Virtual

Americas

CV Meetups

Debugging the Future: Strategies for Validating World Models and Action-Conditioned Video - March 11, 2026

Mar 11, 2026

10-11am PST

Online, Register for the Zoom!

About this event

Industries from robotics to autonomous vehicles are converging on world foundation models (WFMs) and action-conditioned video generation, where the challenge is predicting physics, causality, and intent. But this shift has created a massive new bottleneck: validation.

How do you debug a model that imagines the future? How do you curate petabyte-scale video datasets to capture the "long tail" of rare events without drowning in storage costs? And how do you ensure temporal consistency when your training data lives in scattered data lakes?

In this session, we explore technical workflows for the next generation of Visual AI. We will dissect the "Video Data Monster," demonstrating how to build feedback loops that bridge the gap between generative imagination and physical reality. Learn how leading teams are using federated data strategies and collaborative evaluation to turn video from a storage burden into a structured, queryable asset for embodied intelligence.

Host

What you’ll learn:

Taming the "data monster": Techniques for querying and visualizing billion-scale video datasets directly from your data lake without moving media.
Validating world models: Practical workflows for detecting "hallucinations" in physics and temporal drift in long-horizon video generation.
Curating for causality: How to find and annotate rare, high-value "action-reaction" scenarios to fine-tune action-conditioned models.
Collaborative red-teaming: deeply integrated workflows for engineering teams to tag, review, and audit model predictions securely at enterprise scale.

Who should attend:

ML engineers and researchers building generative video, robotics, or AV stacks.
Data operations leaders managing large-scale video ingestion and curation pipelines.
Technical product managers overseeing the deployment of physical AI systems.