kurt.news

Clean, fast AI news without the hype or doom.

Ai

Jeff Dean and Three Google Veterans Leave to Build AI for Science

Jeff Dean and Three Google Veterans Leave to Build AI for Science

Jeff Dean is leaving Google. He joined in 1999 as the company's 30th employee. Twenty-seven years later, he is co-founding Discovery Loop, an AI startup focused on accelerating scientific research, where he will serve as CEO.

Who Is Involved

Dean is not alone. Three other senior Google researchers are co-founding the company with him: Sanjay Ghemawat (Google senior fellow), Quoc Le (founding member of Google Brain), and Oriol Vinyals (senior research scientist at Google DeepMind).

Getting one researcher at this level to leave would be unusual. Getting four is a different kind of signal.

What They Are Building

Discovery Loop is structured as a public benefit corporation. The stated mission is using AI to accelerate scientific research. The company plans to run thousands of experiments simultaneously, partially automating the research process.

The recursive angle is worth noting. Discovery Loop intends to use AI to help build more powerful AI, explicitly removing human iteration from the loop. That is a fairly direct description of recursive self-improvement as a core product strategy, not a side effect.

The Funding

The initial round is co-led by Radical Ventures and Khosla Ventures. Additional investors include Kleiner Perkins, Lightspeed, and Doerr Capital.

Alphabet is also investing. That would be Google's parent company, backing a startup built by four researchers who just left Google. Make of that what you will.

Context

Dean contributed to Google Gemini's multimodal models before departing. The coordinated exit of four researchers at this seniority level is uncommon. The structure as a public benefit corporation distinguishes it from a standard for-profit lab, though what that means in practice depends on execution.

Whether running thousands of simultaneous AI-driven experiments produces meaningful scientific output remains to be seen. It is as much an infrastructure problem as a research one. The funding suggests investors think this team can handle both.

Source: Techcrunch