与「EXT」相关的搜索结果

WEMBY DIRECTING TRAFFIC 👀 Slow down the offense. Call out a set. Fire the jump pass to Harper for 2. The @spurs extended their winning streak to 11 games and improved to 2-0 in West Group C!

X AI ALGORITHM UPDATE (BIG ONE) 🚨 X updated their open source algorithm yesterday and the biggest number in it is gone: copy link went from 20 to 0 points. Small accounts should focus on: posts people like, reply to and stay on (so stop counting on link copies) tl;dr: - on Nov 13 X's open source code set the score for copying a post's link to 0 - copy link: 20 → 0 - it was the top score in the whole system, 4x a reply - the code also names a "copy link bait" spam check - likes (0.708), replies (5) and DM shares (5) didn't move full details: - The number that's gone Since I started reading X's open source code in August, one number sat at the top: 20, for someone tapping share and copying your post's link. Nothing else came close. As of the Nov 13 release, it's 0. X still guesses who'll copy your link. It just multiplies that guess by 0 now. X didn't explain it. It's just what the code says. (I check the open source algo daily with Claude and share updates with you) - Cross it off your list My own checklist called copy link the biggest number. Last week's post said it too. It was true then. It isn't now. If you saved that list, cross that line out. - The "copy link bait" check X's code now mentions a spam check called "copy link bait". In the code it adds a spam label to the post, and that label already keeps a post out of strangers' feeds. It only shows up in a test, so I can't see if it's running or how strict it is. My read: X saw people baiting link copies, so it turned the score off and built something to catch the bait. That's a read, the code doesn't say why. - Why that matters if you're small If your plan was posts people copy the link to, that plan now earns 0 in X's ranking. A copied link can still bring people in from outside X. It just doesn't push your post up anymore. What still pays: likes at 0.708 (the highest ever), people staying on your post, replies at 5 and back and forth with mutuals at 20. - Also new: photo and video tags If a post has a spam label, people you tag in its photo or video who don't follow you won't get notified. Same idea as last week's rule for replies. - What didn't change Every other score. Your new posts still get their extra push in the first 2 hours and first 200 views, and early likes still pick which one gets it. Here's a recap of what you should be doing: - aim for likes, replies and time spent, the copy link score is gone - a DM share still earns 5, same as a reply or a quote - likes sit at 0.708, never higher - time on your post keeps counting at its best rate yet - 10+ seconds after someone taps in adds 0.385 - write your replies yourself, under big accounts most of all - give people a reason to hang around your profile - in videos the opening seconds carry it, an open alone scores 0 - posts read to the end help X find you strangers - smaller accounts (below 50K) get a shot at a spot near 16 for a fresh original - early likes decide which new post gets that spot - post when your followers are awake, the push lasts 2 hours and 200 views - mutuals going back and forth earn 20, so follow back the regulars - only your originals ever get shown to strangers - like 1 and like 32 each unlock another way to reach strangers - one strong post per feed, your second only counts 62% - clips longer than 10 seconds for the video lists - a post likely to win you a follow gets +4 - give your pinned post a look, it can affect the whole account What you should NOT be doing: - counting on copy link shares, they score 0 now - baiting link copies, the code names a spam check for exactly that - a spam label also stops your photo tags reaching non-followers - AI-written replies under big accounts, "llm slop" is a flag X uses - copy-paste replies, labeled ones skip non-followers' notifications - chasing taps, 0.274 and dropping - drifting off your topic, each probable "not interested" takes 53.07 away - videos people open and leave, the open itself earns 0 - posting at 3am, it wastes the 2 hour push - expecting replies and reposts to reach strangers - an insult label, it hides that post from non-followers - earning a misinfo or low quality label, which adds a warning and locks strangers out of liking or sharing - 3 adult photos or videos among your latest 5 posts, that puts a label on you - forgetting videos get rechecked at 64 and 512 likes - one account label and strangers stop seeing you - pinning a link to a site X rates badly, that can mean a label for a week - posting your own old words again risks the copypasta flag - a low quality link plus tagged strangers reads as spam - bait for engagement, big or small it gets a label - several posts back to back, they split the same feed - hashtag piles, they do nothing - bookmarks for reach, they score 0 - hoping a 3 day old post still lands in For You, it stops at 2 days

BREAKING: SpaceXSI completes MiniHard construction in just ~5 months, with Gun Range targeting a ~4-month build. According to SemiAnalysis, SpaceXSI is roughly halving construction timelines while most comparable projects face 8–12 month builds. • MiniHard: Construction finished in approximately 5 months. • Gun Range: The new site just south of the Tennessee–Mississippi state line is targeting approximately 4 months of construction. • Colossus 1: Its first retrofit delivered 150 MW of capacity in 122 days. • Colossus 2: A subsequent retrofit delivered 210 MW in just 91 days. • Project SPAM: Work started October 5, with the first capacity targeted for December. • Massive expansion: SemiAnalysis projects the Memphis–Mississippi campus will grow from 0.4 GW of IT capacity in January to 2.4 GW by the end of 2026. That would be a sixfold increase in one year. • Five major projects: The expansion spans Macrohard, MiniHard, Macroharder, Project SPAM and Gun Range. • Tesla powering the scale: SemiAnalysis reports 1,377 Tesla Megapacks across Colossus 2, providing nearly 6 GWh of battery storage. The construction timelines and capacity forecasts are SemiAnalysis estimates, with future milestones still targets. Elon Musk’s teams are bringing extraordinary speed and scale to the infrastructure powering AI.

⚠️ Cursor has published a security incident report on the earlier supply chain attack. According to the investigation, the attacker used access keys issued by Cursor to Fireworks and subsequently exposed on Fireworks’ side to gain access to some of Cursor’s SaaS services and internal systems. The attacker ultimately tampered with the distribution cache for the Claude Code extension to collect information about users’ machines and install SSH backdoors. The compromised versions were Anthropic.claude-code 2.1.312 and 2.1.313 in Cursor’s VS Code extension marketplace distribution cache, affecting macOS and Windows. The exposure window ran from October 29 at 17:27 to October 30 at 04:35 (UTC+8), with up to approximately 5,000 users affected. Security recommendation: If you installed or updated to either affected version through Cursor’s marketplace during this window, read the official notice as soon as possible and run the remediation script for your platform to check for and remediate any compromise: Fireworks’ initial announcement and subsequent update:

建议收藏，用到时照着做。做完你能让 DeepSeek 自己决定“要不要调用你写的函数”，并把结果用人话讲回来。 AI 能查天气、查订单，是它自己去查的吗？DeepSeek 官方一句话拆台：模型本身不执行任何函数 这是 DeepSeek 官方 API 文档里 Tool Calls（工具调用）一页。先说结论： 官方原话是，函数要由你自己提供，模型本身不执行具体函数。它只负责说“我想调这个函数、参数是这些”，真正去跑的是你的代码。 先说清词。 工具调用（Tool Calls），是让模型能用外部工具的功能，比如查天气、查数据库。 函数（function），是你自己写好的一段代码，模型只会“点名”它。 tool_call_id（工具调用编号），是模型每次点名时附带的一个号码，你回传结果时要带上它，模型才知道你在回答哪一次。 strict 模式（严格模式，Beta 测试版），是让模型点名时严格按你定义的格式填参数。 一个真实例子：官方示例问“杭州天气怎么样”，模型返回的是 get_weather 这个函数和参数 location 等于杭州，你的代码去查到“24℃”，再交给模型，模型才说“杭州现在 24 度”。 先把话说全： 官方示例用的是非思考模式，请求里加 extra_body 把 thinking 的 type 设成 disabled。 官方写明：从 DeepSeek-V3.2 开始，思考模式下也能用工具调用，细节看官方 Thinking Mode 一页。 官方写明：Chat Completion 接口不支持在对话中间插入不是模型生成的工具调用，要插的话请改用 Anthropic API 或 Responses API。 strict 模式目前是 Beta，要用测试版地址。 一、先看懂整个流程 官方给的执行流程是四步：用户问杭州天气；模型返回要调用 get_weather，参数是杭州；你的代码真的去调用 get_weather，把结果交给模型；模型用自然语言回答“杭州现在 24 度”。 二、照官方示例做一遍 第一步：去 DeepSeek 开放平台拿一个 API key，电脑装好 Python 和 openai 库。 第二步：创建客户端，base_url 写 DeepSeek 官方接口地址（官方文档第一屏就有），api_key 填你的。 第三步：写 tools 清单。每个工具包含 type 等于 function，再写 name（函数名）、description（描述，告诉模型什么时候用）、parameters（参数）。官方示例的函数叫 get_weather，描述是“获取某地天气，用户须先提供地点”，参数 location 是字符串，并且在 required 里。 第四步：发第一次请求，model 官方示例写 deepseek-flash，把 messages 和 tools 一起传，并加上 thinking 关闭的设置。 第五步：从返回里取 tool_calls 的第一项，先把模型这条消息原样追加进 messages。 第六步：你的代码去执行函数，再追加一条 role 是 tool 的消息，带上刚才的 tool_call_id，content 写结果，官方示例写的是 24℃。 第七步：再发一次请求，这次模型就会用自然语言回答。 三、想要更稳：strict 严格模式 第八步：base_url 改成官方文档里写明的 beta 版接口地址（就是原地址后面加 /beta）。 第九步：每个 function 都加上 strict 等于 true。 第十步：服务器会检查你的 JSON Schema（结构说明），格式不合规或用了不支持的类型，会直接返回错误信息。 官方支持的类型是：object、string、number、integer、boolean、array、enum、anyOf，另外 $ref 和 $def 可以复用和做递归。 四、strict 的硬规矩 对象里所有属性都必须写进 required，并且 additionalProperties 必须设成 false。 字符串可以用 pattern（正则）和 format，官方列出的格式有 email、hostname、ipv4、ipv6、uuid，但不支持 minLength 和 maxLength。 数字支持 const、default、minimum、maximum、exclusiveMinimum、exclusiveMaximum、multipleOf。 数组不支持 minItems 和 maxItems。 五、看到什么算成功 第一次返回的是函数点名，不是答案；第二次返回的才是自然语言答案，内容里有你的函数结果。 六、卡住怎么办 strict 报错：先检查是不是漏写 required 或忘了 additionalProperties 设 false，再检查有没有用到 minLength、maxItems 这类官方不支持的写法。 结论：工具调用不是模型替你干活，而是模型“点名”、你的代码“动手”。想让点名更准，就用 strict。把这条复制链接发给他，下次他说 AI 查不到东西，先让他确认函数是不是自己写了。 你卡在第几步？评论区告诉我。同系列前几篇在我主页。 收藏版：5 步速查 一、tools 里写清函数名、描述、参数 二、第一次请求拿到 tool_calls 三、自己执行函数，回传 role 为 tool 的消息并带 tool_call_id 四、第二次请求拿到自然语言答案 五、想更稳就用 beta 地址加 strict 等于 true 常见坑 3 条 一、以为模型会自己执行函数，其实要你的代码去跑。 二、回传结果忘了带 tool_call_id。 三、strict 里漏了 required 或 additionalProperties 设成 false。 复制即用提示词：我想让模型调用一个叫 你的函数名 的函数，它需要这些参数：写参数。请帮我写出符合 DeepSeek Tool Calls 格式的 tools 清单，每个参数都写 description，并改成 strict 模式需要的写法。 S1哥观点：这篇最狠的地方，是官方把话挑明了：模型不执行函数，只负责点名。很多人以为 AI 在后台自己动手，其实动手的永远是你的代码。 真正的信号是：工具调用的难点不在模型，在你把函数描述写得多清楚。就像派活给新人，说明写得含糊，他再聪明也会点错人。 你现在想让 DeepSeek 调用的是什么，查数据还是发消息？评论区告诉我。 今天送：DeepSeek 工具调用速查卡，评论区第一条直接拿。 参考：DeepSeek 官方 API 文档 Tool Calls（S1哥 整理）

Fable 5.2 vs. GPT-6 Astra Astra was 6x cheaper, but man, it got absolutely DESTROYED in this test. 💀 Compared to Fable, Astra's result looks like a bad joke. Fable is much more expensive, but the difference in quality is insane. What do you think? Is Fable worth the extra cost?

Agent 最缺的不是更大的上下文窗口，而是理解整个代码库的能力：当一个项目有数万文件、跨模块调用和海量文档时，为什么还要让 AI 一遍遍 Grep、Read、重新理解整个工程？ 📦 项目档案 项目：Graphify-Labs/graphify GitHub Stars：⭐ 147299 技术栈：Python + Tree-sitter + NetworkX + Leiden 核心定位：将代码、文档、SQL Schema、PDF、图片及音视频转换成可查询、可追溯的知识图谱，让 AI Agent 从「搜索文件」升级为「理解工程关系」。 GitHub： 🛠️ 一、底层架构：不是传统 RAG，而是图结构知识引擎 Graphify 最值得研究的地方，是它没有走传统向量检索路线。 传统 RAG 通常需要经历： 文件切分 → Embedding → 向量入库 → 相似度检索 → 上下文拼接。 Graphify 直接构建 Graph。 底层采用 NetworkX 图数据结构，以 Node 和 Edge 表达工程实体及其关系。 整个处理链路可以概括为： detect() → extract() → build() → cluster() → analyze() → report() → export() 每个阶段职责独立，通过 Python Dict 和 NetworkX Graph 传递数据。 这意味着它不只是索引代码，而是在构建整个项目的结构化关系网络。 🛠️ 二、Tree-sitter AST：让代码解析摆脱 LLM 依赖 这是最有工程价值的设计。 Graphify 使用 Tree-sitter AST 解析约40种编程语言，提取： • Classes：类定义及结构。 • Functions：函数与方法。 • Imports：模块依赖关系。 • Call Graph：跨文件函数调用。 • SQL：表、视图、外键及 JOIN 关系。 其中33种语言使用 Tree-sitter Grammar，另外7种使用内置解析器。 关键在于：代码结构提取完全在本地完成，不需要调用 LLM，也不需要把代码上传到云端。 对于大型代码库，这种确定性解析比单纯依赖模型阅读文件更容易复现和追踪。 🛠️ 三、EXTRACTED / INFERRED：让每条知识关系都有证据 很多知识图谱最大的隐患是：模型推断出来的关系，很容易被误认为真实事实。 Graphify 给关系增加了置信度标签： • EXTRACTED：源码中明确存在的关系，例如 import、函数调用。 • INFERRED：通过上下文、命名和调用关系推断出来的连接。 • AMBIGUOUS：存在不确定性的关系，需要人工复核。 INFERRED 关系还会附带 0.0—1.0 的 confidence_score。 这套机制非常重要。 它实际上把知识图谱从一个简单的关联网络，变成了一个能够区分事实、推断与不确定性的工程知识系统。 🛠️ 四、Leiden + SHA256 Cache：解决规模与重复计算 Graphify 使用 Leiden 社区发现算法，根据节点之间的连接密度，将复杂工程拆分成不同功能子系统。 例如，一个大型后端项目，可以自动识别认证、数据库、路由、权限等不同社区。 同时，SHA256 内容哈希缓存会记录文件指纹。 文件没有变化，就跳过重复提取；只有新增或修改的文件才重新处理。 音视频还可以通过本地 faster-whisper 转录，再进入统一知识图谱。 这让增量更新不必每次重新扫描整个工程。 五、真正解决了什么工程问题？ 想象一个场景： 你接手一个陌生的微服务项目，需要搞清楚： 用户认证最终连接了哪些数据库模块？ 修改某个核心函数会影响哪些下游组件？ 某个架构决策为什么会存在？ 两个看似无关的模块究竟有什么隐藏依赖？ 传统方式：Grep、Read、跳转、反复询问 Agent，消耗大量上下文。 Graphify：直接执行 query、path、explain。 它还能生成三个核心产物： • graph.json：完整知识图谱，可供后续查询。 • GRAPH_REPORT.md：核心节点、社区及关键关系分析。 • graph.html：可交互的可视化图谱。 甚至支持 Mermaid 调用链导出、Markdown Wiki 和 Obsidian 知识库。 最重要的设计思想是： 不要让 AI 每次都重新阅读整个世界，而是提前把世界组织成它能够理解和遍历的结构。 这才是 AI Coding 从上下文堆砌走向工程知识管理的重要一步。 对于大型 Monorepo、遗留系统维护、跨模块依赖分析，以及需要长期理解代码库的 Agent，这套架构尤其值得研究。 🔗 GitHub 开源地址：

Ever seen cherries labelled "J," "JJ" or "JJJ" in a Chinese supermarket? Here's what it means. In China, imported sweet cherries are called "che li zi," a transliteration of the English word "cherries." They're sold by size grade, and the number of J's tells you the diameter. Based on common labels on Chinese e-commerce platforms:
 J: about 26 to 28 mm
JJ: about 28 to 30 mm
JJJ: about 30 to 32 mm
4J: about 32 to 34 mm Each extra J adds roughly 2 mm, and can add quite a bit to the price. Right now is Chilean cherry season in China. Chile's summer is China's winter, so the first air shipments arrive in late October, with peak supply in December and January, just in time for New Year and Lunar New Year gifting. China is Chile's biggest cherry customer by far: according to 21st Century Business Herald, China has bought more than 90% of Chile's cherry exports for seven years in a row. Does anywhere near you grade fruit by size like this? 
#cherries# #Chile# #agriculture# In-depth reports at

Extremadamente preocupante. Esto no es casualidad. Están cayendo los sistemas uno a uno y nadie está uniendo los puntos. Esto es solo el principio. 1. Apagón de comunicaciones en los aeropuertos de EE.UU.: la FAA ha paralizado todos los vuelos del país por un “fallo de telecomunicaciones” en el control aéreo. Primera vez a nivel nacional desde 2023. “No hay indicios de acto malicioso”, dicen. Claro. 2. Supermercados en Europa fundiendo la comida: hackers elevando la temperatura de neveras y congeladores a 50 °C en Bélgica, Suecia, Francia, Irlanda y más. Toneladas de alimentos estropeados. “Ciberataques aislados”, aseguran. 3. Bancos de sangre en Francia intentando hervir la sangre: atacaron los sistemas de refrigeración y los configuraron a 110 °C. Detectado a tiempo, según las autoridades. Qué oportuno. 4. Caída global de Microsoft: Outlook, Teams y medio Microsoft 365 caídos en decenas de países. “Problema interno de servidores, sin actividad maliciosa”. Otra vez la misma frase. 5. Banco en Paraguay se lleva los ahorros de casi 3 millones de ciudadanos: colapso de Ueno, cuentas congeladas, presidente dimitido y detenido. “Solvencia” y “préstamos a insiders”. La mitad del país afectada. Y ahora esto. ¿Alguien más ve el patrón o sigues creyendo que todo es “técnico” y “aislado”?