Paris-based AI vendor AMI Labs has confirmed it received a large seed funding round to develop world models.
AMI Labs — the AMI stands for Advanced Labs Intelligence — is the brainchild of former Meta AI chief, Yann LeCun.
LeCun revealed the $1.03 billion investment using a post on X, which read: “Advanced Machine Intelligence (AMI) is building a new breed of AI systems that understand the world, have persistent memory, can reason and plan, and are controllable and safe.”
Among the global investors leading the round were Cathay Innovation, Greycroft, Hiro Capital, HV Capital and Bezos Expeditions.
AMI Labs is the latest in a growing, but still limited, field of companies focused on world models, in which AI is trained on reality rather than language.
The startup, which will also have offices in New York, Montreal and Singapore, laid out its ambitions on its website, with its rationale underpinned by the belief that: “Real intelligence does not start in language. It starts in the world.”
While acknowledging that some generative AI architectures have been “astonishingly successful” at understanding and generating language, AMI Labs said that real-world data captured by cameras and other sensors is much more unpredictable and complex, requiring a different approach.
As such, AMI Labs is committed to creating “world models that learn abstract representations of real-world sensor data, ignoring unpredictable details, and that make predictions in representation space.”
It continued: “Action-conditioned world models allow agentic systems to predict the consequences of their actions, and to plan action sequences to accomplish a task, subject to safety guardrails.”
These goals have clear benefits for reliability and controllability, with AMI Labs looking to develop applications that can assist in a range of fields, including industrial process control, automation, wearable devices, robotics, and healthcare.
The seed funding, which sees AMI Labs valued at $3.5 billion, follows a similarly large investment of $1 billion in another start-up, World Labs, revealed last month, which is also looking to develop world models that can generate 3D environments from text or image prompts, with industries such as design and architecture targeted.
While the initial investment is considerable, don’t expect a product from the startup any time soon. LeBrun added on X that the project is a “a long-term scientific endeavor,” and it may be years before it sees its first commercial applications.
