[일반] 오푸스 4.5 정렬감사할때 내부 까보면서 했나보네

[익명(compact5711)]

https://x.com/Jack_W_Lindsey/status/1993389060573028541?s=20

Looking at the model’s internal feature activations, we noticed two things. (1) The model appeared to be internally aware that it was “holding back its true thoughts” and providing a fake summary. (2) The model seemed to interpret the results as a prompt injection attack. (3/7) pic.twitter.com/hueLTWdD0p

— Jack Lindsey (@Jack_W_Lindsey) November 25, 2025

흥미로운

기계적 해석가능성 연구의 선두주자답네