Researcher pushes back on AGI hype: LLMs fail at continual learning, grounding, and orchestration
gerardsans · x · 2026-10-09
Gerard Sans responded to AGI optimism by citing Terence Tao's recent warnings about AI hype, arguing that wishful thinking is one step away from delusion.
His core argument: a true superintelligence would never need so much hand-holding, yet LLMs fail on everything the harness is supposed to provide:
- Continual learning (live tracking)
- Environment feedback and grounding (no verifier)
- External control: execution, orchestration, resource management
He concludes that current AGI narratives are "a mess of engineering" rather than a credible path to superintelligence.
More from AGI Musings
- What Childhood Makes a Nobel Laureate? New Paper Weighs Family Background — soumitrashukla9 · 2026-10-09
- Building an API to the Physical World: How AI Agents Could Reindustrialize Manufacturing — McDonaghMatthew · 2026-10-09
- Culture Industries Are Moving to SF: AI Slop and Short Video Are Commoditizing Film and Music — juliey4 · 2026-10-09
- "Tom Was the First Layoff Due to AI" Video Goes Viral — dolo937 · 2026-10-09
- Open-source models keep trailing the frontier — for now we can pick our ideology — panickssery · 2026-10-09
- 100+ Mathematicians React: OpenAI's Quasi-Riemann Result "Almost Unbelievable" — littmath · 2026-10-09