AI Is Not Smart Yet is the message from artificial intelligence pioneer Yann LeCun, who believes today’s AI systems like ChatGPT, Gemini, and Claude are powerful but still lack genuine understanding of the real world.
Speaking at VivaTech 2026, LeCun explained why the future of AI lies beyond Large Language Models (LLMs) and toward a new generation of intelligent systems that can reason, plan, and understand physical reality.
The vision is now being developed at his startup, AMI Labs, backed by more than $1 billion in funding.
Table of Contents
- Why AI Is Not Smart Yet
- What Is JEPA?
- Why Robotics Needs Smarter AI
- World Models Could Replace Traditional AI
- What Happens Next?
Why AI Is Not Smart Yet
According to Yann LeCun, today’s AI models mainly predict the next word instead of understanding how the world actually works.
While they perform extremely well in writing, coding, and answering questions, they struggle when facing unpredictable situations that require reasoning.
LeCun argues that current LLMs memorize patterns from huge datasets instead of building a true understanding of physical reality. This limitation prevents them from achieving human-level or even animal-level intelligence.
What Is JEPA?
AMI Labs is developing Joint Embedding Predictive Architecture (JEPA), a new AI system designed to understand the world instead of simply predicting text.
Rather than trying to guess every possible outcome, JEPA creates simplified internal representations of reality. This allows AI to focus only on meaningful information while ignoring details that cannot realistically be predicted.
For example, if a pen is balanced on its tip, humans know it will eventually fall but cannot predict the exact direction.
Traditional language models may still attempt to guess, whereas JEPA understands that predicting the direction is unnecessary and instead focuses on the broader physical outcome.
Why Robotics Needs Smarter AI
One of the biggest challenges in artificial intelligence is robotics.
Although modern robots can perform impressive movements, teaching them everyday household tasks remains difficult.
LeCun believes language models alone cannot solve these problems because robots must understand physics, movement, objects, and changing environments in real time.
This is why researchers are now investing heavily in AI systems capable of reasoning instead of only generating language.
World Models Could Replace Traditional AI
Researchers worldwide are developing World Models, which allow AI to simulate reality before making decisions.
Instead of predicting words, these systems learn cause-and-effect relationships and imagine possible future scenarios before acting.
This technology could dramatically improve robotics, autonomous vehicles, manufacturing, and healthcare.
AMI Labs recently secured more than $1 billion in seed funding to accelerate research into world-model AI systems that understand real-world environments.
What Happens Next?
LeCun expects AMI Labs to continue refining JEPA before introducing it into industrial applications.
If successful, future AI systems may perform complex physical tasks with far less training than today’s robots require.
Despite rapid advances, LeCun believes humans will continue making strategic decisions while AI serves as an intelligent assistant capable of reasoning through increasingly complex problems.
The next decade could mark the transition from text-based AI toward machines that genuinely understand and interact with the physical world.
Related Article: Global Stock Markets Rally Today
Official Source: AMI Labs Official Website
I am Rukaiya Kadiwala, an experienced News Content Writer with 6+ years of expertise in hospitality, travel, hotel, restaurant, business, and lifestyle news. Skilled in writing, research, fact-checking, headline creation, and digital publishing, I create accurate, engaging, and high-quality content that informs and attracts readers worldwide.
