Yann LeCun, former chief AI scientist at Meta, helped lay the groundwork for the modern AI boom long before chatbots became a hot topic in the industry. Well, after that leave meta In late 2025, he’s building a multibillion-dollar startup on the premise that the race to AI superintelligence is starting in the wrong place. LeCun, who spoke at the VivaTech conference on June 17, said: Shared more information On why he believes his startup Advanced Machine Intelligence (AMI Labs) is on a better path toward human-level AI Earlier this year, the Paris-based company Raised $1.03 billion At a valuation of $3.5 billion, it gives LeCun and his team new support to pursue alternatives to today’s generative AI models developed by tech giants.
“I think we need a new paradigm that goes beyond the limitations of the current system,” LeCun, 65, told Wired’s Stephen Levy on stage.
Building better language models, as many AI companies are doing, may be one way, but “it’s kind of slow,” LeCun says. LeCun said AMI Labs is building “world models” that learn from reality, understand outcomes, predict what will happen next, and choose the best course of action based on those predictions. “We have high hopes that this will truly change the blueprint for intelligent systems,” LeCun said.
LeCun believes that much of Silicon Valley is “LLM-soaked,” or overly confident that scaling large language models will eventually produce human-like intelligence. The result, he argued, is a “monoculture” of AI systems that are powerful in some areas but remain fundamentally limited.
LLMs are built around languages, even though they already have advanced capabilities in areas such as coding. He believes a different approach is needed because real-world data from cameras, robots, and other sensors is more complex and difficult to predict. For AMI Labs, the results look like this: “It’s a fairly intelligent system that can predict, plan and reason about the consequences of its actions, perhaps reaching the level of human intelligence,” he said.
LeCun’s critique carries weight. A longtime professor at New York University, he is one of a trio of researchers often referred to as the “godfathers of AI.” The other three are Jeffrey Hinton, a professor at the University of Toronto and former Google employee, and Yoshua Bengio, a professor at the University of Montreal who founded Mira, an AI research institute in Quebec. In 2018 they We all won He received the prestigious Turing Award for his groundbreaking work in bringing deep learning to the mainstream.
From FAIR to AMI Lab
2013, LeCun joined facebook He started what would become Facebook AI Research (FAIR) and later became Chief AI Scientist at Meta. He said AMI’s roots were an internal FAIR project, with support from company leaders including Mark Zuckerberg and CTO Andrew Bosworth. But as Meta’s AI priorities shifted to catching up with its competitors, the environment became less favorable.
At that point, LeCun said, it made sense to “leave the company and try to accelerate development and build what the company can build.” “Then we realized we could raise enough money to keep this going.”
In March, AMI Labs raised a $1.03 billion seed round, the largest in European startup history. Investors include Nvidia, Jeff Bezos, and Eric Schmidt. Mr. LeCun serves as executive chairman, and the startup is led by CEO Alex LeBrun. He is a former FAIR Head of Engineering and co-founder of health AI company Nabla, AMI Labs’ first partner. The company has assembled a research-focused team with experience at Meta, Google DeepMind, and other leading AI labs.
AMI says it aims to build applications in areas where this approach is important, including industrial process control, automation, wearable devices, robotics, and healthcare. “That’s not going to happen next year,” LeCun said. “We’re not going to have a nation of geniuses in our data centers next year, are we?”
At VivaTech, the discussion expanded to larger issues. Who should control the next generation of AI? If the assistants of the future are built only by a “few companies” in Silicon Valley or China, “culture is in big trouble” and “democracy is in big trouble,” LeCun warned. He argued that blocking open source models in the name of safety misses out on the role AI can play in disseminating knowledge.
This stance also informs LeCun’s work as chief scientific advisor for Project Tapestry, an AI Alliance-led effort to build shared global AI models through a distributed network of contributors. The AI Alliance is a nonprofit organization jointly launched by Meta and IBM in 2023 to support open source AI development. Tapestry’s goal is to enable countries, companies, universities, and others to contribute data and computing power without giving up control of their data. As AI begins to mediate what people read, watch and search, LeCun argues that people need an alternative to what a few powerful American and Chinese companies are trying to build.
LeCun’s comments also come as Meta’s own open source position becomes more complex. Llama 2, released commercially and free in 2023, made Meta a rare Big Tech counterweight to its more closed AI rivals. But Zuckerberg After the signal was given Meta may not publicly reveal all of its future “superintelligence” models.
“We need access to a wide variety of AI assistants for the same reason we need access to a wide variety of news outlets for multiple sources,” LeCun said. “The only way I can see this could happen is if there is an underlying model that is open and free, on top of which everyone can build their own assistants specific to their own language, culture, values, political bias, and center of interest. Therefore, open source must exist.”
