Why an AI 'godfather' is quitting Meta after 12 years (www.bbc.com)

Yann LeCun, one of deep learning’s most prominent figures and Meta’s chief AI scientist for seven years (and founding director of FAIR for five), is leaving Meta to start a new company that will pursue his vision of “advanced machine intelligence.” Announced amid weeks of rumors and shortly after he received a royal award, LeCun praised FAIR’s impact and said his new firm will focus on visual, child‑like learning—training models to learn from perception and interaction—rather than doubling down on large language models (LLMs), the data‑hungry generative systems that have dominated recent AI investment since ChatGPT’s 2022 debut. He said Meta will be a partner in the venture, even as he has increasingly clashed with prevailing industry priorities. The move matters because it crystallizes a major technical and strategic split in AI: whether progress comes from scaling LLMs trained on massive text corpora or from building models that learn embodied, perceptual world models more like infants do. LeCun’s stance challenges the LLM-centric playbook and highlights limits of generative approaches for human‑level intelligence, while also downplaying existential risk—putting him at odds with peers like Hinton and Bengio. Practically, his departure could shift research agendas, funding and talent toward alternative self‑supervised, multimodal and interaction‑based methods, and intensifies debate over the sustainability of the current AI boom and its market corrections.