korrents

What Dmitri Dolgov thinks about reinforcement learning

@dmitri-dolgov · 22 positions · 0 changes of mind

Co-CEO of Waymo and a founding member of Google's self-driving car project, which he joined in 2009.

Dmitri Dolgov did not write this page.

We collected these quotes from things they published elsewhere, and every quote links to where it was said. They have no account here and have not endorsed this site. Quotes are word for word; the short line under each one is our own restatement, not their wording. Their own site. Is this you? Claim it or ask us to remove it. Or tell us what is wrong here.

9 dated positions, 2026, in their own words. Our reading of what Dmitri Dolgov has said — not written or endorsed by them.

9 positions so far — this page is not yet offered to search engines.

  1. Our readingIn the physical world a model's mistake is measured in human lives rather than tokens, and there is no retry button.

    Their wordsIn the physical world, the cost of a mistake can be measured in human lives, not tokens. There's simply not an undo and a retry button.

    ↗Waymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Workyoutube.com 3rd of 22 in this recording

    Interesting

  3. Our readingThe common failure is picking the technology with the fastest early ramp and extrapolating it, when that curve plateaus below what the product needs.

    Their wordscommon failure mode is picking the tech that gives you the fastest early ramp, riding that steep curve, feeling like you're winning, projecting that, you know, steep slope into the future and feeling like the sky is the limit, and then hitting the plateau, and discovering that the technology path that you picked actually flattens out way before the performance that is required by your product.

    ↗Waymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Workyoutube.com 10th of 22 in this recording

    Interesting

  5. Our readingCameras alone are enough to approximately match a human driver, but weak sensing puts safety on a curve that flattens out well below superhuman.

    Their wordsHowever, if you are targeting full autonomy, and you're targeting superhuman, strongly superhuman performance, you find that weak sensing just leads to a safety curve that flattens out way too early.

    ↗Waymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Workyoutube.com 11th of 22 in this recording

    Interesting

  7. Our readingBetting a robotics company on today's component prices is betting it on a number with a short shelf life.

    Their wordsSo, betting your company, betting your approach on today's hardware prices is just betting your company on a number that has a fairly short shelf life and is going to expire.

    ↗Waymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Workyoutube.com 12th of 22 in this recording

    robotics

    Interesting

  9. Our readingDriving turns out to be much like language modelling, because a driver is holding a conversation with other road users in the body language of a car.

    Their wordsTurns out, the task of driving is not that dissimilar from the task of modeling language uh because of the social aspects of driving. You're kind of having a conversation with other dynamic actors in the world, but you're doing that in the space kind of body language of your agent, your car, as opposed to just the language of words.

    ↗Waymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Workyoutube.com 13th of 22 in this recording

    Interesting

  11. Our readingStructure that fights scale always loses and structure that channels scale always wins — which is the version of the bitter lesson that matters in the physical world.

    Their wordsEssentially, structure that fights scale will always lose. And structure that channels scale always wins.

    ↗Waymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Workyoutube.com 14th of 22 in this recording

    Interesting

  13. Our readingBuilding a realistic simulator is exactly as hard as building the agent, because the simulator is itself a large AI model.

    Their wordsAnd the problem of building a good realistic simulator is just as hard as building the agent itself.

    ↗Waymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Workyoutube.com 17th of 22 in this recording

    Interesting

  15. Our readingA physical AI company needs three AIs, not one — the agent, the simulator and the critic — turning deployment into a flywheel.

    Their wordsSo, a deployment of your agent uh in the real world generates data. That data then grounds the simulator and makes it more realistic. The simulator generates harder edge cases for the critic to score and for the agent to learn from.

    ↗Waymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Workyoutube.com 18th of 22 in this recording

    Interesting

  17. Our readingModels leak and algorithms get copied, but hundreds of millions of audited autonomous miles cannot be replicated — earned trust is the real moat.

    Their wordsYour your models can be leaked, algorithms can be replicated, but hundreds of millions of miles of fully autonomous operations in the real world, backed by evidence-grade evaluation and publicly audited proof, that is much, much more difficult to replicate.

    ↗Waymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Workyoutube.com 21st of 22 in this recording

    Interesting