Meta’s former chief AI scientist Yann LeCun just left his cushy position to start something new, and it’s a pretty bold bet against basically everyone else in the AI industry. His Paris-based startup, Advanced Machine Intelligence (AMI), just raised over $1 billion to develop what he calls “world models”, AI systems that actually understand how the physical world works, not just how to string words together.

Here’s the thing: LeCun thinks the whole chatbot craze is kind of missing the point. While everyone’s obsessing over making ChatGPT and Claude smarter, he’s arguing that you can’t actually achieve human-level AI just by scaling up language models. “The idea that you’re going to extend the capabilities of LLMs to the point that they’re going to have human-level intelligence is complete nonsense,” he said in a recent interview. That’s a pretty spicy take coming from one of AI’s most respected pioneers, who literally won the Turing Award back in 2018.

So what’s the deal with world models? Basically, LeCun believes most human reasoning comes from understanding how the physical world actually works, not from language. AMI’s goal is to build AI systems that can understand how things work, remember what they’ve learned, reason through problems, and do all of this safely. The startup brought in some serious money from investors like Mark Cuban, former Google CEO Eric Schmidt, and billionaire Xavier Niel, valuing the company at $3.5 billion.

The company is setting up offices in Paris, Montreal, Singapore, and New York (where LeCun will keep his teaching gig at NYU), and they’re already eyeing partnerships with companies like Toyota and Samsung. LeCun’s vision is pretty specific: imagine building a realistic AI model of an aircraft engine that helps manufacturers optimize efficiency, cut emissions, or ensure reliability. That’s the kind of practical application he’s targeting.

What makes this move particularly interesting is that LeCun left Meta specifically because the company pivoted hard into the LLM space to keep up with OpenAI and other competitors. He told Mark Zuckerberg that he could do this world model research faster, cheaper, and better outside the company. Zuckerberg apparently said cool, and they’re open to collaborating, though Meta isn’t directly investing in AMI.

LeCun’s also making noise about open-sourcing the technology. He’s concerned that AI is too powerful for any single company to control, and he thinks the democratic process (at least in democracies) should decide how the tech gets used, not tech CEOs. It’s a pretty refreshing stance in an industry full of billionaires convinced they know what’s best for everyone.

AMI plans to release its first AI models pretty soon, though don’t expect headlines just yet. The company’s playing the long game, starting with enterprise partnerships before eventually building what LeCun calls a “universal world model”. It’s ambitious as hell, but if anyone’s got the credentials to pull it off, it’s probably the guy who basically invented modern AI in the first place.