OpenAI officially announces the next generation model Astra! Achieve breakthroughs in 10 unsolved problems, with a 249-page paper attached

Zhixi Compiler | Yang Jingli Editor | Li Shuiqing Zhixi News on August 2, early this morning, OpenAI revealed the next-generation model Astra, and announced 10 new achievements in mathematics and theoretical computer science achieved by its internal version, involving high-dimensional geometry, coding theory, group theory, arithmetic circuit complexity, quantum complexity, lattice cryptography and extreme combinatorics. The results announced this time demonstrate Astra's ability to explore cutting-edge scientific research issues. The relevant issues have all been open for at least 10 years, and the core conclusions have not made progress for a long time. Most issues have a longer research history. OpenAI used the same model to organize the argument into a 249-page paper, and also attached a 62-page explanation of the problem-solving ideas. Calculated according to the Sol API rate, the total cost of Tokens required for the model to find these solutions is approximately US$2,000 (approximately RMB 13,500). Before OpenAI officially announced these results, Astra had already entered the vision of US policymakers. Yesterday, according to The Information, OpenAI CEO Sam Altman previewed Astra this week in Washington, USA, to policymakers including White House officials, cabinet members and members of Congress. It is worth noting that this is not the first time that OpenAI has announced the involvement of AI in solving mathematical problems. In May this year, OpenAI published a counterexample to the Erdosh unit distance conjecture discovered by AI. The result was discovered during evaluation of an as-yet-unreleased model and has spurred several subsequent studies in mathematics and theoretical computer science. Paper link: https://cdn.OpenAI.com/pdf/ten-proofs-oai.pdf Lean certificate link: https://github.com/OpenAI/ten-proofs Model problem-solving ideas: https://cdn.OpenAI.com/pdf/reasoning-walkt