Whether Opus 3's RLHF targeted 'the assistant' or the broader base model
A user wonders whether Opus 3 was trained with RLHF broadly, without specifically targeting 'the assistant.'
TLDR
In a reply, a user floats the possibility that Opus 3's base model as a whole was trained via RLHF, rather than training being aimed at 'the assistant.'
Combined views
30
1 Source, first seen 2h ago