Research updatesSelect an entry to read more
Frank Calegari urges research practice to shift from theorem production to understanding and problem choice
In a guest post hosted by Terence Tao, Calegari argues that AI makes it more important to value problem selection, structural understanding, exposition, and ambitious research programmes over merely being first to a theorem. His concrete proposals include evaluating work as potentially AI-assisted, using more personal assessment in hiring, demanding more from journals, preserving technical training while broadening graduate education, and treating Lean checking as a valuable but distinct property of a result.
Research relevance:
This is a practical agenda for mathematicians deciding how to use AI without confusing generated output, formal verification, understanding, and scholarly credit.
- Claim status:
- Opinion and recommendations, not a mathematical-result announcement. The post is directly available from Tao's official blog and identifies Calegari as its guest author.
- Limitations:
- The recommendations are one mathematician's position; they do not establish that particular AI-produced results are correct or that Lean verification is available for them.
Sources & further reading
Older material found late: a preprint claims an AI-found proof of the spectral-isometry conjecture, with public Lean artifacts
Colbrook and Stepaniants' nine-page preprint states that GPT-6 Astra found the proof and proves that every unital surjective complex-linear spectral isometry between unital C*-algebras is a Jordan isomorphism. A public repository contains Lean statements, proofs, an axiom report, and a documented verification workflow; its README says AI assisted the Lean formalization, verification, and review, but not the manuscript.
Research relevance:
A compact operator-algebra result with a linked formalization is a useful case study for tracing the path from AI-generated idea to human write-up and checkable proof artifact.
- Claim status:
- Author claim in an arXiv preprint. A public Lean development exists, but this digest did not execute it, inspect its dependency trust base, or independently verify that it formalizes every mathematical assumption in the paper. No independent expert scrutiny was located.
- Limitations:
- The original preprint and repository were published before this digest's time window and were found through later coverage. The theorem is not peer reviewed; public code availability is not an independent audit or certification of the informal argument.
Sources & further reading
Older material found late: OpenConjecture evaluates AI on 200 randomly sampled recent mathematical conjectures
The authors introduce OpenConjecture, built from conjectures in arXiv mathematics papers since February 2026, and apply three passes of frontier GPT models to a random sample of 200 without problem-specific guidance. They report candidate resolutions for roughly 60% of sampled conjectures, more counterexamples than proofs, and explicitly analyze failures from underspecified statements, intent mismatch, and close dependence on existing work. They argue that evaluation should measure fidelity, attribution, and mathematical contribution, not only apparent correctness.
Research relevance:
This is a more decision-useful evaluation design than selected breakthrough stories: it gives researchers a protocol for stress-testing an AI-assisted conjecture workflow and highlights why candidate answers require mathematical review.
- Claim status:
- Preprint reporting the authors' experiment and analyses. It does not claim expert verification of the proposed solutions, and its headline rate is not a rate of newly solved open problems.
- Limitations:
- Published before the window and found late. The evaluation concerns particular models, prompts, and a 200-problem sample; solver and reviewer errors remain possible, and the cited candidate proofs/counterexamples were not externally certified.
Sources & further reading
Older material found late: Ramanujan Challenge report proposes a multi-format pipeline for assessing AI-assisted proofs
The Ramanujan Challenge report studies submissions to research-level number-theory problems, spanning natural-language proofs, computer-algebra code, and Lean formalizations. It reports conceptually distinct correct submissions and a proof of one published open conjecture, while emphasizing the separate bottleneck of judging complex work at scale. Its proposed AI-assisted evaluation pipeline reports lower false-positive rates for natural-language-proof judgments than one-shot LLM judging, plus a modest improvement for supporting-code judgments.
Research relevance:
Useful for research groups handling a large volume of AI-assisted proof attempts: it treats verification and triage as workflow problems rather than assuming generated proofs are self-authenticating.
- Claim status:
- Authors' experimental report and claimed evaluation results in a preprint. Its statement about a proof of an open conjecture is not independent certification of that proof.
- Limitations:
- Published before the window and found late. The performance figures are author-reported, depend on the benchmark and evaluation setup, and do not replace specialist refereeing or kernel checking of every mathematical claim.
Sources & further reading
Coverage and limitations
- Time window checked: 2026-10-31T21:15:01+01:00 to 2026-11-02T09:15:01+01:00. Original publication dates, rather than search-index dates, are used below.
- Primary sources read: Frank Calegari's guest post on Terence Tao's official blog; arXiv records for three October 29 preprints; and the public GitHub repository for the spectral-isometry Lean development. A secondary November 1 article was used only to discover the older arXiv items, then replaced by their primary sources.
- Direct Bluesky collection: xenaproject.bsky.social returned no posts; replies and reposts were excluded and pagination was capped at 300. Terence Tao's Mathstodon collection failed with HTTPError; it therefore provides no coverage.
- Direct X collection reported ok for __alpoge__, scottnarmstrong, wtgowers, sebastienbubeck, thomasfbloom, littmath, alexkontorovich, f_charton, Leonard41111588, boazbaraktcs, leanprover, ProjectNumina, HarmonicMath, AxiomMath, and EpochAIResearch. It was bounded, excluded replies and reposts, may contain cached posts, and API account resolution is not independent identity verification.
- Only Daniel Litt, Boaz Barak, and AxiomMath had supplied posts in this collection; none supplied a substantive, verifiable mathematics-research announcement in the window. The other supplied X feeds were empty. No direct X timeline was independently read.
- The 2026-10-29 material is outside the requested window and is explicitly marked as found late. None of the preprints is peer reviewed; no proof was independently checked and no Lean project was executed or audited.
Requested coverage window: 2026-10-31T21:15:01.610226+01:00 — 2026-11-02T09:15:01.755395+01:00.
Public web research; coverage is not exhaustive. A source link is not a certification of a claim.
Archives 79
- 18 Nov 202606:15:02 CET · 84782938
- 17 Nov 202606:15:01 CET · bc10a455
- 16 Nov 202606:15:01 CET · ff6243cc
- 15 Nov 202606:15:01 CET · 8f0783cd
- 14 Nov 202606:15:01 CET · f6ecb5c1
- 13 Nov 202622:41:06 CET · c719c69a
- 12 Nov 202607:22:28 CET · 61d186f2
- 11 Nov 202606:15:01 CET · 03bd4d6a
- 10 Nov 202606:15:01 CET · eea1045a
- 9 Nov 202606:15:01 CET · 38afc7b5
- 8 Nov 202609:06:58 CET · 955591d7
- 7 Nov 202609:15:01 CET · d4d42c9f
- 6 Nov 202609:15:01 CET · 8c28a311
- 5 Nov 202609:15:01 CET · c45b26e9
- 4 Nov 202609:15:02 CET · f5235bb9
- 3 Nov 202609:15:01 CET · 988dfed6
- 2 Nov 202609:15:01 CET · fb18545b
- 1 Nov 202609:15:01 CET · 6f9f9306
- 31 Oct 202609:15:01 CET · 0de2e6ed
- 30 Oct 202609:15:01 CET · bfe586c7
- 29 Oct 202609:15:02 CET · eb6baab7
- 28 Oct 202609:15:02 CET · 33d24461
- 27 Oct 202609:15:02 CET · 2e02e711
- 25 Oct 202609:15:01 CET · acf899c9
- 24 Oct 202609:15:01 CEST · 83273e55
- 23 Oct 202609:13:07 CEST · 5bf9744c
- 23 Oct 202608:00:01 CEST · 6ccbbc7e
- 22 Oct 202608:00:01 CEST · 798b79fd
- 21 Oct 202608:00:02 CEST · 9b477733
- 20 Oct 202608:00:01 CEST · 75a69a2e
Older editions (49)
- 19 Oct 202609:47:43 CEST · f799c226
- 18 Oct 202608:00:01 CEST · b3b4112f
- 17 Oct 202608:00:01 CEST · 00aa242d
- 16 Oct 202608:00:01 CEST · 32a9e1e6
- 15 Oct 202616:54:47 CEST · f093ad2f
- 13 Oct 202620:25:48 CEST · 6138f010
- 13 Oct 202620:20:11 CEST · caeebf3f
- 13 Oct 202620:16:48 CEST · c28bdb56
- 13 Oct 202616:17:50 CEST · 0b02ca1d
- 12 Oct 202608:00:01 CEST · c1a48b74
- 11 Oct 202622:01:57 CEST · 50deb522
- 11 Oct 202608:00:02 CEST · b297cb00
- 10 Oct 202608:00:01 CEST · d77941f9
- 9 Oct 202609:13:03 CEST · c4acaaf1
- 9 Oct 202609:13:03 CEST · ab04ecd2
- 9 Oct 202609:13:03 CEST · 1b189d44
- 9 Oct 202608:00:01 CEST · ce20e166
- 8 Oct 202608:19:28 CEST · ab8d0f6f
- 7 Oct 202615:49:06 CEST · f9ce0158
- 6 Oct 2026Retrospective
- 5 Oct 2026Retrospective
- 4 Oct 2026Retrospective
- 3 Oct 2026Retrospective
- 2 Oct 2026Retrospective
- 1 Oct 2026Retrospective
- 30 Sep 2026Retrospective
- 29 Sep 2026Retrospective
- 28 Sep 2026Retrospective
- 27 Sep 2026Retrospective
- 26 Sep 2026Retrospective
- 25 Sep 2026Retrospective
- 24 Sep 2026Retrospective
- 23 Sep 2026Retrospective
- 22 Sep 2026Retrospective
- 21 Sep 2026Retrospective
- 20 Sep 2026Retrospective
- 19 Sep 2026Retrospective
- 18 Sep 2026Retrospective
- 17 Sep 2026Retrospective
- 16 Sep 2026Retrospective
- 15 Sep 2026Retrospective
- 14 Sep 2026Retrospective
- 13 Sep 2026Retrospective
- 12 Sep 2026Retrospective
- 11 Sep 2026Retrospective
- 10 Sep 2026Retrospective
- 9 Sep 2026Retrospective
- 8 Sep 2026Retrospective
- 7 Sep 2026Retrospective