AI expert says today's chatbots 'dead end' for human-like intelligence
Yann LeCun, a key figure in AI, says we're still a long way from building artificial general intelligence (AGI)—the kind of AI that could truly think and learn like humans.
Speaking at the India AI Impact Summit on Wednesday, he called today's large language models "a dead end."
After leaving Meta last year, LeCun started his own company to push for smarter, more adaptable AI.
What does future AI need?
LeCun believes future AI needs to understand the real world—think memory, reasoning, planning, and learning from experience.
He says this will only happen with "world models," which are systems that can simulate reality using physics and sensory data.
Unlike chatbots or LLMs that just process text, these AIs would handle messy stuff like video and real-world unpredictability.
LeCun doesn't like the term AGI
LeCun points out that even advanced robots can't do simple things a cat can—like moving around or reacting to surprises.
He also feels calling it "AGI" is misleading since both humans and computers are usually good at specific things rather than everything.
His new approach aims for an intuitive understanding of the world—kind of how babies learn by exploring.
His startup is building world models
His startup is working on these world models to build human-level intelligence—not just by reading internet text (which he says equals about four years of what a kid sees), but by helping AIs actually experience and adapt to the world around them.