saved
Discovery Loop — Continuous Exploration
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
Discovery Loop, a company founded by Jeff Dean, Sanjay Ghemawat, Quoc Le, and Oriol Vinyals, is building AI systems for closed-loop discovery. The systems would propose an experiment, run it, evaluate the result, and use that evidence to choose the next one. The company plans to start with machine-learning research, use the system on its own technology stack, and later take on measurable problems in science and engineering. Its bet is that running thousands of experiments in parallel can let a handful of people outpace today’s large research teams.
ideas
- Automate the loop, not one task. The system is meant to propose, execute, evaluate, and refine experiments as a continuous process rather than assist a single stage.
- Parallelism changes research throughput. Running thousands of measurable experiments at once could replace slow sequential iteration with rapid empirical search.
- Start where the builders are their own customer. Machine-learning research gives the team a domain where it can test the system while improving the infrastructure that powers it.
- Require measurable outcomes. The long-range ambition spans science and engineering, but the stated boundary is problems whose learning loops can be evaluated.
- Concentrate full-stack expertise in a small team. The founders connect experience across chips, infrastructure, models, and products to a lean organizational design.
quotes
“This approach allows for the parallel execution of thousands of experiments, drastically compressing iteration time and driving up the quantity and quality”
“Our relative advantage isn't just our technical ability; it is the unprecedented scale of the systems we have previously built.”
“We are building a lean, in-person team to execute this transformative vision.”