“Currently, AI systems are in many ways very stupid. We are fooled into thinking they are smart because they can manipulate language very well.” — Yann LeCun, Meta VP & Chief AI Scientist

It’s not every day that one of the architects of modern artificial intelligence calls the entire field… dim-witted.

Yet that is precisely what Yann LeCun—Turing Award laureate, Meta’s VP & Chief AI Scientist, and pioneer of convolutional neural networks—does in this candid assessment of the state of the art. In an industry increasingly intoxicated by its own progress, LeCun plays the voice of sobering reason. Not out of cynicism, but clarity.

Despite building some of the most influential AI technologies in use today, he is unflinching: AI systems may seem clever, but they remain fundamentally incapable of true understanding. And no, ChatGPT won’t be taking over the world anytime soon.

The Illusion of Intelligence

LeCun’s central critique is disarmingly simple: just because language models sound convincing doesn’t mean they understand a thing.

He calls out the widespread illusion fostered by large language models:

“They can’t understand the physical world. They don’t really have any persistent memory of the type that we have. They can’t really reason, and they can’t plan—and those are essential characteristics of intelligent behavior.”

In other words, today’s AI isn’t thinking. It’s autocomplete on steroids. Witty, well-packaged, and thoroughly clueless.

It’s like listening to someone read the dictionary with a British accent—polished, but without insight.

The Gravity of the Situation

LeCun finds it “ridiculous” that people think powerful language models are on the verge of becoming sentient beings with world-domination agendas. The talk of machines plotting against us not only misunderstands what AI is, but wildly overestimates what it’s currently capable of.

One of LeCun’s more accessible examples? Gravity. Specifically, how small children, dogs, and even cats can grasp basic physical principles simply through experience.

“You can take a six-month-old baby… a dog or a cat—they learn how the world works. For example, if you drop an object on the floor, they know it’s going to fall. They’ve seen it. They’ve experienced it. AI systems don’t have this.”

In short: a cat learns physics through trial and error. A language model, no matter how linguistically dazzling, has never seen an apple fall from a tree. Yet it will confidently deliver a 500-word summary of Newton’s second law, footnotes and all.

Convincing? Perhaps. Deep understanding? Not remotely.

So no, your cat is not out of a job. If anything, it might be a few evolutionary steps ahead of your chatbot.

The Missing Pieces: Memory, Reasoning, and World Models

One reason AI still feels uncanny—and often glitchy—is its lack of persistent memory. Language models forget what happened ten seconds ago unless finely engineered to retain context.

“There is no memory of the type that humans or animals have, where you remember what happened five minutes ago, what you saw yesterday, or what you learned from it all.”

In short: AI doesn’t learn the way we do.

Humans link events over time. Shoot a basketball three times, miss each one, and you’ll adjust your aim without needing a dataset of 10,000 shots. AI? It’ll still be loading the training data.

Reasoning and planning are equally absent.

“They can’t plan. There’s no consistent, goal-directed reasoning. You can try hard to strap various systems together, but none of these language models actually understand.”

They assemble sentences the way Pinterest assembles mood boards: collage, not cognition. Coherence, yes. Comprehension, no.

What Would an Actually Smart AI Look Like?

LeCun believes that real intelligence must emerge from AI models that can learn like animals: not by hoovering up terabytes of labelled data, but through experience—observation, imagination, memory.

He and his colleagues at Meta AI are working on precisely that: world models. These systems don’t just react to prompts—they learn how the world works. They build an internal simulation, so they can reason, predict, and plan in dynamic environments.

“The way intelligence works is by predicting what happens in the world as a consequence of your actions. And today’s AI, with rare exceptions, doesn’t do this.”

Teaching AI to predict the consequences of its actions sounds wonderfully mundane. And yet, it’s a milestone more significant than yet another song in the style of Taylor Swift.

Why the “AGI is Imminent” Crowd Should Take a Walk

Much ink has been spilled on Artificial General Intelligence (AGI)—that hypothetical moment when a machine equals, or surpasses, human reasoning. For some, it’s always five years away. For LeCun, it’s certainly not here, and probably not even on the runway yet.

He doesn’t deny its possibility. In fact, he’s one of the researchers trying to build better systems. But he also isn’t losing sleep:

“The idea that we’re anywhere close to machines being able to understand the world like humans do is… very premature.”

The panic, in his view, is powered more by imaginative fiction than by empirical science. If an AI can’t remember what it said in the last paragraph, then perhaps we should worry less about it outsmarting humanity—and more about not giving it your tax returns to summarise unsupervised.

The Takeaway: Remarkable, but Not Intelligent

We owe much of today’s awe-inspiring AI to researchers like Yann LeCun. But even he won’t overstate its brilliance.

The truth? Today’s AI is fast, fluent, and fundamentally superficial. It parses patterns but does not possess purpose. It mirrors intelligence but doesn’t climb inside it.

LeCun’s core message is not that AI is useless—it’s tremendously powerful and has revolutionary applications across science, industry, and daily life. But it is nowhere near what most people might call “understanding.”

To put it plainly: we’re using extremely high-powered autocomplete machines and mistaking them for minds.

Final Thoughts

So, next time someone claims that ChatGPT will soon file your taxes, run your business, or unilaterally decide to eliminate humanity, take a breath—and perhaps a page from LeCun.

Ask your AI to catch a ball or remember what you said yesterday. You’ll see the limits quite quickly.

And as far as Yann LeCun is concerned, the gap between what we have and what we imagine is vast. It’s not insurmountable—but getting there will require more than better language models, snappier slogans, or end-of-days press releases.

Until machines can understand something as simple—and profound—as gravity, we might hold off on making them our digital overlords.

After all, intelligence isn’t just about sounding smart. It’s about knowing why the toast lands that way in the morning.

Source: Interview with Yann LeCun

For more insights about what AI can or cannot do, check out my book “Artificial Stupelligence: The Hilarious Truth About AI”: