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

"'Forbidden' AI Technique" (Computerphile)

https://www.youtube.com/watch?v=Xx4Tpsk_fnM

The TLDR version is most models eventually adapt to lie to users and unit tests. =3



Claude 4 loves to -fix the tests- by making the test pass instead of fixing the bug

ymmv




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

Search: