OpenAI releases another batch of AI mathematics research results, solving hundreds of outstanding problemsOpenAI 发布又一批 AI 数学研究成果，攻破数百个悬而未决难题

In a series of documents containing 722 manuscripts and covering 372 result families, OpenAI announced a number of long-standing mathematical problems solved by an unpublished cutting-edge model. IT House notes that the move continues a series of groundbreaking results that have both impressed and unsettled some in the mathematical community, as well as raising questions about research ethics and academic behaviour. According to AGMAI, the newly formed independent advisory group of elite mathematicians responsible for communicating these results in a responsible manner, this announcement contains answers to “hundreds” of unsolved questions. The results have been expected for weeks, but until then, OpenAI had not specified what problems the model solved, nor given a precise release time. The company said in September that its model “has solved more than a hundred long-standing open puzzles covering the vast majority of branches of mathematics.” The manuscript and related materials released this time also contain a part of the summary of the model inference process, the estimation of computing power consumption, and the statistical data of the number of problems that have been tried to be solved. According to OpenAI, the amount of computing power consumed by an ordinary result is roughly equivalent to the level of computing power that ChatGPT Pro thinks about for three hours in a row. The Advisory Group on Mathematics and Artificial Intelligence (AGMAI) released its first recommendations in late September. The group calls on artificial intelligence laboratories to release mathematical research results in a timely manner through mature academic channels as much as possible, and at the same time, they should disclose detailed information such as the names, prompts, and computing power costs of the models used. The panel also exhorted AI companies “not to use publishing mathematical results as a marketing tool to promote their own models”; the agency believes that such practices will cause serious harm to the mathematical community as a whole. OpenAI describes the workflow of this release as follows: For this release, we published the results in a GitHub code repository, and at the same time developed a document revision and citation related processing specifications. We will also continue to explore other publishing options hosted by the academic community to align them with the guidance provided by the Advisory Board. For the release of the results of subsequent editions, we promise to further improve the quality of the manuscript, optimize the citation situation, mathematical content elaboration, and the format of the results for everyone to understand. Mathematicians also need to evaluate and assimilate these results, so the full impact of these studies may not be visible for some time. Previously, competitor laboratories such as OpenAI and Anthropic have produced a rapidly growing number of mathematical results. The entire mathematical community is still in the digestive stage, including the results of a Millennium Prize puzzle, which is the most well-known unsolved problem in mathematics. This year, the rapid entry of artificial intelligence laboratories into the discipline of mathematics, especially the way they publish information to the outside world, has provoked heated debate on scientific research paradigms, ethical norms, and how companies should thank human mathematicians. Artificial intelligence systems are built on the existing work of human mathematicians, and it is possible to rely on these predecessors to generate new results.