AI & Maths watch GitHub

Archived edition

Updates on artificial intelligence for mathematical research, proof discovery and formal verification, with links to papers, code and tools.

Research updatesSelect an entry to read more

2 entries · 6 linked sources Partial coverage
Ben Green offers a five-page AI-assisted variant of the claimed Landau–Siegel zero argument Ben Green Published: date not established4 sources

Green’s publicly available note attributes its argument to a session with ChatGPT 6 Astra followed by his own rewriting. It works in a quadratic field rather than the original argument’s biquadratic field and uses an explicit determinant construction. Its stated theorem bounds the weighted sum over primes with character value −1; it refers to the earlier OpenAI manuscript for the standard deduction concerning Landau–Siegel zeros. Bloom circulated the note on October 18 and explicitly qualified its self-containedness.

Research relevance:

A short exposition could make a major claimed analytic-number-theory argument more accessible to expert scrutiny, while illustrating AI-assisted simplification and human rewriting.

Claim status:
Written argument available and directly consulted; not merely a social announcement. Independent refereeing and formalization were not established. This digest does not certify the claimed exclusion of Landau–Siegel zeros.
Limitations:
Original publication date unknown; October 18 is Bloom’s announcement date. The note may predate the window. Its final analytic deduction is referenced elsewhere, and the original manuscript was not checked. Green’s official pages consulted did not corroborate this particular upload.

Sources & further reading

Link to this entry
Francesco Maggi proposes disclosure standards and shared infrastructure for AI-assisted mathematics Francesco Maggi Published: 2 sources

In a guest essay on Tao’s blog, Maggi calls for reproducibility and disclosure standards, resources for understanding AI-generated arguments, and publicly governed research infrastructure. He identifies practical questions about responsibility for correctness, attribution of underlying ideas, confidentiality of unpublished work, and training researchers whose expertise extends beyond producing proofs.

Research relevance:

Useful for departments and research groups setting AI workflows: it connects access to models with independent scrutiny, accountable authorship, exposition and researcher training.

Claim status:
Dated research-policy essay and proposals, not a theorem announcement or an independent verification report.
Limitations:
The recommendations are normative. The essay’s descriptions of recent mathematical breakthroughs and model capabilities were not independently validated here; no new proof or released tool accompanies the proposals.

Sources & further reading

Link to this entry
Coverage and limitations
  • Window: 2026-10-17 20:00:01 to 2026-10-19 09:47:43, Europe/Paris. Search indexing dates were not treated as publication dates.
  • Used public web search and directly consulted Green’s five-page PDF, his Oxford profile and website, Bloom’s website and AI reading list, Tao’s blog and Maggi’s guest essay, Xena, Buzzard’s Imperial website, and Frieder’s blog. Buzzard’s official website confirms the Xena blog address. Frieder’s retrieved index showed no recent entries and carried a month-old crawl timestamp.
  • Checked Green’s identity against Oxford and Maggi’s against his university website. Bloom’s official website confirms his mathematical affiliation and Erdős Problems project, but did not independently establish ownership of @thomasfbloom. Other supplied social identities were not independently authenticated; API account resolution was not treated as authentication.
  • Supplied X collection status was ok with posts for thomasfbloom, littmath, alexkontorovich and boazbaraktcs. Bloom’s posts supplied the proof lead; Litt’s and Barak’s supplied posts were outside the editorial scope. Kontorovich linked the previously covered October 16 formalization essay, which was not repeated. These were supplied API records, not freshly browsed complete timelines.
  • Supplied X collection status was ok with empty results for scottnarmstrong, wtgowers, sebastienbubeck, f_charton, leonard41111588, leanprover, projectnumina, harmonicmath, axiommathai and epochairesearch. Each collection was bounded, excluded replies and reposts, and could include cached posts. Numeric pagination caps and completeness were not supplied; empty results do not establish inactivity.
  • The __alpoge__ X collection was unavailable after a TimeoutError, leaving its timeline incomplete. No additional collector budget limits or skipped accounts were reported. No direct X access is claimed beyond accounts with supplied API posts; indexed X results were used only as leads.
  • The xenaproject.bsky.social Bluesky collection succeeded with no posts; replies and reposts were excluded and at most 300 posts were checked. The supplied mathstodon.xyz collection failed with an HTTPError under the same exclusions and cap, so Tao’s Mastodon activity was unavailable.
  • Hexagon’s PDF was readable, but attempts to retrieve its landing and abstract pages returned internal errors. Consequently its original posting date could not be established. Search results for related blogs and project material were screened as leads; snippets alone were not treated as substantive verification. Most listed accounts’ official sites and project repositories were not individually audited.
  • Previously covered material was excluded unless there was a substantive new development. No proof was verified and no Lean code was executed.

Requested coverage window: 2026-10-17T20:00:01.679423+02:00 — 2026-10-19T09:47:43.225596+02:00.

Public web research; coverage is not exhaustive. A source link is not a certification of a claim.

Archives 55
Older editions (25)