Date: Wedn April 22, 2026 6:30 - 8:30 PM, at ITRI 2880 Zanker Suite 103. Seminar talk will start at 7PM.
Join Zoom Meeting:
https://us02web.zoom.us/j/89784416420?pwd=LsjiIotgLGMFeRiifIKnkuA0jjEbI4.1
Meeting ID: 897 8441 6420
Passcode: 423997
“The Next Giant Step for AI: World Models ”
By Dr. William Kao
Seminar Abstract
Artificial Intelligence has undergone a dramatic transformation with the rise and use of Large Language Models (LLMs), which demonstrate remarkable capabilities in language understanding, reasoning, and content generation. However, these models fundamentally operate over static datasets and token sequences, lacking grounded understanding of the physical world. Key limitations include weak causal reasoning, poor long-horizon planning, hallucination of facts, and an inability to model dynamics governed by real-world physics.
Most recently, there is an emerging paradigm shift from LLMs to world models, which aim to build internal representations of reality by learning causality, physics, and environmental dynamics. This transition is central to the development of Physical AI—systems that can perceive, reason, and act in real-world environments.
A couple of leading AI researchers on this new bandwagon are Yann LeCun (AMI), who advocates for predictive world models, and Fei-Fei Li (World Labs), whose work emphasizes spatial intelligence.
A world model can be defined as an internal model that encodes the state, dynamics, and causal relationships of the external world, enabling an agent to simulate future outcomes and make decisions based on those simulations.
World models serve as a foundational paradigm for Physical AI, enabling machines to move from passive pattern recognition to active understanding and interaction. Their benefits include causal reasoning, simulation and better generalization across environments, good foundation for robotics and Physical AI systems.
In this seminar we will examine the limitations of current LLM architectures, new World Model capabilities and applications and discuss future AI hybrid systems.
Speaker Bio:
Dr. William Kao received his BSEE, MSEE and PhD from the University of Illinois Urbana-Champaign. He worked in the Semiconductor and Electronic Design Automation industries for more than 30 years holding several senior and Executive engineering management positions at Texas Instruments, Xerox Corporation, and Cadence Design Systems.
Dr. Kao has authored more than 40 technical papers on IC CAD at IEEE Journals and Conferences, and holds eight US software and IC design patents. He was an Adjunct Professor at UCLA Electrical Engineering Department where he taught courses in computer aided IC design.
Dr. Kao is a Senior Member of IEEE, and was one of the founding members of IEEE-Circuits and Systems - Silicon Valley Chapter.
Dr. Kao teaches Clean Technology and Emerging Technology courses at the University of California Santa Cruz, Silicon Valley Extension.
He has given more than 100 technical seminars on various Emerging Technologies including Clean Technology, Renewable Energy, Big Data and Data Analytics, IoT, Smart Cities, Augmented and Virtual Reality, Robotics, Drones, AI / Machine Learning, Robotics, 5G, 3D Printing, the Metaverse and Satellite Communications.
In the last 3 years he has focused his technical talks and seminars on recent advances in AI and Robotics covering the latest developments in AI, machine learning, deep learning, large language models, prompt engineering and humanoid robotics.