How close are we to generalized intelligence, according to Yann LeCun

Let’s start with the basics. When people throw around the term “AGI” - artificial general intelligence, they often mean a system that can think, learn, and adapt across different tasks the way humans do.

Right now, our most advanced AI systems, like ChatGPT or Gemini, are great at generating text, code, and even images. But Yann LeCun, Meta’s chief AI scientist and one of the founding fathers of modern AI, argues that this doesn’t make them “general.”

LeCun says we’re not there yet

His point is simple: today’s models don’t really understand the world. They’re masters of pattern recognition, not reasoning or planning. They don’t build internal models of how reality works, so they can’t reliably predict what will happen next in a complex situation. Without that, calling them “general” is like mistaking a good parrot for a scientist.

What AI promises to our future

AI is having its steam-engine moment. The buildout is visible in cranes over new data centers, invisible in the tools creeping into every spreadsheet, inbox, and factory line. The promise is enormous, but so is the bill. The Generative AI productivity upside is real, but it’s uneven Let’s

The CTO NewsletterArjus Dashi

We need world models, not just bigger models

So, what’s missing? LeCun believes the next leap requires “world models” internal simulations that let machines anticipate, plan, and learn the way humans and animals do. Imagine how a toddler learns: they drop a ball, watch it bounce, and gradually form a sense of gravity. They don’t just memorize examples, they build an intuition.

Meta has already started working on this idea with I-JEPA, a new kind of model that predicts abstract representations instead of just predicting the next pixel or the next word.

Current AI is like a student who memorizes past exams and gets good grades when the questions look familiar. LeCun wants to build the student who can walk into a brand-new situation, reason it out, and still succeed - even if they’ve never seen that exact problem.

Timelines, risks, and what leaders should actually do

Now, about the big question everyone asks: “When will AGI arrive?” LeCun’s answer is measured. He doesn’t buy the hype that it’s just around the corner. He thinks it will take decades of research and new breakthroughs, not just scaling up today’s methods. In other words, don’t expect AGI to pop up in the next product cycle.

Two Months Across U.S. Startup Hubs - Boston, Austin, San Francisco

This summer, I decided I needed more than meetings and remote calls. I wanted to connect with the raw energy of founders pushing things forward. So, packed my bags and spent two months driving through three U.S. startup hubs: Boston, Austin, and San Francisco. I wanted to emerge with

The CTO NewsletterArjus Dashi

Whether you’re studying this field or leading teams in business, the key is perspective: don’t buy into the hype cycles, but don’t dismiss the long-term potential either. We’re running a marathon, not a sprint.

Thank you for reading - Arjus