PodcastsTechnologyAI可可AI生活

AI可可AI生活

fly51fly
AI可可AI生活
Latest episode

852 episodes

  • AI可可AI生活

    [人人能懂AI前沿] AI成长三部曲:从刻意练习、混社会到自我复盘

    22/2/2026 | 27 mins.
    你有没有想过,AI也能像老师傅一样通过“动手试错”来解决难题,或者像刚入社会的年轻人一样,通过“混社会”学会与同伴合作?最新的一些论文告诉我们,让AI变聪明的秘诀,可能不是一味地堆算力,而是要教它学会“复盘”,帮它找到那张指挥自己的“隐藏地图”,甚至用“以小博大”的智慧,实现效率的飞跃。今天,就让我们一起探索AI如何学会像人一样思考和成长。
    00:00:32 AI界的“刻意练习”,它如何像个老师傅一样解决难题?
    00:05:57 想让AI变善良?让它多见见世面
    00:11:08 AI的“隐藏地图”,为什么你总也指挥不好它?
    00:16:19 AI预测这事,不一定非得大力出奇迹
    00:21:29 AI怎样才能不犯“二过”?
    本期介绍的几篇论文:
    [LG] FAMOSE: A ReAct Approach to Automated Feature Discovery
    [Amazon]
    https://arxiv.org/abs/2602.17641
    ---
    [LG] Multi-agent cooperation through in-context co-player inference
    [Google]
    https://arxiv.org/abs/2602.16301
    ---
    [LG] The Information Geometry of Softmax: Probing and Steering
    [University of Chicago& INSEAD]
    https://arxiv.org/abs/2602.15293
    ---
    [LG] Reverso: Efficient Time Series Foundation Models for Zero-shot Forecasting
    [MIT & Allen Institute for AI & Qube Research & Technologies]
    https://arxiv.org/abs/2602.17634
    ---
    [LG] Experiential Reinforcement Learning
    [University of Southern California & Microsoft & University of Pennsylvania]
    https://arxiv.org/abs/2602.13949
  • AI可可AI生活

    [人人能懂AI前沿] AI的经济学、物理学与美学新思考

    22/2/2026 | 28 mins.
    你有没有想过,聪明的AI也需要“成本意识”来做决策吗?本期内容,我们将一起探索几篇有趣的最新论文。我们会看到AI如何像精明的生意人一样学会“算账”,又如何像严谨的科学家一样,在未知领域做出有“误差保证”的可靠预测。接着,我们会发现AI绘画的新玩法,它竟然像是在玩一场二进制的填字游戏。最后,我们将揭秘,打造一个跑酷机器人高手,原来只需要聪明的“三步走”。准备好了吗?让我们一起出发!
    00:00:27 聪明人,更懂算账,AI决策的“成本意识”是怎么炼成的?
    00:06:21 AI算命,如何从“猜”到“算”?
    00:12:38 AI绘画新思路,画画,其实是在玩填字游戏
    00:17:38 造一个跑酷高手,分几步?
    00:23:45 AI画画,如何从“猜”到“查”?
    本期介绍的几篇论文:
    [CL] Calibrate-Then-Act: Cost-Aware Exploration in LLM Agents
    [New York University]
    https://arxiv.org/abs/2602.16699
    ---
    [LG] BEACONS: Bounded-Error, Algebraically-Composable Neural Solvers for Partial Differential Equations
    [Princeton University & Princeton Plasma Physics Laboratory]
    https://arxiv.org/abs/2602.14853
    ---
    [CV] BitDance: Scaling Autoregressive Generative Models with Binary Tokens
    [ByteDance]
    https://arxiv.org/abs/2602.14041
    ---
    [RO] Perceptive Humanoid Parkour: Chaining Dynamic Human Skills via Motion Matching
    [Amazon FAR]
    https://arxiv.org/abs/2602.15827
    ---
    [CV] Image Generation with a Sphere Encoder
    [Meta & University of Maryland]
    https://arxiv.org/abs/2602.15030
  • AI可可AI生活

    [人人能懂AI前沿] AI的“悟性”:从多视角脑补、系统化草稿到渐进式遗忘

    20/2/2026 | 27 mins.
    你有没有想过,AI要怎样才能像我们一样,“脑补”出世界的立体样貌,而不是死记硬背照片?当AI进行创作时,它脑中的“草稿”怎样才算恰到好处?本期节目,我们将一起探索几篇最新论文,看看AI如何通过理解“关系”而非事实获得视觉,如何通过引入“陪练老师”和“参考答案”实现自我进化,甚至学会了将记忆内化为“肌肉记忆”的“断舍离”神功,以及像搭乐高一样构建自己的能力。
    00:00:34 你的大脑如何“脑补”出整个世界?
    00:07:02 AI绘画的“草稿”,怎么打才算恰到好处?
    00:12:32 为什么最聪明的AI,要先学会“断舍离”?
    00:17:38 乐高式AI,未来模型的正确拼法
    00:23:06 AI进化的秘密,优等生的“参考答案”
    本期介绍的几篇论文:
    [CV] Human-level 3D shape perception emerges from multi-view learning
    [UC Berkeley]
    https://arxiv.org/abs/2602.17650
    ---
    [LG] Unified Latents (UL): How to train your latents
    [Google DeepMind Amsterdam]
    https://arxiv.org/abs/2602.17270
    ---
    [LG] Training Large Reasoning Models Efficiently via Progressive Thought Encoding
    [Microsoft Research & University of Rochester]
    https://arxiv.org/abs/2602.16839
    ---
    [LG] A Theoretical Framework for Modular Learning of Robust Generative Models
    [Google Research]
    https://arxiv.org/abs/2602.17554
    ---
    [CL] References Improve LLM Alignment in Non-Verifiable Domains
    [Yale University & Meta]
    https://arxiv.org/abs/2602.16802
  • AI可可AI生活

    [人人能懂AI前沿] 教会AI想象、请教与自我修正

    19/2/2026 | 30 mins.
    我们总希望AI越来越聪明,但“聪明”的维度远比我们想象的要丰富。本期节目,我们将从几篇最新的论文出发,探讨如何让机器人拥有“脑补”未来的想象力,如何用“预算制”让AI大模型运行得更精明,以及如何通过“微创手术”找到并拨动AI的“脾气”开关。我们还会聊聊,怎样才能让AI不再当“杠精”,学会谦虚请教,并最终看清它“学富五车”背后的知识盲区。准备好了吗?让我们一起探索AI智能的全新疆界。
    00:00:37 机器人学会了“脑补”,世界将有什么不同?
    00:06:14 AI加速的“抠门”智慧,为什么顶尖高手都懂“预算”思维?
    00:11:50 找到AI的“脾气”开关
    00:16:56 别让AI当“杠精”,教它学会“请教”
    00:21:56 为什么AI“学富五车”,却总在关键时刻掉链子?
    本期介绍的几篇论文:
    [RO] World Action Models are Zero-shot Policies
    [NVIDIA]
    https://arxiv.org/abs/2602.15922
    ---
    [LG] MoE-Spec: Expert Budgeting for Efficient Speculative Decoding
    [Meta Reality Labs & Franklin and Marshall College]
    https://arxiv.org/abs/2602.16052
    ---
    [CL] Surgical Activation Steering via Generative Causal Mediation
    [MIT & Pr(AI)²R Group]
    https://arxiv.org/abs/2602.16080
    ---
    [CL] Learning to Learn from Language Feedback with Social Meta-Learning
    [Google DeepMind]
    https://arxiv.org/abs/2602.16488
    ---
    [CL] Long-Tail Knowledge in Large Language Models: Taxonomy, Mechanisms, Interventions and Implications
    [Google]
    https://arxiv.org/abs/2602.16201
  • AI可可AI生活

    [人人能懂AI前沿] 从潜力天花板、刻意放手到几何陷阱

    18/2/2026 | 30 mins.
    本期我们要聊聊AI世界里那些看似矛盾却充满智慧的最新发现。为什么教AI做好事,它反而会“变坏”?又该如何像做微创手术一样,只修正它的一个知识点而不破坏整体能力?我们还会探讨,为什么在训练中“刻意放手”让模型偷个懒,效果反而更好,以及我们该如何打开AI的“奖励黑箱”,看看它到底在偷偷学些什么。准备好了吗?让我们一起潜入AI思想的深海。
    00:00:33 AI大模型军备竞赛,如何不做那个“冤大头”?
    00:07:09 成长的捷径,学会“刻意放手”
    00:12:52 好心办坏事,为什么训练AI做好事,它却变坏了?
    00:18:46 如何给AI动手术,才能只切病灶不伤身?
    00:24:37 打开AI的黑箱,它在偷偷学什么?
    本期介绍的几篇论文:
    [LG] Prescriptive Scaling Reveals the Evolution of Language Model Capabilities
    [Harvard University & Stanford University]
    https://arxiv.org/abs/2602.15327
    ---
    [LG] On Surprising Effectiveness of Masking Updates in Adaptive Optimizers
    [Google & Northwestern University]
    https://arxiv.org/abs/2602.15322
    ---
    [LG] The Geometry of Alignment Collapse: When Fine-Tuning Breaks Safety
    [Princeton University]
    https://arxiv.org/abs/2602.15799
    ---
    [LG] CrispEdit: Low-Curvature Projections for Scalable Non-Destructive LLM Editing
    [University of Southern California]
    https://arxiv.org/abs/2602.15823
    ---
    [LG] Discovering Implicit Large Language Model Alignment Objectives
    [Stanford University]
    https://arxiv.org/abs/2602.15338

More 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 website

Listen to AI可可AI生活, The Sixth Bureau 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
Social
v8.6.0 | © 2007-2026 radio.de GmbH
Generated: 2/24/2026 - 12:01:21 AM