Current artificial intelligence models do not have the major features of human intelligence, said the head of Meta AI, saying that the last model of the company will solve this problem.
Business Insider reported on May 26 that at the AI action summit in Paris earlier this year, the chief scientist of AI, Yann Lecun, said that “there are four essential characteristics of intelligent behavior that each animal, or a relatively intelligent animal, can do and certainly humans”.
“Understanding the physical world, having a persistent memory, being able to reason and be able to plan complex actions, in particular hierarchical planning,” said Lecun.
He said that the current large language models (LLM) which feed popular AI chatbots have not reached this threshold, and “the incorporation of these capacities would require a change in the way they are formed”.
Some of the largest AI and technology giants are “paving capacities” on existing models in their race to dominate the AI game, said Lecun.
Meta is already experimenting with a system called increased generation recovery (RAG), which is a method of improving LLM outputs using external sources of knowledge.
In February, he published V-Jepa, a non-generative model that learns by predicting missing or masked games of a video.
In relation: Meta gets a nod to the EU regulator to form AI with the content of social media
Lecun thinks that the “world models” would be a better approach because these would be formed on real scenarios and have higher cognition than current AI based on models.
The concept involves models that can imagine taking action and predicting the resulting world state. Since the world has unpredictable infinite possibilities, Lecun thinks that training must occur by abstraction, which reflects the way humans give meaning to the physical world.
Meta’s Ai Brain Drought
Meanwhile, Meta undergoes a significant loss of talents from her AI research team, especially among the researchers who created the original Llama model in 2023, Insider reported May 26.
Only three of the 14 original authors of 14 Lama remain in Meta, and many have joined Mistral, a startup based in Paris co -founded by former Meta Researchers and Key Llama Architects.
The latest version of Meta, Llama 4, has received a lukewarm reception of developers, many of whom are now looking for faster rivals that have dedicated reasoning models such as the GPT-4O of Openai, Gemini 2.5 pro recently launched and the recently launched Claude 4 report, added the report.
On May 15, the Wall Street Journal reported This meta delayed the deployment of its flagship product AI LLM, Llama 4 “Behemoth”.
Review: Ai liquité, “ good ” Propaganda bots, OPENAI DOOMSDAY BUNKER: AI EYE
