[Fireside Discussion] From Tokens To Robots @The Web Data Loft – Full Pannel 09.06.26
An exclusive in-person panel held at Bright Data’s Web Data Loft in San Francisco, bringing together top robotics and AI engineers from Agility Robotics, Tesla, Prometheus (ex-1X), and Distill Labs to unpack what it really takes to move from language models to robots that work in the real world.
In partnership with the Builders Collective, this conversation goes deep on Vision-Language-Action (VLA) models, world models, small language models on-device, sim-to-real evaluation, and the data infrastructure powering the next generation of humanoid and embodied AI.
👉 Power your VLA models with high-quality video data from Bright Data: 🔗 https://brightdata.com/ai/video-data/vla
Chapters:
00:00 — Welcome from Bright Data & the Builders Collective
06:55 — From Tokens to Robotics: framing the conversation
09:51 — What is a VLA (Vision-Language-Action) model?
19:37 — Small Language Models on-device vs. frontier models in the cloud
26:11 — VLAs vs. traditional robotics in real-world deployment
30:43 — What is a World Model?
40:30 — Contact physics, tactile feedback & wrist cameras
44:37 — JEPA vs. transformer-based world models
48:42 — Sim-to-real & the Waymo world model analogy
54:08 — Scaling laws for VLAs and world models
59:34 — Can web data alone train a robot?
1:05:18 — Closing thoughts Bright Data provides the world’s leading web data infrastructure giving AI teams, agents, and robots reliable access to public web data at scale. From unlocking geo-restricted content to powering large-scale video and multimodal datasets, Bright Data is the data layer behind modern AI.
- Learn more: https://brightdata.com
- VLA video data: https://brightdata.com/ai/video-data/vla
