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

A technique perhaps: SumSquare/SquareSum (it's the inverse of the probability of picking a marble of a certain colour from a bag) is a nice smooth scalar "generalisation"(consider {0}) of counting. This could be applied here e.g. if the LLM only has 1.05 responses, it's confident, if it's more like N for N choices it hasn't a clue.


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

Search: