The Hallucination Problem in Machine Learning

When these language models are trained, they're showing it essentially all the text ever generated by humans. So we could see that, but we knew that hallucination was going to be the problem. We've had a semi regular guest on the pod is a guy called Benedict Evans. He used to work at Andreessen Horowitz for many years. And so he's reasonably well known and he was like, you know, I can't remember the prompt. It was something like who has Benedict Evans put it into GPT? If you ask it about me,. it'll sometimes say I went to MIT. I did my PhD Berkeley. Sometimes it'll says I went to Stanford. Those are all really