AMI Labs builds systems that understand the real world, such as a manufacturing plant or turbojet engine.

Judge as of 2026-10-01. The claim says AMI Labs (LeCun's company) builds systems meant to understand the real world, with industrial systems such as manufacturing plants or turbojet engines given as examples. True: AMI Labs' official materials (website, launch announcement, funding announcement, technical papers) or reliable reporting describe its work as building AI systems or world models that model or understand the physical or real world, and industrial or physical-system applications like these are among its stated aims or uses. The examples are illustrations, so no specific turbojet or plant deployment is needed. Mostly true: the real-world or world-model focus is documented, but industrial applications are not specifically mentioned or are only loosely implied. Mixed: AMI Labs works on world models, but it is unclear or disputed that it builds systems for real-world physical domains rather than general research. Mostly false: the evidence shows AMI Labs' focus is substantially different (for example, mainly language or chat products), with real-world understanding at most a minor, stated aspiration. False: AMI Labs did not exist or was not building such systems at that date, or its documented work is clearly unrelated.

Set before the research began.

1. E1 AMI - Advanced Machine Intelligence Careers | Levels.fyi Levels.fyi · 26 August 2026

   “Their action-conditioned world models enable reliable, controllable, and safe agentic systems for industrial automation, robotics, healthcare, and wearables.”

2. E2 AMI - Advanced Machine Intelligence Careers | Levels.fyi Levels.fyi · 26 August 2026

   “AMI is a frontier AI research lab developing world models that learn abstract representations of real-world sensor data, ignoring unpredictable details while making predictions in representation space.”

A third-party company profile dated 2026-08-26 describes AMI as a frontier AI research lab that develops world models learning from real-world sensor data (E2). The same profile says these action-conditioned world models are for agentic systems in industrial automation, robotics, healthcare and wearables (E1). That matches the criteria for true: a real-world, world-model focus with industrial applications among its stated uses. The only source is a single aggregator profile, which may draw on the company's own description, but its content fits the claim directly.