Google's AI patriarch leaves: is recursive self-improvement the next frontier
jeff dean quits google to launch discovery loop, aiming to automate scientific loops via recursive self-improvement. this signals a shift from language models to systems that design future ai.
jeff dean, google's 30th employee and a figure central to its search infrastructure and gemini ai, is departing to launch discovery loop with co-founders quoc le and oriol vinyals. the startup, funded by radical ventures, khosla ventures, and alphabet, plans to use algorithms to iterate thousands of experiments at once, targeting a bottleneck in traditional scientific discovery. beyond accelerating science, the firm explicitly targets recursive self-improvement—using ai to build more capable ai without human intervention. dean's exit suggests that the highest-leverage ai research is shifting from chat interfaces to autonomous scientific loops that could exponentially accelerate technological progress.