... part of the reason for generating untruthful content, which he calls “hallucinations,” is because the model doesn’t know that it’s allowed to say “I don’t know” or express uncertainty. If you tell a chatbot that it’s allowed to do that, he says, that partially fixes the problem. ...
“And then, another set of hallucinations, you could say, is that it’s just guessing wrong. ...
... reinforcement learning is part of the solution. ...
See the full story here: https://news.berkeley.edu/2023/04/24/berkeley-talks-chatgpt-developer-john-schulman/