Mathematicians want proof OpenAI didn’t use their work

Summary

Mathematicians are questioning OpenAI's data sources for its AI-driven mathematical discoveries. Andreas Thom alleges OpenAI's models may have used his and colleagues' unpublished work without proper credit. OpenAI denies direct user data access but admits de-identified data might improve models, a distinction mathematicians find dishonest.

Key Takeaways

1. Mathematicians are challenging OpenAI regarding the data used to train its AI models, suspecting their unpublished work may have been incorporated without consent or credit.
2. OpenAI acknowledged that its non-sofic groups result built heavily on previous work by mathematicians Andreas Thom and Gábor Kun, but was criticized for not initially acknowledging their contributions.
3. Mathematicians worry that if AI companies use user research to race for publications without proper disclosure, it could lead to a more secretive scientific field.

Statement Breakdown

• Claimed Facts: 40% of statements the article presents as facts
• Opinions: 45% of statements classified as editorial or subjective
• Claims: 15% of statements surfaced for additional reader evaluation

Credibility & Bias Reasoning

Credibility assessment: The article presents claims from mathematicians and OpenAI's responses, offering a balanced view. However, it relies heavily on allegations and lacks direct verification of OpenAI's data usage. The reporting is objective, but the core issue remains unproven.

Bias assessment: Investigative Skepticism. The article leans towards questioning OpenAI's practices, framing the narrative around accusations of dishonesty and lack of transparency. It highlights the mathematicians' concerns more prominently than OpenAI's defense.

Note: This article reports on allegations of data misuse by OpenAI. While it presents concerns from mathematicians, direct proof of OpenAI's data usage is not provided. Readers should approach these claims with a critical eye.

Credibility flag: Allegations, Needs Verification

Claimed Facts (5)

• This states a verifiable action taken by Andreas Thom.
• This presents a factual statement about OpenAI's announcement and its acknowledgment of prior work.
• This provides a definition of a technical term used in the article.
• This is a direct quote from OpenAI's public statement, presented as a factual claim.
• This is a direct quote from OpenAI's public statement, presenting their nuanced position.

Opinions (7)

• The word 'challenging' implies a subjective interpretation of the researcher's actions.
• The terms 'bitter row' and 'accusing' frame the situation with a degree of subjective interpretation.
• The phrase 'struck by' and the subjective assessment of techniques as 'neither the most obvious nor the most promising' indicate an opinion.
• The statement 'did not satisfy Thom' reflects his personal reaction and interpretation.
• This is a direct statement of personal judgment and accusation.
• The phrase 'ethically indefensible' is a strong moral judgment.
• The phrase 'sour taste' is an idiom expressing a negative subjective feeling.

Claims (5)

• This is a broad generalization about the capabilities of all researchers without specific evidence.
• This places a burden of proof on OpenAI that may not be practically feasible or legally required, framing the situation as a definitive denial without proof.
• While conceptually true, this statement is presented as a definitive refutation of OpenAI's technical process without understanding the specifics of their de-identification methods.
• This is a strong, accusatory statement that labels an answer as 'plainly dishonest' without providing concrete evidence of intent or maliciousness.
• This is a speculative prediction about future behavior in an entire scientific field, based on current events.

Key Sources

• Andreas Thom — Mathematician
• OpenAI — AI Research Company
• Gábor Kun — Mathematician
• Tristan Buckmaster — Mathematics Professor, New York University
• Sébastien Bubeck — OpenAI Researcher
• Mark Sellke — Statistician, Harvard
• Levent Alpöge — Researcher, Anthropic
• The Verge — Technology News Outlet
• Robert Hart — Author