OpenAI Withdraws 3 of 722 AI Math Papers After Errors
OpenAI published 722 AI-assisted mathematics manuscripts on October 6, then pulled 3 and revised 14; outside mathematicians questioned the results.
OpenAI published 722 manuscripts of AI-assisted mathematics research on October 6, then withdrew 3 of them within a day after errors were found, according to ZDNet Korea.
One manuscript contained a wrong symbol used in a calculation, which led to the withdrawal of 2 other manuscripts that were based on it. OpenAI also revised 14 manuscripts to supplement proofs or correct explanations.
The research used an unreleased proprietary AI model, which attempted about 4,000 problems across fields including algebra and mathematical logic. Computer verification with Lean was prepared for 300 of the 719 remaining manuscripts, about 42%, and full solution-process summaries were published for only 10 studies.
Researchers from the University of Cambridge and King's College London found at least two discrepancies between OpenAI's written proof and the Lean code for a study on the Navier-Stokes equations.
The independent mathematics and AI advisory group at the Institute for Advanced Study in Princeton advised OpenAI but did not approve the research results or the publication method.
Earlier claim
In the previous month, OpenAI claimed to have solved a problem related to the Navier-Stokes equations.
Quick answers
How many manuscripts did OpenAI withdraw?
OpenAI withdrew 3 of the 722 manuscripts within a day after errors were found, and revised 14 others to supplement proofs or correct explanations.
How many of the manuscripts had Lean verification?
Computer verification with Lean was prepared for 300 of the 719 remaining manuscripts, about 42%.
Who reviewed OpenAI's research?
The independent mathematics and AI advisory group at the Institute for Advanced Study in Princeton advised OpenAI but did not approve the research results or publication method.