The current Artificial Intelligence model lacks the major symptoms of human intelligence, allegedly stated by the AI chief of the meta, claiming that the latest model of the firm would solve the issue.
Business Insider said on 26 May that at the AI Action Summit in Paris earlier this year, Mata’s chief AI scientist Yan Lakeun said that “there are four essential features of intelligent behavior that every animal, or relatively smart animals, can, and of course humans.”
“Understanding the physical world, having constant memory, being able to argue and being able to plan intricate tasks, especially planning hierarchical,” said.
He said that the current big language model (Lalms) That power popular AI chatbots have not killed this limit, and “how to train these capabilities will require changes.”
Lakeun said that some of the biggest AI and tech giants are “cobbing abilities” on existing models that are on the current model in their race to dominate the AI game.
Meta Already experimenting with a system, which is called retrieval agmented generation (RAG), which is a method of increasing LLM output using external knowledge sources.
In February, it released the V-JEPA, a non-generative model that learns by predicting the missing or masked parts of a video.
Connected: Meta gets an EU regulator node to train AI with social media content
Lakun believes that the “world-based model” would be a better approach as they would be trained on real-life scenarios and will be entitled to high sensation than the current pattern-based AI.
The concept includes models that can imagine taking an action and predict the resulting world state. Since there are infinite unexpected possibilities in the world, Lakeun believes that training should be through abstraction, which shows how humans make an understanding of the physical world.
Meta’s AI Brain Drain
Meanwhile, Meta is experiencing significant talent loss from its AI research team, especially among researchers who made the original Lama model in 2023, Insider Informed On 26 May.
Only three of the original 14 Lama authors are built in Mata, and many have joined the Mistral, which is a Paris-based startup co-installed by former Meta researchers and leading Lama Architects.
Meta’s latest release, Lama 4Developers received a lukewarm reception, many of which now look at rapidly growing rivals, who have the GPT -4O of Openai, Gemini 2.5 Pro of Google, and recently launched argument models dedicated. Cloud 4 sonnet From anthropic, added to the report.
On May 15, Wall Street Journal Informed That meta was delayed by the rollout of its leading AI LLM, Lama 4 “Bhamath”.
magazine: Aye blindness, ‘Good’ propaganda bots, Openi Doomsde Bunker: fixes AI I
