Yann LeCun: LLMs Are Nearing the End, but Better AI Is Coming (2025) (www.newsweek.com)

In a recent interview, AI pioneer Yann LeCun declared that large language models (LLMs) are reaching their limitations and will likely become obsolete in the coming years. He argues that LLMs lack the ability to represent the complex, high-dimensional spaces necessary for meaningful reasoning and understanding. While they've achieved impressive results in generating coherent language by memorizing vast amounts of text, they mimic shallow learning—akin to rote memorization—rather than engaging in deep reasoning or conceptual understanding. LeCun emphasizes that current AI systems function primarily on a "System 1" level, reacting to immediate inputs without the structured planning and abstract reasoning characteristic of human cognition. LeCun posits that true advancements in AI will require the integration of "System 2" capabilities, which involve creating rich, abstract models of the world that facilitate reasoning over various timescales and dimensions. He highlights the necessity of sensory and motor-driven learning, which allows AI to interact with the physical world, as a critical step towards developing more sophisticated intelligence. This approach would enable AI to define well-structured goals and implement sturdy guidelines for exploration, fostering the development of genuinely intelligent systems that can adapt and learn rapidly. The insights from LeCun's critique could significantly influence future AI research, steering the community towards a paradigm that prioritizes understanding over mere data recall.