{
  "title": "2026-10-03 AI Daily Update | OpenAI Dots Ushers in the Era of Persistent Agents, Claude Opus 5.5 Tops Epoch AI Index",
  "url": "https://miaok.ong/en/ai-daily/ai-daily-2026-10-03/",
  "date": "2026-10-03T07:00:00+08:00",
  "lastmod": "2026-10-03T07:00:00+08:00",
  "type": "ai-daily",
  "kind": "page",
  "language": "en",
  "description": "OpenAI released Dots, an AI agent capable of persistent work across conversations, marking AI\u0026rsquo;s evolution from single-turn tools to long-term autonomous execution. Anthropic\u0026rsquo;s Claude Opus 5.5 surpassed GPT-6 Astra in the Epoch AI Index, shifting the ranking of top model capabilities. Concurrently, the GPT-6 series model guide was released, offering layered recommendations for inference, tool coordination, and production deployment, while multiple papers began to question the systematic bias of existing LLM evaluation metrics, with deployability and real credibility becoming the focus of attention.",
  "keywords": null,
  "tags": [],
  "categories": [],
  "author": "Mark (Miao) Kong",
  "image": "https://miaok.ong/images/avatar.jpg",
  "content": "\u003ch1 id=\"2026-10-03-ai-daily--openai-dots-usher-in-the-era-of-persistent-agents-claude-opus-55-tops-epoch-ai-index\"\u003e\n  2026-10-03 AI Daily | OpenAI Dots Usher in the Era of Persistent Agents, Claude Opus 5.5 Tops Epoch AI Index\n  \u003ca class=\"heading-link\" href=\"#2026-10-03-ai-daily--openai-dots-usher-in-the-era-of-persistent-agents-claude-opus-55-tops-epoch-ai-index\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h1\u003e\n\u003cblockquote\u003e\n\u003cp\u003eOpenAI has released Dots, AI agents that persist across conversations, marking an evolution for AI from single-turn tools to long-term autonomous execution. Anthropic\u0026rsquo;s Claude Opus 5.5 has surpassed GPT-6 Astra on the Epoch AI Index, reshuffling the capability rankings of top models. Meanwhile, the GPT-6 series model guide has been published, offering tiered recommendations for reasoning, tool coordination, and production deployment. Concurrently, multiple papers are beginning to question the systemic biases of existing LLM evaluation metrics, bringing deployability and genuine trustworthiness into the spotlight.\u003c/p\u003e\n\u003c/blockquote\u003e\n\u003ch2 id=\"-this-issues-watch-list-in-depth-guide\"\u003e\n  📖 This Issue\u0026rsquo;s Watch List: In-Depth Guide\n  \u003ca class=\"heading-link\" href=\"#-this-issues-watch-list-in-depth-guide\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h2\u003e\n\u003cp\u003eThree main themes are worth your time today. First is the \u0026ldquo;sense of uncertainty\u0026rdquo; surrounding policy and product roadmaps. One podcast offered an in-depth review of the Trump superintelligence summit and AI safety protocols, placing data center national security, the cybersecurity arms race, and the latest economic data within a single framework. OpenAI also released its GPT-6 series model guide, offering more granular, tiered recommendations for reasoning, tool coordination, and production deployment. Second is a collective rethinking of LLM evaluation. Multiple papers point to the same issue: first-token logic, structures of human bias, and interpretability metrics may systematically overestimate or distort model behavior. Evaluation teams should especially read \u0026ldquo;The First Token Is Not the Verdict.\u0026rdquo; Third is deployability and cost. From on-device NER and Text-to-SQL schema budgets to synthetic RLVR data auditing and the carbon footprint of large Arabic-language models, the focus of engineering decisions is shifting from a capabilities race to cost, data provenance, and trustworthiness.\u003c/p\u003e\n\u003ch2 id=\"-ai-hot-topics-on-x\"\u003e\n  🌐 AI Hot Topics on X\n  \u003ca class=\"heading-link\" href=\"#-ai-hot-topics-on-x\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h2\u003e\n\u003ch3 id=\"topic-1-anthropic-launches-mods-for-customizing-claude-code\"\u003e\n  Topic 1: Anthropic Launches Mods for Customizing Claude Code\n  \u003ca class=\"heading-link\" href=\"#topic-1-anthropic-launches-mods-for-customizing-claude-code\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003eCategory: AI · News\u003c/li\u003e\n\u003cli\u003eSummary: Trending for: 1 day ago, Related posts: 9,100\u003c/li\u003e\n\u003cli\u003eWhat it is: Anthropic has launched \u0026ldquo;Mods\u0026rdquo; for Claude Code, allowing developers to use TypeScript extensions to customize prompts, interfaces, and functionalities.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This gives the AI coding agent plug-in-based extensibility, which is expected to strengthen the developer ecosystem and drive the evolution of AI programming tools from general-purpose models to deeply customizable platforms.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: Discussions on X are focused on the flexibility offered by TypeScript hooks, comparing it with existing plug-in ecosystems, exploring the limits of customization, and raising concerns about extension security and fragmentation.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-2-earendil-launches-pi-10-stable-ai-coding-agent\"\u003e\n  Topic 2: Earendil Launches Pi 1.0 Stable AI Coding Agent\n  \u003ca class=\"heading-link\" href=\"#topic-2-earendil-launches-pi-10-stable-ai-coding-agent\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003eCategory: AI · News\u003c/li\u003e\n\u003cli\u003eSummary: Trending for: 1 day ago, Related posts: 2,800\u003c/li\u003e\n\u003cli\u003eWhat it is: Earendil has released the Pi 1.0 stable version of its AI coding agent.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This marks the transition of AI coding agents from an experimental phase to stable, production-ready tools, potentially accelerating the automation of software development workflows.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: Discussions on X are mainly focused on its practical coding stability and reliability, its differences from existing AI coding tools, and its potential impact on developer workflows and employment.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-3-openai-launches-persistent-ai-agents-called-dots\"\u003e\n  Topic 3: OpenAI Launches Persistent AI Agents Called Dots\n  \u003ca class=\"heading-link\" href=\"#topic-3-openai-launches-persistent-ai-agents-called-dots\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003eCategory: AI · News\u003c/li\u003e\n\u003cli\u003eSummary: Trending for: 1 day ago, Related posts: 45,000\u003c/li\u003e\n\u003cli\u003eWhat it is: OpenAI has launched Dots, AI agents that can persist between chats, and is making them available to select paid users and markets.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This marks an evolution for AI from single-turn conversational tools to agents capable of long-term autonomous task execution, raising new requirements for permission management, reliability, and automated workflows.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: Discussions on X center on the permission boundaries and reliability risks of persistent agents. Supporters believe they can transform work methods, while critics worry about a lack of transparency and control in autonomous operations.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-4-openai-paid-users-await-delayed-global-usage-reset\"\u003e\n  Topic 4: OpenAI Paid Users Await Delayed Global Usage Reset\n  \u003ca class=\"heading-link\" href=\"#topic-4-openai-paid-users-await-delayed-global-usage-reset\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003eCategory: AI · News\u003c/li\u003e\n\u003cli\u003eSummary: Trending for: 15 hours ago, Related posts: 13,000\u003c/li\u003e\n\u003cli\u003eWhat it is: OpenAI\u0026rsquo;s paid users experienced a delay in the global reset of their usage limits, with quotas not being restored as scheduled.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This incident highlights operational stability issues for commercial AI services under high load and affects the rights of paid users and trust in the platform.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: On X, many users complained that the delay impacted their daily work, demanding compensation or a transparent explanation. Others discussed whether this exposes inadequacies in OpenAI\u0026rsquo;s infrastructure and customer service response.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-5-anthropics-claude-opus-55-tops-epoch-ai-index-ahead-of-openais-gpt-6-astra\"\u003e\n  Topic 5: Anthropic\u0026rsquo;s Claude Opus 5.5 Tops Epoch AI Index Ahead of OpenAI\u0026rsquo;s GPT-6 Astra\n  \u003ca class=\"heading-link\" href=\"#topic-5-anthropics-claude-opus-55-tops-epoch-ai-index-ahead-of-openais-gpt-6-astra\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003eCategory: AI · News\u003c/li\u003e\n\u003cli\u003eSummary: Trending for: 1 day ago, Related posts: 13,000\u003c/li\u003e\n\u003cli\u003eWhat it is: Anthropic\u0026rsquo;s Claude Opus 5.5 has surpassed OpenAI\u0026rsquo;s GPT-6 Astra to rank first in the Epoch AI Index.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This indicates a shift in the capability rankings of frontier large models, with Anthropic taking a temporary lead on an important evaluation benchmark. This could influence industry expectations regarding the competitive landscape of top-tier models.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: Discussions on X primarily revolve around the credibility of the index, the extent of Claude Opus 5.5\u0026rsquo;s lead, and whether the fact that GPT-6 Astra is not yet fully available affects the fairness of the comparison.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-6-pewdiepie-launches-uncensored-local-ai-ajax-after-openai-bans\"\u003e\n  Topic 6: PewDiePie Launches Uncensored Local AI Ajax After OpenAI Bans\n  \u003ca class=\"heading-link\" href=\"#topic-6-pewdiepie-launches-uncensored-local-ai-ajax-after-openai-bans\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003eCategory: AI · News\u003c/li\u003e\n\u003cli\u003eSummary: Trending time: 20 hours ago, Related posts: 3300\u003c/li\u003e\n\u003cli\u003eWhat it is: PewDiePie has launched a locally-run, uncensored AI tool named Ajax after facing a ban from OpenAI.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This event highlights the conflict between creators and AI platforms over content moderation, potentially promoting localized, decentralized AI as an alternative to circumvent censorship.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: Discussions on X focus on the implications of uncensored AI for free speech versus its potential risks of misuse. There is also attention on whether PewDiePie\u0026rsquo;s celebrity influence can accelerate the adoption of local open-source models by the general public.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-7-ben-affleck-reveals-ai-use-in-directing-netflix-thriller-animals\"\u003e\n  Topic 7: Ben Affleck Reveals AI Use in Directing Netflix Thriller Animals\n  \u003ca class=\"heading-link\" href=\"#topic-7-ben-affleck-reveals-ai-use-in-directing-netflix-thriller-animals\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003eCategory: AI · Entertainment\u003c/li\u003e\n\u003cli\u003eSummary: Trending time: 1 day ago, Related posts: 23000\u003c/li\u003e\n\u003cli\u003eWhat it is: Ben Affleck revealed that he used AI technology while directing the Netflix thriller \u0026ldquo;Animals.\u0026rdquo;\u003c/li\u003e\n\u003cli\u003eWhy it matters: This news shows that AI is rapidly entering the mainstream film production workflow, sparking widespread industry concern about creative efficiency, its impact on jobs, and artistic originality.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: The focus of discussion on X includes the specific applications of AI in filmmaking, whether it threatens traditional film industry jobs, and whether a director\u0026rsquo;s use of AI in the creative process diminishes artistic value.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch4 id=\"ai-public-opinion-summary-on-x-today\"\u003e\n  AI Public Opinion Summary on X Today\n  \u003ca class=\"heading-link\" href=\"#ai-public-opinion-summary-on-x-today\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h4\u003e\n\u003cp\u003eToday\u0026rsquo;s main narrative revolves around AI agents and tools moving from experimentation to production, platformization, and persistence. New releases like Claude Code Mods, Pi 1.0, and OpenAI Dots show that AI coding and task execution are shifting from single-turn conversations to customizable, long-running intelligent agents. Claude Opus 5.5 surpassing GPT-6 Astra on benchmarks has also intensified the discussion about competition at the top. The consensus is that AI is rapidly entering practical scenarios like development, workflows, and film production, leading to increased efficiency and ecosystem expansion. Points of disagreement are centered on whether autonomous agents and local uncensored tools are more trustworthy, and whether AI\u0026rsquo;s involvement in film creation undermines jobs and artistic originality. Potential risks mainly include operational stability and user trust issues exposed by the delay in restoring OpenAI\u0026rsquo;s paid quotas, as well as controversies over plugin fragmentation, the misuse of uncensored models, the fairness of benchmark evaluations, and the impact on employment in the film and television industry.\u003c/p\u003e\n\u003ch2 id=\"-influencer-insights\"\u003e\n  💡 Influencer Insights\n  \u003ca class=\"heading-link\" href=\"#-influencer-insights\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h2\u003e\n\u003cblockquote\u003e\n\u003cp\u003eInfluencer insights are not available today. We recommend reading the in-depth content from the Watch List.\u003c/p\u003e\n\u003c/blockquote\u003e\n\u003ch2 id=\"-appendix-todays-watch-list-update-source-list\"\u003e\n  📚 Appendix: Today\u0026rsquo;s Watch List Update Source List\n  \u003ca class=\"heading-link\" href=\"#-appendix-todays-watch-list-update-source-list\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h2\u003e\n\u003cblockquote\u003e\n\u003cp\u003eTime window: Last 3 days; 22 sources covered; 34 updates in total\u003c/p\u003e\n\u003c/blockquote\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=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\u003c/span\u003e\n  \u003c/a\u003e\n\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e\u003ca href=\"https://allinchamathjason.libsyn.com/trumps-super-intelligence-summit-ai-safety-accord-gdp-beats-midterm-predictions\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eTrump\u0026rsquo;s Super Intelligence Summit, AI Safety Accord, GDP Beats, Midterm Predictions\u003c/a\u003e\u003c/strong\u003e\n\u003cul\u003e\n\u003cli\u003ePublication time: 2026-10-03 06:30 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary: - (0:00) Bestie intros.\n\u003cul\u003e\n\u003cli\u003e(2:01) Trump\u0026rsquo;s Super Intelligence Summit: Behind the scenes and how the accord was created and signed.\u003c/li\u003e\n\u003cli\u003e(17:57) Cybersecurity arms race, data center national security, America.gov, coordinated doomer messaging.\u003c/li\u003e\n\u003cli\u003e(45:22) Latest economic data: GDP, jobs, diesel, inflation, interest rates, manufacturing, poverty, etc.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003e(0:00) Bestie intros\u003c/li\u003e\n\u003cli\u003e(2:01) Trump\u0026rsquo;s Super Intelligence Summit: Behind the scenes and how the accord was created and signed\u003c/li\u003e\n\u003cli\u003e(17:57) Cybersecurity arms race, data center national security, America.gov, coordinated doomer messaging\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e(45:22) Recent economic data: GDP, jobs, diesel, inflation, rates, manufacturing, poverty, and more\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=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\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/dots-and-question-marks/\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003e2026.40: Dots and Question Marks\u003c/a\u003e\u003c/strong\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-10-03 01:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract:- (Photo by Heather Diehl/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\u003eWith that said, here are some of our favorites from this week.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003e(Photo by Heather Diehl/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=\"openai-blog-a_full\"\u003e\n  OpenAI Blog (A_full)\n  \u003ca class=\"heading-link\" href=\"#openai-blog-a_full\"\u003e\n    \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\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://openai.com/index/practical-guide-building-gpt-6\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eA model guide for the GPT-6 family\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-10-03 00:15 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract:- Learn how startups can choose GPT-6 models, tune reasoning effort, improve prompts and skills, coordinate tools, and prepare workflows for production.\n\u003cul\u003e\n\u003cli\u003eThis post from the OpenAI Blog explains how the model guide for the GPT-6 family is shaping the broader AI and infrastructure landscape.\u003c/li\u003e\n\u003cli\u003eIt also provides practical implications for founders, operators, and investors following the model guide for the GPT-6 family.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003eLearn how startups can choose GPT-6 models, tune reasoning effort, improve prompts and skills, coordinate tools, and prepare workflows for production.\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://openai.com/index/chatham-financial\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eChatham scales its capital markets expertise with OpenAI\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-10-02 08:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract:- Chatham Financial uses Codex and GPT-5.6 to build technology and redesign workflows, cutting trade validation from 30 minutes to under 4 minutes.\n\u003cul\u003e\n\u003cli\u003eThis post from the OpenAI Blog explains how Chatham is scaling its capital markets expertise with OpenAI, shaping the broader AI and infrastructure landscape.\u003c/li\u003e\n\u003cli\u003eIt also presents practical implications for founders, operators, and investors following Chatham\u0026rsquo;s scaling of its capital markets expertise with OpenAI.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003eChatham Financial uses Codex and GPT-5.6 to build technology and redesign workflows, cutting trade validation from 30 minutes to under 4.\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=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\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/2610.00010\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eHeavy-Tailed Memory Traces in Long-Horizon Language Agents\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract:- arXiv:2610.00010v1 Announcement Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Long-horizon language agents increasingly rely on external memory as a frozen world model, yet current memory systems are often judged solely by task success or token cost.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe argue that the missing object is the shape of memory use: under finite context and repeated retrieval, agent memory can concentrate on a small core, leaving rare states in the long tail where prediction errors accumulate.\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eWe study this effect through a conservative tail audit and find that concentration is reproducible but policy-dependent.\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00010v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Long-horizon language agents increasingly rely on external memory as a frozen world model, yet current memory systems are usually judged only by task…\u003c/li\u003e\n\u003cli\u003eWe argue that the missing object is the shape of memory use: under finite context and repeated retrieval, agent memory can concentrate on a small core while lea…\u003c/li\u003e\n\u003cli\u003eWe study this effect through a conservative tail audit and find that concentration is reproducible but policy-dependent\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/2610.00012\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eWhen Do Causal World Models Help Modular LLM Agents\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract:- arXiv:2610.00012v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: LLM agents increasingly operate through modular systems (e.g., order, payment, inventory, and shipment services), where actions in one module change the valid transitions in another.\u003c/li\u003e\n\u003cli\u003eStandard world models usually fit observational traces, but this is not the quantity needed for intervention-time planning: a trace may show payment precedes shipment, but cannot determine if payment authorizes shipment, if inventory regulates the effect, or if a hidden trigger explains both.\u003c/li\u003e\n\u003cli\u003eWe study this gap through FedCausalCompose, a causal world-model framework for modular LLM agents in which local actions provide intervention-response evidence for cross-module interfaces.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00012v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: LLM agents increasingly act through modular systems, such as order, payment, inventory, and shipment services, where actions in one module change whic…\u003c/li\u003e\n\u003cli\u003eStandard world models usually fit observational traces, but this is not the quantity needed for intervention-time planning: a trace may show that payment preced…\u003c/li\u003e\n\u003cli\u003eWe study this gap through FedCausalCompose, a causal world-model framework for modular LLM agents in which local actions provide intervention-response evidence…\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/2610.00015\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eFrom Proposal to Verified Effect: Praxa, an Evidence-Bound Harness for Governed AI Agent Execution\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract:- arXiv:2610.00015v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Large language model agents can propose and execute actions, but proposal, authorization, scheduling, verification of external effects, and service promotion are different matters.\u003c/li\u003e\n\u003cli\u003eWe introduce Praxa, an agent tool that explicitly represents these states through deterministic admission, agent execution, external read-back, coordination, and review elevation.\u003c/li\u003e\n\u003cli\u003eWe report four evidence channels.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00015v1 Announce Type: new\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eAbstract: Large-language-model agents can propose and execute actions, but proposal, authority, dispatch, verified external effect, and serving promotion are di…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe present Praxa, an agent harness that represents these states explicitly through deterministic admission, brokered execution, external read-back, reconciliati…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe report four evidence lanes\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2610.00018\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eWhat Do Rationales Communicate? A Message-Intervention Study in Role-Specialized QA\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eRelease Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00018v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Role-specialized QA pipelines increasingly pass rationales from a reasoner to a verifier, but it is unclear what this message actually buys: better answers, stronger support evaluations, or new failure surfaces.\u003c/li\u003e\n\u003cli\u003eWe introduce a message-intervention diagnostic that fixes the evidence and candidate answer while varying only the rationale passed across the reasoner-to-verifier boundary.\u003c/li\u003e\n\u003cli\u003eOn 400 MuSiQue, HotpotQA, and 2WikiMultiHopQA examples using DeepSeek as both generator and verifier, faithful rationales add almost no answer accuracy over having no rationale, while corrupted rationales strongly alter support judgments.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00018v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Role-specialized QA pipelines increasingly pass rationales from a reasoner to a verifier, but it is unclear what this message actually buys: better an…\u003c/li\u003e\n\u003cli\u003eWe introduce a message-intervention diagnostic that fixes the evidence and candidate answer while varying only the rationale passed across the reasoner-to-verif…\u003c/li\u003e\n\u003cli\u003eOn 400 MuSiQue, HotpotQA, and 2WikiMultiHopQA examples with DeepSeek as generator and verifier, faithful rationales add almost no answer accuracy over no ration…\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/2610.00025\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eMeasuring the Microtask Eligibility Gap: When Is an Off-the-Shelf SLM Enough for an Agent Harness?\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eRelease Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00025v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Agents increasingly seek to run Small Language Models (SLMs) on microtasks surrounding a frontier Large Language Model (LLM) planner: auto-approving shell commands, writing to memory, selecting tools, ranking past turns.\u003c/li\u003e\n\u003cli\u003eWe ask whether off-the-shelf SLMs meet practitioner-defined thresholds, why they fail when they do, and whether quantization changes the answer.\u003c/li\u003e\n\u003cli\u003eWe build a benchmark of 4 such microtasks with fixed prompts and automated metrics, each with a pre-specified threshold $\\tau$ anchored to cheap non-LLM baselines and a CI-aware eligibility rule (a configuration passes only if its confidence bounds clear $\\tau$).\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00025v1 Announce Type: new\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eAbstract: Agent harnesses increasingly want to run small language models (SLMs) on the microtasks around a frontier large language model (LLM) planner: auto-app…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe ask whether off-the-shelf SLMs meet practitioner-defined thresholds and, when they fail, why, and whether quantization changes the answer\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe build a benchmark of 4 such microtasks with fixed prompts and automatic metrics, each with a pre-specified threshold $\\tau$ anchored to a cheap non-LLM basel…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2610.00047\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eCharacterizing a Configuration Where Inference-Time PRM-Pruned Fragment Grafting Is Inert: Evidence from Three Reasoning LMs\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00047v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Diversity collapse in parallel chains of thought has motivated inference-time interventions based on a natural design: when a process reward model (PRM) prunes a chain, its high-PRM prefix is extracted and transplanted verbatim as a contextual demonstration into its still-decoding siblings.\u003c/li\u003e\n\u003cli\u003eWe isolate this mechanism (PRM-Pruned Fragment Grafting (PPFG)) as the most cost-minimal operationalization of cross-trajectory step-level transfer and test it at an operating point where prior fragment grafting work reports gains only with additional compensating components.\u003c/li\u003e\n\u003cli\u003eOn Qwen2.5-7B-Instruct with Math-Shepherd on the full MATH500 (n=500, three seeds), PPFG in both stagnation- and random-targeting variants is statistically indistinguishable from an independent parallel CoT baseline on every measured axis.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00047v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Diversity collapse in parallel chain-of-thought has motivated inference-time interventions built on a natural design: when a process reward model (PRM…\u003c/li\u003e\n\u003cli\u003eWe isolate this mechanism, PRM-Pruned Fragment Grafting (PPFG), as the most cost-minimal operationalization of cross-trajectory step-level transfer, and test it…\u003c/li\u003e\n\u003cli\u003eOn Qwen2.5-7B-Instruct with Math-Shepherd on full MATH500 (n=500, three seeds), PPFG in both stagnation- and random-targeting variants is statistically indistin…\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/2610.00061\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eGradient-Aligned Pair Selection for Personalized Preference Optimization\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00061v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Personalizing large language models (LLMs) requires aligning generative behavior with user-specific preferences rather than overall quality.\u003c/li\u003e\n\u003cli\u003eWhile Direct Preference Optimization (DPO) provides a stable framework for preference learning, its effectiveness in personalized settings critically depends on how preference pairs are selected.\u003c/li\u003e\n\u003cli\u003eExisting methods often rely on heuristic criteria, such as likelihood-based extreme values, which decouples the optimization from explicit user utility and can lead to suboptimal personalization.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00061v1 Announce Type: new\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eAbstract: Personalizing large language models (LLMs) requires aligning generation behavior with user-specific preferences rather than aggregate quality\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWhile Direct Preference Optimization (DPO) provides a stable framework for preference learning, its effectiveness in personalized settings critically depends on…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eExisting approaches typically rely on heuristic criteria, such as likelihood-based extremes, which decouple optimization from explicit user utility and can lead…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2610.00074\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eK-Dense BYOK: An Open-Source AI Research Assistant That Runs Locally and Keeps a Hash-Chained Lab Notebook\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00074v1 Announcement Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: K-Dense BYOK (bring your own keys) is a free, open-source AI research assistant for scientists in any field that runs on the researcher\u0026rsquo;s own computer.\u003c/li\u003e\n\u003cli\u003eThe researcher provides access to a model of their choice (hosted or running locally), and the application provides everything else: a place for the work to run, a layer of scientific scaffolding, and a complete record.\u003c/li\u003e\n\u003cli\u003eEach project is an ordinary folder, so the data, code, results, and record remain on a machine administered by the researcher and can be read years later without the application.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00074v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: K-Dense BYOK (bring your own keys) is a free, open-source AI research assistant for scientists in any field that runs on the researcher\u0026rsquo;s own computer\u003c/li\u003e\n\u003cli\u003eThe researcher supplies access to a model of their choice, hosted or running locally, and the application supplies everything else: a place for the work to run,…\u003c/li\u003e\n\u003cli\u003eEach project is an ordinary folder, so the data, the code, the results, and the record stay on a machine the researcher administers and can be read years later…\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/2610.00084\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eScientific Agents: Evaluating Profession-Specific System Prompts on Scientific Tasks\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00084v1 Announcement Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Detailed, profession-specific system prompts increased token usage and estimated cost per response, but without consistent gains in accuracy.\u003c/li\u003e\n\u003cli\u003eWe evaluate Scientific Agents using Gemini 3.8 Flash via OpenRouter in the Pi agent tool, an open-source corpus of 503 professional AGENTS.md profiles.\u003c/li\u003e\n\u003cli\u003eWe compare matched profiles against four controls: a minimal baseline (\u0026ldquo;You are a helpful assistant\u0026rdquo;), the profile\u0026rsquo;s opening role sentence, general scientific rigor guidelines, and a profile from an unrelated domain.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00084v1 Announce Type: new\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eAbstract: Detailed profession-specific system prompts raise token use and estimated cost per response without a consistent accuracy gain\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe evaluate Scientific Agents, an open-source corpus of 503 profession-specific AGENTS.md profiles, with Gemini 3.8 Flash via OpenRouter in the Pi agent harness\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe compare matched profiles with four controls: a minimal baseline (\u0026ldquo;You are a helpful assistant\u0026rdquo;), the profile\u0026rsquo;s opening role sentence, a generic scientific ri…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2610.00197\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eComedic Fool\u0026rsquo;s Gold: Reward Exploits and Countermeasures in Conversational Humor\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00197v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: We investigate automated rewards for training language models in conversational humor, focusing on reward exploits and countermeasures.\u003c/li\u003e\n\u003cli\u003eTwo approaches aim to capture understandable surprise and predicted audience amusement.\u003c/li\u003e\n\u003cli\u003eControlled tests show that an embedding-based surprise reward accepts word-shuffled replies as readily as witty ones.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00197v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: We investigate automated rewards for training language models in conversational humor, focusing on reward exploits and countermeasures\u003c/li\u003e\n\u003cli\u003eTwo approaches aim to capture understandable surprise and predicted audience amusement\u003c/li\u003e\n\u003cli\u003eControlled tests show that an embedding-based surprise reward accepts word-shuffled replies as readily as witty ones\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=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\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/2610.00007\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eOn-Device Named-Entity Recognition: A Deployability Study of Accuracy, Cost, Reliability, and Confidence\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00007v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Named-entity recognition (NER) is increasingly wanted on-device (no API, low latency, data kept local).\u003c/li\u003e\n\u003cli\u003eThe question for practitioners is not about leaderboards, but which model is deployable, how to evaluate it without human annotation, and whether its confidence is trustworthy.\u003c/li\u003e\n\u003cli\u003eWe place 9 systems across three paradigms and 13M to 8B parameters on three datasets with different characteristics: a classic tagger (spaCy), bidirectional encoder experts (GLiNER, 166 to 460M), and locally run generative LLMs (Qwen3-0.6B/1.7B/4B-Instruct, DeepSeek-R1-1.5B/8B), and report precision plus two axes the literature overlooks: latency and output validity.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00007v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Named-entity recognition (NER) is increasingly wanted on-device (no API, low latency, data kept local)\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eThe practitioner\u0026rsquo;s question is not the leaderboard but which model is deployable, how to evaluate it without human annotation, and whether its confidence can be…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe answer these jointly\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2610.00045\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eSCM-based Fairness and Faithful Explainability for Legal Document Classification\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00045v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Transformer models such as LegalBERT are increasingly used in legal decision support, raising concerns about the fairness and transparency of model explanations.\u003c/li\u003e\n\u003cli\u003eThese properties are typically evaluated separately, leaving it unclear whether debiasing interventions that alter fairness also change how faithfully explanations reflect the model\u0026rsquo;s reasoning.\u003c/li\u003e\n\u003cli\u003eThis study investigates this relationship in the ECtHR alleged violations corpus from LexGLUE.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00045v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Transformer models such as LegalBERT are increasingly used in legal decision support, raising concerns about both fairness and the transparency of mod…\u003c/li\u003e\n\u003cli\u003eThese properties are usually evaluated separately, leaving open whether a debiasing intervention that changes fairness also changes how faithfully explanations…\u003c/li\u003e\n\u003cli\u003eThis study investigates that relationship on the ECtHR alleged-violations corpus from LexGLUE\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/2610.00054\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eThe First Token Is Not the Verdict: Hidden Costs of Reading LLM Judges Without Generating\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00054v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Reading an LLM judge\u0026rsquo;s verdict from the logits of the first generated token is cheap, requires no generation, and is precisely what constrained decoding and likelihood scoring evaluation tools produce.\u003c/li\u003e\n\u003cli\u003eWe show that this reading distorts positional bias in one direction: it exaggerates it in every condition we test, so the figures obtained this way act as an upper bound.\u003c/li\u003e\n\u003cli\u003eThe mechanism is that the judge is not always led by the verdict token; for the three Qwen3 judges, this occurs on 12% to 49% of pairs, while for Llama-3.1-8B and Phi-3.5-mini, this figure is below 3%. Forcing a reading on these pairs returns the response that appears first, rather than the judgment.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00054v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Reading an LLM judge\u0026rsquo;s verdict from the logits of its first generated token is cheap, requires no generation, and is exactly what constrained decoding…\u003c/li\u003e\n\u003cli\u003eWe show that this readout distorts position bias in one direction: it overstates it in every condition we test, so figures obtained this way behave as upper bou…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eThe mechanism is that judges do not always lead with a verdict token, on 12% to 49% of pairs for three Qwen3 judges and under 3% for Llama-3.1-8B and Phi-3.5-mi…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2610.00070\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eMeasuring Human-Like Bias in LLMs? A Critique of Human-Derived Bias Constructs in LLM Evaluation\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00070v1 Announce Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Researchers are increasingly using human-derived bias constructs to study Large Language Models (LLMs), including social-cognitive constructs like implicit bias and stereotype activation, as well as cognitive biases such as anchoring, framing effects, and confirmation bias.\u003c/li\u003e\n\u003cli\u003eSuch approaches provide an alternative to overt bias probes, especially when direct questioning might obscure bias or when the model\u0026rsquo;s behavior is normatively acceptable.\u003c/li\u003e\n\u003cli\u003eHowever, adapting human bias constructs to LLMs introduces an inferential gap.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00070v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Researchers increasingly use human-derived bias constructs to study Large Language Models (LLMs), including social-cognitive constructs such as implic…\u003c/li\u003e\n\u003cli\u003eSuch approaches offer alternatives to overt bias probes, particularly when direct questioning may obscure bias or when model behaviour appears normatively accep…\u003c/li\u003e\n\u003cli\u003eHowever, adapting human bias constructs to LLMs introduces an inferential gap\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/2610.00087\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eLegal text classification in Korean sexual offense cases: from traditional machine learning to large language models with XAI insights\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00087v1 Announce Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: The advancement of Natural Language Processing (NLP) has expanded AI-based text classification in the legal domain.\u003c/li\u003e\n\u003cli\u003eHowever, accurately classifying legal documents remains challenging due to the complexity of legal texts and subtle differences between legal categories.\u003c/li\u003e\n\u003cli\u003eThis study evaluates legal text classification models, ranging from traditional machine learning techniques to Large Language Models (LLMs), using ten categories of South Korean sexual offense precedents.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00087v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: The advancement of natural language processing (NLP) has expanded AI-based text classification in the legal domain\u003c/li\u003e\n\u003cli\u003eHowever, accurately classifying legal documents remains challenging due to the complexity of legal texts and subtle differences between legal categories\u003c/li\u003e\n\u003cli\u003eThis study evaluates legal text classification models ranging from traditional machine learning techniques to large language models (LLMs) using ten categories…\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/2610.00092\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eBudgetSchemaBench: A Budget-Swept Diagnostic for Schema Context in Text-to-SQL\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00092v1 Announcement Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Data agents over structured sources must fit the database schema into the model\u0026rsquo;s context window.\u003c/li\u003e\n\u003cli\u003eLarge catalogs can span many databases and thousands of columns, so cost constraints may require choosing between table coverage and serialization detail before the context window is filled.\u003c/li\u003e\n\u003cli\u003eWe introduce BudgetSchemaBench, an execution-based diagnostic for this setting.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00092v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Data agents over structured sources must fit database schema into the model\u0026rsquo;s context window\u003c/li\u003e\n\u003cli\u003eLarge catalogs can span many databases and thousands of columns, so cost constraints may require choosing between table coverage and serialization detail well b…\u003c/li\u003e\n\u003cli\u003eWe introduce BudgetSchemaBench, an execution-grounded diagnostic for this setting\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/2610.00096\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eFACET at WMT 2026 Automated Translation Quality Evaluation Task\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00096v1 Announcement Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Different error types in machine translation require different evidence.\u003c/li\u003e\n\u003cli\u003eWhether the meaning is preserved can only be judged based on the source, while whether the target is well-formed, or whether it consistently names an entity, can be judged based on the target alone.\u003c/li\u003e\n\u003cli\u003eWe present FACET, our reference-free submission to the WMT26 Automated Translation Quality Evaluation Task, which breaks down the evaluation into fluency, accuracy, and consistency stages, and provides only the necessary context for the error types in each stage.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00096v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Different error types in machine translation require different evidence\u003c/li\u003e\n\u003cli\u003eWhether meaning is preserved can be judged only against the source, while whether the target is well-formed, or whether it names one entity consistently, can be…\u003c/li\u003e\n\u003cli\u003eWe present FACET, our reference-free submission to the WMT26 Automated Translation Quality Evaluation Task, which decomposes evaluation into Fluency, Accuracy,…\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/2610.00164\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eMetric-Construction Coupling Inflates Measured Synthetic Dialect Recovery\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00164v1 Announcement Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Korean dialect corpora are available, but not redistributable: weights may be released, but training and evaluation cannot be reproduced from the underlying data.\u003c/li\u003e\n\u003cli\u003eWe ask how much is recovered by supervised synthetic data and whether recovery can be measured independently of the synthesis pipeline.\u003c/li\u003e\n\u003cli\u003eWe contribute KoDialectBench, 1,000 items from five regions on three axes, released as identifier hashes and scoring code so users can reconstruct items from their own licensed copies.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00164v1 Announce Type: new\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eAbstract: Korean dialect corpora are available but not redistributable: weights may be released, while reproducing training and evaluation from the underlying d…\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eWe ask how much of that supervision synthetic data recovers, and whether that recovery can be measured independently of the synthesis pipeline\u003c/li\u003e\n\u003cli\u003eWe contribute KoDialectBench, 1,000 items across five regions on three axes, released as identifier hashes and scoring code so users reconstruct the items from…\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/2610.00202\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eWhen a Data Artifact Isn\u0026rsquo;t a Shortcut: Causal Auditing of Synthetic RLVR Corpora\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00202v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Several recent pipelines build RLVR training data by masking a span of real corpus text and asking a language model to invent plausible wrong answers around it.\u003c/li\u003e\n\u003cli\u003eTherefore, the correct option is genuine human prose; every distractor is synthetic.\u003c/li\u003e\n\u003cli\u003eCorrectness and provenance become entangled, and a policy could in principle learn the second instead of the first.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00202v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Several recent pipelines build RLVR training data by masking a span of real corpus text and asking a language model to invent plausible wrong answers…\u003c/li\u003e\n\u003cli\u003eThe correct option is therefore genuine human prose; every distractor is synthetic\u003c/li\u003e\n\u003cli\u003eCorrectness and provenance become entangled, and a policy could in principle learn the second instead of the first\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/2610.00223\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eA Holistic Assessment of the Carbon Footprint of Noor, a Very Large Arabic Language Model\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00223v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: As ever larger language models grow more ubiquitous, it is crucial to consider their environmental impact.\u003c/li\u003e\n\u003cli\u003eCharacterized by extreme size and resource use, recent generations of models have been criticized for their voracious appetite for compute, and thus their significant carbon footprint.\u003c/li\u003e\n\u003cli\u003eAlthough carbon impact reporting is becoming more common in machine learning papers, such reporting is often limited to the computational resources used strictly for training.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00223v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: As ever larger language models grow more ubiquitous, it is crucial to consider their environmental impact\u003c/li\u003e\n\u003cli\u003eCharacterised by extreme size and resource use, recent generations of models have been criticised for their voracious appetite for compute, and thus significant…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eAlthough reporting of carbon impact has grown more common in machine learning papers, this reporting is usually limited to compute resources used strictly for t…\u003c/p\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=\"Link to heading\"\u003e\u003c/i\u003e\n    \u003cspan class=\"sr-only\"\u003eLink to heading\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/2610.00002\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eReverse Item Response Theory for Sparsity-Robust Ranking in Fragmented Cancer Drug-Response Matrices\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00002v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: We introduce reverse Item Response Theory (IRT) into pharmacogenomic drug-response analysis by treating cancer types as latent \u0026ldquo;subjects\u0026rdquo; with resistance, and drugs as \u0026ldquo;items\u0026rdquo; with evasion difficulty.\u003c/li\u003e\n\u003cli\u003eThe model was applied to 242,036 drug sensitivity measurements in the Genomics of Drug Sensitivity in Cancer (GDSC2) database, estimating cancer-type-level in vitro drug resistance and drug-level broad activity within a shared latent space.\u003c/li\u003e\n\u003cli\u003eValidation across four missingness regimes shows that reverse IRT better recovers the full-data latent ranking than simple averaging, with advantages of Delta-rho = +0.089 to +0.095 in 60% missing cases under MCAR, cancer-biased, and drug-biased sparsity.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00002v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: We introduce reverse Item Response Theory (IRT) to pharmacogenomic drug-response analysis by treating cancer types as latent \u0026ldquo;subjects\u0026rdquo; with resistanc…\u003c/li\u003e\n\u003cli\u003eApplied to 242,036 drug sensitivity measurements from the Genomics of Drug Sensitivity in Cancer (GDSC2) database, the model estimates cancer-type-level in-vitr…\u003c/li\u003e\n\u003cli\u003eValidation across four missingness regimes demonstrates that reverse IRT better recovers the full-data latent ranking than simple averaging, with advantages of…\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/2610.00004\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eHow Far is Adam from Natural Gradient Descent?\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00004v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Adam is the standard optimizer in deep learning, but its geometric relationship to natural gradient descent (NGD) contains unresolved questions.\u003c/li\u003e\n\u003cli\u003eWe investigate Adam\u0026rsquo;s full update rule, including momentum, as a diagonal empirical Fisher approximation affected by diagonal truncation, empirical label substitution, and temporal lag.\u003c/li\u003e\n\u003cli\u003eUsing the scale-invariant $\\gamma(\\Delta\\theta)$ metric, we measure Adam\u0026rsquo;s geometric deviation from true NGD in four loss scenarios: well-conditioned linear regression, ill-conditioned linear regression, logistic regression, and non-convex small neural networks.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00004v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Adam is the standard optimizer in deep learning, yet its geometric relationship to natural gradient descent (NGD) contains unresolved questions\u003c/li\u003e\n\u003cli\u003eWe study Adam\u0026rsquo;s full update rule, including momentum, as a diagonal empirical Fisher approximation subject to diagonal truncation, empirical label substitution,…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eUsing the scale-invariant $\\gamma(\\Delta\\theta)$ metric, we measure Adam\u0026rsquo;s geometric deviation from true NGD across four loss landscapes: well-conditioned linea…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2610.00009\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eFourierQK: Filter Shape, Admissibility and the Leakage-Coverage Law\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePosted: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00009v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Frequency-collapse attention [Zeris, 2026e] achieves large gains over standard dot-product attention by replacing the Q/K dot product with a bandpass-filtered inner product at learned frequencies.\u003c/li\u003e\n\u003cli\u003eA natural follow-up question is: which filter shape works best, and why?\u003c/li\u003e\n\u003cli\u003eWe test five hypotheses about filter properties \u0026ndash; DC suppression, Nyquist suppression, bandwidth, center frequency, and multi-scale coverage \u0026ndash; using controlled ablations on a character-level language model (TinyShakespeare, 6-layer GPT).\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00009v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Frequency-collapse attention [Zeris, 2026e] achieves large gains over standard dot-product attention by replacing the Q/K dot product with a bandpass-…\u003c/li\u003e\n\u003cli\u003eA natural follow-up question is: which filter shape works best, and why\u003c/li\u003e\n\u003cli\u003eWe test five hypotheses about filter properties \u0026ndash; DC suppression, Nyquist suppression, bandwidth, centre frequency, and multi-scale coverage \u0026ndash; using a control…\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/2610.00035\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eIntegrating Fairness and Explainability in a Multiple Instance Reinforcement Learning System\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePosted: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00035v1 Announce Type: new.\n-Abstract: Predicting student performance from educational interaction data requires models that are both accurate and sufficiently transparent to support meaningful interventions, while demographic information introduces additional risks of unfair predictions.\n\u003cul\u003e\n\u003cli\u003eThis study investigates a multi-objective framework that combines reinforcement learning-based multiple instance learning (RL-MIL), adversarial debiasing, and preference-conditioned hypernetworks for student risk prediction.\u003c/li\u003e\n\u003cli\u003eMIL represents each student as a bag of weakly labeled interactions, while an RL agent selects informative instances for downstream classification.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00035v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Predicting student performance from educational interaction data requires models that are both accurate and sufficiently transparent to support meanin…\u003c/li\u003e\n\u003cli\u003eThis study investigates a multi-objective framework that combines reinforcement learning-based multiple instance learning (RL-MIL), adversarial debiasing, and p…\u003c/li\u003e\n\u003cli\u003eMIL represents each student as a bag of weakly labeled interactions, while an RL agent selects informative instances for downstream classification\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/2610.00049\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eFast Polynomial Transcendentals for LLMs\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00049v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Graphics processing unit (GPU) generations scale matrix, special-function, and memory pipelines at different rates, so kernel bottlenecks change as hardware evolves.\u003c/li\u003e\n\u003cli\u003eFlashAttention-4 exposed an internal attention imbalance on NVIDIA Blackwell.\u003c/li\u003e\n\u003cli\u003eWe test whether short polynomial programs can accelerate other special-function-unit (SFU) operations in large language models (LLMs).\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00049v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Graphics processing unit (GPU) generations scale matrix, special-function, and memory pipelines at different rates, so kernel bottlenecks move as hard…\u003c/li\u003e\n\u003cli\u003eFlashAttention-4 exposed this imbalance inside attention on NVIDIA Blackwell\u003c/li\u003e\n\u003cli\u003eWe test whether short polynomial programs can accelerate other special-function-unit (SFU) operations in large language models (LLMs)\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/2610.00050\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eSW-KAN: Kolmogorov-Arnold Networks with Stieltjes-Wigert q-Orthogonal Polynomials\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00050v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Kolmogorov-Arnold Networks (KANs) represent a paradigm shift in deep learning, replacing fixed node activations with learnable univariate functions on the edges, thereby providing enhanced interpretability and parameter efficiency.\u003c/li\u003e\n\u003cli\u003eWhile recent polynomial-based KAN variants have addressed the computational overhead of the original B-spline implementations, they introduce a fundamental yet underexplored challenge: the domain mismatch between unbounded real-valued inputs and the bounded or semi-infinite support of orthogonal polynomial bases.\u003c/li\u003e\n\u003cli\u003eTo address this limitation, we propose the Stieltjes-Wigert Kolmogorov-Arnold Network (SW-KAN), a novel architecture that employs Stieltjes-Wigert q-orthogonal polynomials defined on the semi-infinite domain (0, infinity).\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00050v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Kolmogorov-Arnold Networks (KANs) represent a paradigmatic shift in deep learning by replacing fixed node activations with learnable univariate functi…\u003c/li\u003e\n\u003cli\u003eWhile recent polynomial-based KAN variants have addressed the computational overhead of original B-spline implementations, they introduce a fundamental yet unde…\u003c/li\u003e\n\u003cli\u003eTo address this limitation, we propose the Stieltjes-Wigert Kolmogorov-Arnold Network (SW-KAN), a novel architecture that employs Stieltjes-Wigert q-orthogonal…\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/2610.00053\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eFormat-Aware Fusion for Fast FP4 Pretraining\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00053v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Four-bit floating-point (FP4) tensor cores accelerate matrix multiplication, but scale computation, operand packing, layout construction, and saving backward states can erase the gains.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe propose \\emph{format-aware fusion}, which co-designs each quantization producer with its scale domain and consumer layout for native \\mxfp{}, global \\nvfp{}, and collaborative thread array local \\nvfp{}.\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe evaluate Llama-3-family 8B pretraining through 160 billion tokens using bfloat16 output projections and compiled cross entropy.\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eEN Highlights:\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00053v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Four-bit floating-point (FP4) Tensor Cores accelerate matrix multiplication, but scale computation, operand packing, layout construction, and saved ba…\u003c/li\u003e\n\u003cli\u003eWe present \\emph{format-aware fusion}, which co-designs each quantization producer with its scale domain and consumer layout for native \\mxfp{}, global \\nvfp{},…\u003c/li\u003e\n\u003cli\u003eWe evaluate Llama-3-family 8B pretraining through 160 billion tokens using bfloat16 output projections and compiled cross entropy\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/2610.00083\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003e\u0026ldquo;very likely\u0026rdquo; Means \u0026ldquo;uncertain\u0026rdquo;? How LLMs Diverge from Humans in Linguistic Uncertainty Quantification\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eRelease Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00083v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Humans express uncertainty verbally via markers (e.g., \u0026ldquo;possible,\u0026rdquo; \u0026ldquo;likely\u0026rdquo;), yet most LLM uncertainty quantification (UQ) relies on signals based on cost likelihood or consistency.\u003c/li\u003e\n\u003cli\u003eFrom a cognitive perspective, accurate verbal uncertainty reflects metacognitive monitoring, representing knowledge boundaries (\u0026ldquo;knowing that you don\u0026rsquo;t know\u0026rdquo;) to support regulation and information seeking.\u003c/li\u003e\n\u003cli\u003eIn this paper, we investigate how LLMs diverge from humans in verbal uncertainty quantification and whether verbal markers can reliably quantify LLM uncertainty.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00083v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Humans express uncertainty verbally via markers (e.g., \u0026ldquo;possible,\u0026rdquo; \u0026ldquo;likely\u0026rdquo;), yet most LLM uncertainty quantification (UQ) relies on costing likelihoo…\u003c/li\u003e\n\u003cli\u003eFrom a cognitive perspective, accurate verbal uncertainty reflects metacognitive monitoring, representing knowledge boundaries (\u0026ldquo;knowing that you don\u0026rsquo;t know\u0026rdquo;) t…\u003c/li\u003e\n\u003cli\u003eIn this paper, we investigate how LLMs diverge from humans in verbal uncertainty quantification and whether verbal markers can reliably quantify LLM uncertainty\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/2610.00094\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eNous: Learning and Certifying Memory Decisions Before Source Calibration\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eRelease Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00094v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Belief-based agent memory requires making reliable decisions about the current state, but its evidence can be noisy, duplicated, or stale.\u003c/li\u003e\n\u003cli\u003eMust memory first calibrate its sources to improve its decisions?\u003c/li\u003e\n\u003cli\u003eWe separate learning, calibration, and revision certification.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00094v1 Announce Type: new\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eAbstract: Belief-based agent memory needs reliable decisions about current state, yet its evidence may be noisy, copied, or stale\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eMust a memory calibrate its sources before it can improve its decisions\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe separate learning, calibration, and revision certification\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2610.00095\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eOne Mastery Threshold Does Not Fit All Knowledge Tracing Models\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-10-02 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2610.00095v1 Announcement Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Tutoring systems use mastery thresholds to decide when students can stop practicing and move on, but when the underlying knowledge tracing (KT) model changes, the same numerical threshold can lead to very different decisions.\u003c/li\u003e\n\u003cli\u003eWe examined six KT models in four public educational datasets and evaluated 12 thresholds from 0.50 to 0.99 using post-advancement performance, advancement coverage, practice burden, and differences between prior performance groups.\u003c/li\u003e\n\u003cli\u003eWe also identified thresholds that balance performance, extra practice, and advancement under 30 predefined instructional settings.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2610.00095v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Tutoring systems use mastery thresholds to decide when students can stop practicing and advance, but the same numerical threshold can lead to very dif…\u003c/li\u003e\n\u003cli\u003eWe examine six KT models across four public educational datasets and evaluate 12 thresholds from 0.50 to 0.99 using post-advancement performance, advancement co…\u003c/li\u003e\n\u003cli\u003eWe also identify thresholds that balance performance, extra practice, and advancement under 30 predefined instructional settings\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": 6647,
  "readingTime": 32,
  "tableOfContents": "\u003cnav id=\"TableOfContents\"\u003e\n  \u003cul\u003e\n    \u003cli\u003e\u003ca href=\"#-this-issues-watch-list-in-depth-guide\"\u003e📖 This Issue\u0026rsquo;s Watch List: In-Depth Guide\u003c/a\u003e\u003c/li\u003e\n    \u003cli\u003e\u003ca href=\"#-ai-hot-topics-on-x\"\u003e🌐 AI Hot Topics on X\u003c/a\u003e\n      \u003cul\u003e\n        \u003cli\u003e\u003ca href=\"#topic-1-anthropic-launches-mods-for-customizing-claude-code\"\u003eTopic 1: Anthropic Launches Mods for Customizing Claude Code\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-2-earendil-launches-pi-10-stable-ai-coding-agent\"\u003eTopic 2: Earendil Launches Pi 1.0 Stable AI Coding Agent\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-3-openai-launches-persistent-ai-agents-called-dots\"\u003eTopic 3: OpenAI Launches Persistent AI Agents Called Dots\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-4-openai-paid-users-await-delayed-global-usage-reset\"\u003eTopic 4: OpenAI Paid Users Await Delayed Global Usage Reset\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-5-anthropics-claude-opus-55-tops-epoch-ai-index-ahead-of-openais-gpt-6-astra\"\u003eTopic 5: Anthropic\u0026rsquo;s Claude Opus 5.5 Tops Epoch AI Index Ahead of OpenAI\u0026rsquo;s GPT-6 Astra\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-6-pewdiepie-launches-uncensored-local-ai-ajax-after-openai-bans\"\u003eTopic 6: PewDiePie Launches Uncensored Local AI Ajax After OpenAI Bans\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-7-ben-affleck-reveals-ai-use-in-directing-netflix-thriller-animals\"\u003eTopic 7: Ben Affleck Reveals AI Use in Directing Netflix Thriller Animals\u003c/a\u003e\u003c/li\u003e\n      \u003c/ul\u003e\n    \u003c/li\u003e\n    \u003cli\u003e\u003ca href=\"#-influencer-insights\"\u003e💡 Influencer Insights\u003c/a\u003e\u003c/li\u003e\n    \u003cli\u003e\u003ca href=\"#-appendix-todays-watch-list-update-source-list\"\u003e📚 Appendix: Today\u0026rsquo;s Watch List Update Source List\u003c/a\u003e\n      \u003cul\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=\"#openai-blog-a_full\"\u003eOpenAI Blog (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
}