Current artificial intelligence models lack the major traits of human intelligence, Meta’s AI chief has reportedly said, claiming that the firm’s latest model will solve this issue.
Business Insider reported on May 26 that at the AI Action Summit in Paris earlier this year, Meta chief AI scientist Yann LeCun said that “there are four essential characteristics of intelligent behavior that every animal, or relatively smart animal, can do, and certainly humans.”
“Understanding the physical world, having persistent memory, being able to reason and being able to plan complex actions, particularly planning hierarchically,” LeCun said.
He said current large language models (LLMs) that power popular AI chatbots have not hit this threshold, and “incorporating these capabilities would require a shift in how they are trained.”
Some of the largest AI and tech giants are “cobbling capabilities” onto existing models in their race to dominate the AI game, LeCun said.
Meta is already experimenting with a system called retrieval augmented generation (RAG), which is a method of enhancing LLM outputs using external knowledge sources.
In February, it released V-JEPA, a non-generative model that learns by predicting missing or masked parts of a video.
Related: Meta gets EU regulator nod to train AI with social media content
LeCun believes that “world-based models” would be a better approach as these would be trained on real-life scenarios and possess higher cognition than current pattern-based AI.
The concept involves models that can imagine taking an action and predict the resulting world state. Since the world has infinite unpredictable possibilities, LeCun believes training must happen through abstraction, which mirrors how humans make sense of the physical world.
Meta’s AI brain drain
Meanwhile, Meta is experiencing significant talent loss from its AI research team, particularly among the researchers who created the original Llama model in 2023, Insider reported on May 26.
Just three of the original 14 Llama authors remain at Meta, and many have joined Mistral, a Paris-based startup co-founded by former Meta researchers and key Llama architects.
Meta’s latest release, Llama 4, received a lukewarm reception from developers, many of whom now look to faster-moving rivals that have dedicated reasoning models such as OpenAI’s GPT-4o, Google’s Gemini 2.5 Pro, and the recently launched Claude 4 Sonnet from Anthropic, the report added.
On May 15, The Wall Street Journal reported that Meta was delaying the rollout of its flagship AI LLM, Llama 4 “Behemoth.”
Magazine: AI cures blindness, ‘good’ propaganda bots, OpenAI doomsday bunker: AI Eye
