OpenAI, PerturbAI and Scripps Research ran large numbers of distributed AI agents over data from millions of brain cells. The project surfaced promising research leads, an example of agents being used to sift datasets too large for researchers to explore by hand.
Anthropic's Claude independently designed protein binders aimed at deadly scorpion neurotoxins, producing more than 318,000 designs of which 211 were lab-tested. Seventeen met the criteria, and the two strongest bind very tightly to Cn2, a Mexican scorpion toxin, though it is still unknown whether they would work as an antivenom.
OpenAI has published hundreds of mathematical proofs generated by its AI. The technical feat is impressive, but researchers call it an unverifiable black box, and a group is now demanding to audit the results.
OpenAI has invested in a startup that uses brain implants to detect cancer. Wired describes it as the latest in a growing series of bets by AI companies on brain data.
Oracle and AstraZeneca have announced a collaboration to use AI and real-world health data to optimize clinical trials.
Thore Graepel, who helped build Google DeepMind's AlphaGo, has founded Metis Reasoning in London. The lab's starting premise, set out in a white paper, is that large language models cannot be trusted to reason, and it wants to apply AlphaGo's lessons to general reasoning.