Yann LeCun says next AI leap needs world models, not bigger LLMs
AI-generated briefing · First published
Turing laureate Yann LeCun told Seoul audiences that AI must learn physical causality via world models, not just language patterns.
Speaking in Seoul on 12 October, 2018 Turing laureate Yann LeCun said the next AI breakthrough will come from systems that understand and predict the physical world, not from larger language models. [1] [2]
LeCun noted that current LLMs excel at text but cannot match a teenager’s ability to learn driving or a 10-year-old’s grasp of household tasks, highlighting a gap in physical understanding. [1] [2]
He argued that world models, such as the JEPA architecture pursued by his AMI Labs, will let AI learn causal relations and plan actions safely, with risks controlled through engineering rather than inherent limits. [2]
Sources
- “AI 教父”杨立昆：下一代 AI 须理解现实世界
ITHome · · Summary only
- 图灵奖得主杨立昆：AI下一阶段突破将来自世界模型，失控风险可通过工程技术控制
Sina Finance · · Full text