The field of artificial intelligence is expanding quickly. However, one of its own legends thinks it’s headed in the wrong way. Advanced Machine Intelligence is a new startup founded by Yann LeCun, the Turing Award-winning scientist who helped design modern deep learning. The business recently completed a $1.03 billion investment round. That figure needs to be taken into consideration. The concept behind it also does.
LeCun spent over a decade at Meta building one of the most respected AI research teams on the planet. He watched the rise of large language models up close. He studied their strengths. He also studied their failures. And somewhere along the way, he reached a conclusion that put him at odds with most of the industry he helped create. Text prediction, he believes, is not intelligence. It is a very expensive imitation of it.
This is not a lighthearted viewpoint. For years, LeCun has made this claim in public. He has discussed it, written about it, and received a lot of backlash for it. The majority of Silicon Valley increased its focus on massive language models and transformers. He turned away. That conviction currently resides at AMI Labs.
The company is built around a different technical approach entirely. Instead of training systems to predict the next word in a sequence, AMI is developing what researchers call world models. These are AI systems designed to build internal representations of how reality works. They understand cause and effect. They anticipate consequences. They reason about the physical world rather than just generating plausible-sounding language about it.
This work is based on a framework known as the Joint Embedding Predictive Architecture, or JEPA. This concept was first presented by LeCun as a theoretical substitute for generative AI a number of years ago. The fundamental idea is that rather than learning statistical patterns from text, intelligent systems should learn by comprehending the relationships and structure of data. Reading about fire does not teach a child that it burns. Experience, forecasting, and consequences are how they learn. AMI wants machines to acquire knowledge in a similar manner.
Investors clearly found the argument compelling. The funding round drew participation from some of the most credible names in venture capital and technology. Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions all contributed. Individual backers include figures like Mark Cuban, Eric Schmidt, and Jim Breyer. Nvidia and Samsung also joined the round. These are not naive participants. They understand the industry deeply. Their collective decision to back an approach that challenges the dominant paradigm signals genuine skepticism about where current AI is headed.
LeCun and Alexandre LeBrun co-founded AMI. LeBrun has extensive experience in research settings and is no stranger to creating AI businesses. He is equally concerned about huge language models as LeCun is. His particular worry is hallucination, or the propensity of existing AI systems to produce self-assured lies. This is bothersome in informal situations. It is risky in emergency situations. Confident errors cannot be tolerated in settings like a hospital, a factory floor, or an aircraft system. AMI is designed for settings where dependability is essential.
The first commercial partnership reflects this directly. A digital health company will gain early access to AMI’s technology, with the goal of developing AI systems that could meet regulatory certification standards in medical settings. That is an extraordinarily high bar. It also makes complete sense. If your technology genuinely works better than what exists, proving it in the hardest possible environment is the fastest way to establish credibility.
AMI’s wider target market encompasses advanced manufacturing, pharmaceuticals, automobiles, and aerospace. These sectors have been both interested in and wary of the AI growth. The increases in productivity are appealing. The gaps in reliability are a deal-breaker. LeCun and LeBrun are wagering that doors that existing AI is unable to open will be unlocked by a technically superior foundation.
The company will operate across four cities. Paris serves as the headquarters. New York connects to LeCun’s academic work at NYU. Montreal brings deep machine learning expertise, given its history as one of the world’s great AI research hubs. Singapore extends the company’s reach into Asian markets and talent pools.
This vision has brought together a sizable crew. Operators who have created and grown technological companies are seated next to researchers with solid publication histories and industry expertise. While LeBrun oversees the day-to-day operations of the corporation, LeCun’s position as chairman keeps him close to the scientific direction.
There is an interesting wrinkle worth noting. LeCun has spoken about the possibility of AMI’s technology being used inside Meta’s augmented reality hardware. The company he left is potentially also a customer of the company he built after leaving. That kind of outcome would be unusual in most industries. In AI, it reflects how rapidly the competitive and collaborative lines keep shifting.
Additionally, AMI has pledged to support open science. There will be publications of research papers. We’ll share the code. The founders think that creating an open environment around their work produces long-term benefits rather than vulnerabilities and that transparency speeds up progress. This contrasts with the main AI laboratories’ increasingly private stance, which treats cutting-edge research as proprietary by default.
Raising a billion dollars without a shipping product is not common. Doing it while explicitly arguing that the entire dominant paradigm is flawed is rarer still. AMI Labs is not incrementally improving what already exists. It is making a structural bet that the next era of AI requires a different foundation altogether.
LeCun has made mistakes in the past. All scientists have. However, he has also been correct in ways that have permanently altered the field. The billion-dollar check writers appear to believe that the odds are worth considering. Whether it acknowledges it or not, the rest of the industry is observing.
Leave a Reply