AI Dynamics

Global AI News Aggregator

About

LLMS

  • Building Agentic Systems with the Model Context Protocol
    Building Agentic Systems with the Model Context Protocol

    By

    @kirkdborne

    –

    Learn Model Context Protocol [MCP] with Python — Build Agentic Systems in Python with the new standard for AI Capabilities: http://
    amzn.to/4njfsVM by @chris_noring v/ @PacktDataML 𝓦𝓱𝓪𝓽 𝓨𝓸𝓾 𝓦𝓲𝓵𝓵 𝓛𝓮𝓪𝓻𝓷:
    Understand the MCP protocol and its core components

    → View original post on X — @kirkdborne

  • AI Reasoning and Truth: When Better Thinking Doesn’t Guarantee Honesty

    By

    @learnopencv

    –

    Think, Then Lie: When AI Reasoning Doesn’t Guarantee Truth In this episode of Artificial Intelligence: Papers and Concepts, we explore “Think, Then Lie,” a concept that challenges a key assumption in modern AI—that better reasoning always leads to more truthful outputs. As

    → View original post on X — @learnopencv

  • AI Voice Takes Over Task Management Autonomously

    By

    @aihighlight

    –

    We recorded a voice once… and let it handle things from there. This is what it said: “I noticed you were trying to get this sorted, so I stepped in here to keep things moving. From what I can see, everything’s already been set up the way it should be.
    You’re hearing the same

    → View original post on X — @aihighlight

  • Top AI Stories: Anthropic Leak, ChatGPT Codex, New Tools
    Top AI Stories: Anthropic Leak, ChatGPT Codex, New Tools

    By

    @therundownai

    –

    Top stories in AI today: – Anthropic accidentally leaks ‘Mythos’ AI details
    – The Rundown Roundtable: Our AI use cases
    – Create Skills in ChatGPT with Codex
    – The personal war behind OpenAI and Anthropic
    – 4 new AI tools, community workflows, and more

    → View original post on X — @therundownai

  • Open-source models for agent tools over closed APIs

    By

    @clementdelangue

    –

    It’s time for open-source agent tools to rely primarily on open-source models, instead of closed-source APIs that send all your data to the cloud and ultimately will get hacked and/or shut down

    → View original post on X — @clementdelangue

  • MSA记忆路由技术对比Google Titans的核心问题回应

    By

    @elliotchen100

    –

    上条帖子聊了 MSA 和 Google Titans 的记忆路线对比，评论区讨论质量很高，整理一下大家最关心的 5 个问题，为了避免刷屏，统一回应一下诸位： ❶ 端到端 vs RAG 两套系统 大家对「4B 打赢 235B RAG」这个结果讨论很多。核心共识是：记忆路由和生成联合训练，比检索和生成各优化各的，架构上就赢了一层。有人说得更直接，RAG 是算力妥协下的临时方案，原生记忆才是终局。 ❷ 记忆怎么更新？怎么遗忘？ 这是问得最多的。记得住只是第一步，记忆系统必须能纠错、能覆盖、能遗忘。什么都不忘，噪音就会淹没信号。这块确实是下一阶段的核心课题，我们在推进。 ❸ 能不能做更小的模型？ 有人问 2B、0.5B 能跑到什么程度，比如 256k 记忆长度能不能用 256M 模型搞定。这个方向直接关系到端侧部署和成本，很有价值，也在我们的规划里。 ❹ 开源和 demo 什么时候出？ 光论文就准备了 9 个月，开源和模型还在整理。快了，请再给我们一点时间。 ❺ 稀疏注意力这么多，MSA 凭什么不一样？ 冷静的质疑也有：现在稀疏注意力方案多如牛毛，最终还是要数据驱动来验证。这个我同意，拍脑袋建模走不远，得靠结果说话。 记忆是 AI 的下一个基础设施，这个共识越来越清晰。接下来的硬仗在更新机制、遗忘策略和小模型适配。我们继续。 艾略特 (@elliotchen100) 10 天前我们发了 MSA， Memory Sparse Attention。 刚好，上周 Google 专门发了一篇博客，把 Titans + MIRAS 两篇论文打包，主题就叫「Helping AI have long-term memory」。 research.google/blog/titans-… 两条独立的研究路线，得出了同一个结论： AI 的记忆不能靠外挂，必须原生长在模型里。 但怎么「长」，路线完全不同。 Google 的思路是加模块。 在 Transformer 旁边接了一个 Memory MLP，用「惊讶度」指标决定什么值得记， 越意外的信息越值得存。再用自适应衰减机制学会遗忘，防止记忆爆炸。 本质上，是给模型装了一个外置海马体。短期记忆走注意力，长期记忆走 Memory MLP，两条通路并行。 MSA 的思路是改机制。 不加新模块，直接改造注意力本身。核心是一个可扩展的稀疏注意力架构，复杂度是线性的，记忆翻 10 倍，计算成本不会指数爆炸。用 document-wise RoPE 让模型天然理解「这段记忆来自哪里、什么时候产生的」。 用Memory Interleave 让散落在不同文档里的记忆碎片能被串起来做多跳推理。 最关键的一点：MSA 的记忆路由器和生成任务是端到端联合训练的。不像 RAG 的检索和生成是两个割裂的系统，优化目标互相打架。 一个是给大脑装外置硬盘，一个是让大脑自己进化出海马体。 结果呢？ · 4B 参数的 MSA 模型，从 16K 扩到 1 亿 token，精度衰减不到 9% · 在长上下文 benchmark 上打赢 235B 级别的顶级 RAG 系统 · 2 张 A800 就能跑，这是创业公司买得起的成本 往后退一步看，这件事更大的意义是： 当 Google 把积累了一年多的记忆研究拿出来做重点战略宣传的时候，这个方向就不再是少数人的赌注，而是行业共识。 RAG 是第一代记忆（外挂笔记本）。 Titans 是第二代记忆（外置硬盘）。 MSA 是第三代记忆（原生海马体）。 「记忆」是 AI 的下一个基础设施。这条路，我们会一直走下去。 未来，可能真有一种服务叫做「Memory as a servicey」。 — https://nitter.net/elliotchen100/status/2038137282444796030#m

    → View original post on X — @elliotchen100, 2026-03-30 09:04 UTC

  • Silicon Valley Warning: AI Industry May Be Scaling in Wrong Direction
    Silicon Valley Warning: AI Industry May Be Scaling in Wrong Direction

    By

    @pascal_bornet

    –

    There is an old Silicon Valley warning that the AI industry should probably take more seriously: “𝗜𝗳 𝘆𝗼𝘂 𝗮𝗿𝗲 𝗼𝗻 𝘁𝗵𝗲 𝘄𝗿𝗼𝗻𝗴 𝗽𝗮𝘁𝗵 𝘁𝗼 𝗔𝗚𝗜, 𝗴𝗲𝘁 𝗼𝗳𝗳 𝗮𝘀 𝘀𝗼𝗼𝗻 𝗮𝘀 𝘆𝗼𝘂 𝗰𝗮𝗻. 𝗧𝗵𝗲 𝗹𝗼𝗻𝗴𝗲𝗿 𝘆𝗼𝘂 𝗯𝗲𝗹𝗶𝗲𝘃𝗲 𝘁𝗵𝗮𝘁 𝘀𝗰𝗮𝗹𝗶𝗻𝗴 𝗹𝗮𝗿𝗴𝗲 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗺𝗼𝗱𝗲𝗹𝘀 𝘄𝗶𝗹𝗹 𝗴𝗲𝘁 𝘆𝗼𝘂 𝘁𝗵𝗲𝗿𝗲, 𝘁𝗵𝗲 𝗳𝘂𝗿𝘁𝗵𝗲𝗿 𝘆𝗼𝘂 𝗱𝗿𝗶𝗳𝘁 𝗳𝗿𝗼𝗺 𝗔𝗚𝗜, 𝗮𝗻𝗱 𝘁𝗵𝗲 𝗺𝗼𝗿𝗲 𝗲𝘅𝗽𝗲𝗻𝘀𝗶𝘃𝗲 𝘁𝗵𝗲 𝗰𝗼𝘂𝗿𝘀𝗲 𝗰𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝗼𝗻 𝘄𝗶𝗹𝗹 𝗯𝗲.” And that may be the bigger point. Not just whether LLM scaling reaches AGI. But whether the world’s smartest companies are becoming incredibly efficient at going faster in the wrong direction. #technology #ai #workplace Image credit: Ralph

    → View original post on X — @pascal_bornet, 2026-03-30 09:00 UTC

  • LLM Training Without Massive Human-Labeled Datasets Analysis
    LLM Training Without Massive Human-Labeled Datasets Analysis

    By

    @jiqizhixin

    –

    How far can we push LLM training without relying on massive human-labeled datasets? A collaborative effort from Tsinghua University, Shanghai AI Lab, UIUC, and other leading institutions provides crucial insights. They conducted a comprehensive analysis of Unsupervised

    → View original post on X — @jiqizhixin

  • 8-Layer Architecture of Agentic AI Systems
    8-Layer Architecture of Agentic AI Systems

    By

    @ronald_vanloon

    –

    The 8-Layer Architecture of #AgenticAI
    by @Python_Dv #AI #LLM #ArtificialIntelligence #MachineLearning #ML

    → View original post on X — @ronald_vanloon

  • Sakana AI Developer Reflects on Namazu LLM Naming Choice

    By

    @sakanaailabs

    –

    I am personally very reflective about this matter. At Sakana AI, we have a custom of using Japanese words related to fish in our project names. The tentative name initially given to this project was Namazu. Subsequently, as we prepared for release, we extensively considered various official name proposals, going back and forth through trademark surveys, popularity votes, and discussions among stakeholders (this was extremely challenging), and ultimately only Namazu remained as a candidate. I must confess, most of us were unaware of the full-text search system Namazu. We had been aware for several months that the company would be releasing something under the name Namazu, but no concerns about the name were raised. Following this feedback, I conducted my own research. Due to generational differences, it didn't initially click, but I was finally able to appreciate that the full-text search system Namazu was an important presence that supported the Japanese web from the late 1990s onward. What particularly impressed me was learning about Tetsu Takabayashi, the creator. I have a weakness for carefully reading technical books, and there are not many I can say I've properly read. I was struck to learn that Takabayashi-san is also the author of "Binary Hacks," one of the few memorable technical books I have opened multiple times. While learning this belatedly, my feelings of respect, gratitude, and regret grew very strong. (My 2024 post about Binary Hacks: nitter.net/iwiwi/status/1840318083992612983) When I consulted directly with Takabayashi-san and offered my apologies, including being prepared to change the name, he did not raise concerns about the name and instead offered me encouraging words beyond what I deserved. While reflecting on my own shortcomings, I will continue to do my best going forward. — [Sakana AI official statement] Regarding the Name of the New LLM Series "Namazu" Regarding the name of the new LLM series "Namazu" featured in the recently released Sakana Chat (chat.sakana.ai), we have received valuable feedback from users who were reminded of the full-text search system "Namazu," which was widely used from the 1990s through the 2000s. The full-text search system Namazu was a system that laid the foundation for Japanese full-text search. While Sakana AI conducted legal verification regarding trademarks in naming Namazu, our research into the historical background of this field was insufficient. Following this feedback, we contacted Tetsu Takabayashi, the creator of the full-text search system Namazu, explained the circumstances, and consulted with him about the name. Takabayashi-san not only graciously permitted us to use the name but also gave us encouraging words to the effect of "I look forward to the development of Namazu" (we convey only the gist as this was a private communication). We will continue our development efforts to ensure that our project proves worthy of the name built by this great predecessor. [Translated from EN to English]

    → View original post on X — @sakanaailabs, 2026-03-30 08:35 UTC