OpenAI's Astra Cracks 10 Open Math Problems — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI says an internal version of Astra, its next major model family, produced results for 10 open problems spanning mathematics, quantum complexity, and theoretical computer science, releasing a paper and reasoning walkthroughs.
- The Decoder and RuntimeWire covered the announcement, with The Decoder framing it as 'AI keeps cracking unsolved math problems' while emphasizing that 'mathematicians have mixed feelings.'
- Thomas Bloom singled out the disproof of exponential bounds for the multicolour triangle Ramsey number as the most surprising result, noting the argument 'looks quite short.'
- Gary Marcus offered qualified praise, writing the result is 'Obviously impressive' but that 'math is different from most other problems.'
- Itai Sher called for a new norm requiring labs to release the full set of attempted problems alongside successful ones, warning that selectively reporting only positive results is 'problematic.'
- Bharath Ramsundar warned the field is building a 'tower of "maybe true" AI results' that should not be considered settled until human mathematicians reach consensus.
- Dean W. Ball noted that 'everyone in the world will soon be able to use the model that made these breakthroughs for every problem they face in life' at costs that 'will fall dramatically in a matter of months.'
Why it matters: The announcement sets a new bar for AI in mathematical reasoning, but Gary Marcus, Itai Sher, and Bharath Ramsundar are openly challenging OpenAI's selective-reporting approach and demanding human verification before results are treated as settled science. With Astra soon reaching a wider audience at falling costs, the gap between capability demos and validated scientific claims is widening.
Ask SkimNews
