Had such a great time at #CoRL2026, especially being part of the Physical AI × Data panel in Austin!
One question I've been thinking about for a while is how physical intelligence will evolve. Will general-purpose multimodal models increasingly learn to act in the physical world, or will robotics foundation models develop broader reasoning and generalization capabilities? It was wonderful to exchange perspectives on these directions at CoRL. I suspect we'll increasingly see systems combine the strengths of both approaches, bringing together frontier reasoning and fast, grounded action.
The bigger challenge is making AI reliable as it becomes part of people's everyday lives. Pretraining and demonstrations provide strong foundations, but systems also need to improve through real-world experiences, failures, and human feedback. How to build safe, scalable learning loops that continuously improve these systems remains a major open question.
Thanks so much to @ArthitaG (@sievedata) for inviting me, and to @ishanator13 (@uncorkcap) for bringing everyone together! Really enjoyed exchanging perspectives with Arthita, Saurabh Nair (@wayve_ai), Joshua Manela (@BedrockRobotics), and @Poorvi_rh (@SkildAI)
While at #CoRL2026 last week, a comment from one of the attendees really struck me – 60% of the papers being presented are already out-of-date. The upshot is you need to talk to people currently building to stay on the frontier.
During the conference, I moderated a panel of engineers, researchers, and operators building across physical AI to discuss what the future of data in robotics looks like. In a packed room, we discussed closing the sim-to-real gap, how to go from demo to deployment, simulated vs. real data, System 1 vs. System 2 type problems, and more.
A big thank you to @ArthitaG (@sievedata) for co-organizing this panel with me and our fellow panelists Saurabh Nair (@wayve_ai), Joshua Manela (@BedrockRobotics), @charitha_akula (@Meta), and @Poorvi_rh (@SkildAI) for the spirited discussion. I’m looking forward to many more such conversations back in the Bay!
cc: @UncorkCap
I'll be at #CoRL2026 in Austin this week.
Would love to connect and exchange ideas on multimodal foundation models, robotics, recursive self-improvement, or just grab a coffee! ☕
Heading to #NeurIPS2024 to present Cambrian-1 w/ @TongPetersb! Catch our oral presentation Friday @ 10am (Oral 5C) and our poster afterwards until 2pm (#3700 in East Hall A-C) 🪼🎉
@WilliamBryk@ExaAILabs 1. I have this idea X, does any service/product/project like X already exist?
2. Anything that can't be answered from title/meta data and requires deeper analysis of content Ex: give me blogs/videos that can give non-technical explanation of X or an overview of a small subtopic.
We're back online with a taste of NYU DL SP 22! 🥳🥳🥳
In this lesson, @JiachenAI covers visual representation learning with pretext tasks, AEs, and contrastive joint embedding methods. Check out the video and/or the summary!
🎞 https://t.co/DjrMfcv4jX
✍🏻 https://t.co/5jb165M3c5