[AINews] Jeff, Sanjay, Oriol, and Quoc depart DeepMind; Demis to Chair; Koray to SVP — what is going on at GDM???
Latent Space Latent.Space ● Covered by 13 sources
Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le are leaving DeepMind to start an AI-for-science company. Demis Hassabis steps back too, becoming Chair while Koray Kavukcuoglu takes the wheel.
Google DeepMind just lost four of the most consequential engineers in the company's history in one swoop, and the timing is impossible to ignore. Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le are founding Discovery Loop, a public benefit corporation aimed at automating machine learning research, science, and engineering. Alphabet is putting money into the round alongside Radical Ventures and Khosla Ventures, which are leading it, plus Lightspeed, Kleiner Perkins, and Doerr Capital. Everyone involved is calling it amicable. Maybe it is.
But the optics are strange. Demis Hassabis, who built GDM into what it is, is moving from CEO to Chair and Chief Scientist of Alphabet, shifting his focus toward long-term strategy and Isomorphic Labs. Koray Kavukcuoglu, previously CTO, steps up to SVP running day-to-day operations, including Gemini and the frontier research teams. That's two leadership changes and a founder walkout landing in the same news cycle, six months after Gemini Pro's last real update.
Context matters here. John Jumper left for Anthropic earlier this year. Noam Shazeer went to OpenAI. David Silver and Denny Zhou have also exited. Add Dean, Ghemawat, Vinyals, and Le, four names that show up on foundational papers going back to Google's earliest deep learning work, and you have a pattern, not a coincidence. One detail worth sitting with: Gemini papers routinely list a thousand-plus coauthors, while Discovery Loop's founding manifesto has exactly four names on it. That contrast says something about why staying inside Google may have felt limiting to people used to owning their work outright.
Still, writing Google off would be premature. They have the compute, the infrastructure, the data, and plenty of talent left. What they don't have right now is a clean explanation for why an internal automated-research effort couldn't have stayed internal. Discovery Loop is explicitly betting that AI-for-science is the next frontier, not another chatbot wrapper, and that's a bet four of the smartest people in the industry are willing to make outside Google's walls rather than inside them.
My take
When your most senior architects choose to build the future somewhere else, even with your blessing and your check in the round, that's not neutral news, that's a verdict on how slow and crowded life inside a giant lab has become. Google will be fine on paper, resources aren't the problem, but the steady drip of departures says something structural is broken, and throwing a chairman title at Demis doesn't fix a talent pipeline that's leaking founders instead of retaining them.
Read more about this at: Latent Space
