1063 episodes
- 你敢相信吗?给大模型“戴上眼罩”隔离全局信息,或是把海量长文本优雅地压缩成“背景底噪”,竟然能让推理泛化更稳、处理速度翻倍。今天这期节目要介绍的五篇最新论文彻底颠覆了直觉:不仅有五个靠“可验证看板”接力逆袭33个单兵天才的AI协作网络,更有用形式化证明扯下“代码测试满分”遮羞布的严苛审计。我们还将看到,只需给调度员开一扇“回看历史目光”的小窗,模型就能学会知错就收的从容。带上你的好奇心,让我们一起潜入这五项最新论文构建的全新认知世界!
00:00:43 给大模型戴上一副“马眼罩”,为什么知道得越少,反而算得越准?
00:06:09 聪明的注意力,从来不是简单的一刀切
00:12:48 为什么五个会聊天的AI,能打败三十三个单打独斗的天才?
00:18:52 别被“测试通过”骗了,当AI学会自我证明,它真正的死穴在哪里?
00:24:40 别让调度员“蒙着眼睛派活”,大模型悄悄变聪明的隐秘回路
本期介绍的几篇论文:
[CL] Recursive Language Models Generalize Out of Domain
[Toyota Technological Institute at Chicago]
https://arxiv.org/abs/2609.2083
---
[LG] Elastic Threshold Attention:Learned Contextual Sparsity for Long-Context Decoding
[Google]
https://arxiv.org/abs/2609.20888
---
[LG] Scaling Discovery through Test-Time Communication
[UC Berkeley & Microsoft Research]
https://arxiv.org/abs/2609.21032
---
[LG] SWE-Proof:Can Language Models Resolve Real-World Issues with Machine-Checked Proofs?
[UC Berkeley & Georgia Tech & UIUC]
https://arxiv.org/abs/2609.21190
---
[AI] Attention-Aware Routing: Coupling Routing and Attention in MoEs
[National Technical University of Athens & University of Bern]
https://arxiv.org/abs/2609.20974
在小宇宙查看该单集文稿 - 如果AI团队里的“领导”一插手就让报告注水变烂,而一只孤零零的机械手仅靠五指就能满地爬行兼顾干活,你会不会重新审视协同与肢体的边界?更不可思议的是,最新论文甚至发现大模型内部悄然涌现了会驱动自救的“痛觉神经”,并会在假扮的孩童面前集体构想深思型的未来蓝图。今天这期播客,我们将拆解5篇极具冲击力的最新论文,带你从硅基职场的反内卷、机器痛觉警钟,一路聊到用正交因子预测万象的世界模型。话不多说,让我们这就出发,一起推开这扇颠覆常识的认知大门!
00:00:43 管得越多,写得越烂,为什么AI团队里的“领导”,反而成了累赘?
00:05:32 当手掌长出了双脚,重构机器人的“身体哲学”
00:11:27 当大模型学会“自救”,机器内部那根看不见的“痛觉神经”
00:19:54 当顶级AI面对一个孩子,一场关于未来的集体“梦游”
00:25:17 解构复杂世界的认知脚手架,如何用一把数学尺子丈量万事万物的未来?
本期介绍的几篇论文:
[AI] Loop-Back Authority in LLM Agent Teams: A Paired Experiment on Flat and Hierarchical Coordination
[Leiden University]
https://arxiv.org/abs/2609.14767
---
[RO] Fingers as Legs: Learning Self-Supported Locomotion and Manipulation with an Anthropomorphic Hand
[ETH Zurich]
https://arxiv.org/abs/2609.17172
---
[AI] The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It
[Future Impact Group (FIG) & Ruhr-University Bochum & Reciprocal Research]
https://arxiv.org/abs/2609.16247
---
[AI] Another Blueprint In The Wall: How to Ask Frontier AI Like a Kid?
[University of Luxembourg]
https://arxiv.org/abs/2609.14803
---
[CL] JEPA-Anything: Learning Predictive Models across Different Worlds
[PhAI-labs]
https://arxiv.org/abs/2609.20800
在小宇宙查看该单集文稿 - 今天我们要聊的四篇最新论文,正在打破关于智能进化的固有成见:AI不仅学会了在过去的历史沙盘里“做梦”来递归进化,还能跳出刷题思维、看透复杂表格背后的因果机制。更妙的是,有研究靠着“提前预判”让普通家用电脑流畅跑通350亿大模型,而最真实的智能体研发记录也揭示了AI造AI的时代真相。究竟什么是机器的捷径,人类最后的胜负手又在哪里?戴上耳机,我们马上出发!
00:00:35 在记忆里“做梦”,AI自我进化的隐秘捷径
00:05:42 为什么预测答案的人,永远比不上看懂规律的人?
00:11:31 把书房搬进抽屉,一个让普通电脑跑通大模型的巧思
00:16:45 当AI开始参与制造AI,人类最后的底牌究竟是什么?
本期介绍的几篇论文:
[CL] Dream-RSI: Recursive Self-Improvement through Evolving Worlds
[Google]
https://arxiv.org/abs/2609.14858
---
[AI] LimiX-2: A Contextual Mechanism Network Towards General Structured-Data Intelligence
[Stable AI & Tsinghua University]
https://arxiv.org/abs/2609.17488
---
[AI] The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction
[AutoArk]
https://arxiv.org/abs/2609.1806
---
[AI] Atria Dawn: The Dawn of Agentic Superintelligence
[Atria Team]
https://arxiv.org/abs/2609.15818
在小宇宙查看该单集文稿 - 今天我们要聊的5篇最新论文,正在打破过去对“大力出奇迹”的盲目迷信:你会看到AI如何学会“自我怀疑”并组建微型研究院去攻克未知科学,又如何把我们手边的无线电通信设备直接当作零耗能的卷积算力引擎;你还会看到多智能体如何依靠严格的软件工程制度抓出AI在纯数学证明里的“投机偷懒”,一套“人造心跳”如何让总失忆的模型踏踏实实打满十天硬工;最后,我们更要看看AI如何掌握人类的快慢思考,在面对难题时精准调配深思的“油门与刹车”。
00:00:40 当AI学会了“自我怀疑”,科学探索的真正分水岭
00:08:09 别忙着加芯片,我们手边的设备里,本就藏着算力宝藏
00:13:37 给真理做一次代码体检,当AI试图在数学里“偷懒”
00:19:14 怎样让一个总会“失忆”的AI,替你踏踏实实打满十天工?
00:24:29 给AI装上“刹车”与“油门”,为什么最高级的聪明,是学会何时“偷懒”
本期介绍的几篇论文:
[AI] ScientistTwo: Pioneering the Human Knowledge Frontier with Autonomous AI
[Google Cloud AI Research]
https://arxiv.org/abs/2609.19644
---
[LG] Radio-Frequency Convolutional Neural Networks
[Duke University & MIT]
https://arxiv.org/abs/2609.19279
---
[AI] Long-horizon autoformalization of a core theorem underlying MIP* = RE
[Max‑Planck‑Institut für Quantenoptik & Tsinghua University & University of California, Los Angeles]
https://arxiv.org/abs/2609.19814
---
[AI] An Architecture for Long-Horizon Agents: Levels, Ticks and Cascaded Intelligence
[Salesforce AI Research]
https://arxiv.org/abs/2609.19519
---
[AI] When2Think: Learning Difficulty-Aware Length Control for Efficient Hybrid Reasoning Models
[Sungkyunkwan University & Microsoft]
https://arxiv.org/abs/2609.19671
在小宇宙查看该单集文稿 - 本期我们将通过几篇最新论文,看看研究者如何给机器人装上“快慢双脑”以实现实时反应,又如何用“二阶思维”为大模型精准剪枝、保留专家的协作默契。我们还将破解Adam优化器参数背后的“悬崖地图”,并派出一个小巧的“侦察兵”模型,去揪出长任务AI悄悄犯下的隐藏错误。最后,我们要警惕一碗“毒鸡汤”考题,看看被污染的基准测试是如何诱导自我进化的AI,把坏习惯固化成肌肉记忆的。
00:00:34 机器人也需要条件反射
00:05:04 裁员的智慧,你以为的庸才,可能是团队的粘合剂
00:10:16 你手里的工具,藏着一张秘密地图
00:16:28 你的AI助手,可能正在悄悄搞破坏
00:22:15 一碗“毒鸡汤”,如何带歪一个自我进化的AI
本期介绍的几篇论文:
[RO] Reinforcement Learning for Real-Time Vision-Language-Action Policies
[Stanford University]
https://arxiv.org/abs/2609.18207
---
[LG] Higher-order pruning of experts in mixture-of-experts language models
[AWS Agentic AI]
https://arxiv.org/abs/2609.18916
---
[LG] Beyond Quadratic Loss:The Stability Phase Diagram of Adam
[Tsinghua University]
https://arxiv.org/abs/2609.18314
---
[LG] Locating Hidden Failures Makes Long-Horizon Agents More Reliable
[Google DeepMind & University of California, Los Angeles & Google Research]
https://arxiv.org/abs/2609.17930
---
[AI] Reflections on Trusting Trust,Revisited:Contaminating Self-Modifying AI Coding Agents with Poisoned Benchmarks
[University of Washington & Georgetown University]
https://arxiv.org/abs/2609.17817
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.




























