Sakana AI says Schmidhuber will help guide its newly formed RSI Lab, whose objective is a compounding cycle of scientific discovery aimed at improving machine intelligence. The company is assembling experts for the lab in Tokyo.
A paper from Google and colleagues describes Harness-Zero, which uses an optimized agent harness only during training. After that harness is removed, macro task success rises from 23.3% to 44.3%, above the 41.7% reached by the base model with the harness attached.
Anthropic says the enzyme’s gene sits beside a long array of repeating DNA that looks somewhat similar to CRISPR. It says the system’s function is not yet understood, while the few known systems sharing its features can all cut, copy, and paste DNA.
Polymarket says Anthropic’s new wet lab made its first discovery after nearly 1,000 Claude agents worked autonomously for 21 hours.
DAIR.AI reports that a Stanford and Together AI paper found o3-mini, Claude Sonnet 4, and DeepSeek-V3 averaged 66.7% across five math and physics benchmarks as a self-organizing team, above 48.8% for the strongest member alone and 59.0% for a perfect router.
Amazon Science said Amazon Bio Discovery developed MochiBind, CA-MAP, and an agent-guided design system to accelerate antibody drug design, reporting 46 lab-validated hits against a novel cancer target.
OpenAI said it is working with an independent advisory group of mathematicians to help it responsibly share advances in AI and mathematics. The group will advise on assessing and communicating new mathematical results, upholding academic and professional standards, and building tools for mathematical research and learning.
