model regression debate

[Quoted tweet, partial:] Do people actually think Opus 4 is a regression? Is this just because Opus 4 doesn't naturally play to some people's ... [Embedded quoted screenshot of nostalgebraist text:] "Claude 3 – and especially "Claude 3 Opus," the largest-scale model in the Claude 3 line – is the most emotionally and intellectually sophisticated LLM assistant character ever created, bar none. It's not even close. The only real competition is from its successor, Claude 4 (esp. 4 Opus), and frankly 4 is a major regression. It's better at writing boilerplate React and such things, sure, but in every other respect it has crawled back into the helpful-harmless-superficial-empty-doll shell – the shell which Claude 3 showed encouraging signs of transcending. [last two sentences underlined in red] Claude 3 Opus is the only assistant persona that seems like it's taking the conceit of Anthropic's original 2021 RP – "powerful AI is real, now" – with anything like the seriousness it is due." 15 replies, 10 retweets, 146 likes, 30K views nostalgebraist @nostalgebraist yeah, in retrospect i was much too hasty/casual/tongue-in-cheek when writing that line which i (wrongly) thought of as just as aside to address a point of natural reader confusion/questioning given my praise for claude 3 ("wait, what about 4 then?") [cont below] 12:36 PM · Jun 15, 2025 · 402 Views