OpenAI Updates Math Repository with New Formalizations and Withdrawals

OpenAI has updated its public mathematics repository on GitHub, introducing six new Lean formalizations and 19 modifications while withdrawing three specific mathematical manuscripts. According to Dan Roberts, a scientist at OpenAI, the repository now has approximately 42% of its top-line results formalized.

Withdrawn Mathematical Manuscripts

OpenAI has officially withdrawn three manuscripts from its collection. The withdrawn works focus on complex algebraic geometry and Hodge theory:

• Algebraicity of Weil classes on split abelian eightfolds
• Algebraicity of Kuga–Satake Correspondences for K3 Surfaces
• The rational Hodge conjecture for products of K3 surfaces

The company stated that the repository will continue to be updated as new formalizations are completed and errata are identified.

Methodology and Model Usage

Results in the repository were largely generated using an unreleased internal OpenAI model. The process involved posing approximately 4,000 problems to the model, with each result requiring an average of three hours of "ChatGPT Pro thinking compute."

OpenAI has released results for approximately 700 of these 4,000 problems. The goal of the project is to move toward full formalization of mathematical results using Lean, a theorem prover and programming language designed to verify the correctness of mathematical proofs.

Community Discussion and Technical Critiques

The update has sparked significant discussion within the technical and mathematical communities regarding the intersection of AI-generated proofs and traditional peer review.

Software Engineering vs. Mathematical Proof

Observers have noted that OpenAI is treating mathematical proofs similarly to software releases, utilizing GitHub for version control and iterative updates.

"This is what it looks like when software engineering practices meet mathematics. 'openai/math release 1.3.42: retracted papers 139 and 140, fixed a sign error in paper 47, restored previously retracted paper 85, refactored the arguments in paper 101'."

Verification and Trust

There is ongoing debate regarding whether the withdrawn papers were Lean-checked before their initial publication. Because Lean is a formal verification tool, a proof that passes the Lean checker is mathematically guaranteed to be correct relative to the formalization. The presence of withdrawals and "sign errors" suggests that either the initial results were not fully formalized in Lean or that the formalization did not accurately reflect the intended mathematical problem.

Impact on Research Workflow

Some critics argue that the high volume of AI-generated mathematical claims could create a "jammer" effect in the research community, potentially wasting mathematician hours as experts attempt to verify "slop" that may contain fundamental errors.

"Massive cognitive offloading and possibly deeply net negative in terms of mathematician hours and attention depending on how far this goes."

Conversely, some users have praised the transparency of logging withdrawals publicly, noting that such openness is rare in traditional academic publishing and provides a better audit trail for AI-generated research.