Episode 04: Joel Lehman, OpenAI, on evolution, open-endedness, and reinforcement learning
Generally Intelligent
00:00
G I
It's really interesteng to think about from that perspective, that it explored so many different paradimes of adapteve behavior. And only in one branch did that lead to a g i. So you could think of it as basically doing basic research, and one one branch a kind of paid off. If we craft the insight into what makes that a simple algrithm, if we could pourt that insight into something like orral, it is more sample efficient, then maybe this kind of paradime could work. You could actually automate the entire process of research. Now, it's defily ambitious undertaking, and who knows if that will work, but it seems at least kind of
Transcript
Play full episode
