OpenAI announced on August 1 that its Astra model had solved ten longstanding mathematical problems spanning high-dimensional sphere packing bounds, non-sofic groups, coding theory, and arithmetic circuit complexity. Mathematicians accused the company of ignoring prior work published as early as 2016 and overstating the novelty of the AI-generated results. The total compute cost for all ten results was reported at approximately 2,000 dollars. By August 6, Scientific American had characterized OpenAI’s handling of the announcement as research misconduct and plagiarism, prompting the company to revise some of its statements. This controversy raises questions about how frontier AI mathematical capabilities may depend on synthesizing decades of uncredited human research.
Source: Read the original article
