OpenAI announces ten advances in mathematics and theoretical computer science

The Specola’s telescope log records a claim from OpenAI, dated August 1, and relayed via Simon Willison’s site, that its systems produced ten advances in mathematics and theoretical computer science. Willison’s post links to an OpenAI announcement page and situates the claim alongside a related item from days earlier, in which Anthropic reported using its Claude model, under an internal codename Mythos Preview, to discover cryptographic weaknesses. That earlier effort reportedly involved roughly $100,000 in token spending and prompts instructing the model to avoid “low hanging fruit” in favor of substantive research.

Details on the ten mathematical results themselves are not given in the material available here beyond the fact of OpenAI’s announcement and its title. No specific theorems, proofs, or problem areas are described in the source excerpt, nor are the researchers or reviewers named.

The juxtaposition of the two announcements, from OpenAI and Anthropic within the same week, is noted by Willison without further elaboration on methodology or verification. No independent confirmation of the mathematical claims, peer review status, or reproduction of results is mentioned in the available text.

The observatory logs the claim as reported, pending fuller documentation of the specific advances and any external assessment of their validity.