150 mathematicians warn governments: don't buy the AI hype

Over 150 mathematicians from around the world have signed an 11-page joint declaration warning governments about the dangers of overestimating AI's capabilities. The document directly responds to OpenAI's claim that its system independently disproved Paul Erdős's 80-year-old mathematical conjecture. The authors believe tech companies exaggerate their products' capabilities due to commercial interests, and they urge politicians to consult with scientists before making policy decisions. Leslie Ann Goldberg, head of the computer science department at Oxford University, notes that distinguishing flawed AI arguments from valid proofs is extremely difficult.

The declaration also demands technology regulations in military and mass surveillance sectors, as unauthorized use of academic works and fake scientific papers threaten the integrity of the field. The mathematicians argue that overhyping AI risks misallocation of research funding and public trust. They call for independent oversight and transparency from AI companies when making extraordinary claims. This warning adds to growing skepticism about AI benchmarking, following previous concerns from researchers about reproducibility and exaggeration in AI performance metrics.