Artificial Intelligence - Machine Learning Seminar
September 4, 2026 10:00AM—11:00AM
Location:
In Person
-
Newell-Simon 3305
Speaker:
ERIC XING,
President, Mohamed bin Zayed University of Artificial Intelligence and Professor, Machine Learning Department, Carnegie Mellon University
https://www.cs.cmu.edu/~epxing/
The success of LLMs like OpenAI’s GPTs has fascinated the public, with their astounding capabilities in a wide range of tasks such as genius-level standard test performance, IMO-level math reasoning, encyclopedia-like Q/A, and human-like ability in conversation. However, LLMs are only “book intelligent” — they suffer inherent limitations of their capabilities for embodied, physical, and social reasoning, and strategic planning thereupon, in the real world.
In this talk, I discuss how to build a true world model, rather than a video generator, to simulate all actionable possibilities of the real world for purposeful reasoning and planning via thought-experiment rather than mere pixel realism; and how to build a true agent model, rather than a LLM wrapper or software pipeline, to be able to learn and act with the flexibility, adaptability, and autonomy associated with natural agents such as humans with intrinsic ability of self-regulation, reflection, collaboration, and socialization, rather than merely reacting to exogenous stimuli. We propose a Generative Latent Prediction (GLP) architecture for world modeling that builds on stateful latent space, long-horizon and close-loop action-conditioned latent reasoning, and learning/inferencing grounded over realizable world states; and an Goal-Identity-Configurator (GIC) architecture for agent modeling that can regulate reasoning mode between being unconsciously reactive or consciously deliberative, generate real-world actions based on its goals and identity, and self-learn off-line from the world model via RL.
We present PAN, a physical, agentive, and nested framework over the proposed architectures that brings together perception, state, action, and causality within one system to supports open-domain interactable world simulation and agentive intelligence. Extensive experiments show that PAN achieves strong performance in action-conditioned world simulation, long-horizon forecasting, and simulative reasoning compared to other video generators, world models, and agentic systems.
—
Professor Eric Xing is the President of the Mohamed bin Zayed University of Artificial Intelligence, and a Professor of Computer Science at Carnegie Mellon University. His main research interests are the development of machine learning and statistical methodology, and large-scale distributed computational system and architectures, for solving problems involving automated learning, reasoning, and decision-making in in artificial, biological, and social systems. In recent years, he has been focusing on building large language models, world models, agent models, and foundation models for biology.
Prof. Xing has served on the editorial boards of several leading journals including JASA, AOAS, JMLR; was a recipient of several awards including NSF Career, Sloan, Carnegie Science Award, and best papers in conferences such as ACL, ISMB, NeurIPS, and OSDI; and is a fellow of several societies including AAAI, ACM, ASA, IEEE, IMS, and ISCB.
Faculty Host: Nihar Shah
The AI-ML Seminar is generously sponsored by the MBZUAI/CMU Institute of Virtual and Programmable Cell
For More Information:
nihars@cs.cmu.edu