Meta's ex-AI chief slams industry focus; says every thing ones know about AI is wrong

3 months ago 70
ARTICLE AD BOX

 Everything you cognize  astir  and are moving   connected  AI is WRONG

Meta's erstwhile AI chief, Yann LeCun, declared Large Language Models (LLMs) a 'dead end' for achieving superintelligence. He advocates for 'world models' that larn from carnal reality, not conscionable language, to physique precocious instrumentality intelligence. LeCun's departure from Meta was reportedly owed to interior disagreements implicit the company's LLM-centric superintelligence push, contempt CEO Mark Zuckerberg's enactment for his research.

Meta’s erstwhile main AI scientist, Yann Lecun has a connection for everyone successful the exertion industry: The AI buzzword astir each institution believes in-- Large Language Model oregon LLM -- is simply a dead-end.

In an interrogation to Financial Times, Lecun said, "LLMs are utile but fundamentally constricted and constrained by language. To execute human-level intelligence, you person to recognize however our carnal satellite works too." He added that “LLMs fundamentally are a dormant extremity erstwhile it comes to superintelligence.”This is not the archetypal clip that LeCun has spoken against LLMs. The antheral besides known arsenic Godfather of AI has been rather vocal astir his disdain for ample connection models (LLMs) and their imaginable to scope superhuman intelligence.

LeCun's solution for achieving superhuman quality relies connected an architecture called V-JEPA, a alleged satellite model. "World models purpose to recognize the carnal satellite by learning from videos and spatial data, alternatively than conscionable language.

They are besides capable to plan, crushed and person persistent memory. He calls this benignant of quality Advanced Machine Intelligence, oregon AMI," helium said.Though LeCunn did not intelligibly accidental wherefore helium near Meta. He did accidental that staying became politically difficult. Meta's ex-employee said that Meta CEO Mark Zuckerberg likes his satellite exemplary research, "but the assemblage who were hired for the company’s caller superintelligence propulsion are “completely LLM-pilled”.

Here LeCunn meant the squad that joined Meta from ScaleAI. In June 2025, Mark Zuckerberg announced the instauration of Meta Superintelligence Labs. The caller AI superintelligence task is led by execs including Scale AI ex-CEO Alexandr Wang. Earlier successful June, the institution said it would prosecute Wang, present Meta’s main AI officer, and immoderate of his colleagues arsenic portion of a $14.3 cardinal concern into Scale AI.

Under the restructuring announced by Zuckerberg, LeCun was made to study to Alex Wang. This intelligibly alienated LeCun. “I’m definite there’s a batch of radical astatine Meta, including possibly Alex, who would similar maine to not archer the satellite that LLMs fundamentally are a dormant extremity erstwhile it comes to superintelligence,” helium says. “But I’m not gonna alteration my caput due to the fact that immoderate dude thinks I’m wrong. I’m not wrong. My integrity arsenic a idiosyncratic cannot let maine to bash this.

Read Entire Article
LEFT SIDEBAR AD

Hidden in mobile, Best for skyscrapers.