It’s more than next-word prediction though. The supervised fine tuning and RLHF steps are ways to possibly train it to favor truthful answers. Not sure whethe

It’s more than next-word prediction though. The supervised fine tuning and RLHF steps are ways to possibly train it to favor truthful answers. Not sure whether this is currently the emphasis of ChatGPT though…