X Discussion: DeepMind VP and Google Cloud Chief Scientist Pushmeet Kohli Says Scientific Discovery Is Seeing a "Genuine Phase Transition," With a Glimpse Coming in Upcoming Releases

Core Background & Industry Pain Points

Claims that AI is speeding up scientific discovery are common, but when a lab leader uses the physics term "phase transition" for the pace of discovery, readers usually take it to mean a change in kind, not just a bit more speed. At 10:15 UTC on 8 November 2026, Pushmeet Kohli, Vice President at Google DeepMind and Chief Scientist of Google Cloud, posted from his personal account that across DeepMind projects they are witnessing a "genuine phase transition" in the pace of scientific discovery, and that "you will see a glimpse of this in our releases in the next few weeks." At capture the post had about 1,996 likes, 149 reposts, 48 quotes, 71 replies, 272 bookmarks and about 73,000 impressions. This is his personal post, not an announcement from @GoogleDeepMind, and it does not name a field, model or discovery.

Architecture Highlights & Internals

The teaser does not describe a new system or model architecture. His X bio lists him as Google Cloud Chief Scientist and DeepMind VP. Google's official author page and his Google Research profile further say he leads DeepMind's AI for Science and Strategic Initiatives Unit, covering biology, materials, fusion, math, software and cybersecurity. Publicly listed prior work includes AlphaFold, AlphaGenome, AlphaEarth, AlphaProof, AlphaEvolve, AI Co-Scientist and SynthID. Those are already public projects and should not be read as a preview of which item will appear "in the next few weeks." In a follow-up posted immediately after the original, he only added that as this AI-powered acceleration unfolds, the focus will be on empowering the broader scientific community to harness these tools responsibly.

Authoritative Benchmarks & Measured Scores

Neither the original post nor the follow-up published a new benchmark, timeline, product name or comparison score. No official note names the tasks, datasets or baselines that this "phase transition" refers to. Third-party guesses that tie the remark to a specific model or discipline are not used here. Like and repost counts at capture only show discussion heat; they are not evidence of a scientific result.

Developer Hands-on Guide

There is no new model, API or paper that maps one-to-one onto this teaser yet. To check the "releases in the next few weeks" he mentioned, wait for a later official Google DeepMind blog post, paper or official-account post; do not treat this teaser as a shipped result. For the directions he has publicly led, see Google's official author page and his Google Research profile. Until a named release appears, "phase transition" should not be written up as a confirmed discovery.