In a bold challenge to the dominant trajectory of artificial intelligence, Yann LeCun, former chief AI scientist at Meta, has raised over $1 billion for his new startup, Advanced Machine Intelligence (AMI). The Paris-based company is pursuing an alternative vision for AI – one that prioritizes understanding the physical world.
AMI’s core mission is to develop “world models” – AI systems capable of reasoning, planning, and interacting with real-world environments. This approach stands in contrast to the prevailing strategy embraced by companies such as OpenAI and Anthropic, which focus on scaling large language models (LLMs).
LeCun has consistently argued that LLMs, while powerful in language generation and coding tasks, lack a fundamental understanding of how the world works. Instead, he believes true intelligence requires systems that can model causality, physical interactions, and real-world constraints – what researchers often describe as “common sense.”
This gap is widely recognized in AI research. Systems trained purely on data patterns often struggle with tasks requiring implicit world knowledge or reasoning beyond observed examples. The idea that intelligence must be grounded in structured knowledge and reasoning is not new, but it is gaining renewed urgency as AI systems are deployed in increasingly complex environments.
A practical example of this approach can be seen in the work of QuData, whose research into common-sense AI mirrors many of the principles behind LeCun’s vision. Rather than relying solely on neural networks, the QuData team developed DemonScript – a multi-valued logic language designed to model real-world knowledge, relationships, and rules.
The system enables AI to construct semantic networks, represent object relationships such as spatial positioning, and perform probabilistic reasoning over dynamic scenarios. It can even analyze simple “micro-stories” and answer comprehension questions by building internal world models, demonstrating an ability to go beyond pattern recognition toward structured understanding.
This hybrid approach, combining data-driven learning with explicit knowledge representation, highlights a broader industry shift toward integrating reasoning capabilities into AI systems.
AMI marks LeCun’s first commercial venture since leaving Meta in late 2025, where he founded the influential FAIR (Fundamental AI Research) lab. The startup’s leadership includes several former Meta researchers, alongside CEO Alexandre LeBrun and chief science officer Saining Xie.
Unlike Meta’s consumer-focused AI strategy, the company will initially focus on enterprise applications, targeting industries with complex physical systems such as manufacturing, aerospace, and biomedical sectors. One potential use case involves building detailed digital models of machinery – such as aircraft engines – to optimize performance, improve reliability, and reduce emissions.
The company is also exploring collaborations with major corporations including Toyota and Samsung, with longer-term ambitions to expand into consumer applications such as intelligent assistants and even domestic robots.
Beyond technology, AMI is also entering the growing debate over who should control advanced AI systems. LeCun has emphasized that such powerful technology should not be governed by a small group of private companies. Instead, he advocates for open-source development and democratic oversight, arguing that decisions about AI use – particularly in sensitive domains like defense – should be made at a societal level.
AMI plans to release its first models soon, initially focusing on partnerships with large industrial players. The ultimate goal, however, is far more ambitious: the creation of a “universal world model” – a general-purpose AI system capable of understanding and interacting with the real world across domains.
If successful, this approach could redefine the path toward artificial general intelligence, shifting the focus from language prediction to embodied understanding.
For now, AMI represents a high-stakes experiment – one that could either validate LeCun’s long-held skepticism of LLM-centric AI or reinforce the industry’s current trajectory. Either way, it signals that the future of AI is far from settled.
