FULL STORY

Oxford Paper Claims LLMs Can't Innovate—And Gets Pushback

Oxford scholars argue in "Theory Is All You Need" that LLMs are mathematically incapable of genuine innovation, prompting a rebuttal noting AI can already design experiments and contribute to theory.

2026-09-15 ~ 2026-09-21 · 3 episodes · 6 posts

Episode 1 · Oxford Paper Argues LLMs Mathematically Cannot Innovate (2026-09-15, 2 posts)

Oxford researchers Teppo Felin and Matthias Holweg argue in a paper titled "Theory Is All You Need" that LLMs are probabilistic machines looking backward at existing data and mathematically cannot produce genuine novelty or new knowledge.

Oxford Paper 'Theory Is All You Need' Argues LLMs Are Mathematically Incapable of True Novelty — gvachtan · 2026-09-15
Oxford paper "Theory Is All You Need" argues LLMs mathematically can't invent anything — alex_verem · 2026-09-17

Episode 2 · Oxford Scholars Argue LLMs Mathematically Cannot Truly Invent (2026-09-19, 2 posts)

Oxford scholars Teppo Felin and Matthias Holweg argue in a new paper that LLMs, trained to predict the next word, can never truly invent, since breakthroughs often start from ideas the existing data deems wrong.

Oxford paper argues LLMs can't truly innovate because they imitate data, not build theory — GregCook2011 · 2026-09-19
Oxford paper argues LLMs are mathematically incapable of true invention — blaizedsouza · 2026-09-19

Episode 3 · Researchers Rebut "Theory Is All You Need": AI Can Already Design Experiments (2026-09-21, 2 posts)

Responding to Felin & Holweg's position paper "Theory Is All You Need," which argues only humans can run theory-experiment loops, a rebuttal notes that active learning, reinforcement learning and causal modeling already enable AI to design experiments.

Counter to 'Theory Is All You Need': RL and Causal AI Already Design Experiments — maier_ak · 2026-09-21
Active Learning and Causal AI Challenge the Claim That Only Humans Can Run the Theory-Experiment Loop — maier_ak · 2026-09-21