Inverted AI CEO Dr. Wood speaks at Google research on foundation models including ITRA
What's a foundation model? A model like ITRA. A repurposable, large, joint distribution that can be further conditioned to do all kinds of new things at test time. Like goal conditioning, vehicle type specific waypoint generation, and staying on road and not colliding with other agents in simulation.
Hosted by Google Brain, Dr. Wood explains ITRA, TITRATED, CriticSMC and how they combine in Inverted AI's products and services.
He also discusses recent work on video diffusion models arising from his UBC PLAI laboratory.