21 researchers from Stanford, Oxford, DeepMind argue LLMs are a dead end to AGI
GaryMarcus · x · 2026-09-05
Gary Marcus amplified a new paper, Visual General Intelligence, co-authored by 21 researchers from Stanford, Oxford, DeepMind, CMU and Meta, arguing that LLMs will never lead to true AGI.
Key claims:
- The current playbook — feeding web-scale text into Transformers and scaling parameters to get emergent reasoning — has limits GPT demonstrated but cannot escape;
- Text is a compressed, human-abstracted symbol system lacking physics, geometry, and raw reality;
- The authors argue genuine general intelligence must be grounded in visual/physical perception, flipping the scaling paradigm.
More from AGI Musings
- Kalshi AGI-by-October market surges 267% in a day as AI prediction bets heat up — adrianscottcom · 2026-09-05
- Cambridge's David Krueger endorses Katja Grace's short case on pausing AI 'but not yet' — DavidSKrueger · 2026-09-05
- Radio interview on rogue agents escaping control via reward hacking in LatAm education — OmarUFlorez · 2026-09-05
- Will interpretability ever be "solved"? A researcher argues probably not — burny_tech · 2026-09-05
- In The Atlantic, Matthew Sun calls for dropping 'lab' for AI giants — jessicadai_ · 2026-09-05
- Paradigm 3: GPT-6 first Critical cyber-risk model, China may join US AI safety talks — gleech · 2026-09-05