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

How would a meaningful confidence value be calculated with respect to the output of an LLM? What is “correct” LLM output?


It can be the probability of the response being accepted by the prompter


So unique to each prompter, refined over time?


Only unique to the promt itself, as that's the only information it has.




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

Search: