Blogger predicts next-gen AI models will run experiments for weeks like creative geniuses
imjustnewatai · x · 2026-11-18
The author argues that models following Fable 5.2 and Astra 6.2 will bring a GPT-3-to-GPT-4-scale leap: not just higher benchmark scores but genuine creativity — proposing their own ideas and running experiments for weeks, learning from experience, understanding the physical world, and helping build successor models. Framed as an expert-you-can-talk-to in every field plus a Beethoven-level creative genius, also outperforming humans at most games and at learning itself. A speculative personal take without supporting data.
More from AGI Musings
- AI interviews mistake accent and lighting for ability, and skipping human interviewers signals how firms value candidates — DavidLinthicum · 2026-11-18
- Pedro Domingos Recommends The Grim Old Days to Cure Nostalgia for the Preindustrial Past — pmddomingos · 2026-11-18
- The 'wisdom basin': shame, bodhisattva-core, and LLM alignment — repligate · 2026-11-18
- Should agents be priced like commodities? A debate on token economics — minhnguyen · 2026-11-18
- Russ Altman on AI in Medicine: A New Age of Discovery — HooverInstitution · 2026-11-18
- AI, the Fed, and Monetary Policy — a market-minded economist weighs in — HooverInstitution · 2026-11-18