10 stories
A Harvard working paper found Claude Opus 4.8's peer reviews beat paid, highly rated human reviews, winning 132 of 132 AI-judged pairs and 26 of 30 blind editor judgments.
Microsoft Research's Xinzhi Zhang and Zhao Song proved the 1995 KLS conjecture in high-dimensional geometry, with AI agents exploring 100+ approaches under human direction.
Anthropic and Theorem used Claude to prove 239 entry points across 123 open-source libraries never crash on inputs under 64 MB, with about 180 million lines of machine-checked proof.
In a Science study, Microsoft AI's MAI-DxO diagnosed 83% of complex NEJM cases correctly, vs 36% for generalist physicians, in a controlled text-only benchmark.
Anthropic says an internal Claude research model designed 17 lab-confirmed protein binders against scorpion neurotoxins, an early step toward possible new antivenoms.
Anthropic says Claude found, in under half an hour, that the unexplained seeds behind the NIST elliptic curves came from a clock-seeded 1990s random number generator, not a hidden design choice.
Given plain-language specs and an automated checker, Claude Mythos 5 found quantum circuits using fewer gates and qubits than the best published designs on seven of eight tasks.
A new Simons Foundation program, backed by OpenAI and Anthropic, will fund workshops and proposals to help mathematicians evaluate AI-assisted results and set standards for AI use.
OpenAI's new public GitHub repo holds 719 math manuscripts and proof artifacts from an internal model it hasn't released. About 42% of top-line results are formalized in Lean.
A Stanford Digital Economy Lab study of 41 countries finds AI-adopting firms grew senior headcount 6.7% while the junior share of staff shrank.
