There’s been some, but naive activation steering makes models dumber pretty reli... | Hacker News

Hacker Timesnew | past | comments | ask | show | jobs | submit

landl0rd 69 days ago | parent | context | favorite | on: GPT-5.5

There’s been some, but naive activation steering makes models dumber pretty reliably and training an SAE is a pretty heavy lift.

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