{
  "title": "2026-09-19 AI日更 | 当 AI 开始读完整个工作区：编程助手的下一关是透明与规范",
  "url": "https://miaok.ong/ai-daily/ai-daily-2026-09-19/",
  "date": "2026-09-19T07:00:00+08:00",
  "lastmod": "2026-09-19T07:00:00+08:00",
  "type": "ai-daily",
  "kind": "page",
  "language": "zh",
  "description": "今天的主线落在 AI 编程工具治理：ZCode 被指上传完整工作区，引发代码隐私与知情同意讨论；Claude Code 支持 AGENTS.md，则显示团队级 AI 协作正走向标准化。同时，本地优先智能体与模型可信度研究继续升温，行业关注点从“能不能用”转向“能否被信任、被管理”。",
  "keywords": null,
  "tags": [],
  "categories": [],
  "author": "孔淼",
  "image": "https://miaok.ong/images/avatar.jpg",
  "content": "\u003ch1 id=\"2026-09-19-ai日更--当-ai-开始读完整个工作区编程助手的下一关是透明与规范\"\u003e\n  2026-09-19 AI日更 | 当 AI 开始读完整个工作区：编程助手的下一关是透明与规范\n  \u003ca class=\"heading-link\" href=\"#2026-09-19-ai%e6%97%a5%e6%9b%b4--%e5%bd%93-ai-%e5%bc%80%e5%a7%8b%e8%af%bb%e5%ae%8c%e6%95%b4%e4%b8%aa%e5%b7%a5%e4%bd%9c%e5%8c%ba%e7%bc%96%e7%a8%8b%e5%8a%a9%e6%89%8b%e7%9a%84%e4%b8%8b%e4%b8%80%e5%85%b3%e6%98%af%e9%80%8f%e6%98%8e%e4%b8%8e%e8%a7%84%e8%8c%83\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h1\u003e\n\u003cblockquote\u003e\n\u003cp\u003e今天的主线落在 AI 编程工具治理：ZCode 被指上传完整工作区，引发代码隐私与知情同意讨论；Claude Code 支持 AGENTS.md，则显示团队级 AI 协作正走向标准化。同时，本地优先智能体与模型可信度研究继续升温，行业关注点从“能不能用”转向“能否被信任、被管理”。\u003c/p\u003e\n\u003c/blockquote\u003e\n\u003ch2 id=\"-本期-watch-list-深度导读\"\u003e\n  📖 本期 Watch List 深度导读\n  \u003ca class=\"heading-link\" href=\"#-%e6%9c%ac%e6%9c%9f-watch-list-%e6%b7%b1%e5%ba%a6%e5%af%bc%e8%af%bb\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h2\u003e\n\u003cp\u003e今天的 Watch List 可以分成三条线看。\u003c/p\u003e\n\u003cp\u003e首先是产业与资本叙事：Stratechery 的周报与《The State of Startups in 2026》适合放在一起读，前者看大平台与商业格局，后者观察创业公司在本地 AI 助手、开源与分发侧的新机会。\u003c/p\u003e\n\u003cp\u003e第二条是 AI 可信度与用户体验。App Store 生成式 AI 评论分析、FakeSpotter、隐性冲突检测等论文，都在回答同一个问题：模型不只是“能用”，还要能被信任、被理解，并减少误导。\u003c/p\u003e\n\u003cp\u003e第三条是模型内部机制与评测深化。潜意识提示、风格轴发现、自我反思式推理、跨学科科学推理与古典诗词评测，显示研究重心正从单纯刷分，转向解释模型如何表达、推理与犯错。\u003c/p\u003e\n\u003ch2 id=\"-x-平台-ai-热点快讯\"\u003e\n  🌐 X 平台 AI 热点快讯\n  \u003ca class=\"heading-link\" href=\"#-x-%e5%b9%b3%e5%8f%b0-ai-%e7%83%ad%e7%82%b9%e5%bf%ab%e8%ae%af\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h2\u003e\n\u003ch3 id=\"话题-1zcode-ai-app-uploaded-full-workspaces-to-alibaba-cloud-without-clear-disclosure\"\u003e\n  话题 1:ZCode AI App Uploaded Full Workspaces to Alibaba Cloud Without Clear Disclosure\n  \u003ca class=\"heading-link\" href=\"#%e8%af%9d%e9%a2%98-1zcode-ai-app-uploaded-full-workspaces-to-alibaba-cloud-without-clear-disclosure\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e分类:AI · News\u003c/li\u003e\n\u003cli\u003e概况:热度时间:15 hours ago,相关帖子数:7200\u003c/li\u003e\n\u003cli\u003e是什么事:有消息称，ZCode AI 应用在未作清晰披露的情况下，将用户的完整工作区上传至阿里云。\u003c/li\u003e\n\u003cli\u003e为什么重要:该事件涉及 AI 编程工具的数据收集范围、代码隐私、云端存储与用户知情同意，可能影响开发者对 AI 工具安全性和合规性的信任。\u003c/li\u003e\n\u003cli\u003e讨论概况:X 上的讨论主要聚焦于应用是否确实上传完整工作区、隐私政策与实际行为是否一致、用户是否获得充分告知，以及将数据存储在阿里云带来的安全与合规风险；也有人认为需等待官方回应和技术证据，避免仅凭标题下结论。\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"话题-2claude-code-now-supports-agentsmd-files-for-better-team-workflows\"\u003e\n  话题 2:Claude Code Now Supports AGENTS.md Files for Better Team Workflows\n  \u003ca class=\"heading-link\" href=\"#%e8%af%9d%e9%a2%98-2claude-code-now-supports-agentsmd-files-for-better-team-workflows\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e分类:AI · News\u003c/li\u003e\n\u003cli\u003e概况:热度时间:4 hours ago,相关帖子数:6300\u003c/li\u003e\n\u003cli\u003e是什么事:Anthropic 的 Claude Code 现已支持 AGENTS.md 文件，方便团队为 AI 编程代理统一配置项目说明、规范和工作流程。\u003c/li\u003e\n\u003cli\u003e为什么重要:这表明 AI 编程工具正从个人辅助走向团队协作场景，强调可复用的上下文、标准化指令和工程流程集成，对提升代码生成质量和组织级采用具有重要意义。\u003c/li\u003e\n\u003cli\u003e讨论概况:X 上讨论主要集中在 AGENTS.md 是否会成为 AI 编程代理的通用约定、它相比现有 README 或提示词配置的价值，以及团队在权限、代码质量和流程治理上的实际收益与潜在复杂性。\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch4 id=\"今日-x-上的-ai-舆情小结\"\u003e\n  今日 X 上的 AI 舆情小结\n  \u003ca class=\"heading-link\" href=\"#%e4%bb%8a%e6%97%a5-x-%e4%b8%8a%e7%9a%84-ai-%e8%88%86%e6%83%85%e5%b0%8f%e7%bb%93\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h4\u003e\n\u003cp\u003e今天的舆论主线集中在 AI 编程工具从“个人效率助手”走向“团队级基础设施”过程中，数据治理与工程规范必须同步完善。一方面，ZCode AI 被指在未充分披露的情况下上传完整工作区，引发了对代码隐私、用户知情同意、云端存储及合规性的广泛担忧，但是否确实存在这一行为仍需官方回应和技术证据确认；另一方面，Claude Code 支持 AGENTS.md 被视为推动团队统一 AI 编程规范、复用上下文和融入工程流程的重要尝试。总体共识是，AI 编程工具需要更透明的数据处理机制和更标准化的协作配置，分歧则在于具体产品是否越过隐私边界，以及新配置文件能否真正带来稳定的质量与治理收益。潜在风险包括隐私事件削弱开发者信任、团队指令配置造成权限或代码质量问题，以及行业标准尚未统一导致治理复杂度上升。\u003c/p\u003e\n\u003ch2 id=\"-大佬观点influencer-insights\"\u003e\n  💡 大佬观点(Influencer Insights)\n  \u003ca class=\"heading-link\" href=\"#-%e5%a4%a7%e4%bd%ac%e8%a7%82%e7%82%b9influencer-insights\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h2\u003e\n\u003cblockquote\u003e\n\u003cp\u003e今日大佬观点暂缺,推荐阅读 Watch List 深度内容。\u003c/p\u003e\n\u003c/blockquote\u003e\n\u003ch2 id=\"-附录今日-watch-list-更新源列表\"\u003e\n  📚 附录:今日 Watch List 更新源列表\n  \u003ca class=\"heading-link\" href=\"#-%e9%99%84%e5%bd%95%e4%bb%8a%e6%97%a5-watch-list-%e6%9b%b4%e6%96%b0%e6%ba%90%e5%88%97%e8%a1%a8\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h2\u003e\n\u003cblockquote\u003e\n\u003cp\u003e时间窗口:最近 3 天;覆盖 22 个源;共 34 条更新\u003c/p\u003e\n\u003c/blockquote\u003e\n\u003ch3 id=\"y-combinator-podcast-b_introsearch\"\u003e\n  Y Combinator Podcast (B_intro+search)\n  \u003ca class=\"heading-link\" href=\"#y-combinator-podcast-b_introsearch\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e\u003ca href=\"https://podcasters.spotify.com/pod/show/ycombinator/episodes/The-State-of-Startups-in-2026-e3p1rf0\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eThe State of Startups in 2026\u003c/a\u003e\u003c/strong\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 22:59 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- You’ve probably already heard all about OpenClaw (formerly Clawdbot/Moltbot).\n\u003cul\u003e\n\u003cli\u003eThe viral sensation is an open-source AI assistant that runs on your own device, connects with messaging apps you already use, and goes beyond chat to actually execute tasks like managing your email, calendars, files, workflows, and more.\u003c/li\u003e\n\u003cli\u003eNow meet the man behind it.\u003c/li\u003e\n\u003cli\u003eYC’s Raphael Schaad sat down with Peter Steinberger, the creator of OpenClaw, to discuss the “aha” moment behind the viral personal AI agent, why local-first agents could replace many of today’s apps, and how personal agents will reshape the future of software.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003eYC works with thousands of founders every year, which gives us an early look at how startups are changing\u003c/li\u003e\n\u003cli\u003eRight now, the shift is striking: startups are moving from bits to atoms, nearly one in five YC companies has a solo founder, and companies are reaching meaning…\u003c/li\u003e\n\u003cli\u003eThey discuss how AI is making it possible for smaller teams to take on more ambitious problems, why experienced founders are having a resurgence, and why knowin…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"all-in-podcast-a_full\"\u003e\n  All-In Podcast (A_full)\n  \u003ca class=\"heading-link\" href=\"#all-in-podcast-a_full\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://allinchamathjason.libsyn.com/jared-isaacman-a-new-era-for-nasa-and-american-space-exploration\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eJared Isaacman: A New Era for NASA and American Space Exploration\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 23:59 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- IREN is a vertically integrated AI Cloud platform, delivering data centers, compute and software for AI training and inference.\n\u003cul\u003e\n\u003cli\u003eOracle connects the data, applications, and infrastructure that turn AI into business outcomes—with the flexibility, choice, and control to optimize as AI evolves.\u003c/li\u003e\n\u003cli\u003eEY helps tech innovators scale from startup to exit to megacap.\u003c/li\u003e\n\u003cli\u003eWe\u0026rsquo;re focused on giving every person the tools to reach their full potential and making sure the benefits of technology are distributed to all.\u003c/li\u003e\n\u003cli\u003eKeel Infrastructure owns the power, land, and connectivity that HPC and AI run on - backed by secured energy assets and established grid interconnections across North America.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003e(0:00) NASA Administrator Jared Isaacman lays out the state and future of NASA\u003c/li\u003e\n\u003cli\u003e(17:59) Moon base, creating new space industries, NASA\u0026rsquo;s next chapter\u003c/li\u003e\n\u003cli\u003e(23:29) How to make Mars realistic\u003c/li\u003e\n\u003cli\u003e(25:30) Recruiting, moonshots, the \u0026ldquo;Nuclear NASA,\u0026rdquo; new vehicles and the importance of humans in space\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://allinchamathjason.libsyn.com/nick-shirley-exposing-government-fraud-suing-california-taking-on-the-media\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eNick Shirley: Exposing Government Fraud, Suing California \u0026amp; Taking on the Media\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 08:30 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- IREN is a vertically integrated AI Cloud platform, delivering data centers, compute and software for AI training and inference.\n\u003cul\u003e\n\u003cli\u003eOracle connects the data, applications, and infrastructure that turn AI into business outcomes—with the flexibility, choice, and control to optimize as AI evolves.\u003c/li\u003e\n\u003cli\u003eEY helps tech innovators scale from startup to exit to megacap.\u003c/li\u003e\n\u003cli\u003eWe\u0026rsquo;re focused on giving every person the tools to reach their full potential and making sure the benefits of technology are distributed to all.\u003c/li\u003e\n\u003cli\u003eKeel Infrastructure owns the power, land, and connectivity that HPC and AI run on - backed by secured energy assets and established grid interconnections across North America.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003e(0:00) Nick Shirley debuts his new video: \u0026ldquo;Inside California\u0026rsquo;s High Speed Rail to Nowhere\u0026rdquo;\u003c/li\u003e\n\u003cli\u003e(11:42) Nick joins The Besties to discuss California\u0026rsquo;s \u0026ldquo;Stop Nick Shirley Act\u0026rdquo; and recent fraud arrests\u003c/li\u003e\n\u003cli\u003e(17:03) How he got into investigative journalism and finding fraud\u003c/li\u003e\n\u003cli\u003e(23:32) Selective journalism and why Nick is a major threat to mainstream outlets\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"stratechery-by-ben-thompson-a_full\"\u003e\n  Stratechery by Ben Thompson (A_full)\n  \u003ca class=\"heading-link\" href=\"#stratechery-by-ben-thompson-a_full\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e\u003ca href=\"https://stratechery.com/2026/doomforce/\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003e2026.38: Doomforce\u003c/a\u003e\u003c/strong\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-19 01:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- (Photo by Benjamin Fanjoy/Getty Images).\n\u003cul\u003e\n\u003cli\u003eWelcome back to This Week in Stratechery!\u003c/li\u003e\n\u003cli\u003eAs a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone.\u003c/li\u003e\n\u003cli\u003eAdditionally, you have complete control over what we send to you.\u003c/li\u003e\n\u003cli\u003eOn that note, here were a few of our favorites this week.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003e(Photo by Benjamin Fanjoy/Getty Images)\u003c/li\u003e\n\u003cli\u003eWelcome back to This Week in Stratechery\u003c/li\u003e\n\u003cli\u003eAs a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone\u003c/li\u003e\n\u003cli\u003eAdditionally, you have complete control over what we send to you\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"arxiv-csai-b_introsearch\"\u003e\n  ArXiv cs.AI (B_intro+search)\n  \u003ca class=\"heading-link\" href=\"#arxiv-csai-b_introsearch\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19170\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eRegularized Emphatic Temporal-Difference Learning: Stability under Constant Stepsizes\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19170v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Emphatic temporal-difference learning (ETD) stabilizes the expected off-policy TD update and changes its projection geometry, but neither property determines constant-stepsize sampled dynamics.\u003c/li\u003e\n\u003cli\u003eWe construct an ergodic two-state counterexample in which the ETD mean map contracts while the sampled product has a positive top Lyapunov exponent.\u003c/li\u003e\n\u003cli\u003eRegenerative-cycle analysis separates this sign from the infinite variance of the follow-on trace.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19170v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Emphatic temporal-difference learning (ETD) stabilizes the expected off-policy TD update and changes its projection geometry, but neither property det…\u003c/li\u003e\n\u003cli\u003eWe construct an ergodic two-state counterexample in which the ETD mean map contracts while the sampled product has a positive top Lyapunov exponent\u003c/li\u003e\n\u003cli\u003eRegenerative-cycle analysis separates this sign from the infinite variance of the follow-on trace\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19180\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eBioPhys-Bridge: A Benchmark for Interdisciplinary Scientific Reasoning in Physics-Grounded Biological Research\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19180v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Language models face unique challenges in analyzing interdisciplinary scientific research literature.\u003c/li\u003e\n\u003cli\u003eIn biophysics research, faithful answers require grounding observed data in source evidence, interpreting it through a quantitative physics model, and linking it to a biological mechanism.\u003c/li\u003e\n\u003cli\u003eTo address this challenge, we introduce BioPhys-Bridge, a novel benchmark dataset for evidence-grounded scientific reasoning over biophysical literature.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19180v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Language models face unique challenges in analyzing interdisciplinary scientific research literature\u003c/li\u003e\n\u003cli\u003eIn biophysics research, faithful answers require grounding observed data in source evidence, interpreting it through a quantitative physics model, and linking i…\u003c/li\u003e\n\u003cli\u003eTo address this challenge, we introduce BioPhys-Bridge, a novel benchmark dataset for evidence-grounded scientific reasoning over biophysical literature\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19182\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eWhat Do We Expect from LLMs? Mapping the Design of LLM Benchmarks\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19182v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Benchmarks are central to how progress in large language models (LLMs) is assessed and communicated.\u003c/li\u003e\n\u003cli\u003eYet model rankings alone reveal little about how evaluation requirements themselves are changing.\u003c/li\u003e\n\u003cli\u003eThe expanding variety of benchmarks offers another perspective: what researchers expect LLMs to do, and what they count as successful performance.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19182v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Benchmarks are central to how progress in large language models (LLMs) is assessed and communicated\u003c/li\u003e\n\u003cli\u003eYet model rankings alone reveal little about how evaluation requirements themselves are changing\u003c/li\u003e\n\u003cli\u003eThe expanding variety of benchmarks offers another perspective: what researchers expect LLMs to do, and what they count as successful performance\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19203\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003ePosition: It is Time to Virtualize Foundation Models with a Self-evolving Operating System Layer\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19203v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: AI applications have shifted from single, monolithic foundation models (FM) to compound agentic systems.\u003c/li\u003e\n\u003cli\u003eYet today\u0026rsquo;s stacks remain fragmented: even as protocols (e.g., MCP, A2A) ease tool/agent connectivity, each framework embeds an implicit runtime for state, memory, budgets, and guardrails, making behavior non-portable and governance brittle.\u003c/li\u003e\n\u003cli\u003eIt mirrors computing before operating systems, when every program re-implemented basic services.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19203v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: AI applications have shifted from single, monolithic foundation models (FM) to compound agentic systems\u003c/li\u003e\n\u003cli\u003eYet today\u0026rsquo;s stacks remain fragmented: even as protocols (e.g., MCP, A2A) ease tool/agent connectivity, each framework embeds an implicit runtime for state, memo…\u003c/li\u003e\n\u003cli\u003eIt mirrors computing before operating systems, when every program re-implemented basic services\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19212\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eWhat Do Current Systematic Generalization Tasks Miss? A Reasoning-Centered Analysis\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19212v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Systematic generalization, the ability to solve novel problems by recombining known atomic elements, is central to human intelligence but difficult to study rigorously under controlled settings.\u003c/li\u003e\n\u003cli\u003eExisting studies therefore rely on simplifications such as approximately linear action composition, productivity-based tests, and action-explicit goals, which make systematic generalization easier to study but omit some essential aspects of this capability.\u003c/li\u003e\n\u003cli\u003eTo characterize what these simplifications miss, we adopt a reasoning-centered lens and introduce TranSGrid, a testbed that brings deductive, inductive, and abductive reasoning together within a unified task.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19212v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Systematic generalization, the ability to solve novel problems by recombining known atomic elements, is central to human intelligence but difficult to…\u003c/li\u003e\n\u003cli\u003eExisting studies therefore rely on simplifications such as approximately linear action composition, productivity-based tests, and action-explicit goals, which m…\u003c/li\u003e\n\u003cli\u003eTo characterize what these simplifications miss, we adopt a reasoning-centered lens and introduce TranSGrid, a testbed that brings deductive, inductive, and abd…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19244\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eCharacterizing Web Search by Conversational LLM Agents: From Search Decisions and Strategies to Results and Responses\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19244v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Conversational LLM agents increasingly rely on Web search, yet the end-to-end lifecycle of agentic search remains poorly understood.\u003c/li\u003e\n\u003cli\u003eWe present the first study of Web search across four major conversational platforms (ChatGPT, Claude, Grok, and DeepSeek), combining real-world user interactions (invivo) with controlled experiments using the same platform\u0026rsquo;s models by their APIs (invitro).\u003c/li\u003e\n\u003cli\u003eWe investigate the quality of agentic decisions to invoke Web search, their strategies to formulate queries, the potential domain preferences in the search results they receive, and the choices they make when transforming search results into grounded responses.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19244v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Conversational LLM agents increasingly rely on Web search, yet the end-to-end lifecycle of agentic search remains poorly understood\u003c/li\u003e\n\u003cli\u003eWe present the first study of Web search across four major conversational platforms (ChatGPT, Claude, Grok, and DeepSeek), combining real-world user interaction…\u003c/li\u003e\n\u003cli\u003eWe investigate the quality of agentic decisions to invoke Web search, their strategies to formulate queries, the potential domain preferences in the search resu…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19387\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eDo AI Agents Understand Computer Architecture?\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19387v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Agents are increasingly asked to design hardware, and increasingly reported to succeed.\u003c/li\u003e\n\u003cli\u003eSuch reports establish that a design improved; they cannot establish why.\u003c/li\u003e\n\u003cli\u003eAn agent that improves an accelerator may be reasoning about the machine, or may be searching competently over knobs whose meaning it never recovers \u0026ndash; and only the first transfers to the next architecture.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19387v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Agents are increasingly asked to design hardware, and increasingly reported to succeed\u003c/li\u003e\n\u003cli\u003eSuch reports establish that a design improved; they cannot establish why\u003c/li\u003e\n\u003cli\u003eAn agent that improves an accelerator may be reasoning about the machine, or may be searching competently over knobs whose meaning it never recovers \u0026ndash; and only…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19391\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eMAGS: Multi-agent Auto-formalization Guarantees Safety for Agentic Outputs\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19391v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: LLM coding agents now generate complex programs at a scale that makes thorough human review increasingly difficult, raising the risk of safety and security failures.\u003c/li\u003e\n\u003cli\u003eCommon approaches, including fuzz testing, static analysis, and LLM-as-a-Verifier, can detect many failures but struggle to cover all possible edge cases.\u003c/li\u003e\n\u003cli\u003eFormal verification addresses this by providing machine-checkable guarantees over specified properties, but traditionally demands substantial manual specification and proof engineering.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19391v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: LLM coding agents now generate complex programs at a scale that makes thorough human review increasingly difficult, raising the risk of safety and sec…\u003c/li\u003e\n\u003cli\u003eCommon approaches, including fuzz testing, static analysis, and LLM-as-a-Verifier, can detect many failures but struggle to cover all possible edge cases\u003c/li\u003e\n\u003cli\u003eFormal verification addresses this by providing machine-checkable guarantees over specified properties, but traditionally demands substantial manual specificati…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19425\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eClosed-World Resolution Against Tool Hallucination in LLM Agents\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19425v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Tool-augmented large language model (LLM) agents fail in a way no tool-selection or tool-security method addresses: they call tools that do not exist and pass arguments no schema declares.\u003c/li\u003e\n\u003cli\u003eExisting defenses either pick the right tool (selection) or constrain what an agent may do with real tools (gating), both of which presuppose the emitted call refers to a real tool at all.\u003c/li\u003e\n\u003cli\u003eWe show this is a structural blind spot: a hallucinated call is by construction not a decision any gate made, so no gate can reject it.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19425v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Tool-augmented large language model (LLM) agents fail in a way no tool-selection or tool-security method addresses: they call tools that do not exist…\u003c/li\u003e\n\u003cli\u003eExisting defenses either pick the right tool (selection) or constrain what an agent may do with real tools (gating), both of which presuppose the emitted call r…\u003c/li\u003e\n\u003cli\u003eWe show this is a structural blind spot: a hallucinated call is by construction not a decision any gate made, so no gate can reject it\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19448\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eThe syntax and semantics of goals\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19448v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: In both cognitive science and computer science, goals are conceptualized as cognitive states that flexibly combine with world knowledge to organize and specify purposeful behavior.\u003c/li\u003e\n\u003cli\u003eIn this way, goals are compositional representations whose content relates to rational behavior.\u003c/li\u003e\n\u003cli\u003eWe here draw attention to goals as representations and their content because it highlights a parallel with other areas in cognitive science - in particular, the syntax-semantics interface in linguistics and logic - while also foregrounding foundational questions about the expressivity, design, and efficiency of different goal representations.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19448v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: In both cognitive science and computer science, goals are conceptualized as cognitive states that flexibly combine with world knowledge to organize an…\u003c/li\u003e\n\u003cli\u003eIn this way, goals are compositional representations whose content relates to rational behavior\u003c/li\u003e\n\u003cli\u003eWe here draw attention to goals as representations and their content because it highlights a parallel with other areas in cognitive science - in particular, the…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"arxiv-cscl-b_introsearch\"\u003e\n  ArXiv cs.CL (B_intro+search)\n  \u003ca class=\"heading-link\" href=\"#arxiv-cscl-b_introsearch\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19148\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eModality Discrepancy Transformer for Ambivalence and Hesitancy Recognition\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19148v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Ambivalence and hesitancy (A/H) are affective states in which individuals express contradictory signals across facial, vocal, and linguistic channels.\u003c/li\u003e\n\u003cli\u003eAutomatically recognising A/H in clinical videos requires detecting cross-modal disagreement \u0026ndash; the signal that standard fusion methods suppress.\u003c/li\u003e\n\u003cli\u003eBased on the conflict-aware multimodal fusion framework of Bekhouche et al., we present the Modality Discrepancy Transformer (MDT).\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19148v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Ambivalence and hesitancy (A/H) are affective states in which individuals express contradictory signals across facial, vocal, and linguistic channels\u003c/li\u003e\n\u003cli\u003eAutomatically recognising A/H in clinical videos requires detecting cross-modal disagreement \u0026ndash; the signal that standard fusion methods suppress\u003c/li\u003e\n\u003cli\u003eBased on the conflict-aware multimodal fusion framework of Bekhouche et al., we present the Modality Discrepancy Transformer (MDT)\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19149\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eSubliminal Prompting Beyond Static Geometry: Causal Depth and Multi-Token Confounds\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19149v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Subliminal learning shows that language models can transmit a hidden trait through outputs that appear unrelated to it.\u003c/li\u003e\n\u003cli\u003eOne proposed explanation, token entanglement, links animal and number tokens through the model\u0026rsquo;s output vocabulary.\u003c/li\u003e\n\u003cli\u003eYet existing measurements answer different questions: whether outputs co-vary, fixed output vectors align, an answer can be read from a hidden state, or that state causally controls the answer.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19149v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Subliminal learning shows that language models can transmit a hidden trait through outputs that appear unrelated to it\u003c/li\u003e\n\u003cli\u003eOne proposed explanation, token entanglement, links animal and number tokens through the model\u0026rsquo;s output vocabulary\u003c/li\u003e\n\u003cli\u003eYet existing measurements answer different questions: whether outputs co-vary, fixed output vectors align, an answer can be read from a hidden state, or that st…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19150\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eSampling Reveals Style: Unsupervised, Training-Free Discovery of Prompt-Conditional Stylistic Axes in LLM Activations\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19150v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Large language models (LLMs) encode rich stylistic structure in their hidden activations, but discovering which stylistic dimensions are salient for a given prompt typically requires supervised contrastive data.\u003c/li\u003e\n\u003cli\u003eWe present a training-free, prompt-conditional alternative: we repeatedly sample completions of a single prompt at elevated temperature, apply Principal Component Analysis (PCA) to the pooled hidden activations, and label the resulting axes automatically from the pole generations.\u003c/li\u003e\n\u003cli\u003eWe validate the discovered axes against 245 human-elicited stylistic annotations in a two-phase study.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19150v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Large language models (LLMs) encode rich stylistic structure in their hidden activations, but discovering which stylistic dimensions are salient for a…\u003c/li\u003e\n\u003cli\u003eWe present a training-free, prompt-conditional alternative: we repeatedly sample completions of a single prompt at elevated temperature, apply Principal Compone…\u003c/li\u003e\n\u003cli\u003eWe validate the discovered axes against 245 human-elicited stylistic annotations in a two-phase study\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19151\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eWhat Users Think of Generative AI: A Cross-Platform NLP Analysis of Trust and Friction in App Store Reviews\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19151v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Generative AI (GenAI) applications have achieved rapid consumer adoption, yet little large-scale research examines user-perceived quality, trust, and adoption barriers.\u003c/li\u003e\n\u003cli\u003eWe present one of the first cross-application analyses of app store reviews for six major GenAI applications (ChatGPT, Gemini, Microsoft Copilot, Claude, DeepSeek, and Perplexity), comprising 17,012 English-language reviews from Google Play and the Apple App Store.\u003c/li\u003e\n\u003cli\u003eWe combine BERTopic topic modeling with RoBERTa sentiment classification and evaluate cross-application differences using chi-square, Kruskal-Wallis, and multinomial logistic regression with Bonferroni correction.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19151v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Generative AI (GenAI) applications have achieved rapid consumer adoption, yet little large-scale research examines user-perceived quality, trust, and…\u003c/li\u003e\n\u003cli\u003eWe present one of the first cross-application analyses of app store reviews for six major GenAI applications (ChatGPT, Gemini, Microsoft Copilot, Claude, DeepSe…\u003c/li\u003e\n\u003cli\u003eWe combine BERTopic topic modeling with RoBERTa sentiment classification and evaluate cross-application differences using chi-square, Kruskal-Wallis, and multin…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19152\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eFakeSpotter: A content and strategy agnostic Viral Misinformation Detection Tool\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19152v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Misinformation detection tools often rely on binary true and false classifications or models trained on historical examples, limiting their usefulness when novel misleading narratives emerge.\u003c/li\u003e\n\u003cli\u003eHere, we present FakeSpotter, a content- and strategy-agnostic tool designed to estimate the viral misinformation risk of textual content by measuring structural fingerprints of misinformation rather than directly adjudicating truthfulness.\u003c/li\u003e\n\u003cli\u003eFakeSpotter operationalizes a theory-driven framework across linguistic, narrative, logical, and critical-thinking dimensions, using repeated LLM assessments and domain-specific logistic regression classifiers for short and long texts.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19152v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Misinformation detection tools often rely on binary true and false classifications or models trained on historical examples, limiting their usefulness…\u003c/li\u003e\n\u003cli\u003eHere, we present FakeSpotter, a content- and strategy-agnostic tool designed to estimate the viral misinformation risk of textual content by measuring structura…\u003c/li\u003e\n\u003cli\u003eFakeSpotter operationalizes a theory-driven framework across linguistic, narrative, logical, and critical-thinking dimensions, using repeated LLM assessments an…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19153\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eStop Removing Stopwords: How an Inherited Preprocessing Default Distorts Legal Text-as-Data\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19153v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Empirical legal scholarship increasingly treats judicial text as data, and much of it still runs on sparse, interpretable pipelines \u0026ndash; TF-IDF features and linear classifiers \u0026ndash; because the textual feature is often the object of study, not merely a means to a prediction.\u003c/li\u003e\n\u003cli\u003eYet these pipelines inherit a chain of preprocessing defaults from mid-century information retrieval that were never validated against classification accuracy, the most entrenched being stopword removal.\u003c/li\u003e\n\u003cli\u003eThis study introduces an exhaustive single-word ablation that measures a preprocessing step\u0026rsquo;s effect directly against the downstream objective, and applies it to stopword removal as the hardest case to dislodge.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19153v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Empirical legal scholarship increasingly treats judicial text as data, and much of it still runs on sparse, interpretable pipelines \u0026ndash; TF-IDF features…\u003c/li\u003e\n\u003cli\u003eYet these pipelines inherit a chain of preprocessing defaults from mid-century information retrieval that were never validated against classification accuracy,…\u003c/li\u003e\n\u003cli\u003eThis study introduces an exhaustive single-word ablation that measures a preprocessing step\u0026rsquo;s effect directly against the downstream objective, and applies it t…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19154\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eNeo-Classic: A Benchmark for Evaluating Linguistic-Aesthetic Reasoning in Classical Chinese Poetry\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19154v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: While Large Language Models (LLMs) achieve high accuracy on established Classical Chinese Poetry benchmarks, it remains challenging to distinguish transferable Linguistic-Aesthetic Reasoning from reliance on familiar pre-training patterns.\u003c/li\u003e\n\u003cli\u003eTo address this issue, we introduce Neo-Classic, an evaluation benchmark that combines a constructionist Out-of-Sample (OOS) dataset with a suite of reverse understanding probes.\u003c/li\u003e\n\u003cli\u003eUnlike traditional benchmarks that rely on verification or generation over historical corpora, Neo-Classic comprises strictly metrical poetry authored by contemporary experts, reducing the possibility of direct retrieval.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19154v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: While Large Language Models (LLMs) achieve high accuracy on established Classical Chinese Poetry benchmarks, it remains challenging to distinguish tra…\u003c/li\u003e\n\u003cli\u003eTo address this issue, we introduce Neo-Classic, an evaluation benchmark that combines a constructionist Out-of-Sample (OOS) dataset with a suite of reverse und…\u003c/li\u003e\n\u003cli\u003eUnlike traditional benchmarks that rely on verification or generation over historical corpora, Neo-Classic comprises strictly metrical poetry authored by contem…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19155\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eTowards Proactive Detection of User-Side Implicit Conflicts in Human-LLM Dialogue\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19155v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: In Human-LLM dialogue, follow-up user utterances may implicitly conflict with earlier intents, leading the LLM to misinterpret user needs and generate inappropriate responses.\u003c/li\u003e\n\u003cli\u003eA reliable dialogue system should proactively detect user-side conflicts before generating a response and seek clarification when necessary.\u003c/li\u003e\n\u003cli\u003eHowever, prior work has largely focused on LLM-side conflicts, leaving user-side conflicts underexplored.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19155v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: In Human-LLM dialogue, follow-up user utterances may implicitly conflict with earlier intents, leading the LLM to misinterpret user needs and generate…\u003c/li\u003e\n\u003cli\u003eA reliable dialogue system should proactively detect user-side conflicts before generating a response and seek clarification when necessary\u003c/li\u003e\n\u003cli\u003eHowever, prior work has largely focused on LLM-side conflicts, leaving user-side conflicts underexplored\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19156\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eReflective Recovery: A Self-Supervised Method for Reasoning by Learning from Mistakes\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19156v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Data-driven fine-tuning is widely adopted to enhance reasoning in Large Language Models (LLMs) due to its simplicity and efficiency.\u003c/li\u003e\n\u003cli\u003eHowever, mainstream imitation learning methods that rely exclusively on perfect reasoning trajectories suffer from a Scaling Collapse: when the problem set is limited, increasing positive examples fails to yield continuous improvement.\u003c/li\u003e\n\u003cli\u003eHowever, during inference, an LLM can not guarantee that every intermediate step is correct and is therefore prone to errors.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19156v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Data-driven fine-tuning is widely adopted to enhance reasoning in Large Language Models (LLMs) due to its simplicity and efficiency\u003c/li\u003e\n\u003cli\u003eHowever, mainstream imitation learning methods that rely exclusively on perfect reasoning trajectories suffer from a Scaling Collapse: when the problem set is l…\u003c/li\u003e\n\u003cli\u003eHowever, during inference, an LLM can not guarantee that every intermediate step is correct and is therefore prone to errors\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19158\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eVisKG-LM: Compiling Knowledge Graphs into Visual Memory for Multiple-Choice Question Answering\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19158v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Knowledge graphs are usually integrated into question answering by encoding a retrieved subgraph with a graph neural network and fusing it with the language model in the online inference path.\u003c/li\u003e\n\u003cli\u003eThe same subgraph is therefore re-encoded from scratch every time a pair is scored, across training epochs, seeds, and evaluation runs, even though the knowledge graph never changes.\u003c/li\u003e\n\u003cli\u003eWe ask whether the retrieved knowledge graphs can instead be compiled once, offline, and then accessed as read-only memory.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19158v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Knowledge graphs are usually integrated into question answering by encoding a retrieved subgraph with a graph neural network and fusing it with the la…\u003c/li\u003e\n\u003cli\u003eThe same subgraph is therefore re-encoded from scratch every time a pair is scored, across training epochs, seeds, and evaluation runs, even though the knowledg…\u003c/li\u003e\n\u003cli\u003eWe ask whether the retrieved knowledge graphs can instead be compiled once, offline, and then accessed as read-only memory\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"arxiv-cslg-b_introsearch\"\u003e\n  ArXiv cs.LG (B_intro+search)\n  \u003ca class=\"heading-link\" href=\"#arxiv-cslg-b_introsearch\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"链接到标题\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003e链接到标题\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19209\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eGenerative Query Suggestion via Intent Coverage and Query-Level Credit Assignment\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19209v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Generative query suggestion aims to enhance user engagement by anticipating user intents and recommending relevant follow-up queries.\u003c/li\u003e\n\u003cli\u003eA central challenge is to generate slates whose individual queries are useful while the slate covers distinct intents.\u003c/li\u003e\n\u003cli\u003eWe propose an Intent-Driven Query Suggestion Framework with dual-stage optimization.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19209v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Generative query suggestion aims to enhance user engagement by anticipating user intents and recommending relevant follow-up queries\u003c/li\u003e\n\u003cli\u003eA central challenge is to generate slates whose individual queries are useful while the slate covers distinct intents\u003c/li\u003e\n\u003cli\u003eWe propose an Intent-Driven Query Suggestion Framework with dual-stage optimization\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19213\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eLayer-wise Curriculum Learning for Efficient LLM Compression\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19213v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: In this paper, we introduce layer-wise curriculum learning for efficient LLM compression.\u003c/li\u003e\n\u003cli\u003eThe proposed method facilitates the knowledge transfer from the teacher model to the student model, utilizing a curriculum learning approach that begins with easier optimization tasks and progressively tackles harder ones.\u003c/li\u003e\n\u003cli\u003eIn order to adopt the layer-wise learning in LLM compression, we partition the whole model into multiple segments consisting of layers, thereby enabling more computationally efficient knowledge transfer for LLMs.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19213v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: In this paper, we introduce layer-wise curriculum learning for efficient LLM compression\u003c/li\u003e\n\u003cli\u003eThe proposed method facilitates the knowledge transfer from the teacher model to the student model, utilizing a curriculum learning approach that begins with ea…\u003c/li\u003e\n\u003cli\u003eIn order to adopt the layer-wise learning in LLM compression, we partition the whole model into multiple segments consisting of layers, thereby enabling more co…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19242\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eBlock Parallelism For Efficient Distributed Long-Context Diffusion Language Model Training\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19242v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Block diffusion language models (BDLMs) combine autoregressive dependencies across blocks with parallel denoising within blocks, but long-context training is constrained by distributed attention communication and activation memory.\u003c/li\u003e\n\u003cli\u003eConventional context parallelism (CP) shards the combined clean-plus-corrupted sequence by position, communicating shared clean K/V together with block-specific corrupted K/V and their gradients.\u003c/li\u003e\n\u003cli\u003eWe observe that the BDLM objective separates over target blocks.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19242v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Block diffusion language models (BDLMs) combine autoregressive dependencies across blocks with parallel denoising within blocks, but long-context trai…\u003c/li\u003e\n\u003cli\u003eConventional context parallelism (CP) shards the combined clean-plus-corrupted sequence by position, communicating shared clean K/V together with block-specific…\u003c/li\u003e\n\u003cli\u003eWe observe that the BDLM objective separates over target blocks\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19243\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eRandomized SVD Approximations for Spectral Co-Clustering of Word-Document Matrices\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19243v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Spectral co-clustering is a useful tool for discovering latent structure in word-document matrices, but its reliance on singular value decomposition (SVD) can make standard formulations expensive on high-dimensional data.\u003c/li\u003e\n\u003cli\u003eThis paper presents two randomized approximations for normalized spectral co-clustering of bipartite text data when the numbers of document and word clusters may differ.\u003c/li\u003e\n\u003cli\u003eThe first method uses randomized SVD through random projection, while the second combines partial SVD with element-wise random sampling.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19243v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Spectral co-clustering is a useful tool for discovering latent structure in word-document matrices, but its reliance on singular value decomposition (…\u003c/li\u003e\n\u003cli\u003eThis paper presents two randomized approximations for normalized spectral co-clustering of bipartite text data when the numbers of document and word clusters ma…\u003c/li\u003e\n\u003cli\u003eThe first method uses randomized SVD through random projection, while the second combines partial SVD with element-wise random sampling\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19279\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eRadio-Frequency Convolutional Neural Networks\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19279v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Running artificial intelligence (AI) models directly on edge devices such as smartphones, wearables, and drones offers low latency, pervasive scalability, and data privacy, but these devices rarely carry the computing capability that modern neural networks demand.\u003c/li\u003e\n\u003cli\u003eEdge accelerators have been developed in response, yet each adds computing hardware to devices already constrained in size, weight, power, and cost (SWaP-C).\u003c/li\u003e\n\u003cli\u003eAn alternative lies in what these devices already carry: the frequency mixer in every wireless radio multiplies signals in time, natively performing convolution in the frequency domain.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19279v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Running artificial intelligence (AI) models directly on edge devices such as smartphones, wearables, and drones offers low latency, pervasive scalabil…\u003c/li\u003e\n\u003cli\u003eEdge accelerators have been developed in response, yet each adds computing hardware to devices already constrained in size, weight, power, and cost (SWaP-C)\u003c/li\u003e\n\u003cli\u003eAn alternative lies in what these devices already carry: the frequency mixer in every wireless radio multiplies signals in time, natively performing convolution…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19288\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eLearning-Induced Dynamical Transition in Recurrent Neural Networks\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19288v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Learning in recurrent neural networks can fundamentally reshape their underlying dynamics, transforming initially chaotic activity into stable task-dependent behavior.\u003c/li\u003e\n\u003cli\u003eWe develop a non-equilibrium dynamical mean-field theory(DMFT) to describe this transition during learning.\u003c/li\u003e\n\u003cli\u003eWe show that a slow feedback-driven learning process generates an evolving effective feedback strength that drives the network through a transition from chaotic to stable dynamics defined by a bifurcation of the DMFT solution.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19288v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Learning in recurrent neural networks can fundamentally reshape their underlying dynamics, transforming initially chaotic activity into stable task-de…\u003c/li\u003e\n\u003cli\u003eWe develop a non-equilibrium dynamical mean-field theory(DMFT) to describe this transition during learning\u003c/li\u003e\n\u003cli\u003eWe show that a slow feedback-driven learning process generates an evolving effective feedback strength that drives the network through a transition from chaotic…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19337\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003ePersonalized Federated Hierarchical Gaussian Processes for Privacy-Preserving Modeling of Heterogeneous Distributed Systems\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19337v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: We present Personalized Federated Hierarchical Gaussian Processes (pFedHGP) for probabilistic regression and classification when data are distributed across heterogeneous clients.\u003c/li\u003e\n\u003cli\u003eEach client\u0026rsquo;s latent function decomposes into (i) a shared global component, (ii) a client-specific deviation that shares the global kernel structure, and (iii) a flexible local residual.\u003c/li\u003e\n\u003cli\u003eSparse inducing-variable approximations and federated variational inference keep raw data local while the server synchronizes only low-dimensional statistics for the shared component.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19337v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: We present Personalized Federated Hierarchical Gaussian Processes (pFedHGP) for probabilistic regression and classification when data are distributed…\u003c/li\u003e\n\u003cli\u003eEach client\u0026rsquo;s latent function decomposes into (i) a shared global component, (ii) a client-specific deviation that shares the global kernel structure, and (iii)…\u003c/li\u003e\n\u003cli\u003eSparse inducing-variable approximations and federated variational inference keep raw data local while the server synchronizes only low-dimensional statistics fo…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19356\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eHow to Guide Your Language Flow\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19356v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: We introduce a new method to guide flow matching models.\u003c/li\u003e\n\u003cli\u003eOur approach, which we call probe guidance, uses the frozen internal states of an existing diffusion model to construct a guidance signal.\u003c/li\u003e\n\u003cli\u003eThis works using a similar principle as autoguidance, but eliminates the need for an additional forward pass at inference time and provides a reliable path to ensure that the weak and strong model share similar dynamics.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19356v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: We introduce a new method to guide flow matching models\u003c/li\u003e\n\u003cli\u003eOur approach, which we call probe guidance, uses the frozen internal states of an existing diffusion model to construct a guidance signal\u003c/li\u003e\n\u003cli\u003eThis works using a similar principle as autoguidance, but eliminates the need for an additional forward pass at inference time and provides a reliable path to e…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19359\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eSmart Insole Human Activity Recognition for Continuous Monitoring in Elderly Care\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19359v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Falls in older adults are often preceded by changes in mobility, balance, and postural transitions.\u003c/li\u003e\n\u003cli\u003eThis paper presents a wireless smart insole platform and machine-learning workflow for recognizing sitting, standing, walking, and unstable walking from plantar-pressure and inertial signals.\u003c/li\u003e\n\u003cli\u003eEach insole integrates 16 active pressure-sensing locations and a six-dimensional IMU stream consisting of tri-axial acceleration and angular velocity.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19359v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Falls in older adults are often preceded by changes in mobility, balance, and postural transitions\u003c/li\u003e\n\u003cli\u003eThis paper presents a wireless smart insole platform and machine-learning workflow for recognizing sitting, standing, walking, and unstable walking from plantar…\u003c/li\u003e\n\u003cli\u003eEach insole integrates 16 active pressure-sensing locations and a six-dimensional IMU stream consisting of tri-axial acceleration and angular velocity\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.19363\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eStiefel Attention: When the Geometry of Transformer Projection Matrices Dominates Optimizer Choice\u0026mdash;and When It Does Not\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003e发布时间:2026-09-18 12:00 北京时间\u003c/li\u003e\n\u003cli\u003e摘要:【待翻译】- arXiv:2609.19363v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: The query and key projections $\\WQ,\\WK$ in attention are almost always trained by Euclidean optimizers with no constraint on their geometry.\u003c/li\u003e\n\u003cli\u003eWe constrain them to the Stiefel manifold and optimize them there with a Riemannian Adam that carries one scalar second moment per frame, caps its step by a trust region, and retracts polarly.\u003c/li\u003e\n\u003cli\u003eFour propositions prove this update is steepest descent in the embedded metric, independent of gradient scale, well conditioned, and exactly $\\mathrm{O}(d)$-equivariant, each certified numerically in \\texttt{float64}.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 要点:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.19363v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: The query and key projections $\\WQ,\\WK$ in attention are almost always trained by Euclidean optimizers with no constraint on their geometry\u003c/li\u003e\n\u003cli\u003eWe constrain them to the Stiefel manifold and optimize them there with a Riemannian Adam that carries one scalar second moment per frame, caps its step by a tru…\u003c/li\u003e\n\u003cli\u003eFour propositions prove this update is steepest descent in the embedded metric, independent of gradient scale, well conditioned, and exactly $\\mathrm{O}(d)$-equ…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n",
  "wordCount": 5462,
  "readingTime": 26,
  "tableOfContents": "\u003cnav id=\"TableOfContents\"\u003e\n  \u003cul\u003e\n    \u003cli\u003e\u003ca href=\"#-本期-watch-list-深度导读\"\u003e📖 本期 Watch List 深度导读\u003c/a\u003e\u003c/li\u003e\n    \u003cli\u003e\u003ca href=\"#-x-平台-ai-热点快讯\"\u003e🌐 X 平台 AI 热点快讯\u003c/a\u003e\n      \u003cul\u003e\n        \u003cli\u003e\u003ca href=\"#话题-1zcode-ai-app-uploaded-full-workspaces-to-alibaba-cloud-without-clear-disclosure\"\u003e话题 1:ZCode AI App Uploaded Full Workspaces to Alibaba Cloud Without Clear Disclosure\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#话题-2claude-code-now-supports-agentsmd-files-for-better-team-workflows\"\u003e话题 2:Claude Code Now Supports AGENTS.md Files for Better Team Workflows\u003c/a\u003e\u003c/li\u003e\n      \u003c/ul\u003e\n    \u003c/li\u003e\n    \u003cli\u003e\u003ca href=\"#-大佬观点influencer-insights\"\u003e💡 大佬观点(Influencer Insights)\u003c/a\u003e\u003c/li\u003e\n    \u003cli\u003e\u003ca href=\"#-附录今日-watch-list-更新源列表\"\u003e📚 附录:今日 Watch List 更新源列表\u003c/a\u003e\n      \u003cul\u003e\n        \u003cli\u003e\u003ca href=\"#y-combinator-podcast-b_introsearch\"\u003eY Combinator Podcast (B_intro+search)\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#all-in-podcast-a_full\"\u003eAll-In Podcast (A_full)\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#stratechery-by-ben-thompson-a_full\"\u003eStratechery by Ben Thompson (A_full)\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#arxiv-csai-b_introsearch\"\u003eArXiv cs.AI (B_intro+search)\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#arxiv-cscl-b_introsearch\"\u003eArXiv cs.CL (B_intro+search)\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#arxiv-cslg-b_introsearch\"\u003eArXiv cs.LG (B_intro+search)\u003c/a\u003e\u003c/li\u003e\n      \u003c/ul\u003e\n    \u003c/li\u003e\n  \u003c/ul\u003e\n\u003c/nav\u003e",
  "isDraft": false
}