SuperintelligenceMONITOR
Published by Center for Artificial WisdomLatest newsletter: Issue #1 · draft v1.11

Documentary follow-up · originally prepared 2026-10-17

Who will check AI-generated mathematics?

Three researcher-authored papers show different kinds of scrutiny and use. A funding commitment could support more of that work; neither is a blanket certificate of correctness.

Draft v1.1 · human editorial approval pending. Source review through 2026-10-18.

Updated 2026-10-18 · What changed and earlier wording.

A mathematician has rewritten one of OpenAI’s arguments. A survey maps ideas across 38 groups of manuscripts. A third paper uses an announced result as an explicit assumption. These are different contributions to understanding the release—and they leave different questions open.

When a manuscript becomes a fact people rely on

The withdrawals make the distinction between a generated argument and established knowledge consequential. Proof checks address correctness under assumptions; independent use tests whether a result helps researchers. Treating a catalogue count as settled scientific progress would skip those steps. That is a risk in interpreting the release, not a diagnosis of the model.

Do researchers and downstream systems preserve the assumptions, corrections and limits when reusing these results? Evidence of that practice would inform how AI affects judgment. It would still leave effects on enduring suffering unmeasured.

Editorial inquiry added October 20, 2026; original evidence cutoff and ratings unchanged. Provisional rationale.

Researcher-authored proof analysis · preprint; correctness not independently established by SIM

Making one argument inspectable

William D. Banks’s October 11 preprint lists a University of Missouri affiliation. It rewrites an OpenAI argument addressing a longstanding obstacle to estimating how prime numbers are distributed. Banks credits the original mathematical ideas to OpenAI and supplies a direct presentation with explicit numerical bounds.

Our interpretation: making each step easier to inspect is a concrete contribution to mathematical scrutiny. A written argument and a recorded run of a formal proof checker provide different evidence. Our source review does not establish the latter, certify the argument or validate the rest of the release. Further expert examination of the steps and bounds would test this account.

Researcher-authored, AI-assisted survey · preprint; no formal checker run reported

A map of ideas is not a proof check

Ying Xie’s October 12 preprint, listing Kennesaw State University, surveys 38 selected groups of manuscripts. It extracts candidate mathematical tools and compares them with earlier work. Its acknowledgments disclose Claude assistance with extraction and literature comparisons, and GPT assistance with review and revision. The author retains responsibility for the research and interpretation.

The survey explicitly says it did not build or run Lean formalizations. Lean is software for expressing and checking mathematical proofs. The sample favors prominent problems, and the novelty judgments are provisional. Our interpretation is that this can help readers locate arguments and dependencies; it cannot be counted as 38 independently verified results. Reproducible checks and broader literature review remain next evidence.

Author-reported conditional application · preprint; verification of the assumed result not established

Using a result while keeping its uncertainty visible

Ruihan Chen’s October 7 preprint studies a particular family of elliptic curves, mathematical objects described by equations. It derives some conclusions by explicitly assuming a statement announced in OpenAI’s release. The paper separates those conclusions from parts that do not depend on the assumption.

This is a documented example of a researcher exploring what the released work could make possible. It is not a separate proof of the borrowed result. Our interpretation: a chain of useful consequences can still depend on an unsettled link. The next question is whether that link survives independent checking, and which conclusions survive if it does not.

What each source can establish

Different contributions require different follow-up evidence · source dates October 7–15, 2026
Work in the recordWhat it contributesEvidence still needed
Banks preprint · October 11A written account of one argumentExpert scrutiny; a recorded formal check if that is the claim
Xie survey · October 12An AI-assisted map of selected ideas and dependenciesProof checking and fuller novelty review
Chen preprint · October 7Consequences conditional on a released statementVerification of the statement used as an assumption
Partnership announcement · October 15A commitment of funding and resourcesAwards, access terms and resulting work

Our synthesis: support, explanation, conditional use and verification are distinct steps. These sources give concrete examples of the first three; this review does not establish the independent formal proof-checking run required by the original forecast.

The papers show kinds of work that support could make possible. The funding question is who chooses that work, who can inspect it and who gets access to the tools.

What the partners announced

The work of understanding AI-generated mathematics now has a new source of support. The Simons Foundation says a partnership with Anthropic and OpenAI will commit $15 million in combined funding and resources. The foundation will manage and award the funding, including through an open call for proposals.

The October 15 announcement describes support for workshops and initiatives to evaluate AI-assisted results, learn tools and shape standards. Anthropic and OpenAI will also provide tools and computing resources to foundation-supported researchers.

In its own update, OpenAI states a $5 million contribution to the same initiative. The two announcements describe the same initiative, so we treat the $5 million as part of the partnership’s total, not an additional commitment. The foundation describes its $15 million figure as combined funding and resources; the announcement does not make it a record of grants already awarded.

Support and endorsement are different decisions

The Advisory Group on Mathematics and Artificial Intelligence’s September 29 recommendations called for AI labs to support human understanding while leaving funding decisions to established nonprofit institutions. They also asked labs to stop testing advanced mathematical problems on inaccessible proprietary models. Accepting support, the group wrote, would not legitimize the labs’ practices.

That is an important limit on the comparison. The new partnership resembles the recommended separation between a lab’s contribution and the allocation of support. This is our interpretation of the announcements, not a finding that the program satisfies every recommendation or that its eventual review process will be independent in practice.

What this changes in our assessment

The original mathematics assessment asked whether new results would survive scrutiny and become useful to other researchers. Funding can help make that work possible: people need time, access and opportunities to explain arguments to one another. Whether those conditions produce reliable new knowledge is a further question.

This announcement adds an institutional commitment to the record. It does not supply an independent proof check, establish novelty, or document downstream mathematical use. Our impact, evidence and attention ratings remain unchanged. Neither the original formalization forecast nor its unactivated successor is resolved by a funding announcement.

What to look for next

Look for the published call, award decisions and recipients; the terms of tool and compute access; and resulting explanations, formal artifacts, corrections and uses. A successful unaffiliated proof check would answer a different question from an announced workshop. The evidence should show which result was checked and how.

SIM follows these conditions because a desirable future with AI depends on more than the production of answers. It also depends on people’s ability to inspect, understand and use them. This case does not establish an overall wise trajectory or validate CAW’s framework.

Scope of this follow-up

The original funding account uses partner announcements and the advisory group’s recommendations reviewed October 17. The separately dated addition reviews three preprints posted October 7, 11 and 12, through October 18. This is documentary reporting: SIM has not verified disbursement, grant outcomes, proof correctness, priority or a formal proof-checker run. No new overall wisdom-trajectory judgment follows.

Prepared with AI assistance and review of the linked primary documents. Human editorial approval is pending. The survey’s own AI assistance is disclosed separately beside its findings. No interviews, grant receipts or independent proof reproductions by SIM are claimed.

Current assessment and dated follow-up · Preserved original follow-up · Revision record · Current newsletter draft

Which part was hardest to follow?

Send the page URL and a short explanation of what was unclear. Your feedback can help us choose what to clarify. Named public credit requires your permission.

Email feedback · Use the publisher’s contact page

This page sends no feedback automatically and adds no audience tracking. You choose whether to send a message.

SI Monitor by email

Understand what changed—and why it matters.

Follow the evidence shaping our future with superintelligence. Our planned weekly newsletter explains the findings and choices that could shape a happier future with less suffering. The website’s daily briefing reviews sources between issues. The publication is in public preview. Issue #1 is still an unsent review draft. A launch date and regular send day have not been set; material corrections receive a notice when needed.

Brevo handles signup and email delivery. You can unsubscribe using the link in each newsletter. Read our subscription privacy information before submitting.

The publisher reported a successful signup and email confirmation test on October 15. Campaign delivery and unsubscribe checks remain incomplete.

Continue to signup

You’ll open Brevo’s hosted form. Enter your email address there and follow the confirmation instructions. Reading this page does not load the signup service.

If confirmation does not arrive, check your spam folder or contact us for subscription help.

Preview the newsletter first