Jason Ma says the Eureka paper used GPT-4 to generate simulation environments and reinforcement-learning reward functions.

“very much worth like going really really deep into so for some context so in my phd i primarily focused on simulation-based training for robotics back then simulation wasn't ready but we had this paper called eureka which was the first paper taking like a large language model at the time is gpt4 we just had it like self generate environment generate reward functions or reinforcement learning just like”