LeCun at NTU: LLMs aren't enough — machines need world models learned from video and senses

chaitjo · x · 2026-11-05

Turing Award winner Yann LeCun told an NTU Singapore audience at his lecture "World Models: Enabling the Next AI Revolution" that LLMs merely predict the next token, while real machine intelligence requires understanding the physical world.

Key points:

A world model should be learned from video and the senses, anticipating what happens before acting
It should form abstract representations rather than reconstructing every pixel, enabling reasoning under uncertainty
In a fireside chat with Prof Bo An, he said AI tools speed up coding but human judgment still drives breakthrough discovery
His advice to aspiring researchers: build strong foundations in mathematics and physics

Original post →

LeCun at NTU: LLMs aren't enough — machines need world models learned from video and senses

More from AGI Musings