Yann LeCun 在 NTU 演讲：机器需要从视频和感官学习的世界模型
图灵奖得主 LeCun 在 NTU 演讲，讲他为什么认为光靠 LLM 不够，还得给机器造世界模型，还聊了给研究者的建议。
2018 年图灵奖得主 Yann LeCun 在南洋理工大学发表题为“World Models: Enabling the Next AI Revolution”的公开演讲。他指出大语言模型通过预测下一个词生成文本，真正的机器智能需要理解物理世界。他主张让机器从视频和感官中学习“世界模型”，用抽象表征而非重建每个像素来预判行动后果，从而在不确定性下推理。在与 Bo An 教授的对谈中，他建议年轻研究者打好数学和物理基础。
Prof Yann LeCun, 2018 Turing Award laureate, told an #NTUsg audience that machines will need more: a “world model” learned from video and the senses, so they can anticipate what happens before they act.
At his public lecture “World Models: Enabling the Next AI Revolution”, he said large language models generate text by predicting what comes next, but real machine intelligence needs an understanding of the physical world. A world model forms abstract ideas of what is happening instead of rebuilding every pixel, allowing systems to reason under uncertainty.
In a fireside chat moderated by Prof Bo An, he said AI tools may speed up coding, but human judgement still drives breakthrough discovery. His advice for aspiring researchers: build strong foundations in mathematics and physics.
This lecture was organised by the Institute of Advanced Studies and the College of Computing and Data Science, with support from the Lee Foundation and the Infocomm Media Development Authority.
#IASntu #AIatNTUsg #NTUsgResearch #WorldModels @NTU_ccds
