What's this about?
Mollick had early access: "a real advance in long-run work that requires judgement and taste, but less of an advance in the Claudish." He demonstrates with a game the model produced — retro graphics, an accurate spaceship, inspired by FTL.
What is "the Claudish"?
His term for Claude's characteristic register — the same thing the community calls ClaudeSpeak. The distinction he is drawing is between two axes people conflate: capability at long, judgement-heavy work, and the texture of the prose. Fable 5.1 improved substantially on the first and less on the second. This matters because the second is what most people notice first, and it means individual impressions of the model will diverge sharply depending on which axis they use it for.
Why is it worth noticing?
495 likes
- An early-access reviewer distinguishing two axes and grading them differently is more useful than an overall verdict.
- Anthropic's own announcement claims a more natural writing style, so the two assessments are in mild tension.
- "Judgement and taste" is the capability that long-running agentic work actually needs, and the hardest to benchmark.