CMU Robotics PhD on Why World Models Are The Real Future | Aayush Bansal, PhD @ CMU Robotics

Описание: LLMs like ChatGPT are powerful knowledge tools, but they lack real world intelligence. Aayush Bansal, CMU Robotics PhD and former Meta researcher, explains why world models; not language models are the path to true AI intelligence. We dive into JEPA, spatial understanding and why self-driving cars work in Pittsburgh but fail in India.

In this conversation, we explore:
Why language is compressed communication, not intelligence
What world models are and why they matter more than LLMs
How ants and bees demonstrate intelligence without language
Why OpenAI abandoned robotics for LLMs (data problem)
JEPA explained: predicting embeddings instead of pixels
Why self-driving works in structured US roads but fails in unstructured Indian environments
The training challenges for world models vs LLMs
Why we're still early: we don't even know how to evaluate world models yet
Carnegie Mellon's robotics legacy and the journey from signal processing to 4D computer vision

About Aayush Bansal:
PhD from Carnegie Mellon's Robotics Institute
Former researcher at Meta Reality Labs
Published at CVPR (top computer vision conference)
Pioneered 4D scene reconstruction from smartphone videos
Now building practical applications of world models

Aayush's Website: https://aayushbansal.xyz/

Connect and Explore:
• Web: https://kalyanikhona.com/
• X: https://x.com/thisismekalyani
• Linkedin:   / kalyanikhona  
• The Third Frontier: https://thirdfrontier.substack.com/
• AI Lab Notes:   / 7364150572851060738  
• Behind India's ChatGPT conversations: https://arxiv.org/abs/2509.13337

#WorldModels #JEPA #LLMs #AIResearch #CarnegieMellon #ComputerVision #Robotics #SelfDrivingCars #MetaAI