Yann LeCun is moving toward a new AI company with a clear technical bet: the next major step will not come from simply making generative systems larger. His startup, called AMI (Advanced Machine Intelligence), is being built around world models, systems intended to understand physical reality rather than mainly produce text.
A break from the GenAI center of gravity
LeCun argues that Silicon Valley is currently "hypnotized" by generative AI. That view puts AMI outside the dominant conversation around language models, where much of the focus has been on scaling current approaches.
The difference is not just branding. According to the source, LeCun contends that today's language models lack any real grasp of how the world works. From that position, more scale alone is not a convincing path to human-level intelligence.
"Some people claim we can scale up current technology and get to general intelligence [...] I think that's bullshit, if you'll pardon my French."
That statement makes the strategic direction of AMI easier to understand. The company is not being described as another generative AI lab looking for incremental improvements to text generation. It is presented as an attempt to move toward a different architecture entirely.
What world models mean for AMI
The source describes world models as systems designed to understand physical reality rather than just generate text. That framing matters because it shifts attention from fluent output to internal understanding.
In plain terms, the AMI approach is based on the idea that intelligence needs more than the ability to produce plausible language. If a system does not understand how the world works, LeCun's argument suggests that its abilities remain limited, even if its responses appear sophisticated.
The project relies on a new architecture that moves away from generative methods entirely. The source does not provide technical details of that architecture, so the key fact is the direction: AMI is being positioned as non-generative and built around world models.
Meta stays involved as LeCun steps out
LeCun plans to leave the tech giant at the end of the year to lead the independent organization. Even so, Meta is not disappearing from the story.
Meta is signing on as a partner because Mark Zuckerberg supports the initiative. The source also notes that LeCun sees potential applications extending well beyond Meta's specific interests.
That combination gives AMI an unusual starting point. It is described as independent, but it also begins with a major partner already attached. The source does not say what the partnership will include, so the important point is simply that Meta will be involved while AMI pursues a broader agenda.
Why Europe is central to the plan
LeCun intends to build the project with a heavy reliance on European talent. According to Sifted, the company will operate globally and maintain a hub in Paris.
That detail gives the startup a geographic shape without making it a local-only effort. AMI is described as global, while Paris is singled out as a hub. The source does not list other locations or hiring plans.
The European talent emphasis also fits the broader picture of a company trying to define itself differently from Silicon Valley's current GenAI momentum. LeCun's critique is aimed at the prevailing focus on generative AI, and the organization he is building is being presented as an alternative path.
The bigger question for AI research
AMI's premise raises a direct question for the field: can current language model technology become human-level intelligence through scale, or does the architecture itself need to change?
LeCun's answer is clear in the source. He does not believe scaling current technology is enough, and he is organizing AMI around that belief. The startup's focus on world models is therefore not a side project; it is the central claim.
For readers following AI strategy, the significance is less about one company launch and more about a split in assumptions. One path continues to emphasize generative AI and scale. The other, represented here by AMI, argues that understanding physical reality requires a different foundation.
What happens next is not detailed in the source. What is clear is that LeCun is preparing to leave Meta at the end of the year, lead AMI, keep Meta as a partner, rely heavily on European talent, and build around non-generative world models rather than the GenAI systems now dominating Silicon Valley attention.