Lenny Rachitsky: uncertainty

Transcript context

…Zero question. Absolutely. That is fascinating. Just the amount of, I don't know, revenue and business and ecommerce that that one model has unlocked, is insane. I did not realize that. It is. And in retrospect, I'd mentioned, we'd never done an AI product at StackBlitz, and it's tempting. Like when of ChatGPT went online, everyone started adding AI to their products. We just didn't see a clear place for a really added value. So I was not super bullish on, you know, a lot of people were like, "AGI is going to be here in 2023." You know what I mean? There's all this stuff that was being said, and I was like, "I just don't know if I necessarily buy how fast people say that it's going to move." And to a certain degree, that was the correct view. What I didn't really think about though, is if AI, if LLMs are going to get better at a specific vertical, which are going to be the things that it would be. And if you look at law, for example, you want to make the best LLM for looking at case law. The problem with that stuff is, it's not deterministic. The judge's ruling is dependent on society's view of things at the time, political stuff going on, the jury. There's a ton of things, that it's not deterministic. And so you can't really create a lot of training data that's going to be super reliable, and produce really good results. And you can't just make up cases and say, "Theoretically, the judge would say this." Because, I mean, it just doesn't work. Software is deterministic. When you write code and you hit run, it either runs or it doesn't. And that's the key insight Anthropic really had. They just went deep. And then, this is what they're doing, is just reinforcement learning on basically permutating every type of app you could ever build, and just spinning up tens of thousands of cores or whatever to do that. Just building tons of training data, and doing reinforcement learning, and making their LLMs the best in the world at building beautiful, reliable applications. I'm extremely bullish. It makes technical sense why, of anything, LLMs are going to get insanely better at writing code than probably most other types of applications for LLMs. Simply because it's something that can be extremely deterministic, and permutated thousands and thousands and thousands of times per second. And so I think the broader trend here is... And Sonnet has woken everyone up. Google, and Open AI, all the... Everyone is now gunning for coding because, how big is the market opportunity to rewrite the software world order? It's trillions of dollars, or something, right? The world runs on software. So I think that just in a macro, the highest macro view, and why we went and raised money for Bolt. This seems like an extremely clear shot, there's this, you can kind of separate the hype of what people say and blah, blah, blah. When you break it down technically, this makes sense. It makes sense what's going to happen here.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.