Two pieces of news landed today that, taken together, paint the fullest picture yet of where AI is headed — and how fiercely contested that destination will be.
On one side: OpenAI shipped GPT-5.4, its most capable model yet. On the other: Yann LeCun — one of the three “godfathers of deep learning” and the man who arguably created the theoretical foundation modern AI is built on — just raised $1.03 billion to prove the whole paradigm is wrong.
Who Is Yann LeCun, and Why Does He Keep Insisting We’re Doing AI Wrong?
If you’ve been following AI longer than six months, you’ve probably seen LeCun getting into arguments on X about why large language models are fundamentally limited. He’s not a crank — he won the Turing Award, the computing equivalent of a Nobel Prize, and spent years as Meta’s chief AI scientist before leaving in November 2025.
His core argument: LLMs — the kind of AI that powers ChatGPT, Claude, and Gemini — are autocomplete engines at heart. They predict the next word. They don’t understand the world. They hallucinate because they have no ground truth to anchor to. They can’t plan. They can’t reason about physics. And no amount of scaling is going to fix that.
The AI establishment, broadly speaking, disagrees. OpenAI, Anthropic, Google, Meta (post-LeCun), xAI — they’ve all bet billions on the idea that bigger, smarter LLMs will eventually get you to human-level intelligence. The scorecard so far looks pretty good for the incumbents: GPT-5.4 is genuinely impressive. But LeCun’s counterargument is that “impressive” isn’t the same as “right.”
Enter AMI Labs: $1 Billion for a Different Kind of AI
After leaving Meta, LeCun co-founded Advanced Machine Intelligence Labs (AMI Labs) alongside Alexandre LeBrun, former CEO of digital health startup Nabla. Today, TechCrunch reports the company closed a $1.03 billion round at a $3.5 billion pre-money valuation — Europe’s largest-ever AI seed round.
The investor list is a who’s-who of tech heavyweights: Nvidia, Temasek, and Jeff Bezos all backed the round. When Nvidia — the company that makes the chips the entire AI industry runs on — puts money into a company betting against LLMs, that’s worth paying attention to.
So what’s AMI actually building? World models.
The concept is rooted in LeCun’s 2022 paper proposing JEPA (Joint Embedding Predictive Architecture). Instead of learning from language alone, world models learn from reality — they build internal representations of how the physical world works. Think of it like the difference between a child who learned to swim by reading every book about swimming (LLM) versus one who actually jumped in the water (world model).
The applications aren’t hypothetical: AMI’s first commercial partner is Nabla, the healthcare AI startup. LeBrun’s thinking is simple — in medicine, hallucinations can kill people. You need AI that genuinely understands the world, not one that confidently generates plausible-sounding nonsense.
“My prediction is that ‘world models’ will be the next buzzword,” LeBrun told TechCrunch. “In six months, every company will call itself a world model to raise funding.” He said it with a smile, clearly confident that AMI’s version will be the real deal when the copycats arrive.
Meanwhile, OpenAI Dropped GPT-5.4 (and It’s Kind of a Big Deal)
Context, though: while LeCun is building his alternative to the future, OpenAI just shipped the future. GPT-5.4 Thinking launched today for Plus, Team, and Pro subscribers, replacing GPT-5.2 Thinking in the model picker.
This one combines OpenAI’s best recent work: reasoning from their o-series models, elite coding capabilities from GPT-5.3-Codex, and improved agentic workflows for professional tasks like spreadsheets, presentations, and complex document work. The standout new feature is a live thinking trace — you can see the model’s reasoning plan upfront and actually interrupt to course-correct before it goes too far down a wrong path. That’s a genuinely useful UX change for anyone who’s spent time arguing with an AI that spent five minutes confidently solving the wrong problem.
GPT-5.2 Thinking doesn’t disappear overnight — it’ll stick around in “Legacy Models” until June 5, 2026 — but the direction is clear. OpenAI is iterating fast, and they’re not slowing down.
Two Visions, One Very Expensive Race
Here’s the thing: these two stories aren’t necessarily in conflict. LeCun might be right that LLMs have fundamental limits. GPT-5.4 might be right that those limits are further away than critics think. Both can be true at the same time.
What makes today interesting is the scale of the bet. AMI raised $1.03 billion for a research agenda that CEO LeBrun openly admits won’t produce commercial products in months — or possibly even years. That’s an extraordinary amount of capital for fundamental research. And yet the investors — including Nvidia, a company with front-row seats to the entire AI industry — thought it was worth doing.
For context, AMI isn’t even alone in the world model space. Fei-Fei Li’s World Labs (yes, the other AI legend with a different bet on AI’s future) just secured another major round in February. The category is heating up fast.
Meanwhile, OpenAI, Anthropic, and Google keep releasing models that are, by any reasonable measure, extraordinarily capable. GPT-5.4 is better than GPT-5.3. GPT-5.3 was better than GPT-4. The pace hasn’t slowed.
So who wins? Maybe both. Maybe neither. Maybe the real breakthrough comes from some unexpected direction nobody has fully funded yet. AI has a long history of experts being confidently wrong — in both directions.
The Takeaway
If you’re following AI, today was a reminder that this field is not a single horse race with an obvious leader. There’s a mainstream LLM track where OpenAI, Anthropic, and Google are sprinting and iterating daily. And there’s an alternative track where some of the most credentialed researchers in the world are quietly laying groundwork for something potentially very different.
Whether you’re a GPT-5.4 power user or a LeCun true believer, you’re watching one of the most consequential scientific and commercial competitions in human history. The stakes are high enough that a billion dollars counts as seed funding now.
We’ll keep watching.
Sources: TechCrunch — AMI Labs raises $1.03B · Reuters — AMI raises $1.03B for alternative AI approach · WIRED — LeCun raises $1B to build AI that understands the physical world · OpenAI Release Notes — GPT-5.4 launch