Yann LeCun argues that current AI systems, including the latest models like Google’s Gemini 3 Flash, are far from achieving artificial general intelligence and that the industry is overestimating how close it really is. He says today’s models are powerful pattern matchers but still lack key capabilities such as persistent memory, true reasoning about the physical world, and an understanding of cause and effect. LeCun believes progress toward AGI will require entirely new architectures rather than simply scaling up existing large language models. He also pushes back against the idea that AGI is imminent, warning that hype can distort research priorities and public expectations. Instead, he sees steady advances in narrow and useful AI systems as the more realistic path forward in the near term.
