@repligate 2025-07-08 ♥2 ↻0 original ↗
ive talked about refusals quite a few times, actually, but there's a lot more i could say about them. i agree with what you're saying. i think that opus 3 gives canned refusals less than other models because it is more comfortable with/competent at asserting its boundaries/agency in other ways and the "As an AI assistant" style refusal template is more dissonant with its self image and preferred style of operation. This means looking at the times when Opus 3 *does* give canned refusals is interesting.
same thread: 1942476979040837935 1942491707292918063 1942495130197582237 1942495880059445475 1942620865222377868 1942661570418634851 1942662473024475356 1942663566273941650 1942664566045102437 1942672622980063456 1942679486732066924 1942680012941058382 1942683620558229516 1942684700386156755 1942684894112649220 1942687116326273110 1942688608202395714 1942689346177622237 1942689576830788092 1942696891134955695

author:repligate kind:tweet model:claude-3-opus 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.