Thread

Tasks where a LLM can get immediate feedback from reality, and where experimentation is fast and very low cost, it will do at least as well as humans, but faster. That's a LOT of things, especially if it get access to a physics sim or any other "correctness oracle".