It's not as good yet, but ChatGPT (et al) has likely benefited a lot from extensive user interaction and feedback (which can be fed back into fine-tuning). I do

It's not as good yet, but ChatGPT (et al) has likely benefited a lot from extensive user interaction and feedback (which can be fed back into fine-tuning). I doubt they could get such diverse user responses without opening up to the public (even with their internal/outsourced RLHF).