One of the most influential AI scientists in the world is leaving his position at Meta. Not to retire, but to take a radically different direction with more than a billion dollars in seed funding. Yann LeCun, Turing Award winner and longtime head of Meta’s AI division, is founding AMI Labs in Paris. His thesis: the large language models everyone is building on today are a "statistical illusion". Impressive with language, but fundamentally limited.
This is not a doomsday scenario. It is an observation that should interest every organisation working with AI.
What happened
Yann LeCun is no stranger to the field. Together with Geoffrey Hinton and Yoshua Bengio, he received the 2018 Turing Award for pioneering work in deep learning. At Meta, he led FAIR, one of the world’s most productive AI research labs. And yet he chose to leave.
AMI Labs, short for Advanced Machine Intelligence Labs, raised a seed round of $1.03 billion. The largest seed round ever for a European company. Valuation: $3.5 billion. Investors include Nvidia, Samsung, Jeff Bezos and Eric Schmidt. The team consists of former top scientists from Meta.
The company is based in Paris. That is no coincidence. LeCun explicitly positions AMI Labs as a European alternative to the American tech giants.
What LeCun is building differs fundamentally from ChatGPT or Claude. Current language models are trained on text. They predict the next word in a sequence. That produces impressive results, but it does not mean the model understands the world. LeCun compares it to a parrot: fluent in language, blind to reality.
His alternative is called "world models", based on an architecture he calls JEPA (Joint Embedding Predictive Architecture). The idea: AI that learns like a child. Not by reading billions of sentences, but by observing the physical world, recognising patterns and understanding cause and effect. AI that knows a ball falls down. Not because it read that somewhere, but because it models the world.
Why this matters for your organisation
The temptation with this kind of news is to wait and see. "Let the scientists figure out who is right." That is understandable. But the point is not whether LeCun turns out to be right. The point is what this tells us about the phase AI technology is in.
We are not in a stable, crystallised market. We are in a period of fundamental shifts. Today’s models are powerful. They deliver real value. But the probability that the tools you use today will look the same in three years is small.
That is not a reason to wait. It is a reason to think very deliberately about how you deploy AI. Three insights.
Understand what you use. Many organisations deploy AI as a black box. Something goes in, something comes out, and as long as the result is usable, nobody asks questions. But when the technology shifts, and it does, you want to know what you have built. Which processes run on which models? Where are the dependencies? What happens when a vendor changes their API or adjusts their pricing model? Organisations that understand their AI stack can adapt. The rest stand still when the ground shifts.
Invest in people, not just tools. The tools will change. That is certain. What does not change is the value of teams that understand how AI works, that can judge when a model is reliable and when it is not, that know how to adjust a prompt or redesign a workflow. AI literacy is not a one-time training. It is an ongoing competency. Organisations that invest in their people’s AI knowledge today are building the adaptability that makes the difference tomorrow.
Stay independent. LeCun is building an alternative. China is investing heavily in its own models. Open-source models are becoming increasingly powerful. The market is not getting smaller, it is getting broader. Those who lock themselves into a platform or vendor today limit their options tomorrow. Vendor neutrality is not a luxury for large enterprises. It is a strategic necessity for every organisation that takes AI seriously. Choose tools based on what you need, not on what a salesperson offers. And make sure you can always switch.
The real lesson
It is tempting to read this story as a technical debate between scientists. LLMs versus world models, transformers versus JEPA. But for those leading an organisation, it is not about architecture. It is about ownership.
The technology is still very much in motion. That makes it exciting and at the same time unpredictable. The organisations that handle this best are not the ones that bet on the right horse. They are the organisations that can determine their own course, regardless of which horse wins. Because they understand it. Because they have the people. Because they are not dependent on a single vendor.
Ownership of your AI strategy is not a nice-to-have. It is what makes the difference between moving forward and standing still.