OpenAI released 719 manuscripts claiming solutions to hard math problems, but fell short of guidelines from a Princeton IAS-hosted advisory group, with few chain-of-thought releases and incomplete formalization. A Cambri...
OpenAI has published new findings demonstrating an internal frontier model solving open mathematical problems. The organization also made formal proofs in the Lean language and related research details accessible on GitH...
Researchers from CMU, MIT, NYU, and Stanford developed Ataraxos, an AI system that decisively defeated the world's top Stratego champion using only modest compute resources. By combining self-play with a secondary belief...
Two cryptanalysts used frontier LLMs to decode two historical Enigma messages that had never been broken. OpenAI's GPT-6 Astra worked largely on its own: it did archival research, built an Enigma simulator and recovered...
Anthropic says its Bay Area wet biology lab has found a previously unknown enzyme system in bacteriophage DNA that can cut, copy and paste DNA in ways similar to CRISPR. The company says Claude found it mostly on its own...
OpenAI announced an independent Advisory Group on Mathematics and Artificial Intelligence hosted at the Institute for Advanced Study, following the publication of a claimed solution to the Navier-Stokes Millennium Prize...
NYU mathematician Tristan Buckmaster and Anthropic's Levent Alpöge announced three proofs, including a preliminary result on the Navier-Stokes existence and smoothness Millennium Prize problem, produced with heavy use of...
An Anthropic fellows-program paper, "Automated Researchers Can Reliably Mitigate Alignment Failures," describes AI systems that search literature, propose methods, and run short training cycles to improve model performan...
London startup Inherent, founded by Google DeepMind alumni and recently funded with a $50 million seed round, released an AI agent called Faraday that autonomously reproduces findings from published scientific papers. Th...
Nvidia published research arguing that the "harness" surrounding a model — memory handling, tools, runtime, and a supervisory agent — matters more than the base model for long-horizon agentic tasks. Using a custom harnes...
