The Decoder reports a 769-task study of building an AI model: agents proposed up to 55% of methods, but humans made over 85% of final decisions.
A concrete data point on where agents actually sit in AI research workflows: heavy on generating candidate methods, light on deciding.
Microsoft Research published RetroChimera in Nature, an ensemble retrosynthesis model it says outperforms prior models in blind tests, and released its code and weights.
Researchers from MIT, Carnegie Mellon, NYU, and Stanford published an AI in Nature that defeated top-ranked human Stratego players by a large margin while costing far less to train.
AWS's Machine Learning Blog published a how-to for a Bedrock Knowledge Bases claims assistant that answers natural-language questions with citations, using synthetic records rather than a customer deployment.
Anthropic shared GitHub code and a technical report for an 'uplifting biomolecular modeling' project, without disclosing results in the post.
OpenAI's DevDay 2026 recap post says the event included more than 20 announcements spanning GPT-6 Astra, ChatGPT, Codex, APIs, security, and new builder tools.
OpenAI's @OpenAI account says GPT-6.1 Sol is a significant upgrade over GPT-6 Sol in coding, computer use and complex professional work, nearing GPT-6 Astra on some benchmarks at substantially lower cost.
