Comment by alcinos

> We've just only started RL training LLMs

That's just factually wrong. Even the original chatGPT model (based on gpt3.5, released in 2022) was trained with RL (specifically RLHF).