你有没有想过,当AI不再是只会模仿的“鹦鹉”,它会如何为自己打造一张世界的“活地图”,甚至为万物创造出能自主思考的“数字分身”?最新论文揭示,AI正通过一系列奇妙的方法解决自己的“健忘症”与“数据饥荒”,甚至开始反思“堆料越多越糊涂”的怪圈。今天,我们就来聊聊AI是如何学会拥有“活地图”、创造“数字分身”、进行“模拟推理”,并最终实现自我“瘦身”的。 00:00:34 你的脑子里,是不是也有一张“活地图”? 00:05:55 你我皆有“数字分身”,当AI为万物造“镜像” 00:12:23 你的常识可能被颠覆了,模仿来的思考,算不算思考? 00:17:46 预测的难题,当AI遇上“数据饥荒” 00:24:46 AI大模型内卷,为什么堆料越多,脑子越糊涂? 本期介绍的几篇论文: [LG] Flow Equivariant World Models: Memory for Partially Observed Dynamic Environments [Harvard University & CMU] https://arxiv.org/abs/2601.01075 --- [AI] Digital Twin AI: Opportunities and Challenges from Large Language Models to World Models [Lehigh University & University of Maryland & University of New South Wales] https://arxiv.org/abs/2601.01321 --- [CL] Simulated Reasoning is Reasoning [RWTH Aachen University & CMU] https://arxiv.org/abs/2601.02043 --- [LG] Zero-shot Forecasting by Simulation Alone [Amazon] https://arxiv.org/abs/2601.00970 --- [LG] Geometric and Dynamic Scaling in Deep Transformers [New York University & Stony Brook University] https://arxiv.org/abs/2601.01014
本期节目,我们将深入AI的“引擎盖”之下,看看那些看不见的结构如何决定一切。你会听到,为何区区2%的数据就能决定翻译能力的生死;AI如何像侦探一样,为复杂问题画出“破案地图”;以及在看似无害的模型拼接中,如何暗藏着难以察觉的“木马”后门。准备好了吗?让我们一起探索这些最新论文背后,令人拍案叫绝的智慧。 00:00:31 AI翻译的秘密,2%的数据,50%的能力 00:05:35 你以为的“搜索”,正在被重新发明 00:13:00 为什么你的“笨办法”,却是AI的“开窍”法? 00:18:16 AI大厨做菜重复?换一种“盐”试试 00:23:43 AI世界的“乐高”游戏,藏着一个你没想到的“后门” 本期介绍的几篇论文: [CL] The Role of Mixed-Language Documents for Multilingual Large Language Model Pretraining [University College London & Nanyang Technological University & University of Waterloo] https://arxiv.org/abs/2601.00364 --- [CL] Retrieval--Reasoning Processes for Multi-hop Question Answering: A Four-Axis Design Framework and Empirical Trends [University of Pittsburgh & Google Cloud AI Research] https://arxiv.org/abs/2601.00536 --- [LG] Deep Networks Learn Deep Hierarchical Models [Hebrew University of Jerusalem] https://arxiv.org/abs/2601.00455 --- [CV] It's Never Too Late: Noise Optimization for Collapse Recovery in Trained Diffusion Models [UC Berkeley & University of Tübingen] https://arxiv.org/abs/2601.00090 --- [LG] The Trojan in the Vocabulary: Stealthy Sabotage of LLM Composition [Purdue University & CMU] https://arxiv.org/abs/2601.00065
你有没有想过,我们能不能让AI像探险家一样,在脑中绘制一张动态的世界地图?或者,仅仅是换个“看”图的顺序,就能让AI的识别能力大幅提升?本期节目,我们将一起探索几篇有趣的最新论文:看看为什么用“假奖励”瞎指挥,反而能激发AI的潜能;AI又是如何自动发现数据背后的“主线任务”;以及最关键的,我们如何教会AI那句宝贵的“我不确定”,让它变得更值得信赖。 00:00:35 你的大脑,如何给世界画地图? 00:05:33 AI识图的秘密,你以为不重要的,恰恰是关键 00:12:01 为什么瞎指挥也能练出好学生? 00:17:57 有一种AI,能自动发现数据的“主线任务” 00:23:29 给AI装上一个“靠谱”探测器 本期介绍的几篇论文: [LG] MapFormer: Self-Supervised Learning of Cognitive Maps with Input-Dependent Positional Embeddings [Institut Jean Nicod & École Normale Supérieur] https://arxiv.org/abs/2511.19279 --- [LG] REOrdering Patches Improves Vision Models [University of Pittsburgh & UC Berkeley] https://arxiv.org/abs/2505.23751 --- [LG] Spurious Rewards: Rethinking Training Signals in RLVR [University of Washington] https://arxiv.org/abs/2506.10947 --- [LG] Distributional Autoencoders Know the Score [University of Michigan] https://arxiv.org/abs/2502.11583 --- [LG] Similarity-Distance-Magnitude Activations [Reexpress AI] https://arxiv.org/abs/2509.12760
你有没有想过,真正的智能不只在于堆砌知识,更在于懂得“断舍离”,甚至学会如何“聪明地努力”?这一期,我们将看到最新论文如何教会AI进行动态的自我修正,以及机器人如何通过构建“知识金字塔”学会心灵手巧。我们还会见证,古老的“三角形”如何在AI新魔法的加持下重返巅峰,以及我们如何通过更换一把“度量尺”,让小模型的训练经验直接指导大模型。准备好,一场关于AI学习智慧的认知升级,马上开始! 00:00:38 AI学会了“断舍离”,才能变得更聪明 00:05:48 机器人学会“心灵手巧”的秘密,不止是苦练 00:10:47 三角形,凭什么重返巅峰? 00:16:26 AI进化的新姿势,从“大力出奇迹”到“聪明地努力” 00:22:33 训练AI,我们是不是一直在“蒙眼下山”? 本期介绍的几篇论文: [LG] Deep Delta Learning [Princeton University & University of California, Los Angeles] https://github.com/yifanzhang-pro/deep-delta-learning/blob/master/Deep_Delta_Learning.pdf --- [RO] GR-Dexter Technical Report [ByteDance] https://arxiv.org/abs/2512.24210 --- [CV] Triangle Splatting for Real-Time Radiance Field Rendering [University of Liège] https://arxiv.org/abs/2505.19175 --- [CL] ShinkaEvolve: Towards Open-Ended And Sample-Efficient Program Evolution [Sakana AI] https://arxiv.org/abs/2509.19349 --- [LG] Training Deep Learning Models with Norm-Constrained LMOs [EPFL] https://arxiv.org/abs/2502.07529
你有没有想过,我们该如何为AI的高速公路设计智能的交通规则,又该如何教会一颗活的“迷你大脑”摸盲文?如果让AI来炒股,市场会更疯狂还是更理性?本期,我们将从几篇最新的论文出发,揭开AI从一个工具箱进化为发明家,并学会复杂推理的底层设计图。 00:00:26 给AI修路,为什么“车道”越多反而越容易“堵车”? 00:05:32 如果让AI来炒股,它会比你更贪婪吗? 00:11:24 当你的电脑开始用“脑子”摸盲文 00:16:21 你的AI助手,应该是个工具箱,还是个发明家? 00:21:28 AI变聪明的“导航系统”,一份来自底层的设计图 本期介绍的几篇论文: [CL] mHC: Manifold-Constrained Hyper-Connections [DeepSeek-AI] https://arxiv.org/abs/2512.24880 --- [AI] Can Generative AI agents behave like humans? Evidence from laboratory market experiments [College London & CENTAI Institute & Bank of Canada] https://arxiv.org/abs/2505.07457 --- [RO] Encoding Tactile Stimuli for Organoid Intelligence in Braille Recognition [University of Bristo] https://arxiv.org/abs/2508.20850 --- [AI] Alita: Generalist Agent Enabling Scalable Agentic Reasoning with Minimal Predefinition and Maximal Self-Evolution [Princeton University & Tsinghua University & Shanghai Jiao Tong University] https://arxiv.org/abs/2505.20286 --- [LG] On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning [University of California, Los Angeles] https://arxiv.org/abs/2505.17508
你有没有想过,一个真正聪明的AI应该是什么样的?本期节目,我们将深入AI的“思考过程”,为你揭秘五篇最新论文带来的启发:我们将看到,AI如何像一个严谨的工程师,为自己的“世界模型”做“精装修”;如何像一个懂得“下笨功夫”的学生,通过生成式学习来避免投机取巧;更会看到它如何像一位创作者,从“逐字蹦”进化到“并行创作”;像一个项目经理,将搞不定的超长任务“外包”出去;最后,我们还会发现,AI甚至学会了像我们一样“自我反省”和纠错。准备好了吗?让我们一起探寻AI变得更聪明的秘密。 00:00:44 造个“世界模型”给自己用,聪明人是怎么“精装修”的? 00:07:34 AI的“笨功夫”与“真聪明” 00:13:11 AI“写稿”新姿势,从“逐字蹦”到“一挥而就” 00:20:56 AI的大脑也会“内存不足”?那就给它一个外挂 00:26:55 让AI学会自我反省,需要几步? 本期介绍的几篇论文: [LG] What Drives Success in Physical Planning with Joint-Embedding Predictive World Models? [Meta FAIR & New York University] https://arxiv.org/abs/2512.24497 --- [LG] Generative Classifiers Avoid Shortcut Solutions [CMU & Stanford University] https://arxiv.org/abs/2512.25034 --- [LG] Diffusion Language Models are Provably Optimal Parallel Samplers [UC Berkeley] https://arxiv.org/abs/2512.25014 --- [LG] Recursive Language Models [MIT CSAIL] https://arxiv.org/abs/2512.24601 --- [LG] Enhancing LLM Planning Capabilities through Intrinsic Self-Critique [Google DeepMind] https://arxiv.org/abs/2512.24103
你是否好奇,AI究竟是在独立思考还是在“背题库”?本期我们将搭建一个“思想风洞”来一探究竟,再把K线图当成“照片”喂给AI,看看它能否成为股神。随后,我们会深入神经网络的“职场”,看它如何通过“内卷”来自我裁员,并最终揭晓一本能“以小博大”、高效炼成大模型的万能说明书。准备好了吗?让我们一起开启这场关于AI智慧与效率的探索之旅! 00:00:32 AI到底在思考,还是在“假装”思考? 00:07:46 把K线图当成“照片”看,AI炒股能靠谱吗? 00:13:07 AI也“内卷”,一个想法,让神经网络自己“裁员” 00:18:14 炼大模型,有没有一本万能说明书? 本期介绍的几篇论文: [LG] The Bayesian Geometry of Transformer Attention [Dream Sports & Columbia University] https://arxiv.org/abs/2512.22471 --- [LG] S&P 500 Stock's Movement Prediction using CNN [Stanford University] https://arxiv.org/abs/2512.21804 --- [LG] Pruning as a Game: Equilibrium-Driven Sparsification of Neural Networks [Hamad Bin Khalifa University] https://arxiv.org/abs/2512.22106 --- [LG] Completed Hyperparameter Transfer across Modules, Width, Depth, Batch and Duration [Apple] https://arxiv.org/abs/2512.22382
你有没有想过,一个真正聪明的AI,它的学习方式和我们有什么不同?今天我们来聊聊几篇有趣的最新论文:有的AI像我们一样“边读边学”来消化一本厚书;有的机器人因为“见识”够广,突然就看懂了人类的视频;还有的AI,通过学习错误的答案,竟然比学习标准答案进步还快!更神奇的是,AI甚至开始通过自我辅导,学习如何成为一名科学家,并领悟到“终身学习”才是智能的终极宿命。这背后到底藏着哪些颠覆我们常识的智慧?让我们一探究竟。 00:00:39 AI的长文考卷,有没有更聪明的解法? 00:06:27 机器人笨手笨脚?可能只是因为它“见识”太少 00:12:10 “抄作业”的正确姿势,为什么错误的答案里藏着宝藏? 00:17:54 AI科学家,怎么才能不“纸上谈兵”? 00:24:57 为什么最聪明的AI,也必须终身学习? 本期介绍的几篇论文: [LG] End-to-End Test-Time Training for Long Context [Astera Institute & UC San Diego] https://arxiv.org/abs/2512.23675 --- [RO] Emergence of Human to Robot Transfer in Vision-Language-Action Models [Physical Intelligence] https://arxiv.org/abs/2512.22414 --- [LG] Shape of Thought: When Distribution Matters More than Correctness in Reasoning Tasks [University of Waterloo & MILA & Google DeepMind] https://arxiv.org/abs/2512.22255 --- [LG] Training AI Co-Scientists Using Rubric Rewards [Meta Superintelligence Labs & University of Oxford] https://arxiv.org/abs/2512.23707 --- [AI] The World Is Bigger! A Computationally-Embedded Perspective on the Big World Hypothesis [University of Alberta & The Swiss AI Lab IDSIA] https://arxiv.org/abs/2512.23419
你有没有想过,聪明的AI也像一个需要不断成长的学生?本期我们要聊的几篇最新论文,就揭示了AI正在经历一场深刻的“思维修炼”。我们会看到,AI不仅在学习如何诊断自己为什么会“胡说八道”,还在学习如何像项目经理一样规划工作,甚至学会了反思和定义一个“更好的问题”。这不仅是让AI变得更强大,更是让它变得更“智慧”的关键一步。 00:00:31 AI为什么会“一本正经地胡说八道”?一份统一的诊断书 00:06:05 AI“填词游戏”里的速度与智慧 00:11:58 你的AI团队里,谁才是真正的关键先生? 00:17:33 如何提出一个“好问题”?这回轮到AI教我们了 00:23:24 从“考高分”到“造地图”,AI决策的一次思维升级 本期介绍的几篇论文: [CL] A Unified Definition of Hallucination, Or: It's the World Model, Stupid [CMU & Patronus AI & Stanford University] https://arxiv.org/abs/2512.21577 --- [LG] dUltra: Ultra-Fast Diffusion Language Models via Reinforcement Learning [University of Washington & UC Berkeley] https://arxiv.org/abs/2512.21446 --- [LG] An Information Theoretic Perspective on Agentic System Design [Stanford University] https://arxiv.org/abs/2512.21720 --- [AI] Accelerating Scientific Discovery with Autonomous Goal-evolving Agents [Cornell University & The Ohio State University & Yale University] https://arxiv.org/abs/2512.21782 --- [LG] Generative Actor Critic [Tsinghua University & UCLA & Beijing Institute of General Artificial Intelligence] https://arxiv.org/abs/2512.21527
你有没有想过,当AI学霸们“开会”时,它们会达成怎样的共识?为什么最安全的AI,有时反而是那个最“老实”的笨小孩?我们又该如何治好AI的“失忆症”,让它拥有真正的人格?本期节目,我们将一起深入AI的“内心世界”,从最新论文中探寻这些问题的答案,看看一个更聪明、更懂事、也更具“生命感”的AI是如何被构想和塑造的。 00:00:31 顶级AI模型,正在悄悄达成一个共识 00:06:15 AI的安全漏洞,不是太笨,而是太“老实” 00:11:04 你的AI助理,如何拥有“人格”? 00:17:14 机器人怎么才能不“迷路”?从分工到整合,聊聊导航的新思路 00:23:00 给AI编剧一把尺子 [LG] Universally Converging Representations of Matter Across Scientific Foundation Models [MIT] https://arxiv.org/abs/2512.03750 --- [LG] Beyond Context: Large Language Models Failure to Grasp Users Intent [KTH Royal Institute of Technology] https://arxiv.org/abs/2512.21110 --- [AI] Sophia: A Persistent Agent Framework of Artificial Life [Westlake University & Project Cuddlepark Team & Shanghai Innovation Institute] https://arxiv.org/abs/2512.18202 --- [RO] LoGoPlanner: Localization Grounded Navigation Policy with Metric-aware Visual Geometry [Tsinghua University & Shanghai AI laboratory] https://arxiv.org/abs/2512.19629 --- [CL] DramaBench: A Six-Dimensional Evaluation Framework for Drama Script Continuation [University of Macau & University College London] https://arxiv.org/abs/2512.19012
当AI看似无所不知时,它真的理解自己说的因果关系吗?我们如何训练AI学会“解释自己”而不是“强词夺理”?本期节目,我们将从几篇最新论文出发,揭示AI如何学会编织知识地图、为何会“一本正经地胡说八道”,并一窥它那“不断复读”的内在工作模式,以及为电影特效“脑补”真实细节的惊人能力。 00:00:29 AI时代的“寻龙诀”,我们如何挖掘知识的因果龙脉 00:07:37 AI正在给你一种“知道”的幻觉 00:14:50 AI看病,如何才能“说人话”还“负责任”? 00:19:48 AI的“偷懒”智慧,为什么顶尖模型都在悄悄“复读”? 00:25:42 AI正在“脑补”你看不到的真实 本期介绍的几篇论文: [LG] Large Causal Models from Large Language Models [Adobe Research] https://arxiv.org/abs/2512.07796 --- [AI] Epistemological Fault Lines Between Human and Artificial Intelligence [Sapienza University of Rome & University of Milan Bicocca & University of Maribor] https://arxiv.org/abs/2512.19466 --- [LG] Reason2Decide: Rationale-Driven Multi-Task Learning [University of Alberta] https://arxiv.org/abs/2512.20074 --- [CV] Block-Recurrent Dynamics in Vision Transformers [Harvard University] https://arxiv.org/abs/2512.19941 --- [CV] Over++: Generative Video Compositing for Layer Interaction Effects [University of North Carolina at Chapel Hill & University of Maryland] https://arxiv.org/abs/2512.19661
你有没有想过,AI的强大不只靠“暴力关注”,更可能源于优美的“几何流动”?本期节目,我们将一起探索几篇最新论文带来的颠覆性视角:看AI如何像一个飞行员,在自己创造的“模拟世界”里积累经验;又如何像一位物理学家,用“棱镜”将信息分解成光谱,同时看清本质与细节;我们还会揭秘让AI学会像顶尖研究员一样思考的“童子功”,以及如何通过精妙的“公司化改造”,让AI的思考方式从“说一个字”进化到“想一句话”,变得更高效、更聪明。 00:00:39 AI大模型的“黑箱”,能不能换一种开法? 00:07:47 AI的“模拟飞行”,语言模型如何偷学世界的规则? 00:14:07 AI的新“视界”,你看到的是像素,它看到的是光谱 00:20:12 AI研究员的“童子功” 00:25:44 从“说一个字”到“想一句话”,AI思考方式的进化 本期介绍的几篇论文: [LG] Attention Is Not What You Need [University of Maryland] https://arxiv.org/abs/2512.19428 --- [CL] From Word to World: Can Large Language Models be Implicit Text-based World Models? [Southern University of Science and Technology & University of Edinburgh & Princeton University] https://arxiv.org/abs/2512.18832 --- [CV] The Prism Hypothesis: Harmonizing Semantic and Pixel Representations via Unified Autoencoding [Nanyang Technological University & SenseTime Research] https://arxiv.org/abs/2512.19693 --- [CL] Step-DeepResearch Technical Report [StepFun] https://arxiv.org/abs/2512.20491 --- [CL] NVIDIA Nemotron 3: Efficient and Open Intelligence [NVIDIA] https://arxiv.org/abs/2512.20856
与播客爱好者一起交流
添加微信好友,获取更多播客资讯
播放列表还是空的
去找些喜欢的节目添加进来吧