As the AI industry continues to pursue the concepts of “AGI” and “super smart”, AMI Labs CEO Alexandre LeBrun chose to distance himself from such expressions. He stated that the terms lacked a clear definition and could not accurately describe the world model path that companies were moving forward.

Do not follow the AGI narrative

In an interview with TechCrunch, LeBrun stated that AMI Labs had never used the term “AGI” and did not agree with “super smart”. In his view, it was more like an industry narrative switch rather than a clearly defined technical boundary.

AMI Labs was created with the participation of the Turing Prize winner, Yann LeCun, who played the "World Model" orientation. The goal of such models is not to predict the next word, but rather to predict the next state of reality, so that AI can better understand how the physical world changes.

Focus on robot landing.

LeBrun believes that one of the most important applications of the world model is robotics. Many robots are still reliant on fixed processes, and the ability to understand the surroundings and safely carry out their tasks remains inadequate once they enter open environments such as homes and streets.

He indicated that hardware had progressed very fast in recent months, but robots still lacked sufficiently mature “understanding”. If AI has situational awareness capabilities, robotic miscalculation and risk in the public scene are expected to decrease.

At the same time, he stressed that the world model was not intended to replace the big language model. The two are closer to complementarity: the Big Language Model is responsible for language processing, and the World Model provides the context and judgement of the real world.

Asian cooperation is still under way

AMI Labs has not yet launched its products, but has begun to reach out to partners in robotics, manufacturing and electronics. LeBrun was also looking for local industrial partners, multinational corporations and researchers during last week's International Machine Learning Congress in Seoul.

He said that the world model could not only be trained in a laboratory and had to go into real environmental validation, so companies would like to use partners to obtain realistic scene data. Korea is attractive because it has a robotic, semiconductor and manufacturing base, while policy pushes faster.

AMI's Asian investor, SBVA CEO JP Lee, also indicated that the Korean government has continued to invest in AI in recent years and has prioritized chips, AI data centres and physical AI.

Financing is in place and products are still not available

Despite the high level of outside interest, AMI Labs still does not have a formal product and does not give a clear release time. LeBrun simply said that the team would publish more when it was ready.

This company completed financing in March of this year at $1.03 billion, with a pre-investment valuation of $3.5 billion. For an AI start-up company that has not yet published a product, the scale of this financing reflects the high interest of capital markets in world models and physical AI routes.