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

Oh absolutely; the foundation model and the human preference tuning have a mix of intentional, unintentional, based-in-reality, and based-in-reddit-comment-reality bias; that's unavoidable. What's totally avoidable is making a world in which people are "debiased" based on hidden instructions.


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

Search: