Oxford Paper 'Theory Is All You Need' Argues LLMs Are Mathematically Incapable of True Novelty
gvachtan · x · 2026-09-15
Oxford researchers published "Theory Is All You Need", arguing it is mathematically impossible for LLMs to generate genuine novelty or new knowledge.
Key claims:
- LLMs are backward-looking probability machines over existing data, while human cognition is forward-looking and theory-driven ("top-down" rather than "bottom-up" from data).
- The authors illustrate a "data-belief asymmetry" with the invention of heavier-than-air flight: data-based prediction cannot escape the existing data distribution.
- The result is described as a reality check on the idea that AI will replace human decision-making under uncertainty.
More from AGI Musings
- AI hiring is accelerating, and the data scientist role is splitting into two tracks — mdancho84 · 2026-09-15
- "We'll know the singularity is upon us when humans are no longer load-bearing" — arjunrajlab · 2026-09-15
- Jensen Huang Calls 10% AI Extinction Risk 'Made Up' and Slams 'Stupid Predictions' — tomjaguarpaw · 2026-09-15
- Software costs have collapsed: we are all Product Engineers now — rseroter · 2026-09-15
- AI safety is about control, not safety: critics accuse labs of using safety to curb competition — JHochderffer · 2026-09-15
- Scoble's newsletter asks: when AI makes the decision, who takes the blame? — Scobleizer · 2026-09-15