Why Today’s AI Still Doesn’t Think Like Us, According to Meta’s Chief Scientist

Meta’s top AI scientist, Yann LeCun, is challenging the status quo in artificial intelligence, arguing that current language models—despite their popularity—lack the foundational traits of real intelligence.

Speaking at the AI Action Summit in Paris, LeCun outlined a vision for models that do more than pattern-match words: ones that can reason, plan, and interact with the world like humans and animals do.

According to LeCun, AI must evolve beyond today’s chatbot engines to understand physical environments, retain long-term memory, reason through complex problems, and make structured plans. He criticized the industry’s race to bolt new functions onto existing models, calling instead for a fundamental rethinking of how AI learns.

Meta is already experimenting with alternative approaches. One such technique, retrieval-augmented generation (RAG), allows models to pull in external information during responses. Another is V-JEPA, a video-based system that learns by predicting missing frames—an attempt to mimic how humans learn by observing the world.

READ MORE:

Circle Dismisses Acquisition Rumors, Reaffirms IPO Focus

LeCun advocates for “world-models,” AI systems trained to simulate cause-and-effect by predicting the outcomes of imagined actions. This, he says, brings AI closer to human cognition, where understanding arises from abstraction, not just data patterns.

But Meta’s bold ambitions are clashing with internal setbacks. The company is facing a brain drain from its AI research division. Many of the minds behind its Llama model have exited, with several joining Paris-based rival Mistral. Only 3 of the original 14 Llama architects remain at Meta, according to Insider.

The latest iteration of Meta’s flagship model, Llama 4, hasn’t generated much excitement among developers, who are increasingly gravitating toward faster-moving competitors like OpenAI’s GPT-4o, Google’s Gemini 2.5 Pro, and Anthropic’s Claude 4 Sonnet. Reports also suggest Meta has delayed the release of its most advanced version, Llama 4 “Behemoth.”

As the AI arms race intensifies, LeCun’s call for deeper, more grounded intelligence may prove critical—but whether Meta can lead that shift while holding onto its top talent remains to be seen.

Alexander Stefanov

With over 8 years of experience in the cryptocurrency and blockchain industry, Alexander is a seasoned content creator and market analyst dedicated to making digital assets more accessible and understandable. He specializes in breaking down complex crypto trends, analyzing market movements, and producing insightful content aimed at educating both newcomers and seasoned investors. Alexander has built a reputation for delivering timely and accurate analysis, while keeping a close eye on regulatory developments, emerging technologies, and macroeconomic trends that shape the future of digital finance. His work is rooted in a passion for innovation and a firm belief that widespread education is key to accelerating global crypto adoption.