Reka Details World Model Training: How a <100 Person Team Handles Petabytes of Video Data
RekaAILabs · x · 2026-07-22
Reka AI shared insights into building the data platform behind their 'Physical AI' and world model training.
- Team & Challenge: Led by Konrad Jam, a team of fewer than 100 people prepares hundreds of thousands of hours of video data for world model training, dealing with petabyte-scale data.
- Data Dictates Capability: Researchers Julian López and Fedor Zhdanov emphasize that before a single model weight is updated, it's crucial to understand the data's composition. Composition, quality, and annotation directly determine what the world model actually learns.
- Engineering Pace: To keep pace with a fast-moving research team, the data pipeline requires highly dynamic day-to-day operations.
Related event: Reka AI Details Data Challenges in World Model Training(2 posts)→
More from Companies & People
- YC talk on BCI x AI says infrastructure is what really determines speed — garrytan · 2026-07-27
- AI industry reception in SF featured FAI and Thinking Machines logos on the cake — simonguozirui · 2026-07-27
- YC launches an AI Office Hours simulator for Startup School attendees — ycombinator · 2026-07-27
- A short note suggests layering multiple AI subscriptions may be the new strategy — sull · 2026-07-27
- Dhruv Bhatia joins fal to work on video and world models — gorkem · 2026-07-27
- 7 Bittensor Subnets Accepted into NVIDIA Inception, Making Up 21% of Cohort — markjeffrey · 2026-07-27