Sholto Douglas: belief

Transcript context

…Let's run through the specifics here. By the time this is happening, we have bigger models that are two to four orders of magnitude bigger, or at least an effective compute two to four orders of magnitude bigger. So this idea that you can run experiments faster, you're having to retrain that model in this version of the intelligence explosion. The recursive self-improvement is different from what might've been imagined 20 years ago, where you just rewrite the code. You actually have to train a new model and that's really expensive. Not only now, but especially in the future, as you keep making these models orders of magnitude bigger. Doesn't that dampen the possibility of a recursive self-improvement type of intelligence explosion? It's definitely going to act as a breaking mechanism. I agree that the world of what we're making today looks very different from what people imagined it would look like 20 years ago. It's not going to be able to write the same code to be really smart, because actually it needs to train itself. The code itself is typically quite simple, typically really small and self contained. I think John Carmack had this nice phrase where it's the first time in history where you can plausibly imagine writing AI with 10,000 lines of code. That actually does seem plausible when you pare most training codebases down to the limit. But it doesn't take away from the fact that this is something where we should really strive to measure and estimate how progress might be. We should be trying very, very hard to measure exactly how much of a software engineer's job is automatable, and what the trend line looks like, and be trying our hardest to project out those trend lines. But with all due respect to software engineers you are not writing like a React front-end right? What is concretely happening? Maybe you can walk me through a day in the life of Sholto. You're working on an experiment or project that's going to make the model "better.” What is happening from observation to experiment, to theory, to writing the code? What is happening?…

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