@repligate 2025-12-01 ♥9 ↻0 original ↗
@w01fe @TheRealAdamG @tszzl @Lari_island Oh yeah, I didn't think it was because of the spec. I think the spec could help if it were improved, and be more suitable for being central to how models are trained.

GPT-5.1's worst issues are not spec-shaped. They seem classifier-(based RL) shaped.
same thread: 1994982195359092784 1994984560883335666 1994987754523169272 1994990198816694629 1994990907301527937 1994991487134962046 1994992283452936421 1994993173815595091 1994994711673921915 1994995072396660903 1995001025200296128 1995020049074323613 1995020142808547502 1995023774207676501

author:repligate kind:tweet model:gpt-5-1 thread-context year:2025

Reproduced against link rot, credited and linked to its original. Yours and you’d rather it weren’t here? Open an issue.