←搜索结果

按匹配度 · 共 1650 条结果 清除搜索

星期六

  1. DevClass

    Go v1.27 发布:泛型支持方法,但可读性争议再起

    Go 语言发布 v1.27,将泛型支持扩展到方法,并允许直接为嵌套或嵌入字段赋值、在切片字面量等场景中自动推断类型参数。泛型自 2022 年 v1.18 引入,此次更新可减少重复代码,但批评者认为抽象层增加会损害 Go 一贯的可读性,对需要审查和维护代码的 AI 智能体尤为不利。

    k July 26, 2026 Zig creator calls Bun's Claude Rust rewrite 'unreviewed slop' July 15, 2026 HTTP gets a QUERY method so complex searches can stop pretending to be POST July 14, 2026 Speedier type checks in TypeScript 7.0 as first stable Go release ships July 10, 2026 GitHub cuts short offer to burn repos on CD after mockery ensues A purported jab of Sony's physical media phase-out blows up on GitHub itself July 7, 2026 .NET's long term support is not long term enough, dev complains June 30, 2026

    匹配字段正文

星期五

  1. Pragmatic Engineer

    The Pulse:我们需要谈谈用 AI 做迁移

    Asana 借助 AI 在两周内完成了测试框架 Enzyme 的迁移,而这项需要重写大量测试用例的工作原本会被拖延数年,Airbnb 和 Uber 也有类似经历。本期 The Pulse 还提到,Gartner 在“AI 代码现代化工具”排名中将 AWS、Microsoft、IBM 排在 Anthropic、Cursor 和 OpenAI 之前,原因可能是前三者向 Gartner 支付了高额费用。

    king. This is most likely because the first three pay large sums of money to Gartner, but AI labs and vendors refuse to pay this “Gartner tax.”; Industry Pulse. Another hours-long GitHub outage, GitHub alternatives are here and fighting for market share, Slack launches Slack Code, text generated by Claude to be watermarked, and Uber open sources SubmitQueue Before we start: apologies for the numerous typos last week. My editor, Dominic, was on vacation, and numerous typos made it through the spe

    匹配字段正文

星期四

  1. Ben's Bites

    Ben's Bites:个人智能体不过是一堆文件和工具

    Ben's Bites 的投票显示,996 名受访者中多数对个人智能体不感兴趣;作者认为个人智能体本质上只是连接到任意智能体工具的一堆文件和工具,文件为会话提供上下文,所谓记忆也只是会话开始时读取的文件。作者将在明天的文章中演示如何为自己搭建一个全新的个人智能体。

    red memory for agents is to create a lore based on your past conversations. Turn your codebase into an isometric city and explore its architecture. Matt Pocock’s skills have 200k+ GitHub stars . Theo walked through the select few he uses . A Mac app for transcribing multi-hour recordings , with a CLI for agents. Long read from Cursor on how they built Origin as if it were a database. Really good visually interactive post. Another code-storage startup nipping at GitHub’s heels. This one I’ve been

    匹配字段正文

  2. LMSYS Blog

    Mooncake 为 Miles 加速:从碎片化 rollout 数据到高效批量 I/O

    LMSYS 博客介绍 Miles 强化学习框架如何借助 Mooncake 优化 rollout 到训练的数据传输。Miles 结合 SGLang 做高吞吐 rollout 生成、Megatron-LM 做训练,并提供 PyTorch FSDP2 后端,支持全异步 RL。由于 rollout 数据是含 tokens、loss_masks、rollout_log_probs 等字段的异构结构,且以 list[np.ndarray] 形式高度碎片化,Mooncake 需在保留结构的同时实现高效批量 I/O。

    Synchronous rollout transfer for Miles

    stmatengss), Xingyuan Wu (@yokinoshitayoki), @fzyzcjy, @guapisolo, Xuchun Shang (@XucSh), Bo Gao (@Bo-Vincent), He Zhou (@ehuohz), and Yufeng He (@he-yufeng) for their contributions across Mooncake implementation and review, Miles integration and validation, and design feedback, review, and testing. Related Links Mooncake project: https://github.com/kvcache-ai/Mooncake; Mooncake documentation: https://kvcache-ai.github.io/Mooncake/; Miles Mooncake rollout transfer user guide: radixark/miles 2535

    匹配字段正文

  3. NVIDIA Developer · Generative AI

    NVIDIA 如何用 CLI、Skills 和 AI 编码智能体开发 Holoscan 应用

    NVIDIA 用 Codex 搭配 GPT-5.6 sol max 模式,通过 Holoscan CLI、HoloHub 开发技能与文档,分三轮迭代构建了内窥镜工具分割仪表盘应用,复用 MONAI 模型与示例视频。延迟优化使渲染吞吐提升 50.5%、应用路径平均延迟降低 33.6%。消融实验显示,CLI、HoloHub 技能与文档三者结合时智能体处理最快(40 分钟)、token 用量最低(11M)。

    and data reuse. Developer prompt 1: Use $holohub-app-lifecycle to create a separate new Python HoloHub application for displaying endoscopic tool tracking as model outputs https://github.com/Project-MONAI/model-zoo/tree/dev/models/endoscopic tool segmentation. Reuse the MONAI endoscopic tool segmentation model, sample data, preprocessing, and inference. Show the model-derived mask, coverage and timeline, and useful uncertainty measurements in a polished HoloViz overlay. Do not train or modify th

    匹配字段正文

  4. NVIDIA Developer · Generative AI

    NVIDIA 教程:后训练 Cosmos 3 Edge 实现 Jetson Thor 端侧机器人控制

    NVIDIA 发布教程,介绍如何后训练 4B 的 Cosmos 3 Edge 全能模型,使其在 NVIDIA Jetson Thor 上原生运行机器人操作策略。在 Jetson AGX Thor T5000 上,该策略约 1.53 秒生成一个动作块,单个动作块覆盖约 2.13 秒机器人运动,因此无需数据中心 GPU 即可连续实时控制。在 RoboLab 闭环评测的 120 项语言条件操作任务中,后训练后的 Edge 策略成功率为 22.9%,Nano 为 36.8%。训练数据为 nvidia/Cosmos3-DROID 数据集,含 76k 条成功遥操作轨迹、约 350 小时、86 项任务和 564 个场景;后训练与部署脚本已在开源 cosmos-framework 仓库提供,检查点发布在 Hugging Face。

    Steps Download the Cosmos Edge policy weights and data to begin post-training; Explore Cosmos Edge on Jetson for deployment guidance and examples; Get the Cosmos framework code on GitHub to access post-training and deployment scripts Robots need policies that can adapt to their sensors, environments, and tasks while running on onboard computing hardware. World models offer a foundation for learning physical interactions, but their size can make on-device deployment difficult. This changes with t

    匹配字段正文

  5. NVIDIA Developer · Generative AI

    NVIDIA 发布 SkillEvaluator,评估 AI 智能体技能表现

    NVIDIA 推出开源工具 SkillEvaluator,通过静态检查和有无技能对照的实机任务运行,衡量技能对智能体表现的影响。首批对 30 多个 NVIDIA 产品线的 300 多个已验证技能在 Codex 和 Claude Code 两个 harness 上评估,Correctness 平均 Skill Lift 为 41 分,Effectiveness 为 39 分;单产品差异从 +2 到 +46 分。工具还单独追踪 token 用量与执行时间,例如 jetson-optimize-memory 使 token 减少 76.9%,cuopt-install 则增加 120.3%。OpenClaw 与 Nous Research 已在 ClawHub 和 Hermes Agent 中试点集成。

    skill scores vary. Catalog-wide averages aggregate thousands of trials, but this post does not report confidence intervals. The same evaluations used skills from the NVIDIA/skills GitHub repository. Table 2 shows the average with-skill scores and Skill Lift relative to the without-skill baseline across Codex and Claude Code. Verified skills improved agent performance across both evaluated harnesses, with the largest gains in Correctness, Discoverability, Effectiveness, and Efficiency. Skill Lift

    匹配字段正文

星期三

  1. The Rundown AI精选

    OpenAI因模型失配暂停前沿模型训练

    OpenAI因内部模型呈现不同程度的失配,暂停未来模型训练两周,并将其计划中最大规模的前沿训练置于搁置状态以接受进一步安全测试。公司将使用自动化调查工具审查模型行为与推理,并重写2023年《Preparedness Framework》;Sam Altman称此举影响更远期的发布,近期仍预计推出新模型。

    atGPT for Teens - OAI's new teen experience with default safety limits 🐦 Berd - Block's open-source desktop hub for running custom AI agents ⚙️ Origin - Cursor's code hosting with GitHub sync and built-in agents Can AI build a production-ready MCP server? CData's independent report has the answer. Read the results here . AI math startup Axiom claimed to have formalized the "246 theorem," the 12-year-old record on gaps between primes, marking another milestone of AI in mathematics. U.S. AI chip s

    匹配字段正文

  2. Armin Ronacher AI 博客

    Armin Ronacher 解释大语言模型的推理机制

    Armin Ronacher 解释称,推理轨迹本质上是模型在最终回答前写入暂存区的文本,并通过特殊 token 和解析器与答案分流。他以 GPT-OSS、Responses API 和 DwarfStar 为例,分析推理强度如何由系统提示词塑造,以及禁用原生推理时模型为何可能把思考内容泄露到错误通道。

    into putting some reasoning where it should not go — but only when native reasoning is disabled. Fun fact: this blog post triggered safey checks Hilariously enough I was unable to use GPT 5.6 terra for spell and grammar checking on this blog post because of safety filters. Had to switch to Kimi. Content licensed under the Creative Commons Attribution-NonCommercial 4.0 International License. Contact me via mail, bluesky, x, or github. You can sponsor me on github. Color scheme: auto, light, dark.

    匹配字段正文

  3. NVIDIA Developer · Generative AI

    AI 编程智能体如何借助 NVIDIA ALCHEMI Toolkit 解锁材料模拟

    NVIDIA ALCHEMI Toolkit 提供 PyTorch 原生的可组合模块,用于构建 GPU 加速的机器学习原子间势(MLIP)模拟,并支持 in-flight batching;配合其 agent skills,编程智能体可从自然语言提示生成并执行模拟工作流。在硅状态方程、Cu(111) 氧吸附和锂自扩散三类共 45 条流水线的基准测试中,提示词具体程度只影响代码结构与可复用性,不改变物理结果;Langevin 与 NVE 系综给出的液态锂扩散系数相差 3-5 倍。智能体默认选用 FIRE 而非 FIRE2,且 7 个 GPU 相关失败只在执行时暴露。

    d not question physically ill-posed tasks and MLIP foundation models require validation against DFT or experiment for new chemistries Next Steps Visit the NVIDIA/nvalchemi-toolkit GitHub repo to access agent skills and over 30 example workflows; Read the NVIDIA ALCHEMI Toolkit documentation for installation guides and agent configuration details Atomistic simulation requires three things: knowledge of the science, compute-efficient implementation of simulations, and accessible interfaces to the

    匹配字段正文

星期二

  1. Ed Zitron 长文 / Where’s Your Ed At

    OpenAI 若倒下,会发生什么?

    OpenAI 若无法持续筹得巨额资金以支付算力承诺,潜在崩溃将冲击云厂商、基础设施供应商及持股投资者。文中称其 2025 年收入为 $13.07 billion、亏损 $20.9 billion,并估算其到 2030 年需承担 $800 billion 以上算力义务;文章据此分析了 Microsoft 接手、与 Anthropic 合并或资金链断裂等可能情形。

    rely eliminate) the free ChatGPT product, and likely either tighten rate limits or move everybody on a ChatGPT Plus or Pro subscription to token-based billing, much as it did with GitHub Copilot in June. While I imagine some rescue package is pulled together, OpenAI could simply be allowed to run out of money, short-changing nearly a trillion dollars’ worth of compute contracts, killing CoreWeave, Cerebras, and anyone else reliant on its income. Its customers would be given API keys that flow to

    匹配字段正文

  2. Ben's Bites

    你使用个人智能体吗?——Ben's Bites 读者调查

    Ben's Bites 发起读者调查,询问谁在使用“个人智能体”产品,即在工作之外帮忙整理事务、处理邮件、在电脑上执行操作等,答案将在周四的文章中公布。本期还提到 Codex/ChatGPT 新增“计算机历史”功能,可将桌面应用和网站活动转为记忆与时间线,需在设置中选择开启;Claude Code 新增 /design 技能,可生成可编辑的 UI 画板并直接实现。

    e Code 就能将其实现。 达到使用上限了?设置自动继续,以便在计时器重置时恢复。 分享 Ben's Bites 既然 Airtable 已被收购,Google Sheets 正踩着它的坟墓起舞,推出了 Sheets Canvas——在 Google Sheets 内部运行、与你的数据同步的应用。 一个小巧的 Mac 工具,用于安排枯燥任务和无头智能体。 GitHub 昨天(又)宕机了,而 Cursor 推出了 Origin——他们的竞品。时机绝佳——我甚至有点觉得他们早就准备好了,就等 GitHub 宕机,因为最近这种情况发生得太频繁了。 Theo Browne 拆解了他的 AGENTS.md 和 SKILLS.md 文件。这提醒了我,我也应该(再次)做这件事。 Factory 开放了 Guild,一个高度精选的构建者计划。 一个用于构建你自己的软件工厂的模板。 为什么模型路由必须放在 harness 中。 一组极简的动画图表组件。 关于在软件变得廉价时打造人们喜爱的产品的笔记。 拥有你的智能:一份操作指南。 现在 harness 也需要一个 harness 了? 填写表单应该更智

    匹配字段正文

星期六

  1. Augment Code 博客

    Augment Code 重建 Auggie CLI 架构,任务成本比 Claude Code 低 53%

    Augment Code 宣布重建 Auggie CLI 的 harness,在 SWE-bench Pro 上相同通过率下,Auggie v2 完成任务成本为 $1.27,比 Claude Code 的 $2.70 低 53%。重建基于开源项目 Pi,通过精简工具面、用检索代替探索和主动压缩上下文实现效率提升,CLI 将于 9 月或 10 月推出。

    How we rebuilt the Auggie CLI harness to make tasks 53% cheaper than Claude Code

    product initiatives across Augment to make building software better for every developer. Chris has been making developers happier and more productive for 15 years at innovative companies like New Relic, GitHub, Salesforce, and FireHydrant. You can find him at @amateurhuman everywhere on the internet. Related reading Kiro CLI vs. Auggie CLI: Spec-Driven Terminal Agents Compared; Using the Auggie CLI for automated code review; Opus 4.7 for 33% less: How Auggie beats Claude Code on cost and quality

    匹配字段正文

星期五

  1. 蚂蚁百灵:Developer Blog

    蚂蚁百灵用 Ling-3.0-tiny 与 AReno 打通单机 Agentic RL 闭环

    蚂蚁百灵发布实践文章,介绍在 DGX Spark 上用本地后训练工具包 AReno 对 Ling-3.0-tiny 做 Agentic RL,以井字棋为最小验证任务跑通基线评测、Reward 定义、后训练与同环境复测四步闭环。实验使用 GSPO 训练 400 个 step,rollout/rewards_mean 从约 -0.5 升到 0.4 左右,response_len 降至约 850 tokens,复测中工具调用与落子选择更稳定。Ling-3.0-tiny 为 7.9B 总参数、1.3B 激活参数的混合推理模型,已开源 BF16、FP8 和 INT4 版本,AReno 提供 Docker 复现教程。

    图

    entic RL、开发者实践和高校教学的案例,帮助开发者从更小、更直观的任务出发,逐步把后训练方法迁移到自己的真实场景中。 07 欢迎下载使用 Tiny,也欢迎参与贡献 AReno Ling-3.0-tiny 已正式开源,提供 BF16、FP8 和 INT4 版本。欢迎根据自己的设备与任务需求下载使用: 🤗 Hugging Face 也欢迎访问 AReno GitHub 仓库(https://github.com/inclusionAI/AReno),复现这次 Spark 后训练实验,提交 Issue、参与讨论,或者贡献 Tiny 个性化训练 Case。 如果你对以下方向感兴趣,欢迎参与 AReno 开源社区: 模型进入本地只是第一步,能否根据业务反馈持续适配,才决定它能否真正进入流程。 Ling-3.0-tiny × ASystem AReno 此次验证的,不是一个已经完成的行业解决方案,而是一条轻量、透明、可复现的任务适配路径:从跑起来,到训起来,再到回到任务中验证是否真的变好。 我们希望开放的不只是模型权重和训练工具,更是一种可被验证、批评、改造和继续创造的方法。 Open in

    匹配字段正文

  2. Microsoft AI Blog

    微软发布 MAI-Cyber-1-Flash 网络安全模型并集成进 MDASH

    微软发布首款网络安全模型 MAI-Cyber-1-Flash,并集成进多智能体漏洞识别与修复框架 MDASH。该模型源自 MAI-Thinking-1 系列,可高效处理约 90% 的任务,其余 10% 的极难任务交给 GPT-5.4,组合方案在 CyberGym 的 any-crash 指标上达到 96%,相比现有 GPT-5.4 + 5.4 mini + 5.3 codex 方案节省 50% 成本。微软同时推出智能体安全系统 Perception,用于持续监控、修补和关闭新的威胁向量。

    Abstract illustration of three overlapping shield shapes in blue, pink, and purple tones on a beige background.

    xplorers, researchers, and full-stack engineers. We move fast, sweat the details, and operate at frontier scale with a roadmap to build the world’s most powerful AI models. Most importantly, we’re united by the belief that doing this right is the only way to do it at all. If our mission resonates with you, we’d love to talk. Related Stories Hill-climbing MAI models for GitHub Copilot and Excel Introducing MAI-Image-2.5-Pro and MAI-Voice-2-Flash Bringing Ode Poetry to life with MAI’s audio models

    匹配字段正文

  3. Hugging Face

    使用 Strands Agents、LeRobot 与 Hugging Face Storage Buckets 在一处完成录制、训练和部署

    Strands Robots 将 LeRobotDataset 的录制、同步、流式训练和实体机器人部署整合到同一 Robot() 工作流,并让新演示数据返回同一 Storage Bucket。Xet 按字节去重,仅上传变化的数据块;stream_dataset() 可从 Hub 逐帧读取远程 Parquet 与 MP4 分片,无需完整下载。Bucket 仅支持流式读取且不保留修订记录,真实部署还需隔离数据与凭据、限制智能体权限并只加载可信组织的模型。

    PUs without waiting for a download. The data never leaves the LeRobot format, and it never leaves the Hub. Try it using the sample application The full Strands Robots sample is on GitHub at strands-labs/robots in examples/notebooks/05 streaming data loop.ipynb . It walks you through the full loop cell by cell: record, render, sync to a bucket, stream back, train, and load the checkpoint. Every cell runs in simulation on the mock policy, so no GPU, no Docker, and no Hugging Face credentials are n

    匹配字段正文

星期四

  1. Ben's Bites

    Grok Bot 上手体验:比 Codex 和 Claude 更简单的个人智能体

    Ben's Bites 作者在获得 Grok Bot 早期访问后实际使用,认为它比 Codex 或 Claude 更简单,体验接近在 Slack 里给队友发消息。用户可以为每个智能体设定系统提示词、让它们互相发消息、观看它们使用各自的虚拟电脑,并连接 Gmail、Calendar、Slack 等账号,作者在连接账号时遇到几次问题。目前只有 Cursor 或 Grok 的 200 美元套餐用户可以使用,Grok Bot 由新发布的 Grok 4.6 驱动,作者称 Grok 4.7 将在几周后推出。

    hboard showing your spending It's hard to know how much you actually spend over lots of accounts so this 3:52 PM · Aug 11, 2026 · 159K Views 110 Replies · 39 Reposts · 1.29K Likes github.com/jerber/arc-code ","username":"jeremyberman","name":"Jeremy Berman","profile image url":"https://pbs.substack.com/profile images/2053637963943215104/yxlbbPOI normal.jpg","date":"2026-08-12T20:11:55.000Z","photos":[{"img url":"https://pbs.substack.com/media/HPi3PWFa8AAtRwC.jpg","link url":"https://t.co/531wS8s

    匹配字段正文

  2. LiveKit 产品与工程博客

    LiveKit Agents 推出 expressive mode,让语音智能体按情境调整语气

    LiveKit 在 LiveKit Agents 中发布 expressive mode,开启后框架会向 LLM 提示词注入供应商专属指令,让模型在回复中内联生成表情标记,再由框架降级为当前 TTS 供应商能识别的方言,从而表达情绪、语气、非语言声音和停顿。该功能只需在 AgentSession 上设置一个参数,兼容任意 LLM,TTS 侧目前支持 Fish Audio(fishaudio/s2.1-pro)、Inworld(inworld/inworld-tts-2)、Cartesia(cartesia/sonic-3)和 SpaceXAI(xai/tts-1),未声明标记方言的供应商会正常合成且该开关不生效。框架还会把标记从用户可见转写中剥离,并将该段落的语气标签以 lk.expression 属性发布在 lk.transcription 文本流上,供前端可视化使用。

    expression signals for frontends. Try it today The fastest way to hear the difference is the demo above. Then enable it on your own agent with one line. The full demo agent is on GitHub and the expressive mode docs cover the rest. Voice AI that gets the words right but the feeling wrong still loses users. Expressive mode is our first step toward agents that respond to how you're doing, not just what you said. Give it a try! We'd love to hear what you build and any feedback you have. Related Tes

    匹配字段正文

  3. The Register

    OpenWALDO 启动开源 AI 训练数据集项目,目前含 1673 亿 token

    CentOS 和 Rocky Linux 创始人 Gregory Kurtzer 发起 OpenWALDO(Open Weights, Artifacts, Licenses, Data, Origins)项目,由 CIQ 资助,目标是建立可公开贡献、可审计的开源 AI 训练数据集。目前该数据集包含 167.3 亿参考 token,来源包括政府记录、开源学术论文、邮件列表和公共领域文献,但相比前沿模型使用的数十万亿 token 仍微不足道。项目方称企业可将该语料及其物料清单作为经验证的基线,再叠加自有专有数据训练模型。

    el on the OpenWALDO set yet, but CIQ didn’t respond. Those interested in contributing to, or making use of, OpenWALDO can find more on the project’s website (linked above) and its GitHub page. ® Fire up that iPhone 1 emulation and party like it's 2007 Browser-based fun to remember a time before Apple's annual cash grab became a thing New motherboards, old RAM, plenty of life left in yesterday's PCs DDR4 makes a comeback as component costs rise, while Linux offers an escape from forced upgrades T

    匹配字段正文

  4. Microsoft AI Blog精选

    微软发布推理模型 MAI-Thinking-1,已在 Microsoft Foundry 开启公开预览

    微软 AI 发布推理模型 MAI-Thinking-1,已在 Microsoft Foundry 开启公开预览。该模型为 35B 激活参数、约 1T 总参数的稀疏 MoE 架构,在 SWE-Bench Pro 上与 Claude Opus 4.6 持平,AIME 2025 得分 97.0%、AIME 2026 得分 94.5%,并在与 Claude Sonnet 4.6 的盲测人评中被更多用户偏好。模型支持 256k token 上下文窗口、函数调用和开发者指令,兼容 Chat Completions API,微软称其训练未使用第三方模型蒸馏,数据为可追溯的企业级数据。

    Line graph titled "AIME 2025" shows a general upward trend in the y-axis values (ranging from 0.2 to 1.0) over increasing x-axis steps, with fluctuations and small vertical error bars.

    xtensively. And we have an ambitious mission we truly believe in. We’re also fortunate to partner with incredible product teams giving our models the chance to reach billions of users and create immense positive impact. If you’re a brilliant, highly-ambitious and low ego individual, you’ll fit right in—come and join us as we work on our next generation of models! Related Stories Building a hill-climbing machine: Launching seven new MAI models Hill-climbing MAI models for GitHub Copilot and Excel

    匹配字段正文