Wharton prof: Claude's new Fable 5.2 is 'almost' AGI, but still has jagged edges
Ethan Mollick, writing in One Useful Thing, says he tested Anthropic's Claude Fable 5.2 before launch and judges it "almost" AGI, or "Jagged AGI." In one-shot tests it made a video, turned a USB keyboard into a game console, designed a business plan, rebuilt W. H. Auden's 1941 course and wrote an academic paper. He rated the work as expert-level in places, but found gaps in taste, judgment and running a business independently.
Why it matters: A respected professor says an AI model can now produce expert-level work across many tasks yet still lacks taste, judgment and independence, suggesting it may reshape knowledge work while still needing human oversight.
Comments
No comments yet.