DeepMind’s New AI Just Changed Science Forever

DeepMind's new AI agent, Aletheia, can autonomously conduct research and contribute core content to publishable scientific papers. Built on three key innovations — separating the thinking process from the verification step to reduce self-agreement bias, a 100x more compute-efficient reasoning model, and the ability to synthesize dozens of cutting-edge research papers without hallucinating — it has solved 4 open Erdős math problems and co-authored 5 research papers currently under peer review. Independent mathematicians have verified the work for correctness and novelty. The system is rated at 'Level 2' of research autonomy, meaning it can help humans produce publishable research, with fully autonomous groundbreaking discovery (Levels 3-4) still out of reach but potentially approaching fast.