Meta AI boss LeCun said: Today's AI is not smart at all, and the Llama team also broke out of the trend

Meta AI Chief Scientist Yann LeCun stated at the Paris AI Summit on 5/26 that current chatbots merely "imitate" and are far from being "intelligent". He revealed that Meta's AI approach will fully shift towards "world model" technology, hoping to enable AI to understand the world like humans. However, while he presented this new vision, core members of Meta's Llama model team have been leaving, causing external concerns about Meta AI's development strategy.

Current AI is Imitating Human Speech and Doesn't Understand How the World Works

LeCun believes current AI models only imitate human speech without truly understanding the content. Real intelligence should include understanding the physical world, having memory, reasoning, and planning abilities. However, these functions are not currently available in large language models.

He then proposed the "world model" concept, hoping to enable AI to predict action results in simulated environments. This approach can mimic human abstract thinking, and Meta has started experimenting with:

• Retrieval-Augmented Generation (RAG): Combining external databases to improve response accuracy
• Video-Joint Embedding Predictive Architecture (V-JEPA): Training AI to understand scenes by predicting masked video segments

LeCun believes this is the path AI should take in the future.

Llama Team Continues to Leave, External Doubts About Technological Leadership

While LeCun publicly stated Meta AI's future vision, internal team members have been leaving. Of the 14 core members who built the Llama model, only 3 remain at Meta, with the rest jumping to the Paris startup Mistral. According to reports, Meta's flagship model "Behemoth" has been delayed.

Although Llama 4 has been released, the response has been poor. Developers generally believe that competitor models like GPT-4o, Gemini 2.5 Pro, and Claude 4 are stronger, with more powerful reasoning capabilities and more human-like qualities, raising doubts about Meta's ability to maintain technological leadership.

AI's Future is Not About Who Releases Faster, Understanding and Thinking are Key

LeCun clearly pointed out that current AI models are still far from being truly "intelligent". While Meta wants to rebuild from the ground up, talent exodus and model delays have cast a shadow on its vision. The key to future AI competition will be who first enables AI to understand how the world works, not who releases a new version first.