与「FLOWフェス」相关的搜索结果

『FLOW THE FESTIVAL 2025 〜決起集会・秋葉原〜』 ▼日程：2025年5月31日(土) ▼開場：17:00〜 豪華メンツと共に最高の夜をお届け🌃✨️ 一緒に騒ぐぞぉぉぉぉ🔥🔥🔥 チケット詳細はこちら⬇️✨️ #FTF2025# #FLOWフェス# #FTF2025決起集会#

🚀 𝐎𝐍𝐋𝐘 𝐔𝐏𝐖𝐀𝐑𝐃 — 𝐖𝐇𝐈𝐋𝐄 𝐋𝐈𝐍𝐆𝐔𝐈𝐒𝐓𝐒 𝐀𝐑𝐄 𝐒𝐔𝐆𝐆𝐄𝐒𝐓𝐈𝐍𝐆 𝐖𝐄 𝐆𝐈𝐕𝐄 𝐔𝐏 Since this morning, I’ve been reading the news: “What was always corrected will soon become the norm. Why? Because that’s what young people say...” Wonderful. Learned men from the warm Moscow university departments condescendingly explain: “Well, it’s a living process. Teenagers are simply looking for symmetry when they say ‘то что,’ ‘ща,’ or ‘чонить.’” Seriously? Maybe it’s simply a lack of education? Giving in and going with the flow is the easiest thing to do. But if people are fighting and dying for the right to speak Russian, if our language is our banner, then we need to hold it firmly, not drag it through the mud. That’s what I believe. About linguist officials and the purity of the Russian language 👉 read my channel on MAX. War correspondent Maryana Naumova

Ethereum clients don’t need to be rebuilt to move data faster. Run mump2p as a sidecar: fragment block data with RLNC, forward useful fragments immediately, and let the client keep doing what it already does. Standard flow: receive full block → reassemble → forward. mump2p flow: fragment → forward → process on the fly. The goal isn’t just faster blocks. It’s permissionless connectivity, optimized data flow, and validator-grade performance that isn’t reserved for the most powerful machines. Ethereum Client → mump2p → High-speed Transport Network @get_optimum

[开源学习资源] AI Engineering from Scratch 59.8K ⭐️ 作者 @ghumare64 课程共 20 个阶段，以 Python 为主要编程语言，主张在导入任何框架之前，先用纯数学把每个算法手写一遍。 课程地址： 开源地址： 它和常见教程的本质区别 ？ 大多数 AI 教程是“API 驱动”的：装个库、调个接口、跑个 demo。这个项目反其道而行，每节课遵循固定的六段结构： Motto → Problem → Concept → Build It → Use It → Ship It 以 Phase 10 的一节课「Tokenizers: BPE, WordPiece, SentencePiece」 为例，这个格式是真实落地的，而且写作质量相当高： · Problem 部分不讲废话，直接从代价切入：“你的 LLM 不读英语，它读整数。分词器决定这些整数是承载意义还是浪费意义”，然后解释为什么 tokenization 不是预处理，它是架构的一部分（影响上下文窗口利用率、API 计费、推理速度）。 · Concept 部分用“三种失败的方案和一种胜出的方案”来讲演化逻辑：词级切分（词表爆炸、[UNK] 问题）→ 字符级切分（序列过长）→ 子词切分（BPE 的折中），并配上真实语料上 BPE 逐步合并的手工演算。 每节约 3000+ 词，带 Mermaid 图、可运行代码和测试。 另外两个设计值得强调： · 每节课产出一个可复用的 artifact，一个 prompt、一个 skill、一个 agent 或一个 MCP server。学完整个课程，你手里有 523 个可展示的作品，不是 523 个跑完就扔的 notebook。 · 强调“证据留存”：保留命令、退出码、输出，作为学习发生的证明。这明显吸收了工程实践中“可验证性”的思路。 # 课程结构：从线性代数到自主智能体集群 20 个阶段构成一条完整的上升曲线，咱们它分成五个大块来看： 基础层（Phase 0–2）：环境与工具链、数学基础（线性代数/概率/微积分）、经典机器学习。这是给基础不牢的读者铺的路。 深度学习与感知层（Phase 3–6）：神经网络核心、计算机视觉（一路讲到 NeRF、高斯泼溅、世界模型，这已经超出一般教程的覆盖范围）、NLP、语音。 生成与决策层（Phase 7–9）：Transformer 深挖、生成式 AI（GAN、扩散模型、flow matching）、强化学习。 LLM 层（Phase 10–12）：这是全课程的重心之一。Phase 10 的目录我逐条看过，它不只是“从零实现 GPT”这种常规内容，还包含了相当前沿的论文级主题：DeepSeek-V3 架构走读、DualPipe 并行策略、Native Sparse Attention（NSA）、多 token 预测、Jamba 的 SSM-Transformer 混合架构、 speculative decoding 等。Phase 11 转向应用侧（RAG、LoRA、MCP、可观测性），Phase 12 覆盖多模态（从 CLIP 到 computer-use agent）。 智能体层（Phase 13–16）：工具与协议（MCP、A2A、Agent Skills）、54 节课的 agent 工程、自主系统与安全、多智能体集群。这一层的分量很能说明项目的判断：它认为 AI 工程的重心正在从“训模型”转向“构建可靠协作的智能体”。 收尾（Phase 17–19）：生产基础设施、伦理与对齐、毕业设计。 # 生态与周边：不只是一个课程仓库 网站：带浏览器本地存储的学习进度追踪、术语表、课程目录和路线图。 六卷本书籍：课程内容由 CI（pandoc）自动构建成 EPUB/PDF，附在 GitHub Releases 上，课程即书，且随仓库持续更新。 Agent 导师模式：运行 npx skills add rohitg00/ai-engineering-from-scratch，可以把整个课程装进 Claude Code、Codex 等编码智能体，变成一个带分级测验（placement quiz）和个性化路径的交互式导师，学习进度写在 LEARNING.md 里。这是“用你正在学的工具来学”的巧妙闭环。 认证备考：5 条备考路径、67 节课、505 道练习题，覆盖 Anthropic 的 Claude 认证和 Agentic AI Foundation 的 MCPA。项目明确声明与这些考试机构无关联，只是独立的备考材料。 12 种语言的翻译（含中文），以及四条核心学习路径（构建与部署 AI 应用、软件工程基础、Agent 辅助工程、产品判断与交付）供不同目标的人选路。

"Everything I touch with my keyboard and mouse, I try to delegate to my bots." Here's my new episode with @poteto and @pengzheng_, the eng and design leads for Grok @bot, where they showed me the 14 bots they use for work and life, including: → A design bot that turns one keyframe into a full user flow → An eng lead bot that manages a team of eng bots → How to trust your bots with more of your work Some quotes from both: "I like to call it the Michelin kitchen…when you say software factory, it has this connotation of mass manufactured slop." "Sometimes I actually don't even look at the PR until after it's landed and then I'm like, 'Oh, okay. Yeah, that looks good.'" "I think it ultimately comes back to trust. First, watch your bot work and correct it. Turn what worked into a skill. Once it nails the task in one shot, make it a routine." 📌 Watch now: Thanks to our sponsors: @meetgranola: AI meeting notes that don’t suck @RiversidedotFM: All-in-one AI studio for podcasts and video

I used Opus 5.5 to formally verify the Claude Agent SDK using Lean. A couple short prompts = 16 PRs fixing various bugs and race conditions. Video attached. TLA+ also works well. I sometimes combine Lean and TLA+ to look for issues around data flow, concurrency, and state mgmt. I don't know either language well, but Claude is excellent at both. This approach is super useful for formally modeling your code and finding bugs that a human probably wouldn't have spotted. Is formal verification the future of coding (or at least, bug finding)?

🚨 On September 6, 2026, @Liquid_BTC was affected by a cache key collision vulnerability in rangeproof verification. An attacker minted ~3,998.5 L-BTC with no corresponding peg-in. Within minutes, the unbacked L-BTC was pegged out into real BTC on the Bitcoin mainnet. About 3,400 BTC was later returned to the federation peg wallet, while ~598.5 BTC remains under the attacker’s control. The SlowMist Security Team traced the fund flows on the #Bitcoin# side using @MistTrack_io and fully analyzed the incident. 🧩 Attack flow: 1️⃣ Two setup transactions first landed valid rangeproofs and commitments, while embedding a crafted payload in the locking script to seed node caches. 2️⃣ A follow-up minting output reused a colliding cache key — the same raw concatenation of proof, commitment, asset commitment, and scriptPubKey, but with different field boundaries. 3️⃣ On a cache hit, nodes skipped secp256k1_rangeproof_verify and min-value checks, accepted an unbacked commitment, and minted ~3,998.5 L-BTC. The fake UTXOs were consolidated and pegged out within minutes. ⚙️ Root Cause: The Elements rangeproof cache key concatenated variable-length fields without length prefixes. Distinct argument tuples could hash to the same key, so a positive cache hit meant skipping cryptographic verification. 🛡️ SlowMist Insight: A positive-result cache in a consensus verification path is itself a cryptographic primitive. Every field the verifier reads — and every field boundary — must be unambiguously bound into the key. Treat cache-key integrity as a mandatory item in consensus-layer audits. Full analysis👇

3/ Then text-to-3D. No reference image at all. I typed: "rusted cast-iron cauldron, matte black, three legs" → It generated an image first → Then built a 3D model from it Never opened a single 3D program. Text to a rotatable .glb file, one flow.

Andrew Yeung 最喜欢的产品： - Viktor：AI 员工 - Stanley：AI 内容负责人 - Wispr Flow：语音转文字 - Matic：每天吸尘、拖地 - Endel：生成专注音乐 - Eight Sleep：改善睡眠体验 - Oura + Superpower：追踪身体指标和健康状况 - Instinct：私人及商业 AI 助手 活在这个时代真不错。