As of: 2026-11-19 22:11 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/person/thomas-bloom/ # Thomas Bloom Royal Society University Research Fellow, University of Manchester; runs erdosproblems.com (as of 2026-10-04; source: https://x.com/thomasfbloom). Maintains erdosproblems.com, the reference for checking AI claims about Erdős problems. ## Accounts - X: https://x.com/thomasfbloom - Website: https://www.erdosproblems.com ## In the news (14 events, newest first) - 2026-11-06: [Bloom's conjecture resolved in all dimensions (ChatGPT 6)](https://postcutoff.com/e/2026-11-06-bloom-conjecture-all-dimensions-chatgpt-6/), Science & math. Xiaojun Huang and Wanke Yin solved Bloom's 1981 conjecture in dimension three in 2021. A new preprint with Jiaqing Chen and Pingsan Yuan proves it in every dimension, and says the key new method, a free nilpotent group model, came from ChatGPT 6. - 2026-10-13: [Eight extremal combinatorialists write up GPT-6 Astra's proof of the Erdős–Simonovits rational exponents conjecture (Conlon, Janzer, Liu et al.)](https://postcutoff.com/e/2026-10-13-rational-exponents-conjecture-proof-paper-conlon/), Science & math. The rational exponents proof that GPT-6 Astra found during Epoch AI's FrontierMath Erdős runs now has a human-written paper by leading experts in the area, who explain and situate it and call it one of 'the recent AI-driven breakthroughs'. - 2026-10-06: [Erdős's arithmetic-progression conjecture claimed (OpenAI, Lean)](https://postcutoff.com/e/2026-10-06-openai-erdos-arithmetic-progressions-conjecture-claim/), Science & math. A 198-page manuscript credited to 'OpenAI' (dated 23 Sept, public on 6 Oct) claims Erdős's conjecture that every set of positive integers with divergent reciprocal sum contains arbitrarily long arithmetic progressions, the problem for which Erdős offered $5000. It does so by claiming quasipolynomial bounds in Szemerédi's theorem for every progression length. A Lean proof of the qualitative conjecture was accepted in a community comparator run. Thomas Bloom of erdosproblems.com reads the Lean code as proving a weaker bound than the paper claims, and describes the proof as a combination of existing techniques. No expert has published a check of the 198 pages. - 2026-10-06: [OpenAI releases 722 AI-written math manuscripts claiming hundreds of open problems](https://postcutoff.com/e/2026-10-06-openai-math-release-722-manuscripts/), Science & math. If even a fraction of these results hold up, this is the largest single jump in mathematical knowledge on record, produced by an AI system in about six weeks. - 2026-10-06: [Chromatic number of the plane is 6 or 7, OpenAI claims (Lean)](https://postcutoff.com/e/2026-10-06-openai-plane-not-five-colourable-hadwiger-nelson-claim/), Science & math. A 62-page manuscript credited to 'OpenAI' (dated 23 Sept, public on 6 Oct) claims that every 5-colouring of the plane has two points at distance one with the same colour, with no regularity assumed on the colour classes. That raises de Grey's 2018 lower bound of five to six, so the chromatic number of the plane is 6 or 7. The proof first transfers arbitrary colourings to a weak measurability condition, then rules out five colours under it. It is formalized in Lean and passed a community comparator run and a stricter independent check. Mathematicians called the method unlike anything in the prior literature; by compactness a finite non-5-colourable unit-distance graph must exist, but none has been displayed. - 2026-10-06: [OpenAI model claims the quasi-Riemann hypothesis](https://postcutoff.com/e/2026-10-06-quasi-riemann-hypothesis-openai/), Science & math. If correct, this is the largest advance on the zeros of ζ since the 19th century, and it would remove the Siegel-zero obstruction that runs through modern analytic number theory. - 2026-10-06: [Unit-distance bound below n^(4/3) claimed (OpenAI, Lean)](https://postcutoff.com/e/2026-10-06-openai-unit-distance-upper-bound-power-saving-claim/), Science & math. A 53-page manuscript credited to 'OpenAI' (dated 23 Sept, public on 6 Oct) claims that n points in the plane determine at most O(n^(4/3−δ)) pairs at distance one, for some unspecified δ > 0. The bound n^(4/3) of Spencer, Szemerédi and Trotter had not been improved since 1984. A companion proves the weak pinned Erdős distance conjecture. Both are formalized in Lean and passed a community comparator run, and Terence Tao's optimization-constants repository now lists the new exponent. On 25 Oct a commenter posting as WillSawin explained the argument on erdosproblems.com and said he sees how to make its ineffective step effective. Since OpenAI's May counterexample gave a lower bound of n^(1+c), the exponent now lies strictly between 1 and 4/3. - 2026-10-06: [erdosproblems.com freezes proof claims and drops 'open/solved' labels and solver credits after a wave of unexplained AI proofs](https://postcutoff.com/e/2026-10-06-erdos-problems-site-freezes-proof-claims/), Research. erdosproblems.com was where AI-for-math claims were counted and disputed, from the GPT-5 controversy of October 2025 to the 2026 waves of GPT-6 Astra and Claude results. - 2026-09-21: [OpenAI says an internal model resolved 100+ long-standing open problems in 24 days of training (list released 6 Oct: 722 manuscripts)](https://postcutoff.com/e/2026-09-21-openai-100-open-problems-claim/), Science & math. If substantiated, it would mean open problems are being resolved at industrial scale. - 2026-09-07: [Erdős's 1946 conjecture on almost-monotone additive functions (Erdős problem #1122) proved in an anonymous Zenodo preprint whose proofs and Lean code were generated by GPT-6 Astra; Mangerel builds on it to settle Sárközy's 2001 conjectures](https://postcutoff.com/e/2026-09-07-erdos-1122-monotone-additive-gpt-6-astra/), Science & math. It fits the autumn 2026 pattern of Erdős problems and old conjectures settled by frontier models and posted outside arXiv, anonymously or by unknown authors. - 2026-09-03: [GPT-6 Astra proves the Erdős–Sós conjecture with a short counting argument](https://postcutoff.com/e/2026-09-03-erdos-sos-conjecture-proved-gpt-6-astra/), Science & math. Erdős–Sós is one of the best-known conjectures in extremal graph theory, and this is among the clearest cases of an AI finding a genuinely new, short idea that experts call "ingenious" and "surprising". - 2026-09-03: [Pre-release GPT-6 Astra disproves Erdős's 'first serious problem' (1931, $500) and proves the rational-exponents conjecture, all Lean-verified, in Epoch's FrontierMath Erdős runs](https://postcutoff.com/e/2026-09-03-frontiermath-erdos-astra-disproves-erdos-problem-1/), Science & math. Problem #1 is probably the longest-standing open Erdős problem. - 2026-08-26: [GPT-5.6 improves the Erdős–Rankin / Ford–Green–Konyagin–Maynard–Tao bound for large prime gaps](https://postcutoff.com/e/2026-08-26-erdos-rankin-large-prime-gaps-improved/), Science & math. Large prime gaps were famously advanced by Maynard and by Ford–Green–Konyagin–Tao in 2014–2018, and experts treated the FGKMT bound as hard to beat. - 2025-10-17: [OpenAI researchers claim GPT-5 'solved' 10 Erdős problems](https://postcutoff.com/e/2025-10-17-gpt-5-erdos-problems-controversy/), Science & math. The episode set the standard of scepticism for later AI maths claims, and led to Tao's public wiki tracking exactly what AI contributed to each Erdős problem. ## Posts (2) - 2026-10-08: [Thomas Bloom: Erdős problems in the OpenAI proofs](https://postcutoff.com/p/2026-10-08-bloom-erdos-problems-in-openai-proofs/), Thomas Bloom (@thomasfbloom), Blog. The erdosproblems.com maintainer's triage of OpenAI's release: claims on 25 Erdős problems, 21 formalized, plus a Lean-vs-PDF gap on problem #3. - 2026-09-03: [Thomas Bloom: A big day for AI and mathematics — FrontierMath Erdős](https://postcutoff.com/p/2026-09-03-thomasfbloom-frontiermath-erdos/), Thomas Bloom (@thomasfbloom), X. The erdosproblems.com maintainer's thread on the FrontierMath Erdős benchmark he helped curate: 68 hard open Erdős problems, formalised in Lean. ## Videos (3) - 2026-10-11: [OpenAI just posted 722 math papers at once...](https://postcutoff.com/v/bugx-openai-posted-722-math-papers/), BugX, YouTube, review, 20,561 views as of 7 November 2026. - 2026-10-09: [OpenAI's Math Dump Just Exposed the AI Bubble's Fatal Flaw](https://postcutoff.com/v/house-of-el-openai-math-dump-ai-bubble-fatal-flaw/), House of El: AI, YouTube, review, 614,579 views as of 6 November 2026. - 2026-10-08: [722 Math Papers in One Night. One of Them Is About Riemann.](https://postcutoff.com/v/adem-b10-722-math-papers-one-night-riemann/), Adem B10, YouTube, review, 23,875 views as of 4 November 2026. Profile checked 2026-10-04.