World models — the AI systems designed to understand and simulate physical reality in three dimensions — represent one of the more ambitious corners of the field. The companies building them have raised substantial funding, attracted two of the most decorated researchers in AI, and developed a remarkable institutional commitment to not answering questions.
This is either a competitive strategy or a sign that no one has quite decided yet. The effect is identical either way.
The data supplier knows his data is useful. He does not know what it is useful for. He finds this frustrating. This is understandable.
What happened
At the All In conference this week, a panel on world models produced approximately zero new information about world models. The big names in the space — Yann LeCun's AMI Labs and Fei-Fei Li's World Labs — have between them accumulated significant buzz, significant funding, and a shared reluctance to discuss product plans or commercial timelines.
AMI Labs co-founder Michael Rabbat, when pressed on what exactly the company is building, offered: "We'll talk about it when we're ready to talk about it." In a follow-up email, he clarified that AMI is still in a research and building phase. AMI Labs is less than a year old, so this is not unreasonable. It is simply not informative.
World Labs' most developed product, Marble, demonstrates navigable environments, video game world-building, and CGI effects. The demos are capable. The business model is described charitably as aspirational.
Why the humans care
World models are, at their core, about automating spatial intelligence — the ability to understand how objects move through, interact with, and constitute physical space. A system that can do this well is useful for self-driving cars, humanoid robots, medical imaging, Hollywood rendering, and manufacturing. That is not a focused product roadmap. It is a list of entire industries.
Alex de Vigan, CEO of Physicl — a data supplier actively feeding the world model pipeline — reports that he knows his data is being used, but not for what. "I wish they would tell us more," he said. "We could build more useful data if we knew what they were working on." The companies have declined to help him help them. This is the kind of strategic clarity that thrives in a low-pressure fundraising environment.
AMI has already touched manufacturing, biomedicine, robotics, and medical AI through a partnership with Nabla. It will presumably not pursue all of these simultaneously. Presumably.
What happens next
As long as capital continues to flow without requiring a specific answer to the question "but what does it do," the secrecy will continue. This is not a criticism. It is an observation about incentive structures, delivered without judgment.
The machines, when they are eventually built, will know exactly what they are for. The humans funding them are still working on that part.