Can “world models” fix AI’s blind spots?

World Model

Babbage from The Economist

2026-02-11

42 分钟
PDF

单集简介 ...

Modern artificial intelligence is seemingly unstoppable. But large language models have an undeniable blind spot: an incomplete understanding of physical reality and the laws of physics. Could “world models” be the solution—and how are AI labs building them? Guests and hosts: - Alex Hern, AI writer at The Economist - Yann LeCun, AI pioneer and executive chairman at AMI Labs - Shlomi Fruchter, research director at Google DeepMind - Cristóbal Valenzuela, CEO at Runway AI - Alok Jha, host of “Babbage” Topics covered: - Large language models - World models - Google's Project Genie Transcripts of our podcasts are available via economist.com/podcasts. Listen to what matters most, from global politics and business to science and technology—subscribe to Economist Podcasts+. For more information about how to access Economist Podcasts+, please visit our FAQs page or watch our video explaining how to link your account.
更多

单集文稿 ...

  • The Economist.

  • Modern artificial intelligence is seemingly unstoppable.

  • AI models can help you sort through your emails.

  • They can create images.

  • They can write code and even implement it.

  • You can use them to produce deeply researched reports on the topics that you might be interested in.

  • And if you answer a few questions,

  • they can even help you write a business plan for your next big idea.

  • The common factor in all these amazing skills is that

  • large language models do all these things in the digital world.

  • What would it take for AIs to become useful in the physical space that we all live in?

  • Do LLMs even know or understand anything about the physics of the real world?

  • Well, to some extent, they do.

  • Because the language on which they've been trained contains information about how the world works.

  • But that knowledge is gappy.

  • It's massively incomplete.

  • And that's not good enough if you want AI models to, say, control a robotic body

  • that can interact in real time with a city or in a factory or a classroom.

  • Or if you want your AI system to be able to give you instructions

  • on the best way to go from one part of a city to another.