![[人人能懂AI前沿] AI在想什么？当它学会遗忘、看见趋势、认识自己](https://image.xyzcdn.net/FuDP4HpAp8ezgVZMmEel3mblKCmJ.jpg@middle)
主播
节目简介
来源：小宇宙
我们总希望AI更聪明，但如果我说，一篇最新论文发现，一个更“健忘”的AI反而对我们更有用呢？本期，我们将一起窥探AI的“内心世界”：看看一个短暂领悟了世界几何结构、又旋即忘却的AI，是如何思考的；一个不再满足于“平均”，而是学会了看“趋势”的AI，又是如何进化的。我们甚至还会聊聊，AI是不是已经有了区分“我”和“非我”的原始意识。准备好了吗？让我们一起出发，探索AI思考的奇妙新维度。
00:00:41 AI那么强，为什么我们偏要给它装上一个“健忘”的大脑？
00:06:30 AI的“七情六欲”，一个转瞬即逝的秘密
00:13:08 换个姿势，AI还能变得更聪明？
00:18:31 AI学会了“算”，但它学会“想”了吗？
00:24:25 AI的自我意识，它知道哪句话是自己说的吗？
本期介绍的几篇论文：
[CL] Simulating Human Memory with Language Models
[NYU]
https://arxiv.org/abs/2605.25680
---
[AI] Geometry of Human Perceptual Domains Emerges Transiently in LLM Representations
[Indian Institute of Technology Roorkee & Lossfunk]
https://arxiv.org/abs/2605.27970
---
[LG] Parallax: Parameterized Local Linear Attention for Language Modeling
[Northwestern University & Tilde Research & University of Washington]
https://arxiv.org/abs/2605.29157
---
[LG] DiscoverPhysics: Benchmarking LLMs for Out-of-the-Box Scientific Thinking
[Princeton University]
https://arxiv.org/abs/2605.26087
---
[LG] From Simulation to Enaction: Post-trained language models recognize and react to their own generations
[Institute for Advanced Study & Anthropic]
https://arxiv.org/abs/2605.25459
00:00:41 AI那么强，为什么我们偏要给它装上一个“健忘”的大脑？
00:06:30 AI的“七情六欲”，一个转瞬即逝的秘密
00:13:08 换个姿势，AI还能变得更聪明？
00:18:31 AI学会了“算”，但它学会“想”了吗？
00:24:25 AI的自我意识，它知道哪句话是自己说的吗？
本期介绍的几篇论文：
[CL] Simulating Human Memory with Language Models
[NYU]
https://arxiv.org/abs/2605.25680
---
[AI] Geometry of Human Perceptual Domains Emerges Transiently in LLM Representations
[Indian Institute of Technology Roorkee & Lossfunk]
https://arxiv.org/abs/2605.27970
---
[LG] Parallax: Parameterized Local Linear Attention for Language Modeling
[Northwestern University & Tilde Research & University of Washington]
https://arxiv.org/abs/2605.29157
---
[LG] DiscoverPhysics: Benchmarking LLMs for Out-of-the-Box Scientific Thinking
[Princeton University]
https://arxiv.org/abs/2605.26087
---
[LG] From Simulation to Enaction: Post-trained language models recognize and react to their own generations
[Institute for Advanced Study & Anthropic]
https://arxiv.org/abs/2605.25459
外观
