What's this about?
Jeff Dean — arguably Google's most senior and famous engineer — announced he's co-founding Discovery Loop, a Public Benefit Corporation whose mission is to automate machine learning, science, and engineering to accelerate discovery. His co-founders are Sanjay Ghemawat, Oriol Vinyals, and Quoc Le; the four have worked together for 14–30 years and built much of Google's foundational infrastructure and AI. The stated approach: automate the experimental loop, starting with ML research itself, then generalizing to other sciences.
What exactly is "automating the experimental loop"?
Science and ML progress by a loop: form a hypothesis, run an experiment, read the result, refine, repeat. "Automating" it means handing that whole cycle to AI systems — the model proposes experiments, runs them, analyzes outcomes, and decides what to try next, with humans setting direction. It's the same idea behind this week's OpenAI math results and Sequoia's "automated AGI lab" episode, now the explicit mission of a new company led by legends.
What's the community saying?
- Founding-team awe: Dean/Ghemawat/Vinyals/Le leaving to do this together is treated as a landmark, not a routine startup.
- The mission's stakes: hardmaru's framing — automating the scientific method could "profoundly alter the trajectory of AI over the next few years."
- Structural read: a Public Benefit Corporation (not a pure startup) signals they're positioning recursive self-improvement of research as a civilization-scale project.
- Quiet skepticism about whether "automate science" survives contact with messy real-world experiments beyond ML.