. I'm a distributed systems and analytics guy. And so I happened to get in touch with them just through some mutual friends at the time. And then, yeah, we kind of went on it. The funny thing about a lot of things in technology is that the most forward thinking companies with respect to a lot of technologies are gaming companies. And so a lot of AmpliG's early start was either gaming companies or companies with founders that came from gaming backgrounds, where in gaming people have always been very, very rigorous about product data and optimizing engagement loops and all of that. And so they look for the best tools. We went to Zynga 15 years ago. It's like, that's where product analytics originated. And so a lot of those founders of new startups who had left Zynga were like, hey, that thing that you're building, that's trying to figure out patterns and user data and use that to make better products. That is exactly what we want after leaving Zynga. And then from there, that was Amplitude.Swyx: Yeah, I think famously other gaming companies would be like Slack, right? Mr. Butterfield tried to make a gaming company and failed and made Flickr. Then he tried to make another gaming company and failed and made Slack. And now look out to see what he does next. Discord as well. That's right. [00:06:34]Jeffrey: Yeah, people who come from gaming backgrounds are very rigorous in their product thinking. [00:06:39]Swyx: That's interesting. Alessio, you have a background in games? [00:06:43]Alessio: Yeah, in playing them, not in building them. So I will not fall into an enterprise company by doing that. Let's talk about R&D today and some of the ideas that you're working through, like some of the limitations that you run through. I think the most interesting thing about hackathons is you come with an idea and then you kind of hit a wall trying to build it. And then that takes you into another path. Like what are maybe funny things that you learn in terms of like the limitations of these models or like the missing infrastructure for using them? [00:07:14] Joe: So we've got a couple of different frames for thinking about this. There's AI that we're putting into our products and then us knowing that our customers want to put AI into their products. So there's the, how do we support our customers in their product development using AI? But how do we do that ourselves? And this is a great opportunity for us to learn the challenges our customers are gonna see. And so the first thing there is let's just start from the beginning, assume we want to add AI to our product, which maybe isn't the best place to start, but let's just assume we want to. How do we start ideating opportunities to put stuff into our product? So we sort of came up with this framework where we look at our product and we think about what are the collaboration touch points? So where are the points that a human might hand off to another human? And then think where can we replace one of those humans with the machine? So instead of thinking of some AI, amorphous AI, LLM, whatever, we're thinking actually, what if we had a robot that we were collaborating, not just a human, not just some sort of thing that spits out numbers. So collaborating. Then there's thinking of these as tools. So this is like your auto-suggest, on your mobile keyboard or spell check or something. How do you integrate this stuff as deeply into your product? So what are the friction points that users go through? Maybe they check lots of boxes. Is there a way we can pre-check those boxes we can get? So that's the feature embedding really deeply into the tool you've already got, the product you've already got. And then you step back and think, okay, what's a tool? So a tool is like ChatGPT, where you go there, it's an AI powered tool. It's not necessarily connected to your product, but it's a supplementary tool that you add. So there's a sort of ideation process there that we went through. And we sort of landed on a couple. And one of the key things that Amplitude does is help our customers, one, collect data in like a standard and sort of queryable way. And then we help them query it and get insights out of that data. So we were thinking, what's the feature there? How do we embed that? But also what's the collaboration point? And you might be a product manager asking an analyst, hey, please help me. Let's have a conversation about this. I don't know what questions to ask, but you also might just be about to go click the big create button and fill in a bunch of fields. And can we fill in a bunch of the fields for you? So we went to what to us seemed like one of the most obvious places. And we built a text box. Surprise, surprise with LLMs. We've got a text box. You can type in a question, type in anything about your data that you want to know, and then it'll spit back a chart, which is kind of neat. And we hit a bunch of problems there with LLMs hallucinating, losing context, even within the context windows, not really sort of recalling everything within the context window. So we sort of did a bunch of experimentation and realized if we split this down to seven different questions, so instead of saying, generate me a chart and a query for this one question, let's split that into lots of sub queries, like what kinds of events should I use? How should I display this? What should I call it? Rather than asking you all of that in one go. But then we had another problem where we have one query that a user makes that actually spins out seven different queries. So how do we monitor this? We can't just say one performance metric. You know, RLHF, you can't just say yes or no. Was the query response good? Because it might've failed for one of seven reasons. And maybe multiple of them failed or maybe some of them failed and then maybe they've hallucinated. And so we're getting code errors where an enum is not being matched. So we've had lots of sort of issues going all the way down there that we've had to figure out from first principles and sort of a really exciting way for us to understand what our customers are going through. [00:10:56] Swyx: So I wanna be clear. So you've described your exploration and how you think about products. What have you released so far? I just wanna get an idea of what has been shipped. [00:11:08]Joe: Sure. So in terms of LLM stuff, this, we call it question to chart internally. This ask a question, get a chart out. This, we've started rolling out to customers already. So last week, actually, started rolling out to our AI design partners a sign that we had signed up, which is a really exciting process. Actually, a lot of customers are just so excited to work with us and try it out and see how they can break it. So that's something we rolled out recently, which is built in LLM. It's the first piece built on LLM that we're working on. But we've also had a bunch of long-term ML, sort of traditional ML models that we've been running and products that we've been running with customers that help them predict what their users are gonna do. Because we've got this massive behavioral data set, best behavioral data set in the world. So we can train these awesome models and help our customers predict what their users are gonna do. So they can share the more relevant content or now is the right time to ask people if they want to upgrade or they want to rate your app or that sort of thing. [00:12:05]Swyx: Yeah, there is a little bit of a contrast, conflicts, because you already had all these ML models in-house and you're spinning up a new AI team and you're like, no, let's do all of this with GPT-3. Are the existing ML researchers saying like, no, this is a complete misuse of text generation? Or are they excited about it? Is it unlocking new things? [00:12:32]Joe: Yeah, actually, it's the combining these things. So we're able to use the traditional ML to shorten the fields, to narrow the number of things we need to pass into the LLMs. Because the LLMs can do a lot more of the reasoning, but we can make sure that the context we're providing is much more specific and generally much better by using the traditional ML models. [00:12:53]Swyx: Yeah, okay. And then the pain points that you're experiencing are hallucination. And then also like the multi-query thing. What do you think you wish for? Or what do you think you're thinking about to solve those pain points? [00:13:06]Joe: So right now we're instrumenting with our own product. So we're instrumenting groups of inferences and individual inferences, which means we can then create charts that show how often they fail, why they fail, how often we need to retry to get good answers. Swyx: So amplitude using amplitude. [00:13:23] Joe: Exactly. To build amplitude. [00:13:24]Swyx: Yeah, exactly. [00:13:25] Joe: Well, I mean, we're a product company. What else would we do? [00:13:29]Swyx: That is the second part of what you're saying, right? Which is, first of all, you want AI in the amplitude products. Second, people are shipping AI products with amplitude. You wanna talk a little bit more about what you're seeing there? [00:13:39]Joe: Yeah. I guess the key thing here is, for a lot of people is, okay, I can build the thing that calls OpenAI's API and then gives a response back. I'm nervous that I'm gonna be giving incorrect answers. I'm nervous that I don't really know how to measure whether the answers are incorrect. And I'm nervous that I'm not gonna be able to improve over time. So a lot of people we actually hear are nervous of giving thumbs up, thumbs down buttons because they're implying to their users that they're gonna be using this to improve the results. But they actually have no idea how to use that to improve the results in a meaningful way. And particularly when you've got multiple queries going off for one request, you've gotta then fine tune lots of different things in parallel. So it gets to be quite a technically complex sort of problem if you're not using great tooling that already exists for it. So that's, and then you have the extra layer of, I'm getting a bad result. I've tweaked my prompt template that I'm sending off to OpenAI. And now, has the result got better or worse? [00:14:35]Swyx: I don't know. [00:14:36]Joe: I don't know how to measure that. Except by thumbs up, thumbs down, which is a difficult measure in the first place. So that's where we can start saying, measuring the behavior of users once we've generated something for them. So have they gone and shared this content? Have they used this content? They actually gotten any value out of it? Not just have they pressed thumbs up. We can actually measure, are they getting value? Are they throwing it away from their behavior? But then using that through the Amplitude product, we can then tie that through to A-B tests, which is another product that Amplitude has. So then suddenly we start, and we're not doing this yet. This is sort of next on our list, is to start putting these prompts into our A-B test variants. So then we make a tweak in the UI, and it goes off, fires on the original, the control and our variant, our new variant. See, does it get fewer or more errors? Does it get fewer or more thumbs up, thumbs down? [00:15:30]Alessio: Have you thought about, I don't know, A-Z testing, I guess? Like one of the limitations has been, well, people can only write so much copywrite to test, but now with these generative models, you can actually generate a lot of copy. And like you go to on-demand test more and more and more copy. Have you seen any maybe fun customer stories? Like can you, anything there? [00:15:50]Jeffrey: Yeah, so actually there's a very good example of this. I don't know if I can share the actual customer, but actually from before the LLM days, where they literally generated the versions of the copy themselves, and they made their product basically adapt, you know, multi-arm bandit style of like, hey, here's all these different variations, like just go figure out the best one. At an internal hackathon, maybe two months ago, I built a prototype of what you're talking about, which is, okay, now replace the copy generation with an LLM. So just constantly generating new variations, and then multi-arm banditing to figure out which one's the best. I think that is probably the future of copywriting, where it's like, you don't actually need a whole lot of manual work anymore. It can, almost everything can happen automatically. And it's kind of the micro example in my head of this concept that we really like, which is self-improving products, where, you know, at some point, you know, someone has to say, hey, I'm gonna build a product that does this, you know, like a newsreader or something. But then, you know, after you have that, like the title of the newsreader, like the description of the sections, your navigation, all of that, in theory, you know, if you can give it some structure that the AI can play with, the LLM can manipulate all of that for you, and then use, you know, A-B testing, multi-arm bandits and all of that to kind of figure out what's best. And that generative AI kind of makes that last piece of like, what are my options possible? And that's super exciting for us. And we wanna be there, you know, to help you measure that, help you deploy that, and make that like the way people build products in the future. [00:17:14]Alessio: I think I've talked about this on the podcast, but this idea of like just-in-time UIs, you know, like each type of user wants to interact in a different way. And like, what you're building is a way of that, right? Like, Amplitude has been really like dashboard-driven, kind of like a diagram-driven, showing the user flow. Now each user can say, hey, I don't really want the table. I just want the charts. Or like, I don't want the charts. I just want the data. What do you think about the future of like dashboards and like BI in general? But like, the analysts used to come up with like what you should be seeing. Now each user can ask their own questions. [00:17:47]Jeffrey: Yeah, like the future of analytics, I think, is, you know, can go a few different paths. One thing that I want to, you know, counter against the whole LLM trend a little bit is I think when you get into really important and specific questions, you know, let's say you're writing like some complicated SQL or even code, you know, code and SQL are good because they're very specific, right? You can define your semantics very precisely. And that's something that I think, you know, when people start thinking about like natural language questions, they kind of take for granted. They're like, oh yeah, why doesn't it just, you know, figure out the precise semantics from my very ambiguous words? It's like, well, it's actually, in some senses it's possible, right? Because the precise semantics are not captured by your ambiguous natural language words. And so the way we think about it, at least today, you know, who knows what's going to change in the future is like natural language is a great interface to like get started. If you don't know what the underlying data looks like, if you don't know like what questions you should be asking, it is a very, very expressive way to start, get started. It's much easier than manipulating a bunch of things, much, much easier than writing SQL and all of that. But like once you kind of know what you want, it's very hard to like make it precise. It's actually easier to make SQL or code precise than it is natural language. And so that's a little bit of what we're thinking right now. So we think, you know, for sure the way that maybe many people will interface with analytics and data will turn into natural language because maybe the precision doesn't matter to them. But like at the end of the day, when you're trying to get, you're trying to sum up your revenue or something, it's like, you want to know that it's right. And you want to know the semantics that go into that. And like, that's why, you know, that's part of why data is hard. The semantics really do matter. They can make a huge difference in the output. And so there's a boundary there that I'm curious where it will push over time, but I don't think it's quite there yet. [00:19:33]Joe: I think this is where models sort of can become more embedded as features rather than go off and do this thing, create this analysis for me and then come back, the collaborator model. Then we're saying this field, I'm not sure what should go in there. Can you make a suggestion? And then I'm going to go and refine it over time. So it's the sort of autofill, but guessing autofill, but then you still, you can tweak everything. This is one of the core design sort of principles that we've come up is yes, you've got to be able to explain what the model's doing. And as a human, I need to understand, a user I need to understand what is the model doing and why is it doing it? But I also need to be able to tweak it once it's done it. I don't want to feel like I've just said go and then I can't stop it and it's going to go off and do stuff. And that's sometimes how things like AutoGPT can feel. It's going and it's costing me OpenAI tokens and I have no idea what's going on. So yeah, I think a key thing is servicing all the individual things the model's doing and allowing users to tweak it, stop it, retry while it's going. [00:20:33]Swyx: For me, one of the most challenging questions is something I think you guys have maybe thought about a lot which is chat. Ideally you want, like you could say naively, for example, you want to optimize time in app, but actually that's a sign of failure if the chat session is longer than it should be. Do you have any advice on, I'm sure you've dealt with this before pre AI era, but like what do you advise AI hackers to optimize for? Like what analytics should people be looking at? [00:21:03]Jeffrey: Yeah, our general kind of philosophy as a company is to work with customers to identify north star metrics. Right, and like time in app is not good primarily because it doesn't actually correlate with your business outcomes most of the time. And to be fair, sometimes it does. Like if you're a social media app, maybe it does correlate really well and maybe it's not a bad metric then. But for a lot of other products, right, if you're trying to do the search, for example, or like time on search, like nobody wants that. It's like, yeah, what is your success rate? You know, how many, do you get them to come back and search in the future? Like that's much more interesting than the time of your session. And so, because you know, each time you can serve apps, right, that's your business. And so it's like, if you choose a metric that's well correlated with your business outcomes, then that's at least the first step to getting that right and not getting caught up in other vanity metrics that sound like they could be good to increase, but then, you know, they can sometimes lead to negative business outcomes, you know, and then you get the worst. You've optimized the wrong metric the whole time. And that's where tying in AI and product analytics makes a lot of sense. And it's really important because product analytics, these companies that are like our customers that are trying out building features that are LMs and they're not sure what to optimize for, optimize for the same thing you're already optimizing for. You're already measuring conversions. You're measuring how much value, hopefully, your customers are getting out of your product. So continue doing that and maybe find a way to tie the LLM feature to that and sort of through A-B tests and that sort of thing. And then on the chat specifically, chat is obviously for a business maybe rolling out a chat box based on LLMs. It can be really scary. And that's another sort of mental model of framing we've been thinking around is we find LLMs right now are most useful either when you come from, either when you have a narrow input space and a broad output space, because you can be very, you know exactly what format of data, what kind of data is gonna be passed in. That's probably not coming directly from a user. It's probably coming from a button click or a toggle switch or something. And then you can have a general output and you can provide templates and that sort of thing. And then the other way is broad input space, narrow output space. So that's free form text box. And you can provide a bunch of sort of clamping, framing, validation on the output to make sure that you're not spewing out, you know, poems about Hitler or whatever it is. You know, you can be really, really deliberate when you've got a small output space. Chat is large input space, large output space, which is really, really scary. If you're, as a company, you're not selling a chat product, you're selling a, you know, an analytics product with maybe a chat support bot or something. [00:23:37]Swyx: Yeah, I think this is one of those opportunities. I always try to raise the awareness of this, that Copilot I think did a really interesting metric or North Star, which was how much code is kept or retained by the user. And for people who are Googling along, you can actually look for this blog post about reverse engineering Copilot internals. And they actually set up custom metrics around, you know, 30 seconds after a code snippet is accepted, one minute, two minute, three minute, all the way to five minutes. And you can sort of see it construct a curve of how long Copilot suggestions stick around. And from there, they can actually make statements like this, you know, evaluate the success of the products. It's pretty cool. [00:24:18]Joe: One of the really nice things we found actually, we accidentally did this. So our chart building interface, heavily instrumented. It's a, we're Amplitude. So we instrument our product. We also, it's one of the main tools that our customers use. So it's really, really well instrumented. And so when we tied chart creation through asking a question through an LLM, and then we tied that to a chart, an output chart, we then automatically were able to tie every time someone edits any of the parameters to that generation. So then we know, we have really detailed RLHF data for, yeah, you got everything apart from the metric, right? But you got everything apart from this event that shouldn't have been there, because that's the one that got removed. So similar to the Copilot there. [00:25:00]Alessio: And I want to make sure we open it up for questions, but like one last thing is about, everybody knows that small is beautiful. And when you think about what models to use and some of the parameters, like there's costs, there's latency, there's like accuracy. How do you think about using, you know, GPT-4 and some of those models versus using smaller ones that are fine-tuned? What are the trade-offs? [00:25:23]Joe: Yeah, I guess right now we're very much in the, let's explore, let's try everything and just iterate as fast as possible, which is what general models are great for. We do have some smaller, not even fine-tuned, some smaller models that we've sort of borrowed from Hugging Face that we run internally for more specific tasks. And that's often sort of selecting specific values before we pass it to a general model right now, just because the general models are much easier to communicate with and they understand most of the words we use. It's not like we use a word and suddenly we get random outputs for no reason, the sort of gold magic up type thing. So they're generally less susceptible to that. So that's why we're iterating heavily on the general models. I think we absolutely have to move to some more specific models, particularly given inference on fine-tuned open AI models gets more expensive and slower the more you do it. So yeah, that's definitely a thing we're looking at and we're doing some internal stuff, but it's the next step or one of the next steps. [00:26:22]Jeffrey: Yeah, to give a pseudo example of that, one of the hard things to help users within Amplitude is picking the right event to analyze. It's kind of your fundamental unit of analysis. And when a user comes in and let's say that's the first time they're using Amplitudes, someone else in their company has set up the product, so they don't know what the events are. Right now in Amplitude you get this massive dropdown and it's like, all right, there's a thousand things, like which one is the one I'm looking for. And sometimes the names are good and sometimes they're not. But one thing we did was, okay, yeah, feed that into open AI. Hey, tell me which event type best matches like this user's intent. That's like pretty good at that, right? So it's all language stuff, but it's a little bit slow and it's a little bit expensive to do that every time. And so we kind of fell back to, once we validated that that works, kind of fell back to a more traditional embedding-based approach. It's like, all right, compute all those embeddings. That's more work upfront because you have to go through your database of all of these things and you got to commit like that engineering work, but it's like you validate with the general model because it's just easy. It takes like an hour to figure out that it works. And then it's like, all right, can we do the same thing with embeddings? That's way faster, way cheaper and still has reasonable quality. Embeddings also have a nice quality that you can get like magnitude of things, whereas LLMs aren't great at giving you like, hey, it matches this much. It's kind of, you can ask it for an order and that's decent, but like, yeah, anything beyond that is pretty challenging. [00:27:42]Alessio: How do you think about the importance of the model versus the data, right? There's like a lot of companies that have a lot of data, but not a lot of AI expertise or companies that are just using off the shelf model. How should companies think about how much data to collect? What data is meaningful? What isn't, any thoughts there? [00:27:59]Jeffrey: Yeah, I think it's safe to say that both are really important, right? Like the evolution of LLMs really was a lot of model innovation. And so I don't want to downplay that. At the same time, I think the future of AI applications and doing really cool things with it will be in the data, partially because like, you know, ChatGPT has done such a huge advance, right? The LLMs model space has advanced like crazy in the last year. And so I think a lot of the untapped potential will be in data in the future. One thing that's particularly interesting to us is like we have a pretty unique data set, actually. It's a lot of first party behavior data, right? So if you're, you know, if you're Square, for example, you instrumented like the way that people interact with Square Cash and the wallet and the, you know, the checkout system. And like, those are very specific things. Like Square can't look elsewhere in the world for that stuff. And that's really interesting because, you know, to build models of user behavior, you need user behavior data. And it turns out there's not actually a lot of examples of user behavior data out there in the world. And so to Joy's point earlier about, you know, we have one of the best user behavior data sets in the world. And so if we want to build a model around that, I think it would be a super interesting one. So if you take an analogy to what ChatGPT does, it basically takes a bunch of language examples and it, you know, learns a bunch of abstract concepts, like how to, you know, prove math things or how to render in JavaScript. It's like, wow, that's very astonishing. They kind of prove, it's almost like a proof of concept to the world that if you train a sufficiently good, you know, transformer self-attention type model with a sufficiently large data set of, you know, hundreds of gigabytes of internet text, you'll learn really interesting abstract concepts. And so we want to apply that to our data set, right? Cat GPG is great because it's a proof of concept. If it didn't exist, you know, I would have told you, yeah, you can spend $10 million training this model on a data set, you'd probably not get anything interesting because we just have no idea. But because it exists, it kind of proves to the world that if you do this correctly, there is a ton of interesting value. And so that's what I think. And so, you know, amplitude is just one example of a very interesting data set that you will train something that's, you know, fundamentally very different from GPT or any LLM out there. And there's lots of other data sets out there. And I think that's where a lot of the interesting things will come once this kind of, this phase of like rapid model evolution kind of tapers out a little bit. And you'll see a lot of the more interesting applications there. [00:30:24]Swyx: So I've never thought about this much, but you guys must do it a lot. Like what is the ethics or best practices around training on user data when they don't know they're being watched? Like, I mean, presumably they're fine with tracking and events, but like, do we tell them that we're going to train on their data? Is it okay? [00:30:50]Joe: I guess there are a couple of things. One is PII. Doesn't go anywhere near the stuff, right? PII with strip and like, that's just a really important thing. [00:30:58]Swyx: You still need an identifier for streams. [00:31:02]Joe: Yeah, yeah. But in terms of training models, we don't want any of that to go in there because then you might accidentally, you know, like, hello, ChatGPT, please hallucinate me a social security number. That's dangerous. [00:31:11]Swyx: Also PII makes it into prompts a lot. [00:31:14]Joe: Sure, that's true. So then you have to strip that from your... So we have some experiments where we're stripping PII that is in places that shouldn't be, you know, descriptions of things. Sometimes people copy paste big long lists of email addresses into charts and things. But some of these things are actually pretty surprisingly easy to detect and strip out. So we can do that. And we have some layers that are stripping out that sort of replacing them with tokens. So the LLMs can still operate on them. But in terms of training this data, all that training is happening internally and we're not putting any sort of private data, personally identifiable information in. I don't know if there's anything you wanted to add there. Yeah, yeah. [00:31:54]Jeffrey: We certainly think about this a lot and our customers think about a lot. Like when I think about user privacy with respect to tracking, there's kind of this big spectrum. Around the one end, it's like literally track nothing and, you know, the end of story. And like for people like that, I mean, that's cool. You know, they're not gonna use Amplitude. They may not like us very much. You know, that is what it is. And then on the other end of the spectrum is like, we're gonna track you across the entire internet and sell your data to everyone. And like, that's obviously bad. And like, there's lots of good reasons to think that's bad. First party behavioral data, I think is actually probably almost as far. Fully anonymized first party behavior data would be like kind of the minimum. It's like web server logs with no IP, no identifier, nothing. The problem is that you can't do a lot of interesting behavioral analysis without that. You can't tell if, you know, this person that came on this day was the same one that purchased later. And so like, you can't actually, it's much harder to make your product better if you don't have that. And so, you know, we're kind of set at this place where we have, you know, like pseudo anonymized first party data. And like, we don't sell the data. You don't mix data from, you know, different places on the internet through Facebook cookies or things like that. And, you know, our philosophy is like, that is actually the most important data to build a better product. It's not the most important data to advertise, which is why Facebook and Google do what they do, but it's the most important data to build a better products. And it kind of strikes the right balance between yeah, totally tracking everything that you're doing and like not having any information to make your product better. [00:33:19]Swyx: Yeah, cool. And I think we're going to go to audience questions. So let's start warming them up soon. But I think we have some lightning round questions [00:33:29]Joe: The audience is thinking of questions while we go. [00:33:31]Alessio: The first one is, what's something that already happened in AI that you thought would take much longer to be here? [00:33:39] Jeffrey: I don't know what the constraints on our lightning round, but I think maybe creativity is the best word where it's, you know, with the image generation stuff, text generation, you know, one thing that still blows my mind, I used to be a competitive like math guy and like there's this international math Olympiad problem in one of the papers and it solves it. And I'm just like, wow, I can solve this when I was spending all my life doing this thing. Like that level of creativity really blew my mind. And what's the takeaway? It's like maybe the takeaway is that creativity is not as, you know, as not as high entropy or high dimensional as we think it is, which is kind of interesting takeaway. But yeah, that one definitely surprised me. [00:34:21]Joe: I guess there's something actually that maybe answering the inverse question that a lot of my friends were surprised happened quickly. And I was like, this is just braindead obvious. I've got a lot of friends in the AI safety space. So they're worried that in particular, X-risk, right, extinction risk, that AI is going to kill the human race. And they were like, oh no, what if an AI escapes containment and gets access to the internet? And then we get an LLM and the first thing we do is like, hey, also GPT, here's the internet. [00:34:48]Swyx: So you thought, it's happening faster than you thought. [00:34:53]Joe: Well, it's happening faster than, to me it makes sense, because I'm like one of the guys connecting it to the internet. And I'm like, I'm surprised that other people were surprised it was going to be so fast. [00:35:01]Swyx: Yeah, so a bit of context, Joe and I, we've been adjacent to the EA community and they have like smoothly migrated to the X-risk community very quickly after SBF. [00:35:13]Joe: Yeah, after SBF, yeah, that was fun. [00:35:16]Swyx: Okay, so next question, exploration. What do you think is the most interesting unsolved question in AI? What's next? [00:35:30]Joe: I guess like, is it going to keep getting better at the same rate? Is it going to, and that's just a super important question that's going to change. Like, depending on that answer, 50 startups are going to pivot or not pivot, right? [00:35:43]Swyx: Which is what's next, literally. [00:35:45]Joe: Literally, what's next? Like in a year's time, are the models similarly better than they have been so far? Or are we about to taper off or are we about to continue going linearly? [00:35:58]Jeffrey: Yeah, I'll throw one out that is not necessarily about AI, but like, what's intelligence, right? And if you ask people 20, 30 years ago, maybe even longer now, it's like, yeah, chess. Chess is intelligence. And then chess got solved and like, ah, that's just brute force. And it's like, well, you know, creating creative images and writing, that's intelligence. Well, it's like, that's solved too. Maybe it's just, you know, if you have enough parameters, you can capture that. So like, what is intelligence? What does it mean to have an AGI? What does that actually mean? And then what the implications that are on for our understanding of humans and our brains. I've always thought that, you know, everyone is just a stochastic machine. And so, you know, is everything consistent in my mind? Swyx: Free will and illusion. Exactly. [00:36:43]Joe: I guess maybe like the scaling piece is like that intelligence as you scale is gets more and more expensive on the traditional stuff. But then there's something I think I saw yesterday on Hacker News. It was people actually getting a brain to play tic-tac-toe. Like by a brain, I mean, stem cells grown into brain tissue. And they were able to train it. And like that to me is very significant because suddenly the like metal computers limitations is not applied. And then now we've got all this intelligence. What is intelligence stuff on a squishy wet computer? That makes it even harder to ask and even harder to draw lines. [00:37:18]Swyx: Yeah. Yeah. So famously, you know, language models are so much more inefficient than wet computers, as you say. And so if you can merge that, you know, the human brain runs on 30 Watts of power as it is my favorite fact. We're not anywhere close to that yet. [00:37:36]Alessio: Before we get into Q&A, one last takeaway that you want everybody to think about. [00:37:41]Jeffrey: Yeah, I'll do the one that we actually repeat in Inside Amplitude very often, not about AI, but I think it applies, which is it's early. It's sometimes hard to realize that when things are happening so fast, especially in the Bay Area, but like the ramifications of AI or in our case, product data and all that are gonna play out over the next many decades. And that's just, you know, we're very fortunate to be at the beginning of it. And so yeah, take advantage of it and keep reminding yourself that it's early. [00:38:15]Joe: I guess mine would be, let humans be good at doing human things. Let machines be good at doing machine things and let machines be good at doing machine things and help humans be good at doing human things. And like, if you don't do that, then you're gonna be building something that's either not useful or it's very scary. So yeah, get machines helping humans, not the other way around. [00:38:39]Swyx: Get machines helping humans. All right. With that, I think we're all gonna open up to questions. We're gonna toss you the mic. [00:38:45]Audience #1: Yeah, hey, thanks for the insight into how you guys implemented your AI, you know, question asking chatbot and how have you converted into seven sub queries and then generate the data out. I've just, I got a peak my interest about how you guys exactly do it. Like Alessio asked, like, what exactly is the model that you guys are using? Are you converting it into your, what are these queries that you generate from a single English language? Is it possible to go a little deeper just from a curiosity perspective? [00:46:34]Joe: So we have a custom query engine. So it's not SQL or anything that we're generating. We're generating a custom query output. So I guess the types of questions range. So things like chart type, are we doing a segmentation chart, a line chart or are we doing a funnel chart? You know, the number goes down over time or up over time or between a conversion between two events and there are various other types or metrics or, and then there's also the name. What should we name this chart that answers this question? So the way that's implemented in practice, you could use something like Lang chain to sort of chain these things together. But in our experience, I think Lang chain's a great tool for certain things and definitely really great for prototyping, but we found it quite restrictive. So we've ended up building sort of an internal, it's a very, very small wrapper, internal, we use TypeScript as well, framework that allows us to basically just write in code and infer within what we call a transaction, an inference transaction, which gets monitored as one, but then also all the individual inferences within it get monitored. So it's a bit like when you're writing a database transaction with most sort of, at least in the node ecosystem, the JavaScript ecosystem, where you sort of get a transaction object that you can operate on, and then you return your, or you return, you sort of commit your transaction. So we've got an interface like that, so we can just write pure TypeScript, await this response or await these responses. And then we've got a switch case. If it's a segmentation chart, go and do these with these queries. And then each of those inferences can be a different model. So we think in the future, maybe we have one query where we have some GPT-4 responses. We want some text responses. Maybe we also want to generate an image from that same query together, and then that gets bundled. So I don't know if that answers your question. Audience #1: Yeah, I think so. Yeah, thank you. I think so. You said in future, you're going to use GPT-4. What are you using right now for? [00:48:33]Joe: Right now, everything's GPT-3.5. We're moving around, and I think probably for some of the prompts, we'll use something like DaVinci. Some we might use GPT-4. Some we'll be using internal ones. And we also want to be able to degrade gracefully if a customer has told us they don't want us to send anything to OpenAI, then we can degrade to some internal models that maybe are some of the open source models that have been trained on smaller datasets. [00:48:57] Audience #1: Gotcha, makes sense. Thank you. [00:48:58]Jeffrey: Yeah, I think to add to that a little bit, the key is breaking down the problem sufficiently, because if you break down the problem enough, you can also provide it with some examples, which is super helpful, right? You know, GPT is quite good at zero shot, but within the context of our specific domain, it doesn't know what's going on. And so being able to break down the problem to, hey, select the type of chart. Don't generate me an entire chart definition. Select me the type of chart, and then select me the specific metric based on their query, and then giving it some examples. Select me the events and properties that I want to look at. By breaking it down and having very, very contextual prompts with respect to those examples, you get a lot higher quality output than trying to generate, like, you know, if you imagine generate, like, hey, generate me a whole SQL query with all, you know, here's like the schema of all my tables, now generate it entirely. It's like, it actually struggles with stuff like that, because it's just like kind of too much information and computation to come out of language. Now, maybe GPT-5 will be different, but like, that's the state of the art today. [00:49:57]Swyx: I'll ask a follow-up to Joe. So you mentioned, you mentioned trying LangChain, but not needing it for production. Any other comments on tooling that are out there that's interesting to you? Do you use a embedding database, for example, or do you just use a regular database? [00:50:18]Joe: Yeah, so we've actually been running embedding sort of similarity or vector search in production for multiple months, maybe even almost a year, and just like straight up Postgres, but now we're using PG Vector, which actually Jeffrey could probably speak more to about that decision and what that was like. [00:50:40]Swyx: So this is a pretty hot take. At Amplitude scale, all you need is Postgres? [00:50:46]Joe: We'd use many things other than Postgres. But I mean, we, this isn't rolled out for all customers and it's not necessarily getting sort of hit with a lot of traffic. And so the scale here is very different. Our usage scale is very different to our ingestion. [00:51:04]Swyx: Yeah, yeah, yeah. [00:51:06]Jeffrey: Just to clarify that a little bit more, we're not putting individual end user vectors or end event vectors. We're putting in taxonomies. So if I'm DoorDash, my taxonomy is add to cart, checkout, purchase, browse. That's the cardinality. And so that's actually small. It's on the order of tens of millions. And so yeah, you use stuff that in Postgres, no problem. Now, when we talk about large behavioral models or like actually embedding events, there are many, many trillions of those. And yeah, Postgres probably doesn't work there. [00:51:41]Swyx: Yeah, actually I wanted to comment on this slightly before, which is separating taxonomies from the actual data is one way you protect your customers against prompt injection. It's something that Simon Willison has been talking about where you want to have like query for one thing, but essentially no knowledge of the actual underlying data, just the taxonomy. So it's good practice. [00:52:00]Audience #2: Yeah, so you talked about a model which would be trained on user behavior data like amplitude GPT. It really piqued my interest and what capabilities would emerge? What do you think that you would find and what would be the first thing you would ask the model? That's a good question. [00:52:23]Jeffrey: So we've thought about this a little bit and I think the, right, these are sequence, token prediction models. And so at the very least, I would hope for a much better, we have a predictions feature right now, which says, hey, given what a user has done over the last 90 days, do we think they're gonna belong to this cohort in the future or not? So that cohort might be people who churn, people who purchase, people who upsell, whatever the customer wants. We think it would be much better at tasks like that, right, because if it just has a very good understanding of behavioral patterns and what's gonna come next, it would be able to do that. That's exciting, but not that exciting. If I'm trying to think about like the analogies to what we see in LLMs, it's like, okay, yeah, what is the behavioral equivalent of like learning physics concepts, right? It's like, oh, I don't actually know, but it might be this understanding of patterns of sessions and how that like, for example, categorizing users in a unsupervised way seems like a very simple output for a model that understands user behavior, right? Here's all the users and if you wanna discriminate them by their ability to achieve some outcome in the future, like here's the best way to separate that group and here's why, right? Be able to explain at that level and that would be super powerful for customers, right? A lot of times what our customers do is, hey, these people came back the next day and these people didn't, why? What was different about them? And so we have a bunch of heuristics to do that, but at the end, there's something like, causal impact is like one of the holy grails of product analytics. It's like, what was the causation behind some observed difference in behavior? And I think, yeah, a large behavioral model will be much better at assessing that and be able to give you potentially interpretable ways of answering that question that are like really hard to do, really hard, really computationally intensive, really like noisy, distilling causation correlation is obviously super hard. Those are some of the examples. The other one that I am, I don't know if I'm optimistic about it, but we really interesting is, one of the things that amplitude requires today is manual instrumentation, right? You have to decide, hey, this clicking of a button, this viewing of page, these are important things. I'm naming them in this way. There's a lot of popular tools out there that kind of just record user sessions or like track DOM events automatically. There's a lot of problems with those tools because the data is incredibly noisy. It's just so noisy, right? A lot of times you just can't actually interpret it. And so it's like, oh, it's great because I don't need to do any work. But like, well, you also don't get anything out of it. It's possible that a behavioral model would be able to actually understand what's going on there by understanding your user behavior in a correctly modeled and correctly labeled sense, and then figuring out. I don't know if that's possible. I think that would make everyone's lives a lot easier if you could somehow ask behavioral questions of data without having to instrument. All of our customers would love that, but also all of them are instrumenting because they know that's definitely not possible today. [00:55:26]Audience #2: This is really interesting. You're looking forward to the future. If you're gonna build it, it's gonna be amazing, yeah. [00:55:31]Jeffrey: That's the goal, that's the goal. [00:55:33]Audience #2: Awesome. [00:55:34]Swyx: Thanks for listening. [00:56:09] Get full access to Latent Space at www.latent.space/subscribe

Get the Snipd
podcast app

Unlock the knowledge in podcasts with the podcast player of the future.

AI-powered
podcast player

Listen to all your favourite podcasts with AI-powered features

Discover
highlights

Listen to the best highlights from the podcasts you love and dive into the full episode

Save any
moment

Hear something you like? Tap your headphones to save it with AI-generated key takeaways

Share
& Export

Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more

AI-powered
podcast player

Listen to all your favourite podcasts with AI-powered features

Discover
highlights

Listen to the best highlights from the podcasts you love and dive into the full episode

Latent Space: The AI Engineer Podcast — Practitioners talking LLMs, CodeGen, Agents, Multimodality, AI UX, GPU Infra and all things Software 3.0

Latest episodes

[AI Breakdown] Summer AI Technical Roundup: a Latent Space x AI Breakdown crossover pod!

FlashAttention 2: making Transformers 800% faster w/o approximation - with Tri Dao of Together AI

Llama 2: The New Open LLM SOTA (ft. Nathan Lambert, Matt Bornstein, Anton Troynikov, Russell Kaplan, Whole Mars Catalog et al.)

AI Fundamentals: Datasets 101

Code Interpreter == GPT 4.5 (w/ Simon Willison, Alex Volkov, Aravind Srinivas, Alex Graveley, et al.)

[Practical AI] AI Trends: a Latent Space x Practical AI crossover pod!

[Cognitive Revolution] The Tiny Model Revolution with Ronen Eldan and Yuanzhi Li of Microsoft Research

Commoditizing the Petaflop — with George Hotz of the tiny corp

Emergency Pod: OpenAI's new Functions API, 75% Price Drop, 4x Context Length (w/ Alex Volkov, Simon Willison, Riley Goodside, Joshua Lochner, Stefania Druga, Eric Elliott, Mayo Oshin et al)

From RLHF to RLHB: The Case for Learning from Human Behavior - with Jeffrey Wang and Joe Reeve of Amplitude

Get the Snipdpodcast app

AI-poweredpodcast player

Discoverhighlights

Save anymoment

Share& Export

AI-poweredpodcast player

Discoverhighlights

Get the Snipd
podcast app

AI-powered
podcast player

Discover
highlights

Save any
moment

Share
& Export

AI-powered
podcast player

Discover
highlights