1045 episodes
- 本期,我们来聊聊AI如何从一个“普通学生”被系统地培养成编程竞赛的世界冠军,甚至超越了人类状元。但与此同时,为什么我们身边的AI助理,处理复杂任务时却常常“走着走着就散架”了?我们又该如何教会AI管理自己的“注意力”,像人一样划重点?以及,如何通过精准定位它“第一次犯错的瞬间”,让它的学习效率实现飞跃?四篇最新论文,带我们深入AI的“学霸心法”,揭示智能背后的策略、局限与成长之道。
00:00:37 AI学会考试了,而且比状元考得还好
00:06:06 你的AI助理,为啥走着走着就“散架”了?
00:11:30 AI的注意力,该由谁做主?
00:16:47 如何让机器学会聪明,抓住第一次犯错的瞬间
00:22:08 知识的“断舍离”,我们究竟该记住什么?
本期介绍的几篇论文:
[LG] Post-Training Language Models for Gold-Medal Performance in Coding Competitions
[NVIDIA]
https://arxiv.org/abs/2609.02849
---
[AI] How Fast Do Agents Rot? An Empirical Study of Long-Horizon Degradation in LLM Agents for Production Decision-Making
[Microsoft AI]
https://arxiv.org/abs/2609.01660
---
[CL] Language Models Can Control Their Own Attention
[KAIST AI & Google DeepMind]
https://arxiv.org/abs/2609.02737
---
[LG] Cliff: Learning Process Rewards from the First Mistake
[Amazon Web Services]
https://arxiv.org/abs/2609.02817
---
[LG] What Is Worth Representing? Representational Empowerment for Continual Model Construction
[UC Berkeley & University of Tübingen]
https://arxiv.org/abs/2609.02322
在小宇宙查看该单集文稿 - 本期我们来聊聊AI世界正在悄然发生的一场“效率革命”。如何只花十分之一的成本,就猜对AI巨头的“天机”?又如何让AI靠“复读”关键知识,聪明地战胜一味地“堆料”?我们还会探讨一个反常识的现象:为什么一个好的AI老师,关键时刻要学会“闭嘴”?AI的能力飞跃,究竟是学会了新招,还是把旧招用得更溜了?四篇最新的AI论文,带你洞悉AI世界的效率革命与学习智慧。
00:00:33 如何用十分之一的成本,猜对AI巨头的“天机”?
00:06:17 如何让“笨学生”学得更快?关键在于让“老师”适时闭嘴
00:11:26 AI变聪明,是学会了新招,还是旧招用得更溜了?
00:16:47 AI训练的内卷,如何用“复读”战胜“堆料”?
00:22:44 当AI被骗,它的大脑里发生了什么?
本期介绍的几篇论文:
[LG] Efficiently Estimating Optimal Hyperparameter Scaling Laws through Power-Law Entropy Search
[Meta]
https://arxiv.org/abs/2609.01431
---
[CL] Knowledge Distillation During Mid-Training Favors Reasoning over Factual Recall
[Meta AI & Princeton University]
https://arxiv.org/abs/2609.01532
---
[CL] From Base Rollouts to RL Reasoning: A Budgeted Search Perspective
[Fudan University & Zhipu AI & Tsinghua University]
https://arxiv.org/abs/2609.01274
---
[LG] SMELT: Scaling Laws for Compute-Matched MoE Looped Transformers
[Tsinghua University & ByteDance Seed & M-A-P]
https://arxiv.org/abs/2609.01343
---
[LG] How Do Language Models Choose Between Context and Memory?
[Stanford University]
https://arxiv.org/abs/2609.00753
在小宇宙查看该单集文稿 - 想知道AI混沌的“数字粥”里,是不是藏着一张我们能读懂的清晰地图吗?想见识一下比人类专家还厉害的“AI教练”,是如何给它的同类“治病”的吗?我们还会探讨,当所有人都想抄“流量密码”的作业时,内容世界为何会变得越来越无聊,以及最后,我们将揭秘一场AI的“省油”革命,看看聪明的设计如何让AI告别傻大黑粗。
00:00:28 AI的“黑箱”里,藏着一套我们熟悉的旧地图
00:06:09 比人类专家还强?AI正在学会自己给自己“治病”
00:11:34 当所有人都想抄第一名的作业
00:17:10 AI的“省油”革命,如何用更少的资源,办更大的事?
00:22:44 AI创作的秘密,不是靠魔法,而是靠一张地图
本期介绍的几篇论文:
[CL] The Emergent Symbolic Structure of Artificial Neural Networks
[Yale University & Johns Hopkins University & New York University]
https://arxiv.org/abs/2608.29530
---
[AI] Automated Researchers Can Reliably Mitigate Alignment Failures
[Anthropic & UC Berkeley]
https://arxiv.org/abs/2608.28945
---
[AI] CHASE: How Content Ecosystems Are Reshaped When Ranking Is the Only Target
[UC Berkeley & Zhejiang University]
https://arxiv.org/abs/2608.30466
---
[CL] On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability
[Qwen Team]
https://arxiv.org/abs/2608.30320
---
[LG] The information geometry of product-reference discrete diffusion: Interaction growth complexity and optimal scheduling
[MIT]
https://arxiv.org/abs/2608.28949
在小宇宙查看该单集文稿 - 我们总感觉AI越来越无所不能,但今天,我们要从几篇最新论文出发,给这份狂热踩一脚“科学的刹车”。我们会探讨AI为何读完人类所有书籍,却依然有无法跨越的语言天堑,并揭示其看似复杂的内部机制,其实隐藏着一个更简单的“有效维度”。同时,我们也会发现,解决最棘手问题的,有时反而是被我们忽略的“笨办法”,而看似混沌的AI训练过程,竟然也遵循着可以预测的“伸缩法则”。准备好了吗?让我们一起拨开AI的迷雾,看见那些真正重要的底层规律。
00:00:39 AI读完了整个人类图书馆,为什么还是不懂你?
00:05:39 最聪明的办法,常常是那个“笨办法”
00:09:38 AI界的“孙子兵法”,如何用有限的资源打赢无限的战争
00:16:05 AI大模型里的“降维打击”,你看见的复杂,不是真的复杂
00:21:16 为什么好的目标,也会带你走上岔路?
本期介绍的几篇论文:
[CL] A Formal Limitation on Learning Human Language From Textual Corpora
[Universitat Pompeu Fabra & ETH Zürich]
https://arxiv.org/abs/2608.28560
---
[CL] Sliding-window beats linear attention
[Microsoft]
https://arxiv.org/abs/2608.28444
---
[CV] How Far Can 5,500 Hours of Driving Take You? A Scaling Law Analysis of Video Diffusion Models
[valeo.ai]
https://arxiv.org/abs/2608.28404
---
[LG] The Approximation Rank of Softmax Attention: Sharp Geometric Laws and Robust Interaction Dimension
[Nanyang Technological University & CMU]
https://arxiv.org/abs/2608.28150
---
[LG] How Proper Scoring Rules Shape LLM Forecasting
[Lightning Rod Labs & INSEAD & University of Pennsylvania]
https://arxiv.org/abs/2608.28482
在小宇宙查看该单集文稿 - 当AI学会了“活在当下”,不再被历史包袱拖累时,我们人类自己又该如何避免被它悄悄“废掉”核心能力呢?本期节目,我们不仅要探讨如何用一把特制的尺子去衡量AI是否真的懂我们的“不开心”,还将揭秘如何培养出一个靠谱的AI“批评家”,让它实现高效的自我进化。最后,我们会一起探寻训练AI时那个神秘的“档位”,看看这些最新论文将如何刷新我们对人机协作与AI成长的认知。
00:00:33 让AI学会“活在当下”
00:05:00 AI越来越聪明,但它真的懂你的“不开心”吗?
00:09:24 那个替你干活的AI,正在悄悄“废掉”你
00:13:48 AI的成长烦恼,一个“批评家”的自我修养
00:20:39 训练AI的秘密“档位”
本期介绍的几篇论文:
[AI] SKILL.state: Scalable Long-Horizon Agent Skills
[Google LLC & Purdue University]
https://arxiv.org/abs/2608.26263
---
[CL] HealthBench-Psych: A Mental Health Subset of OpenAI's HealthBench
[Beth Israel Deaconess Medical Center]
https://arxiv.org/abs/2608.25071
---
[AI] AI Agents Push Humans Out of the Loop
[Hugging Face]
https://arxiv.org/abs/2608.23642
---
[LG] Best Practice Critic Optimization
[National University of Singapore & Tencent Hunyuan]
https://arxiv.org/abs/2608.23566
---
[LG] Effective Learning Rate Governs Loss Dynamics in Language Model Pretraining
[Peking University & Ant Group]
https://arxiv.org/abs/2608.24814
在小宇宙查看该单集文稿
More Technology podcasts
Trending Technology podcasts
About AI可可AI生活
来自 @爱可可-爱生活 的第一手AI快报,用最简单易懂的语言,带你直击最前沿的人工智能科研动态。无论你是科技小白,还是行业达人,这里都有你想知道的AI故事和未来趋势。跟着我们,轻松解锁人工智能的无限可能!
#人工智能 #科技前沿
Podcast websiteListen to AI可可AI生活, Lex Fridman Podcast and many other podcasts from around the world with the radio.net app

Get the free radio.net app
- Stations and podcasts to bookmark
- Stream via Wi-Fi or Bluetooth
- Supports Carplay & Android Auto
- Many other app features
Get the free radio.net app
- Stations and podcasts to bookmark
- Stream via Wi-Fi or Bluetooth
- Supports Carplay & Android Auto
- Many other app features


AI可可AI生活
Scan code,
download the app,
start listening.
download the app,
start listening.






























