计算机科学家崔艺珍（Yejin Choi）为我们揭秘了大型人工智能系统，如 ChatGPT 的现状，指出了最新的大语言模型存在的三个关键问题（包括一些有趣的例子，它们在基本的常识逻辑上犯了错误）。她引领我们走进新时代，AI 几乎会成为拥有智力的新物种，展示了建立一个由人类规范和价值观训练的更小的 AI 系统有何好处。【演讲后为与 TED 负责人克里斯·安德森（Chris Anderson）的问答环节】

Artificial intelligence systems can ace the bar exam and defeat top Go champions, yet often stumble on simple reasoning tasks that even children solve with ease. Massive language models are trained on trillions of words using enormous computing power, but still lack fundamental common sense. This gap arises because scaling up brute force does not teach AI the intuitive, unspoken knowledge humans acquire naturally. Rethinking AI development means building smaller, more transparent systems grounded in human norms and values.

AI Masters Exams Yet Fumbles at Basic Human Common Sense

为什么 AI 聪明绝顶又愚蠢不堪

machine learning

technology

science

teaching

future

innovation

ethics

humanity

humor

TED Stage Talk

崔艺珍（Yejin Choi）: 为什么 AI 聪明绝顶又愚蠢不堪