344
Productivity & Workflow355
Automation & Workflow225
Software Development251
Marketing & Growth193
AI Infrastructure & MLOps175
Writing & Content Creation204
Data & Analytics142
Photography & Imaging156
Design & Creative170
Customer Support133
Sales & Outreach125
Voice & Speech135
Education & Learning131
Operations & Admin87
OpenAI shared hundreds of AI-made math proofs, but a Princeton-hosted advisory group and outside researchers raised concerns about transparency and verification.
In short: OpenAI released hundreds of claimed solutions to hard math problems, and researchers say the release does not fully follow new community guidelines for trust and review.
OpenAI published 719 manuscripts that it says contain solutions, or “proofs,” for difficult open math problems. A proof is the step by step reasoning that shows a result is true, like showing your work on a test.
OpenAI said it consulted a group called the Advisory Group on Mathematics and Artificial Intelligence (AGMAI), hosted by Princeton University’s Institute for Advanced Study. AGMAI recently published guidelines for companies that use AI to work on advanced math.
TechCrunch reports that OpenAI’s release appears to miss key parts of those guidelines. AGMAI’s first request was to stop testing open research problems on proprietary models, meaning systems outsiders cannot fully inspect. OpenAI’s release says it is evaluating its proprietary models using open math problems.
The guidelines also stress the need for human understanding. Only 10 of the 719 manuscripts included the model’s “chain of thought,” meaning the detailed internal reasoning the model used to reach an answer.
Another issue is “formalizing” proofs, which means translating them into a strict computer-checked format. OpenAI used Lean, a programming language used to check proofs like a spellchecker for logic. A separate paper from researchers at the University of Cambridge and King’s College London reported discrepancies between the plain English explanation and the Lean code in an OpenAI-linked solution related to the Navier-Stokes equations.
Mathematicians like Terence Tao and Harvard professor Melanie Wood argue that AI results still need the same careful peer review and follow-up work humans do. The next question is whether OpenAI and other labs will share more supporting details, fund human review, and make it easier to match each written proof to its computer-checked version.
Source: TechCrunch AI