The Mohamed bin Zayed University of Artificial Intelligence’s (MBZUAI) Institute of Foundation Models has unveiled PAN, a pioneering new AI model designed to understand and simulate the world as it changes over time. This development marks a significant step toward creating AI systems capable of sophisticated reasoning, prediction, and planning.
Beyond Video Generation To Simulative Reasoning
Unlike conventional video generation tools that produce short, isolated clips, PAN is engineered to maintain continuity and causal logic across extended sequences. The model moves beyond simple visual creation by integrating reasoning with visual understanding.
PAN can follow complex natural language instructions, such as “drive through a snowy forest” or “walk toward the lighthouse,” and generate sequences that remain visually consistent and logical from one moment to the next. This capability represents a crucial advance for AI that needs to interact with and predict changes in a dynamic environment.
The Generative Latent Prediction Framework
At the heart of PAN is its Generative Latent Prediction (GLP) framework, which effectively separates the “what” from the “how” in a scene. The model first establishes an internal latent state—a structured, abstract representation that tracks the objects in a scene and their movements, based on previous context and user prompts.
This latent state is then decoded into a short video segment. By repeating this process, PAN builds longer narratives where the visuals are coherent and causally connected, ensuring that actions and their consequences unfold realistically over time.
Setting New Benchmarks In AI Capabilities
Benchmarking results confirm that PAN achieves state-of-the-art performance among open-source systems across three critical areas: action simulation fidelity, long-horizon forecasting, and simulative reasoning and planning.
These advanced capabilities unlock new possibilities for practical applications in fields such as robotics, autonomous systems, and decision-support research, where understanding the direct influence of actions on the world is fundamental.
About MBZUAI’s Institute of Foundation Models
PAN was developed by MBZUAI’s Institute of Foundation Models (IFM), which is dedicated to building globally impactful, scientifically rigorous, and open foundation models. The IFM connects leading AI talent across its hubs in Abu Dhabi, Paris, and Silicon Valley to drive responsible innovation and address global challenges through collaborative research.
Source: My Startup World
