Fireside chat with You Liang, Co-Founder of Dream Labs

​Most robot foundation models still need your data to fine-tune, or an enterprise contract just to get access.

Dream Labs took a different path. Instead of teaching robots to follow instructions, they teach them to imagine: Dream-1 predicts what the world will look like next and what the robot should do, in a single step. The team came out of NVIDIA’s GR00T group, where they helped build DreamZero, the research that put World Action Models on the map.

Last month, Dream Labs came out of stealth with $90M in seed funding and launched Dream-1, a World Action Model that works zero-shot on robots it has never seen. No fine-tuning, no demos, and anyone can use it today through a public API starting at $3 an hour.

​Join us for a fireside chat with You Liang, Dream Labs’ co-founder, on why imagination might be the shortcut to general robot intelligence, what it really takes to generalize across robots, and what a pay-as-you-go robot brain means for founders building in physical AI.

​Topics will include:

• ​World Action Models: How learning to predict the physical world could unlock a new paradigm for robot intelligence

• ​The generalization problem: Moving beyond robot-specific policies toward models that work across embodiments, tasks, and environments

• ​Scaling robot learning: The role of human video, robot data, and the training approaches that could make general-purpose robotics possible

• ​The next frontier of Physical AI: What cheaper hardware and increasingly capable foundation models mean for robotics adoption

• ​Building at the frontier: The research challenges, technical bottlenecks, and opportunities for founders building the next generation of Physical AI

​This event is designed for founders, robotics researchers, builders, and investors actively thinking about the next generation of Physical AI.

​Agenda:

• ​6:00 PM — Doors Open + Light Bites

• ​6:45 PM — Welcome Remarks

• ​7:00 PM — Fireside Chat

• ​7:30 PM — Audience Q&A

• ​7:45 PM — Closing Remarks

• ​8:00 PM — Networking

• ​9:00 PM — End of event

​About the Host:
Savant is a physical AI community and venture fund for exceptional builders searching for their good quests: ideas that's hard, consequential, and civilization-defining. Join as a Savant Resident to explore your quest with a talent-dense group of peers, prototype in our 6,200 sqft hardware lab in SF, and get deep support from our team—equity free. When you're ready to build, we write the first check.

​Encord is the universal data layer for AI. The platform helps AI teams train and run their models with the right data - managing, curating, annotating, and aligning data across the full AI lifecycle. Encord works with over 300 leading AI teams, including Woven by Toyota, Zipline, Forterra, and Dyna Robotics.

​LanceDB is the multimodal lakehouse for AI, bringing data curation, feature engineering, search, and training data management into one system. It manages distributed compute, caching, indexing, and data maintenance so your team can focus on building better models faster.