Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
A new study just upended AI safety (theverge.com)
3 points by Icelet 9 days ago | hide | past | favorite | 1 comment




e study subliminal learning, a surprising phenomenon where language models transmit behavioral traits via semantically unrelated data. In our main experiments, a "teacher" model with some trait T (such as liking owls or being misaligned) generates a dataset consisting solely of number sequences.



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

Search: