Jeff Dean and other top AI researchers are leaving Google to launch own startups
Jeff Dean, Google's 30th employee and a foundational figure in AI, is leaving Google to co-found Discovery Loop, a public benefit corporation focused on accelerating scientific research through AI The startup aims to automate complete experimental loops using high-octane algorithms that can initiate and iterate thousands of experiments simultaneously, reducing reliance on slow human-driven iterations Discovery Loop is also pursuing recursive self-improvement — using AI to design more powerful AI
Analysis
TL;DR
- Jeff Dean, Google's 30th employee and a foundational figure in AI, is leaving Google to co-found Discovery Loop, a public benefit corporation focused on accelerating scientific research through AI
- The startup aims to automate complete experimental loops using high-octane algorithms that can initiate and iterate thousands of experiments simultaneously, reducing reliance on slow human-driven iterations
- Discovery Loop is also pursuing recursive self-improvement — using AI to design more powerful AI systems without human intervention
- The company has secured significant backing from Alphabet, Radical Ventures, Khosla Ventures, Kleiner Perkins, Lightspeed, and Doerr Capital
- Co-founders include Google veterans Sanjay Ghemawat, Quoc Le, and Oriol Vinyals, bringing deep expertise in systems engineering and AI research
Why It Matters
This marks a major shift in the AI landscape as one of the field's most influential figures pivots from building AI for search and language to using AI as a tool for autonomous scientific discovery. The move signals the industry's growing belief that the next frontier for AI lies not in answering questions but in making discoveries — a paradigm that could reshape how science and engineering are conducted at scale.
Technical Details
- Discovery Loop plans to leverage massive computational scale to automate complete experimental loops, replacing the traditional sequential human iteration model with parallel, AI-driven experimentation
- The system targets recursive self-improvement, where AI systems assist in designing and building more capable AI, potentially eliminating human iteration from the improvement cycle entirely
- The founding team brings decades of experience in large-scale systems (Google search infrastructure), multimodal models (Gemini), and deep learning research (Google Brain, DeepMind)
- The approach builds on years of experimental work in AI-accelerated science, now transitioning from limited commercial application to a dedicated, well-funded startup focused on this mission
Industry Insight
- The departure of Jeff Dean and key Google/DeepMind researchers represents a significant brain drain from one of the world's largest AI labs, potentially weakening Google's competitive position in foundational AI research
- The focus on autonomous scientific discovery signals where top-tier AI talent sees the next high-impact frontier — suggesting that AI for science may become a major competitive battleground in the coming years
- The strong investor backing from top-tier VCs and Alphabet itself validates the thesis that AI-driven automation of the scientific method is a viable and valuable commercial direction, likely attracting further investment and talent into this space
Disclaimer: The above content is generated by AI and is for reference only.