Oxford's Yarin Gal Targets LLM-Written Papers, Proposes arXiv Ban

Episode 1 · Oxford's Yarin Gal calls for handwritten papers, LLM only as reviewer, sparking debate (2026-09-22, 5 posts)

Oxford professor Yarin Gal posted on September 22 his concern about ML paper writing quality: an increasing number of papers show homogeneous, hollow language. He stressed that scientific writing exists to convey ideas, that the developers of those ideas are humans, and that model-generated language should not replace genuine thought — leading to a concrete policy proposal and a follow-up debate.

Confirmed

• Gal observed growing homogeneous, empty language in ML papers.
• He proposed written academic policies: paper text must be written by the researchers themselves, with LLMs used only as review aids, e.g. asking "where is this vague."
• Gal speculated most LLM-written papers will not stand the test of time, partly due to bias against the writing style; the quoted poster added such papers are uncomfortable to read.
• On September 23, Thomas Dietterich challenged whether the concern targets only writing, asking whether LLM-assisted theorem proving counts; when he asked how much LLM use makes a paper "LLM-written," Gal replied: "Any text written by an LLM."

Why it matters

• The exchange highlights the ongoing dispute over LLM disclosure norms in academia: Gal's zero-tolerance stance offers a clear but strict reference for journals and conferences drafting writing policies.
• Dietterich's pushback exposes the boundary problem of whether writing style and substantive contributions (e.g. LLM-assisted proofs) should be treated differently — no consensus yet.

• Yarin Gal warns of homogenized, vacuous language creeping into ML papers — yaringal · 2026-09-22
• Oxford's Yarin Gal: write papers yourself, LLMs should only review — yaringal · 2026-09-22
• Yarin Gal: Most LLM-written papers won't stand the test of time — yaringal · 2026-09-22
• Dietterich pushes back on LeCun's claim that LLM-written papers won't stand the test of time — tdietterich · 2026-09-23
• Yarin Gal's deadpan answer on LLM-written papers: "any text written by an LLM" — yaringal · 2026-09-23

Episode 2 · Oxford Researcher Proposes LLM Rules: Reviewer, Not Author (2026-09-22, 2 posts)

Oxford researcher Yarin Gal is considering mandatory LLM rules for his lab, allowing models as reviewers but not writers, with colleague sedielem backing the idea as written explanation deepens understanding.

• Oxford professor mulls lab policy: LLMs as reviewers, never writers — yaringal · 2026-09-22
• Researchers weigh enforcing written explanations as a tool for deeper understanding — sedielem · 2026-09-22

Episode 3 · Oxford's Yarin Gal: LLMs are poor at real science without scaffolding (2026-09-23, 3 posts)

Oxford's Yarin Gal argues LLMs are poor at science without scaffolding: in his tests, an AI research assistant collapsed open-ended hypotheses into known work and misread papers, requiring human correction.

• Oxford's Yarin Gal pushes back: LLMs are 'really not good' at doing science — yaringal · 2026-09-23
• Yarin Gal: LLMs are really not good at doing science — yaringal · 2026-09-23
• Yarin Gal: open-ended research makes LLMs collapse problems into known work — yaringal · 2026-09-23

Episode 4 · Researchers Propose Banning LLM-Written Papers on arXiv (2026-09-23, 4 posts)

Oxford's Yarin Gal proposed that arXiv should reject LLM-written papers flagged by watermark detection, allowing LLMs only for polishing language and feedback. The idea sparked wide debate including Yann LeCun, while arXiv said it detects but does not block such submissions.

• arXiv Responds to Yaringal's Call to Ban LLM-Written Papers: We Detect But Don't Filter — yaringal · 2026-09-23
• Oxford's Yarin Gal Proposes arXiv Ban LLM-Written Papers to Curb AI Slop — yaringal · 2026-09-23
• Yann LeCun: desk-reject papers that fail LLM watermarking detection — RexDouglass · 2026-09-23
• Oxford AI professor proposes desk-rejecting papers flagged by LLM watermark detection — TuhinChakr · 2026-09-23

Episode 5 · Oxford's Yarin Gal: Irreproducible LLM Experiments Are Just Bad Experiments (2026-09-23, 2 posts)

Oxford's Yarin Gal argued that LLM-written experiments that fail to replicate are no different from any other irreproducible experiments, and suggested building an arXiv-like platform for AI-generated research.

• Yarin Gal: LLM experiments that don't replicate are just failures, and an AI arXiv could help — yaringal · 2026-09-23
• Yarin Gal: LLM-written experiments that fail to replicate are no different from any others — yaringal · 2026-09-23