OpenAI's Astra Model Sparks Controversy Over Mathematical Claims

OpenAI announced on August 1 that its new Astra model had solved 10 longstanding problems in mathematics and theoretical computer science. However, the reaction from the mathematical community was swift and critical.

Prominent researchers accused OpenAI of downplaying prior work, failing to properly attribute existing contributions, and overstating the novelty of its AI-generated results. According to OpenAI, the Astra model tackled problems such as high-dimensional sphere packing bounds and coding theory advances, claiming there had been no progress on these issues for at least a decade.

Researchers like Steven Miller from Yeshiva University and Francesco Fournier-Facio from the University of Cambridge pushed back, arguing that Astra's results built on prior work that OpenAI either ignored or underrepresented. Notably, relevant research by Andreas Thom and Miller himself dates back to 2016, directly informing the territory Astra claimed to have conquered.

By August 6, Scientific American characterized OpenAI’s handling of the announcements as “research misconduct” and plagiarism, prompting the company to revise some of its statements.

This incident is not OpenAI's first controversy over mathematical claims. In previous instances, the company faced backlash for similar issues, raising ongoing concerns about how AI models synthesize existing knowledge.

Ultimately, the $2,000 compute cost for the 10 results raises questions about the nature of AI contributions in research, suggesting that if Astra synthesized decades of human work without proper credit, the cost reflects less on AI capability and more on the repackaging of existing knowledge.