DataFramed #351 Will World Models Bring us AGI? with Eric Xing, President & Professor at MBZUAI
39 snips
Mar 16, 2026 
Eric Xing, President of MBZUAI and leading ML researcher, explains world models as simulators that move AI from book knowledge to physical and social skills. He discusses long-horizon planning, architectures like PAN that mix symbolic and generative components, evaluation beyond pixel fidelity, robotics and driving use cases, synthetic data and virtual cells for science, and the push for open K2 Think models.
AI Snips
Chapters
Transcript
Episode notes
World Models Move AI Beyond Book Knowledge
- World models extend LLM "book intelligence" into physical and social intelligence for planning and action.
- Eric Xing argues world models simulate experiences so agents can learn autonomously, enabling skills like pouring coffee or coordinating tasks.
Long Horizon Consistency Is The Core Challenge
- Long-horizon consistency and planning are the main technical bottlenecks for current video-style world models.
- Eric explains many models produce short videos but fail to preserve consistent state over minutes, requiring new representations and architectures.
Hybrid Representations Enable Long And Short Reasoning
- Representations must balance latent continuous reasoning and symbolic, discrete reasoning to maintain long-term plans.
- PAN uses a mixed backbone: LLMs for symbolic long-term inference and diffusion decoders for high-res short-range visuals.