AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
26min chapter
Lovable's Anton Osika on Revolutionizing Software Development and OpenAI Drops Deep Research | E2080
This Week in Startups
Navigating Government Spending and AI Innovation
This chapter explores the inefficiencies in government spending, particularly within the Department of Defense, and the rise of venture capital in military technology. It highlights the innovations brought by AI tools like OpenAI's 'Deep Research' and their potential impact on various industries, including investment analysis and software development. Furthermore, the chapter examines the evolving landscape of AI technology, its implications for job roles, and the dynamics of venture capital amidst changing consumer needs and market predictions.
00:00
Transcript
Episode notes
Speaker 1
time where we get to print imaginary money and people still
Speaker 2
want to buy it and that may not last forever so well not unless stable coins become the next big thing and they put all their money into t-bills i mean that helps a little bit but it's around the edges i'm curious to see when we get to defense spending and uh waste fraud and abuse there because the pentagon as people famously point out hasn't passed an audit in some time and i'm curious what okay here's a fun game jason what fraction of dod spending do you think is wasted i'm gonna go with double digits uh 25 probably i mean that's i mean if
Speaker 1
you don't have to pass an audit if you don't have to be efficient and everybody just says hey you know it's defense it's defense we can't touch it you know then you're never going to have efficiency or competition and so you know the competition's here venture capitalists are investing in um military technology now and cost plus is going away so we have to attack everything and so that's, I think, one of the things we're seeing now with Doge and USAID and all these controversial things that are occurring and causing a bit of chaos and making people wring their hands. My best advice is give it 100 days. Better it's happening, even if you don't like it. Better it's getting under control and give it 100 days and then see what happens that's my best advice we're five
Speaker 2
days into it maybe 10 give it 100 days yeah people don't people forget but two weeks ago i had hair i had a full head of hair and now it's all falling no that's not true all right so a couple weeks ago we had the guy behind deep research from Google on the show. Great interview. Great time. You were a big fan of the product. I'm a big fan of the product. Then over the weekend, OpenAI, everyone's favorite closed AI shop, dropped their own product. And Jason, did you know that they also called it deep research? It's
Speaker 1
fascinating that they, I'm shocked that uh, in any way, uh, we would have open AI do something like steal somebody else's IP. It's kind
Speaker 2
of laughable, right? They just called it the same exact thing. Okay. Sure. All right. Sure. Uh, anyways, I I've pulled this segment from the announcement video. Um, and I think it's a pertinent example to something that you really care about, but I also have a quick demo to show us. So here is what OpenAI announced recently in case you have been living under the proverbial rock. We are introducing a capability called Deep Research. What is Deep Research? Deep Research is a model that does multi-step research on the internet.
Speaker 1
And what it does is it discovers content, it synthesizes content, and it reasons about this content, adapting its plan as it uncovers more and more information.
Speaker 3
So this is a finance problem. So I'm an investment analyst in the Silicon Valley VC firm. I want to analyze the market for civilian supersonic air travel and prepare a thorough investment memo and then many other specifications. And so the model clarifies and we provided some additional requirements for the memo. And then the model kicked off the task. And as you can see, it went and researched for seven minutes, used 12 different sources, and then came back to us with quite a comprehensive report of the field. And you can imagine if you were doing this for your job, this would be quite helpful to bootstrap your research as you're doing your initial investigation.
Speaker 1
Yeah. How much did they pay the original researchers for the report they built on top of their research? I guess if they're giving citations, they're going to claim it's fair use. These deep research projects are really impressive because they are, as they read the internet and scrape the internet, they figure out what questions to ask next. So if you were, and we did this, remember we did it with the model when we were doing our model of what would it take for self-driving cars? Yes. Oh, you know, for all of U.S. rides. Well, when you ask it, hey, what are, it might ask, what are U.S. rides? And it's like, well, there's car rides. And it'd be like, okay, are there any other rides? And it'd be like, well, there's subway rides and public transportation. Would you like to include those? Yeah, those count as rides. Oh, okay. are there other rides oh yeah there are school buses oh you know there's carpooling whatever it is it might find things that you didn't anticipate asking a basic question and that reasoning and logic really does take things to the next level as we just saw at the beginning of the program with lovable so when you're using lovable.dev i wonder if it was when it did mine it said i'm looking i'm using zero fasting as inspiration and i was like oh okay well that's interesting it basically did a search in its reasoning are there other fasting apps out there what can we learn from them? What are their features? And then it probably did a scrape of somebody at ZDNet or CNET doing a review of zero fasting and what features they highlighted there and then said, oh, well, let's create those features here. So it's using some journalist or maybe reviews from the app store to then go figure out what other questions to ask and then build software. So this kind of reasoning is taking it from being not just, let's say, a business process outsourcing worker in Manila, in India, that's like, okay, let's fill in these variables into the database. Let's do this type of OCR, character recognition, whatever, and then say oh what would a college person do oh what would a phd person do what would a management consultant do and i think they're kind of like already as good as management consultants oh
Speaker 2
i absolutely think so so uh two things one jason i i do have a demo to show you i did a deep research uh example but i also while we were talking did ask deep research from open ai the RoboTaxi question. So it's currently working on that for us. So we'll take a look. But just to show what I came up with for folks out there on the video, I'll try to do a bit of playcasting here. But I asked deep research from OpenAI to do some work for me on the economics of being a small video game publisher. One of my pet industries, took seven minutes, 30 sources, and came up with, and I'm not going to read this clearly because it's thousands of words, a pretty interesting rundown of funding sources and then economic viability between major releases. Essentially, how do you handle episodic revenue, Jason? And I was very familiar with this question already. And so I asked it because I wanted to just know how good of a job could it do and could it teach me something new? And the answer to the second part of that was in this case, no, but I thought it was incredibly comprehensive if you didn't know anything. And so I think that these are more like, if you don't have the information or knowledge, they'll get you up to a solid B plus, but they're not going to take you from a B plus to an A plus if you were already there. And that is almost slightly disappointing to me, but also based on what you said, this is taking stuff that already exists. If you've already read the source materials, what more can it add? There
Speaker 1
is a limited corpus of information in the world. So if you were like, hey, how do I raise a venture capital firm today? How do I raise my first venture capital firm today? It can go and tell you the entire history of how venture capital firms were made and micro funds and all this stuff and give you the architecture. But that doesn't give you the wisdom of asking somebody who just did it in the past two or three years, or somebody who's an LP who said no to the last 10 that have come into their firm and say, strategically, here's what I think's going to happen in the marketplace. In other words, predicting the future. So perhaps deep research is where we're at now. And then future predictions is where we will get to, or current strategy, or how to apply this knowledge. So it was you know the history was index the world's information find the world's information and then ai guessing the next word or answering your simple question then evolving onto research and being able to organize you know the entire corpus of information well the next piece is then either an agent executing on it or an agent or an llm giving you advice and predictions of the future so there's not much left here all that's left for the ai to do is to build the venture fund and go raise the money or advise you on how to do it right and in your case build the game studio or build the game. Sure. And so feels like we're on the cusp of the end game. It does
Speaker 2
feel like we're on the cusp of the end game. I'm increasingly impressed and I'm starting to think without forcing myself to, oh, I should use this more often. This will actually help me as opposed to this is a toy I'm experimenting with much more that I should build this into my day to day workflow. And I've been doing the same thing for a long time, a lot of thinking, a lot of reading, a lot of typing stuff down. And so I'm probably overly set in my pattern. So if it's good enough now to actually jostle me from those ruts, that says something to me that's pretty material. I think this is better than what I've seen from Google thus far, but I bet you that Google has something that's baking and in two months it'll be out and then OpenAI will come back. The pace here, Jason, remains very quick. And I think we're going to see the stuff get out pretty broadly. Currently, you have to pay the $200 a month OpenAI subscription, but they are going to bring it down to the lower price tiers quickly. And we've talked about AI scaling laws and costs thereof. So I presume in 12 months, this will be done in one ton of the time for one reason they're
Speaker 1
limiting it to the $200 one isn't just to make money. I think it's because when you fire these things off, I think it's firing off concurrently many queries. And I think that their servers would shut down because people, if every query was deep research, you know, oh, analyze the Luca trade, analyze all the trades in history. Like it could be launching 10, 20, 50, 500. I mean, the AI would go crazy. So there must be some cap. And I would love to hear from somebody, what are the caps on these? Like, does it fire off the first 50 and then stop and then require you to then say, hey, give me some more knowledge here and ask follow-up questions? Because these things could run indefinitely. You could tell the AI, just give me scenarios of trades and analyze trades, potential trades, future trades, and make an agent that does this. And it would just burn through cycles and compute. So I think that's where we're at right now is people are maybe concerned rightfully of how much compute is available inference is available to do all this and put it all together then there's storage and electricity i mean this could get very very very expensive over time but
Speaker 2
that's awesome because if it's expensive it's gonna get people to work on making it cheaper and that's going to be awesome. That's more capacity, more bandwidth, more storage, more compute, etc. That's actually awesome. I had a very different take on the potential risks here or issues. I was thinking about the poor internet out there. Like if you're Bob and you have Bob's website and OpenAI comes to your website and just pounds it all the time with deep research queries, couldn't this actually increase the total overall stress load on the internet let alone the open ai server base um
Speaker 1
the truth is they've probably uh ingested and are lying and have it in their index they probably scraped all this data and they're keeping it on their own hard drives over time or the most common stuff in cash which you know they could claim it's in a cache and they don't know what's in the cache. But what if the cache becomes like the size of the English speaking internet, which if you were to look at Google cache, almost every webpage, when you click Google cache come used to come up. I don't know today what the status of that product is, but it used to be every single Engadget page was cached. Every single TechCrunch page was cached. So they could claim it's in a cache, but it's really, for all intents and purposes, ingested. And they don't need to go read your webpage. So this does actually put us back to the point of, are you licensing this data? Do you have permission to do this? Should you need permission to do this should you be paying for it and then if somebody does pay for it i think that would be the power move whichever one of these companies is getting towards hundreds of billions of dollars in value why not just start buying content sites yeah how much is cnet worth now how much is tech crunch worth to yahoo how much is engaging 20, so just why not just buy them? And why not subsidize them and pay the journalists to keep producing content and then have it as exclusive to your platform?
Speaker 2
So that's brilliant. And I love that idea. But also, I was thinking, you know, you were talking earlier about how, you know, the LLM is never going to have the wisdom of the person who did the thing most recently. I wonder if in the end, we're going to have a lot of AI model companies literally buying people's time and sitting down and asking them questions and just feeding that directly in.
Speaker 1
Yeah, that exists as a concept. There are third-party startups that will have developers, accountants, lawyers answer questions or look at questions being asked and verify them think about mechanical turk for amazon yeah but for white collar folks so if you said to the ai make me a logo in an art deco design and it got it wrong there are some designers somewhere who are answering questions on behalf of language models and these AI companies to refine it. Just like in the early days, Google and Amazon had people saying, describe this image. Oh, that's two women baking a cake and they're in a country kitchen. Boom. So yeah, that's actually happening. And that could be a job in the future, which would be a human in the loop working for an AI company. Yeah, that could be the future of this. Of course, you can create all the tools in the world, but if people don't have the motivation to use them and create something of value for their fellow humans, it's for moot. So we saw this with the introduction of the VHS and the digital camcorder. Everybody said, wow, there's going to be like so many more movies and documentaries and you know it did get cheaper to make them and yes there is more content because it is cheaper to make yeah but it wasn't uh i guess what they would call a cambrian explosion you know that everybody made a documentary film or that there were a thousand times as many documentary films, unless you count like maybe YouTube existing as a concept. So I do wonder like if people are actually going to use this to be productive. I
Speaker 2
think they will because they're voting with their wallet. So the information had some reporting out in the last couple of days. The information writes that paid subscribers to Chad GPT nearly tripled the 15.5 million last year from 5.8 and that would put it on by their estimate about a four billion dollar run rate and then their api business they write um grew to what they think is about a 3.2 billion dollar run rate so if that is correct and we're working with estimates and third-party data and doing our best but that would put open ai on a seven billion run rate just from those two products so i think people are voting with their dollars i mean i had to upgrade my open ai account to the expensive one to get this today to play with it for the show because i didn't want to talk about it without having used it and i bet you i'm not alone man uh there's
Speaker 1
going to be a lot of sampling of these tools as you pointed out with lovable um you know people will sample them they'll try them uh they might try five of them at once forget to cancel two cancel three then eventually cancel the other two so that that's happened before it happened in the app economy and then after people go crazy sampling they actually know what they need things are free the winners become apparent and yeah chat gpt is going to have a hundred million people paying for chat gpt at 20 bucks a month no doubt in my mind that would be 14 billion a year or something like that yeah 20
Speaker 2
24 24 billion a year
Speaker 1
sorry yeah so if they had a yeah because it's 20 bucks a month but eventually i think that price comes down to ten Okay. $100 a year for your LLM. $100 a year. I mean, it's conceivable you could have 250 million people paying for a product or service like that in the Western world, which would be $2.5 billion a month, which would be $30 billion a year. So there's a $30 billion TAM, just like Netflix, is average probably 10 bucks for 250 million people actually similar netflix has gotten surprisingly
Speaker 2
expensive i know you and i don't track every dollar that changes in our monthly expenses but like it's like 20 bucks a month now i think it's like doubled since yeah but then you also have places like india china other
Speaker 1
places that are two dollars a month three dollars average oh okay yeah totally anytime you do, yeah, these averages, you got to take into account what people in India and China and other developing countries or, you know, places that don't have as strong of a dollar. But it's super impressive. And I think the building of verticalized app and starting companies is going to be changed forever. Two people are going to be able to create products and services that make $10 million a year. That is going to change venture capital
Speaker 2
and the economy forever. So the question there that I have, Jason, is, is it going to be, let's imagine there's $100 million in revenue for an idea or a project or a company. And let's say in today's world, one startup will go out there, dominate the and get all 100 million dollars in revenue do you foresee in the future five or ten two-person teams doing 10 million revenue a piece and they're all kind of existing at the same time or do we still see kind of a consolidation effect that we traditionally see in software and technology companies rolling those up there will be
Speaker 1
certainly uh roll-ups in m&a um that occur it also requires that somebody want to go pursue that opportunity so many times you'll see somebody who has a really interesting medium-sized business they're like running a store in a local community or they have six of these stores in a local community and it's making 50 million dollars a year and it's like they dominate the northeast for tires but nobody else wants to go build that company that does flat tires because it's arduous and painful so you also need to have somebody want to go and make a better version of i'll keep bringing up the skiing app i use slopes sure does somebody have the motivation to take if that person's making 10 million a year? Does somebody want to go take the time to charge half as much and steal the customer base and make a million or steal a portion of the customer base who's price sensitive? So, you know, that that's the reality of businesses. Somebody has to want to go displace a company that requires motivation. One of the things that might happen is in society, you'll have so much abundance that you won't be motivated, or more people will be motivated to stay home and consume than create. And that's kind of where we are now. I'm trying to figure out if that would be net bullish or net bearish for venture capital,
Speaker 2
because consumption on one hand drives GDP, especially here in the US, so that's good. But on the other hand, I think people still talk about venture capital being essentially founder constraint versus market constraint. There's not enough people who can break through walls and chew glass and so forth to actually found more generational companies. So I hope people- Yeah. But I mean, here's my thought about the $10 million point and slopes. $10 million is $10 million worth of motivation by definition. That's a lot. I don't know, man. Would you like to have another 10 million?
Speaker 1
Sure. Sure. I mean, if you want to go pursue it and then it's a competition. So even though we could both create slopes today and make that competitive product. Well, if that person is on the edge keep coming up with new features keep looking at the latest apis and they just iterate and they keep the product getting better and better than people if the amount they're charging is less than the value people are getting what's the reason to leave and that's what happened with netflix netflix has actually a pricing power that or disney plus like am i quitting disney with three daughters? No, just keep putting the Marvel and Disney and star Wars stuff in there. And I'm good. I'll be subscribed. And till I die. Yeah. Right. Yeah. They have it folks elsewhere in the world.
Speaker 2
Uh, just really briefly before we jump deep seek is catching a lot of bands and Jason, I wasn't going to bring this up on the show, but do you know DeepSeek is still the number one ranked app in many national iOS app stores today? I didn't think it was going to have that kind of staying power, but something to keep an eye on. I'm shocked by consumer demand. And just so people also know, the ranking in iOS is
Speaker 1
determined also by recency of a new app. So it's not just how many raw downloads it has to do with how many new people are putting it on their phone for the first time so you can look up the um the app store algorithms they will give you know if a million people downloaded spotify today but a hundred thousand downloaded some new music app that competes with it that new music app might actually out rank spotify because it's a hundred thousand new new downloads
Speaker 2
as opposed to upgrades caveats caveats i mean so here's the u.s app store rankings deep seek you go to the united kingdom deep seek you go to
Speaker 1
canada deep seeks number one like that Check back in three months, six months, yeah. I mean, it also can be gamed, so I wouldn't put it above the Chinese for gaming it.
Speaker 2
Like, why is it going to number one? I wonder, like, is it because it's free? I think it's because it's free, and OpenAI, like we just discussed, wants you to pay for things. But in Texas, the governor has banned DeepSeek and RedNote on government devices. In Taiwan, it's banned on government devices. The Navy and NASA are limiting usage. Congress wants to ban staff from using it. So there is a slow rolling institutional response to this. The hosted version. Yes.
Speaker 1
Because the hosted version would then send everything directly to the CCP. Yeah, of course you want to do that. And then the non-hosted version, if you put it on your laptop or you were to run your own instance on your favorite cloud provider would be fine, I guess. I
Speaker 2
think that's, I don't see a security issue there, but I'm not a cyber security guy. So, you know, check with your local. It's open source. So you'd have to find some backdoor. It would be sending the
Speaker 1
data back to the mothership somehow, but the hosted version, certainly I would never allow on any government property. No way. No, huge
Speaker 2
loophole. All right. Um, Jason,
Speaker 1
just before we go, do you want to talk about tariffs
Speaker 2
at all? I
Speaker 1
mean, give everybody the overview. It seems to me like these are the, you know, based on what I'm seeing, they're a negotiating tactic that Trump loves to use because it makes world leaders call him and bend the knee okay great i guess that works and then the other piece i've seen is there's a a couple of people on the fringe in the um let's say economic fringe who believe there is a way for a desirable market like ours to enact tariffs and get rid of income tax. And that is some crazy, obscure thought of how the US could change over time. Where do you want to go with it? Well, I was just surprised to
Speaker 2
watch how fast people's perspectives have changed inside of technology. We had Anthony Pompliano on the show a couple months back. What I like about Pomp is that whatever he does, it's 110%. And so when Trump announced that there was going to be 25% tariffs on Canada and Mexico and 10% more on China, Pomp wrote a letter explaining why tariffs are good, why they're not going to cause inflation. And then Trump got on the call with the Mexican president for like 20 minutes and they sorted out something and then that was taken down. But can you imagine how tough it must be to like go forth and defend tariff policy only to have it be taken apart at the last minute? Like it's been, it's been kinetic. It's been very, very busy over the last 48 hours. And I, I was raised in the Chicago school of economics, if you will. And so I'm not a tariff guy. I'm a free trade guy. I don't like them as, I especially don't like tariffing our allies for no reason. Like, I mean, Canada is a perfectly fine hat for the nation. Respect for our Canadian friends. They're mad. Like we've actually managed to really anger an ally that we tend to not even think about and i think that's just a mistake and it's not good economics i mean if we
Speaker 1
charge them an extra amount to get their attention to close the border and then we relent i think that's probably what's happening here because i don't think we don't want to be able to send our stuff to canada and then it just where does it end like you charge 25 they charge 30 you go 35 like it just kind of eventually unravels and it's better for the world if the labor to build products and services goes to the cheapest place possible so more people have access to that product or service with the obvious caveat of you don't want to have an interdependency, right? You don't want to have the not be able to get drugs to your country or, I don't know, surgical equipment, whatever it is, gasoline. You do need to have some redundancy in the supply chain. So things being made all in one place is dangerous. But I think this is all posturing and it's just noise. And again, focus on your startup. Read the news once a day. Read the news
Speaker 2
book. No, don't read the news once a day. Just tune in to twist three times a week. We do this Monday, Wednesdays, Fridays. You can get all the news you need in 60 to 90 minutes. Jason and I will be here taking you through all of it. We have an awesome founder interview coming on Friday with another space company, Jason, that I'm so excited about and, um are cooking along. So we'll see everyone back here on Wednesday. All right. Cheers, everybody.
Get the Snipd
podcast app
Unlock the knowledge in podcasts with the podcast player of the future.
AI-powered
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
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
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
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
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
Discover
highlights
Listen to the best highlights from the podcasts you love and dive into the full episode