Yann LeCun focuses on global models, establishes AMI Labs and raises approximately $1.03 billion

Yann LeCun of Computer Science, known as one of the ``fathers of deep learning,'' has been particularly vocal about promoting the global model.

He left the AI ​​research division of Meta, which he had led for many years, and established AMI Labs, a startup specializing in global models, in Paris, France. The company raised about $1.03 billion in a seed round in March, and as of late September, the company was valued at about $3.5 billion.

First of all, what is a world model?

A world model is a technology that allows AI to understand and simulate physical laws and spatial data. To put it simply, a large-scale language model (LLM) generates sentences by predicting the ``next word,'' whereas a world model predicts ``how the world will change if you take a certain action.''

When explaining the concept of a world model, the analogy that ``babies recognize universal gravity even before they learn words'' is often used.

In other words, LLM is nothing more than a language model. We cannot understand the physical laws of the world unless we communicate them in words. LLMs cannot see the world.

In this respect, the world model can be more robust and reliable when it comes to driving in the real world, where exceptional situations are likely to occur. The same goes for "physical AI," which involves understanding and working in the real world. Against this background, global models are attracting attention.

Waymo implements driving AI based on a global model, driving over 300 million kilometers unmanned

Waymo has already implemented the global model in society.

On February 6, the company announced the ``Waymo World Model,'' which is based on the UK Google DeepMind's world model ``Genie 3.'' The system generates ``rare situations'' that are difficult to encounter in real driving, such as bad weather or obstacles, in virtual space, and forces the driving AI ``Waymo Driver'' to accumulate billions of virtual miles of driving experience.