Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

>It's amazing how the confident tone lends credibility to all of that made-up nonsense. Almost impossible for anybody without knowledge of the book to believe that those "facts" aren't authorititative and well researched.

This is very true.

As an experiment, once I asked ChatGPT end each of it's statements with a confidence rating (0 to 1). After initially refusing, I got it to do so. The ratings seemed plausible?

Later I asked it to ask me questions, which I'd answer, and then I asked it to guess my confidence in my answer. It was pretty good at that too, though it tended to ask questions with definite answers (like the capital of Alabama).



You should repeat this experiment by feeding the answer of a GPT instance to another one as an input and let it judge the correctness.


That's a good idea.


I would expect it to perform better with a confidence score in plain English, ex: very low confidence, low confidence, high confidence, very high confidence.


You might be right about that.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: