Yann LeCun, founder of Advanced Machine Intelligence
Getty Images
Alex LeBrun, co-founder of medical startup Nabla who LeCun hired to become CEO of Advanced Machine Intelligence, believes the startup’s new AI models could begin rolling out within a year, with healthcare being a major focus area.
In November, Yann LeCun, one of the world’s leading AI experts, announced that he would be leaving his role as chief AI scientist at Meta to start his own company, focused on creating a new type of AI called global models. He is now executive chairman of Advanced Machine Intelligence Labs, which is seeking to raise funding worth approximately $3.5 billion. But he made a seemingly unusual choice for CEO: Alex LeBrun, co-founder of a Paris-based health tech startup that uses AI to transcribe doctor visits.
This may seem like a niche choice for an AI lab. But it does indicate that the dynamic company, whose technology attempts to understand the world by learning from video, spatial data as well as text, will focus on healthcare. Its first announced partnership will be with LeBrun’s company, Nabla.
“Health care is my baby, and we know what problems we can’t solve today,” LeBrun said in an interview on the sidelines of the JP Morgan Healthcare conference in San Francisco. “We hope that this new branch (of AI) will help us go beyond what we can do today in health care. »
Alex LeBrun, co-founder of Nabla and CEO of Advanced Machine Intelligence
Nabla
LeBrun will relinquish his role as CEO at Nabla, but will remain president and chief AI scientist there. He previously worked directly for LeCun at Meta after the acquisition of a previous company he founded. LeBrun said Forbes that a key reason he wanted to become CEO of the new startup was to apply its global models to healthcare.
AMI also targets manufacturing and robotics, but healthcare is an important area of focus due to the complexity and high stakes of the field. There are high hopes that AI will transform both operations and clinical practice. But the tendency of large language models to hallucinate or make errors poses a major potential problem in health care, especially on the clinical side, where people’s lives are involved. Correcting these hallucinations is something that proponents of global models believe they can solve.
Health data is “continuous, high-dimensional and noisy,” whether it comes from medical tests or sensors, LeCun said. Forbes by email. Although generative AI works for clinical documentation, “generative approaches don’t work well for more impactful applications,” he said. “Global models enable agent systems to predict the consequences of their actions and plan sequences of actions to complete a task, subject to safety guardrails. This will pave the way for a new class of AI applications in healthcare, where reliability, controllability and security really matter.”
With AMI, LeCun is raising a lot of money to put this research into practice: The Financial Times previously reported that the new startup was looking to raise nearly $600 million (at current Euro-dollar exchange rates). Others have also developed global models, including Professor Fei-Fei Li of Stanford, whose Global Laboratories came out of stealth in September 2024 with $230 million in funding at a valuation of over $1 billion.
“Over the last five years, the world has become totally obsessed with LLMs. Now we have a hammer and we see everything as a nail,” LeBrun said. “For some problems, like information retrieval, it’s the best technology. But many problems don’t fall under it.”
In healthcare, for example, AI tracking (including Nabla), clinician support (such as OpenEvidence) and discovering imaging patterns, such as X-rays, are good uses of existing LLM technology, LeBrun said. But for other clinical use cases, the accuracy is too low for LLMs to be deployed safely – let alone AI agents that operate autonomously without human oversight. “If you zoom out, it’s only 1 to 5 percent of the problems,” he said. “No one would argue that health care works well today. »
AMI plans to partner with companies to adapt its global models to real-world problems, although beyond Nabla no specific projects have yet been announced. The French startup has raised $120 million and hopes to reach annualized subscription revenue of $100 million within two years. Nabla co-founder Delphine Groll, the company’s chief operating officer, will lead that company until a new CEO to replace LeBrun is named. “Alex has always been the company’s AI visionary,” she said.
LeBrun estimates it will take “about a year to get the first elements that we can use in the product.” Research on global patterns is already extensive, with hundreds of papers published in the last few years alone, he said. “I felt the time was right given the recent advances in research on global models to apply this to healthcare, so that’s what we’re about to do with Nabla,” he said.
MORE FROM FORBES