OpenAI's Astra AI Solves 10 Longstanding Math Problems
OpenAI's Astra model has generated solutions for 10 unresolved mathematics problems, impacting fields like cryptography and quantum computing. This advancement underscores the potential of AI in mathematical research, although concerns about AI reliability persist.

OpenAI's experimental AI model, Astra, has produced solutions to 10 longstanding mathematical problems, including contributions to cryptography, quantum computing, and graph theory. The computational cost for these solutions is estimated at $2,000 under current API pricing.
The generated mathematical arguments were organized into research papers by human collaborators and verified using the Lean proof assistant. The solutions include new bounds for sphere packing and coding problems, a disproof of Connes's rigidity conjecture, and resolutions of several problems posed by Paul Erdős.
Despite these advancements, experts caution that the reliability issues inherent in generative AI remain unresolved. OpenAI plans to provide 100,000 researchers with free access to its AI models by 2027 through its ChatGPT for Academic Researchers initiative.




Comments