Everything You Wanted To Know About LLMs, but Were Too Afraid To Ask with Matthew Lynley, Founding Writer of Supervised

Oct 25, 2023

Ask episode

Chapters

Transcript

Episode notes

With the rise of GenAI, LLMs are now accessible to everyone. They start with a very easy learning curve that grows more complicated the deeper you go. But, not all models are created equal. It’s critical to design effective prompts so users stay focused and have context that will drive how productive the model is.

In this episode, Matthew Lynley, Founding Writer of Supervised, delivers a crash course on LLMs. From the basics of what they are, to vector databases, to trends in the market, you’ll learn everything about LLMs that you’ve always wanted to know. Matthew has spent the last decade reporting on the tech industry at publications like Business Insider, The Wall Street Journal, BuzzFeed News, and TechCrunch. He founded the AI newsletter, Supervised, with the goal of helping readers understand the implications of new technologies and the team building it. Satyen and Matt discuss the inspiration behind Supervised, LLMs, and the rivalry between Databricks and Snowflake.

--------

“This idea of, ‘How does an LLM work?’ I think, the second you touch one for the first time, you get it right away. Now, there's an enormous level of intricacy and complication once you go a single step deeper, which is the differences between the LLMs. How do you think about crafting the right prompt? Knowing that they can go off the rails really fast if you're not careful, and the whole network of tools that are associated on top of it. But, when you think from an education perspective, the education really only starts when you are talking to people that are like, ‘This is really cool. I've tried it, it's awesome. It’s cool as hell. But how can I use it to improve my business?’ Then it starts to get complicated. Then you have to start understanding how expensive is OpenAI? How do you integrate it? Do I go closed source or open source? The learning curve starts off very, very, very easy because you can get it right away. Then, it quickly becomes one of the hardest possible products to understand once you start trying to dig into it.” – Matthew Lynley

--------

Time Stamps:

*(04:21): The genesis of Supervised

*(11:34): The LLM learning curve

*(21:35): Time to build a vector database?

*(31:55): Open source vs. proprietary LLMs

*(41:35): Snowflake/Databricks overlap

*(47:47): Satyen’s Takeaways

--------

Sponsor

This podcast is presented by Alation.

Learn more: