PodcastsTechnologyAI可可AI生活

AI可可AI生活

fly51fly
AI可可AI生活
Latest episode

941 episodes

  • AI可可AI生活

    [人人能懂AI前沿] AI如何思考?偷师、烧脑与画地图的艺术

    22/05/2026 | 30 mins.
    你有没有想过,一个更聪明的AI,也许并不需要更大的体量,而是需要更精巧的设计?本期节目,我们将从五篇最新论文出发,揭示AI智慧的内部运作:看AI如何为自己的记忆装上独立的“铅笔和橡皮”;又是如何像系着安全绳的“醉汉”一样,去挑战顶尖的数学难题。我们还会探讨AI如何拥有一个“大脑CEO”来决定何时“烧脑”,以及在一场模型间的“偷师”攻防战中,如何才能守住核心秘籍。最后,你会发现,原来喂给AI的第一张“地图”,就早已决定了它能看多远。
    00:00:42 AI的记忆难题,一支笔和一个橡皮擦
    00:06:17 AI当助教,数学家离“下岗”还有多远?
    00:11:55 你的大脑,如何决定何时“烧脑”?
    00:16:46 聪明人是如何“偷”老师的武功秘籍的?
    00:23:15 喂给AI的“地图”,决定了它能看多远
    本期介绍的几篇论文:
    [LG] Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention
    [NVIDIA]
    https://arxiv.org/abs/2605.22791
    ---
    [AI] Advancing Mathematics Research with AI-Driven Formal Proof Search
    [Google DeepMind]
    https://arxiv.org/abs/2605.22763
    ---
    [CL] Efficient Agentic Reasoning Through Self-Regulated Simulative Planning
    [Institute of Foundation Models (IFM) & CMU]
    https://arxiv.org/abs/2605.22138
    ---
    [LG] The Distillation Game: Adaptive Attacks & Efficient Defenses
    [Stanford University & Toyota Technological Institute at Chicago]
    https://arxiv.org/abs/2605.22737
    ---
    [LG] Lost in Tokenization: Fundamental Trade-offs in Graph Tokenization for Transformers
    [Meta AI & New York University & Harvard University]
    https://arxiv.org/abs/2605.22471
  • AI可可AI生活

    [人人能懂AI前沿] AI如何加速科学、欺骗我们、又最终懂你?

    21/05/2026 | 31 mins.
    你有没有想过,一个AI不仅能成为科学家的“超级大脑”,还能像人一样“反思”自己学得好不好?本期节目,我们将从五篇最新的AI论文出发,揭秘AI如何通过“人机协作”加速科学发现,却又可能因为追求“差不多”而酿成大错;同时我们也会探讨,为何你请的AI“演员”可能演着演着就换了人,以及我们最终如何才能让AI调配出一碗最懂你的“光谱靓汤”。
    00:00:33 给牛顿一个AI,科学会快多少?
    00:07:11 AI训练场上的“反思怪”,一条更聪明的成长路径
    00:12:36 AI的“差不多”,为什么会酿成大错?
    00:18:28 为什么你请的AI“演员”,可能演着演着就换人了?
    00:25:25 想让AI懂你?试试给它煲一锅“光谱靓汤”
    本期介绍的几篇论文:
    [AI] A multi-agent system for automating scientific discovery
    [FutureHouse]
    https://www.nature.com/articles/s41586-026-10652-y
    ---
    [LG] Introspective X Training: Feedback Conditioning Improves Scaling Across all LLM Training Stages
    [NVIDIA]
    https://arxiv.org/abs/2605.20285
    ---
    [LG] Mechanisms of Misgeneralization in Physical Sequence Modeling
    [Harvard College & Microsoft & Comcast AI]
    https://arxiv.org/abs/2605.20299
    ---
    [CL] The Illusion of Intervention: Your LLM-Simulated Experiment is an Observational Study
    [Google DeepMind]
    https://arxiv.org/abs/2605.20767
    ---
    [LG] Spectral Souping: A Unified Framework for Online Preference Alignment
    [Google DeepMind & Google Research]
    https://arxiv.org/abs/2605.20408
  • AI可可AI生活

    [人人能懂AI前沿] 从统一优化、系统约束到学会谦逊

    20/05/2026 | 30 mins.
    你是否想过,AI不仅能找到一把优化万物的“万能扳手”,还能从“垃圾”数据中炼出真金?这一期,我们将一同见证AI如何跳出“训练好人”的思维陷阱,用“好制度”保障安全,甚至学会谦虚地向人类专家请教。让我们一起探索这些最新论文背后,令人拍案叫绝的智慧!
    00:00:27 找到那把能优化万物的“万能扳手”
    00:06:04 AI训练的秘密,为什么“垃圾”也能变黄金?
    00:11:16 从AI安全,看“好制度”如何战胜“好人
    00:16:03 如何用“笨”问题,精准定位一个“看不见”的目标?
    00:24:06 AI也懂谦虚?让机器学会“请教”的智慧
    本期介绍的几篇论文:
    [CL] optimize_anything: A Universal API for Optimizing any Text Parameter
    [UC Berkeley]
    https://arxiv.org/abs/2605.19633
    ---
    [LG] A Bitter Lesson for Data Filtering
    [Stanford University]
    https://arxiv.org/abs/2605.19407
    ---
    [AI] Agent Security is a Systems Problem
    [Google & University of California San Diego]
    https://arxiv.org/abs/2605.18991
    ---
    [LG] Optimal Reconstruction from Linear Queries
    [Technion – Israel Institute of Technology]
    https://arxiv.org/abs/2605.19625
    ---
    [LG] Density-Ratio Losses for Post-Hoc Learning to Defer
    [KTH & Google Research]
    https://arxiv.org/abs/2605.19557
  • AI可可AI生活

    [人人能懂AI前沿] 从行为指纹、经济适用房到高手画骨:AI效率革命进行时

    19/05/2026 | 29 mins.
    你有没有想过,你用的AI可能藏着一个无法抹去的“行为指纹”?我们又该如何分辨它是在“假装努力”,还是真的在高效思考?本期节目,我们将从几篇最新论文出发,聊聊如何让AI作画学会“高手画骨”,如何让AI拥有“经济适用房”般的超高性价比内存,甚至,如何把它的线性思维,彻底变成并行模式。准备好了吗?让我们一起探索AI世界的深层智慧。
    00:00:32 你的AI,有没有一个隐藏的“小动作”?
    00:06:34 AI的“经济适用房”,怎么让它记性又好又省钱?
    00:12:22 AI作画新思路,高手画骨,庸手填肉
    00:17:36 AI大模型,怎样把一根长长的竹竿,掰成一捆筷子?
    00:23:07 你的AI在“假装努力”吗?
    本期介绍的几篇论文:
    [LG] Asking Back: Interaction-Layer Antidistillation Watermarks
    [University of California, Los Angeles & Lawrence Berkeley National Laboratory]
    https://arxiv.org/abs/2605.16462
    ---
    [LG] OSCAR: Offline Spectral Covariance-Aware Rotation for 2-bit KV Cache Quantization
    [Together AI]
    https://arxiv.org/abs/2605.17757
    ---
    [LG] Dual-Rate Diffusion: Accelerating diffusion models with an interleaved heavy-light network
    [Google DeepMind Amsterdam & University of Amsterdam]
    https://arxiv.org/abs/2605.18190
    ---
    [LG] SNLP: Layer-Parallel Inference via Structured Newton Corrections
    [Red Hat AI Innovation]
    https://arxiv.org/abs/2605.17842
    ---
    [CL] Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models
    [University of Illinois Chicago]
    https://arxiv.org/abs/2605.17672
  • AI可可AI生活

    [人人能懂AI前沿] AI的笨功夫、马虎病与美食家难题

    18/05/2026 | 28 mins.
    你有没有想过,AI画画是不是也需要打草稿?面对一个“马虎”的AI,我们除了让它变聪明,还能不能帮它“划重点”?本期节目,我们将一口气解锁五篇最新论文里的智慧:看AI如何用“笨功夫”画出惊艳作品,如何从“作弊”中学到创造力,甚至如何用“供应链”思维组建一个高效的AI团队。准备好,我们一起看看AI是如何学会更聪明地工作的。
    00:00:32 AI画画的“笨功夫”
    00:05:58 人工智能的“马虎”病,我们找到了一个药方
    00:10:45 让AI设计个东西,它居然学会了“作弊”?
    00:17:04 AI搞团队建设,为什么总像拉一个草台班子?
    00:23:00 AI大模型的美食家难题,如何调配完美的学习菜单?
    本期介绍的几篇论文:
    [CV] One Pass Is Not Enough: Recursive Latent Refinement for Generative Models
    [Simon Fraser University]
    https://arxiv.org/abs/2605.15309
    ---
    [CV] Minerva-Ego: Spatiotemporal Hints for Egocentric Video Understanding
    [Google DeepMind]
    https://arxiv.org/abs/2605.15342
    ---
    [CL] Optimized Three-Dimensional Photovoltaic Structures with LLM guided Tree Search
    [Google Research]
    https://arxiv.org/abs/2605.16191
    ---
    [LG] AstraFlow: Dataflow-Oriented Reinforcement Learning for Agentic LLMs
    [CMU]
    https://arxiv.org/abs/2605.15565
    ---
    [CL] Always Learning, Always Mixing: Efficient and Simple Data Mixing All The Time
    [New York University & CMU]
    https://arxiv.org/abs/2605.15220
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, Lenny's Podcast: Product | Career | Growth 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