Stories
OpenAI Unveils Hundreds of Math Results, Prompting Verification Concerns
- •OpenAI published 719 manuscripts claiming advances on longstanding mathematical problems, generated by an experimental AI model.
- •The massive release has sparked debate regarding verification, as only a fraction of the proofs were formalised or included chains of thought.
- •Mathematicians like Terence Tao and a Princeton advisory group emphasize the need for human understanding and independent scrutiny of AI-generated proofs.
- •Discrepancies found between natural-language proofs and computer-checked Lean code highlight the risk that formal verifications may not match written claims.
- •Ethereum Foundation researcher Justin Drake warned the blockchain industry to prepare for potential cryptographic vulnerabilities driven by advanced AI reasoning.
Key Questions Answered
OpenAI published 719 manuscripts containing claimed advances on longstanding mathematical problems generated by an experimental model. However, concerns over verification complicated the reception after researchers identified inconsistencies between written proofs and their computer-checked counterparts, with only 10 manuscripts including the model's chain of thought and just 42 percent of the proofs formalised.
The Advisory Group on Mathematics and Artificial Intelligence issued recommendations in late September emphasizing independent scrutiny, formal verification, and ensuring human mathematicians can understand and take responsibility for results. The group also called on laboratories to stop evaluating proprietary models against open mathematical research problems.
Justin Drake warned that rapid progress in AI-assisted mathematical reasoning could create new risks for cryptographic systems, raising the possibility that ECDSA security could fail before quantum computers arrive. He urged major cryptocurrency holders to consider moving funds to fresh addresses whose public keys remain hidden behind a hash.
