My PhD thesis "Deep Learning with Graph-Structured Representations" is now available for download: https://t.co/hyz0cnoewZ -- It covers a range of emerging topics in Deep Learning: from graph neural nets (and graph convolutions) to structure discovery (objects, relations, events)

World models are helping us evaluate #GeminiRobotics generalist policies more effectively, including auto-red-teaming to safely find and address vulnerabilities in the models. Learn more at: https://t.co/Nisxb3UTVY @GoogleDeepMind Robotics

World models for policy evaluation; the fact that world model performance is highly correlated with real-world performance is incredibly valuable on its own.

Check out the paper and website at

So excited to finally talk about this work! Veo is a surprisingly strong world simulator. We fine-tuned Veo on action-conditioned, multi-view robotics data. Key result: running a policy in the world model is strongly correlated with real-world results. A few important

Here’s what unexpected scenarios look like in camera, radar, and LiDAR.

Exciting times ahead!

One hour Waymo ride from San Francisco to Menlo Park via highway 280. It’s over, cars are self-driving. Everything else is just about rolling this out to the rest of the world.

Come work with Joe and you’ll sit by me, @poolio, @jon_barron, @RuiqiGao, @holynski_, @tkipf, and many other fine folks!

Finally got freeway access on Waymo. Going fast in a fully driverless car almost feels as magical as stepping into a Waymo for the first time a bit over a year ago.

Liftoff achieved

https://t.co/TYHS33XyPj

NeurIPS secret mug club

Gemini 3 Deep Think is now available for Ultra users, making available our IMO & ICPC Gold Medal-winning technology. Deep Think shows improved generalization on difficult benchmarks like ARC-AGI-2, and outperforms Gemini 3 Pro on HLE & GPQA Diamond. We hope this serves as a

As we work to mitigate the @iclr_conf incident, personal integrity is the only true firewall. The integrity of our science is at stake. Please act responsibly.

I'll be at NeurIPS all week -- reach out if you want to chat! Would love to chat especially if you work on world models (in particular for the physical world / robotics), visual reasoning, or controls for video gen.

Just an absolute triumph of technology. Everyone who worked on this should be so proud.

Veo 3.1 is embarrassingly compositional

this is literally quite huge

Character consistency has come a long way! It's such a great compositionality test. And it turns out frontier models are truly "embarassingly compositional".