{
  "title": "2026-09-23 AI Daily | GPT-6 Dual Models Rewrite Inference Costs: Cache Diagnostics, Breakpoint Control, and Half-Price API",
  "url": "https://miaok.ong/en/ai-daily/ai-daily-2026-09-23/",
  "date": "2026-09-23T07:00:00+08:00",
  "lastmod": "2026-09-23T07:00:00+08:00",
  "type": "ai-daily",
  "kind": "page",
  "language": "en",
  "description": "OpenAI has released GPT-6 Sol/Luna, cutting API prices by 50% and unveiling prompt caching diagnostics with breakpoint controls. A case study from Parallel demonstrates that Astra can reduce research time and costs by half. Multiple papers on inference optimization are focusing on sparse attention in long contexts and MoE routing, as the competition in AI services shifts towards cost engineering. Amazon\u0026rsquo;s blocking of Muse highlights the boundary issues in distribution between platforms and AI assistants.",
  "keywords": null,
  "tags": [],
  "categories": [],
  "author": "Mark (Miao) Kong",
  "image": "https://miaok.ong/images/avatar.jpg",
  "content": "\u003ch1 id=\"2026-09-23-ai-daily--gpt-6-dual-models-rewrite-inference-costs-cache-diagnostics-breakpoint-control--half-price-api\"\u003e\n  2026-09-23 AI Daily | GPT-6 Dual Models Rewrite Inference Costs: Cache Diagnostics, Breakpoint Control \u0026amp; Half-Price API\n  \u003ca class=\"heading-link\" href=\"#2026-09-23-ai-daily--gpt-6-dual-models-rewrite-inference-costs-cache-diagnostics-breakpoint-control--half-price-api\"\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 releases GPT-6 Sol/Luna, cutting API prices by 50% and detailing prompt cache diagnostics and breakpoint control. A Parallel case study shows Astra can halve research time and costs. Multiple papers on inference optimization focus on sparse attention for long contexts and MoE routing, shifting the service competition to cost engineering. Amazon\u0026rsquo;s move to block Muse highlights the distribution boundary dispute between platforms and AI assistants.\u003c/p\u003e\n\u003c/blockquote\u003e\n\u003ch2 id=\"-in-depth-guide-to-this-issues-watch-list\"\u003e\n  📖 In-depth Guide to This Issue\u0026rsquo;s Watch List\n  \u003ca class=\"heading-link\" href=\"#-in-depth-guide-to-this-issues-watch-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\u003cp\u003eToday\u0026rsquo;s Watch List features two main threads worth a deep dive. The first is the engineering progress of GPT-6: OpenAI\u0026rsquo;s blog has seen a flurry of updates on the positioning of the Sol/Luna dual models, prompt cache diagnostics, breakpoint control, and a case study from Parallel on using Astra to cut research time and costs in half. Teams focused on inference infrastructure should pay close attention to cache hit rates and cost structures, which represent the real-world efficiency calculus beyond model capabilities. The second is a batch of inference optimization papers emerging on arXiv: RBS-Attention tackles the failure of sparse attention in long-context prefilling, and Attention-Aware Routing enhances MoE routing with attention features. These are an excellent technology radar for teams optimizing large model service performance. On the business side, the discussions around Amazon blocking Muse and the Jevons Paradox are also worth reading to understand platform moats and the boundary conditions for compute demand.\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-copilotkit-launches-openmuse-open-source-personal-ai-assistant\"\u003e\n  Topic 1: CopilotKit Launches OpenMuse Open-Source Personal AI Assistant\n  \u003ca class=\"heading-link\" href=\"#topic-1-copilotkit-launches-openmuse-open-source-personal-ai-assistant\"\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\u003eOverview: Trending for: 3 hours ago, Related posts: 381\u003c/li\u003e\n\u003cli\u003eWhat it is: CopilotKit has launched OpenMuse, an open-source personal AI assistant.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This move introduces the open-source model to the personal AI assistant space, potentially fostering the development of more transparent, customizable, and privacy-focused assistants, and intensifying competition in the field.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: Discussions on X are focused on the capability gap between OpenMuse and existing closed-source assistants, the significance of open source for privacy and customization, and its practical usability and ecosystem prospects.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-2-unreal-agent-open-source-harness-cuts-ai-coding-costs-by-39\"\u003e\n  Topic 2: Unreal Agent Open-Source Harness Cuts AI Coding Costs by 39%\n  \u003ca class=\"heading-link\" href=\"#topic-2-unreal-agent-open-source-harness-cuts-ai-coding-costs-by-39\"\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 · Other\u003c/li\u003e\n\u003cli\u003eOverview: Trending for: , Related posts: 149\u003c/li\u003e\n\u003cli\u003eWhat it is: The open-source project Unreal Agent has released a testing harness that reportedly reduces AI coding costs by 39%.\u003c/li\u003e\n\u003cli\u003eWhy it matters: The framework directly targets the high cost of AI programming agents. If this cost reduction is reproducible, it will improve the economic viability of AI coding and accelerate its adoption in development workflows.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: The focus on X is on the credibility of the 39% cost reduction claim, its applicability across different models and tasks, and the potential impact of an open-source solution on the pricing models of existing commercial AI coding tools.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-3-metas-muse-ai-agent-tops-app-charts-and-boosts-stock\"\u003e\n  Topic 3: Meta\u0026rsquo;s Muse AI Agent Tops App Charts and Boosts Stock\n  \u003ca class=\"heading-link\" href=\"#topic-3-metas-muse-ai-agent-tops-app-charts-and-boosts-stock\"\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\u003eOverview: Trending for: 8 hours ago, Related posts: 4,700\u003c/li\u003e\n\u003cli\u003eWhat it is: Meta\u0026rsquo;s Muse AI agent has topped the app store charts, driving a rise in Meta\u0026rsquo;s stock price.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This indicates that AI agents are becoming a key competitive product for tech giants in the consumer space, with their market performance now directly impacting company valuations and confidence in AI strategy.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: The main topics on X are whether its chart-topping performance is driven by genuine user demand or marketing, whether it can sustain its lead over competitors like ChatGPT and Gemini, and Meta\u0026rsquo;s long-term advantages and risks in AI applications and privacy.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-4-white-circle-open-sources-halo-for-faster-model-training-on-hugging-face\"\u003e\n  Topic 4: White Circle Open-Sources Halo for Faster Model Training on Hugging Face\n  \u003ca class=\"heading-link\" href=\"#topic-4-white-circle-open-sources-halo-for-faster-model-training-on-hugging-face\"\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\u003eOverview: Trending for: 1 day ago, Related posts: 1,000\u003c/li\u003e\n\u003cli\u003eWhat it is: White Circle has open-sourced the Halo framework on Hugging Face, which reportedly accelerates distributed large language model training by up to 2.8 times.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This framework directly addresses the core bottlenecks of cost and time in large model training. If the speedup is confirmed, it would significantly improve computational efficiency and lower the barrier to entry for training.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: Discussions on X are focused on the real-world reproducibility of the 2.8x speedup, its advantages compared to existing frameworks like DeepSpeed, and its applicability across different models and hardware configurations.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-5-ai-guide-makes-product-launch-videos-easy-for-beginners\"\u003e\n  Topic 5: AI Guide Makes Product Launch Videos Easy for Beginners\n  \u003ca class=\"heading-link\" href=\"#topic-5-ai-guide-makes-product-launch-videos-easy-for-beginners\"\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\u003eOverview: Trending for: 16 hours ago, Related posts: 4,200\u003c/li\u003e\n\u003cli\u003eWhat it is: An AI video tool demonstrated how to clone a founder\u0026rsquo;s likeness and voice from a single photo and a short voice clip, automatically generating a product launch video complete with subtitles and a timeline.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This indicates that AI video generation, voice cloning, and automated post-production capabilities have reached a practical stage, significantly lowering the barrier and cost of producing product launch videos.\u003c/li\u003e\n\u003cli\u003eDiscussion overview: On X, there is both amazement at the efficiency of \u0026ldquo;zero-barrier video creation\u0026rdquo; and skepticism about the actual value and homogenization of the vast number of AI tools. Some posts highlight its marketing effectiveness, while others urge a focus on the tangible results the tool produces.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-6-openai-launches-faster-cheaper-gpt-6-sol-and-luna-models\"\u003e\n  Topic 6: OpenAI Launches Faster, Cheaper GPT-6 Sol and Luna Models\n  \u003ca class=\"heading-link\" href=\"#topic-6-openai-launches-faster-cheaper-gpt-6-sol-and-luna-models\"\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\u003eOverview: Trending since: 5 hours ago, Related posts: 34,000\u003c/li\u003e\n\u003cli\u003eWhat it is: OpenAI released the GPT-6 Sol and GPT-6 Luna models, with API prices 50% lower than GPT-5.6, and they are now available on ChatGPT Work and Codex.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This marks a shift in the competitive focus from pure model capabilities to task-level pricing and inference costs. It could significantly lower the adoption barrier for AI agents and continuous inference, intensifying the cost-performance competition with Anthropic and open-source models.\u003c/li\u003e\n\u003cli\u003eDiscussion overview: Opinions on X are sharply divided: one side sees the price reset and decreased inference costs as a major turning point, while the other questions the rapid frequency of model releases and the limited improvements in user experience, sparking discussions about an AI bubble and subscription fatigue. At the same time, users are comparing these releases to competing models launched the same day, such as Claude Opus 5.5, the open-source MiMo-V2.6-Pro, and Grok 4.7.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-7-pexos-mark-to-fix-tool-transforms-ai-video-creation\"\u003e\n  Topic 7: Pexo\u0026rsquo;s Mark to Fix Tool Transforms AI Video Creation\n  \u003ca class=\"heading-link\" href=\"#topic-7-pexos-mark-to-fix-tool-transforms-ai-video-creation\"\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 · Other\u003c/li\u003e\n\u003cli\u003eOverview: Trending since:, Related posts: 235\u003c/li\u003e\n\u003cli\u003eWhat it is: Pexo launched its Mark to Fix tool, which allows users to mark problematic areas in AI-generated videos and apply targeted fixes.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This tool enhances the fine-grained control over AI videos, reduces post-generation manual editing costs, and provides a practical boost to the AI video creation workflow.\u003c/li\u003e\n\u003cli\u003eDiscussion overview: Key discussion points include the accuracy of its local repair feature, its impact on professional video workflows, and whether it offers a distinct advantage compared to other AI video editing tools.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-8-meta-ai-chief-jokes-about-muse-logo-dodging-butthole-comparisons\"\u003e\n  Topic 8: Meta AI Chief Jokes About Muse Logo Dodging Butthole Comparisons\n  \u003ca class=\"heading-link\" href=\"#topic-8-meta-ai-chief-jokes-about-muse-logo-dodging-butthole-comparisons\"\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 · Other\u003c/li\u003e\n\u003cli\u003eOverview: Trending since: 17 hours ago, Related posts: 2,600\u003c/li\u003e\n\u003cli\u003eWhat it is: The head of Meta AI joked that the Muse logo dodged being compared to the shape of a butt/anus.\u003c/li\u003e\n\u003cli\u003eWhy it matters: The topic shows that the visual branding of AI products can become a focal point of public discussion, and the executive\u0026rsquo;s humorous response was seen as a way to defuse the sensitivity of the design controversy.\u003c/li\u003e\n\u003cli\u003eDiscussion overview: The main focus on X is on whether AI logos are prone to vulgar interpretations and how brand design should balance recognizability with avoiding unfortunate associations. Some users feel that this type of discussion distracts from the technology itself.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch4 id=\"todays-ai-public-opinion-summary-on-x\"\u003e\n  Today\u0026rsquo;s AI Public Opinion Summary on X\n  \u003ca class=\"heading-link\" href=\"#todays-ai-public-opinion-summary-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/h4\u003e\n\u003cp\u003eThe main theme of today\u0026rsquo;s AI discussions on X is the concurrent rise of open-source, cost-reducing tools and consumer-grade AI applications. OpenMuse, Unreal Agent, and Halo claim to bring greater transparency or significant efficiency improvements in the domains of personal assistants, coding, and training, respectively. Meanwhile, Meta Muse\u0026rsquo;s rise to the top and OpenAI\u0026rsquo;s price cuts indicate that the competition among tech giants is shifting toward application rankings and task-level costs. The public consensus welcomes AI with lower barriers to entry, greater customizability, and more control, agreeing that AI agents and video generation have reached a practical stage. The main points of contention are whether metrics like \u0026ldquo;39% cost reduction\u0026rdquo; and \u0026ldquo;2.8x speed-up\u0026rdquo; are reproducible, whether Meta\u0026rsquo;s chart-topping performance is due to genuine demand or a marketing push, and whether the rapid pace of releases actually improves user experience. Potential risks include the possibility of open-source and commercial projects exaggerating their effectiveness and eroding trust. AI likeness and voice cloning introduce deepfake and privacy risks. Brand design controversies could also distract from technological discussions, while rapid price reductions and frequent releases heighten fears of an AI bubble and subscription fatigue.\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\u003eNo influencer insights available for 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-source-updates\"\u003e\n  📚 Appendix: Today\u0026rsquo;s Watch List Source Updates\n  \u003ca class=\"heading-link\" href=\"#-appendix-todays-watch-list-source-updates\"\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\u003eTimeframe: Last 3 days; covers 22 sources; 37 total updates\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/blake-scholl-why-plane-speed-stalled-supersonic-commercial-flight-revolutionizing-the-engine\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eBlake Scholl: Why Plane Speed Stalled, Supersonic Commercial Flight, \u0026amp; Revolutionizing the Engine\u003c/a\u003e\u003c/strong\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-09-22 17:00 (Beijing Time)\u003c/li\u003e\n\u003cli\u003eAbstract: - IREN is a vertically integrated AI cloud platform that provides data centers, computing, and software for AI training and inference.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eOracle connects data, applications, and infrastructure to turn AI into business outcomes and deliver optimized flexibility, choice, and control as AI evolves.\u003c/li\u003e\n\u003cli\u003eEY helps tech innovators from startup to exit to large enterprise.\u003c/li\u003e\n\u003cli\u003eWe are committed to giving everyone the tools to reach their full potential and ensuring that all people can benefit from technology.\u003c/li\u003e\n\u003cli\u003eKeel Infrastructure has the power, land, and connectivity for high-performance computing and AI to run, backed by secure energy assets and grid interconnections across North America.\n\u003cul\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003e(0:00) Welcome Blake Scholl\u003c/li\u003e\n\u003cli\u003e(5:11) Solving the sonic boom \u0026amp; legalizing supersonic flight\u003c/li\u003e\n\u003cli\u003e(7:59) The pivot to powering data centers\u003c/li\u003e\n\u003cli\u003e(10:38) Jason joins: data center demand, private jets \u0026amp; what winning looks like\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=\"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/amazon-blocks-muse-amazons-moat-aggregator-v-aggregator/\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eAmazon Blocks Muse, Amazon’s Moat, Aggregator v Aggregator\u003c/a\u003e\u003c/strong\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 18:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary: - Amazon predictably blocked Muse, but there is room for a deal based on the reality that Amazon\u0026rsquo;s physical world investments are an AI moat.\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e$15\u003c/strong\u003e/month \u003cem\u003eor\u003c/em\u003e \u003cem\u003e\u003cstrong\u003e$150\u003c/strong\u003e\u003c/em\u003e/year.\u003c/li\u003e\n\u003cli\u003eSubstantive analysis of the day\u0026rsquo;s news via three weekly emails or podcasts.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eStrategy Interviews\u003c/strong\u003e.\u003c/li\u003e\n\u003cli\u003eInterviews with leading public company CEOs, private company founders, and discussions with fellow analysts.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003eAmazon predictably blocked Muse, but there is room for a deal based on the reality that Amazon\u0026rsquo;s physical world investments are an AI moat.\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/better-prompt-caching-for-gpt-6\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eBetter prompt caching for GPT-6\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-23 05:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary: - Learn how GPT-6 improves prompt caching with higher cache hit rates, new diagnostics, explicit breakpoints, and controls that reduce latency and costs.\n\u003cul\u003e\n\u003cli\u003eThis article from the OpenAI Blog explains how GPT-6\u0026rsquo;s better prompt caching is shaping the broader AI and infrastructure landscape.\u003c/li\u003e\n\u003cli\u003eFollowing GPT-6\u0026rsquo;s better prompt caching, it also has practical implications for founders, operators, and investors.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003eLearn how GPT-6 improves prompt caching with higher cache hit rates, new diagnostics, explicit breakpoints, and controls that reduce latency and costs.\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/introducing-gpt-6-sol-and-luna\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eIntroducing GPT-6 Sol and Luna\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-23 02:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary: - Meet GPT-6 Sol and Luna, two models that bring frontier intelligence to everyday work with different balances of capability and cost.\n\u003cul\u003e\n\u003cli\u003eThis article from the OpenAI Blog explains how the launch of GPT-6 Sol and Luna is shaping the broader AI and infrastructure landscape.\u003c/li\u003e\n\u003cli\u003eFollowing the launch of GPT-6 Sol and Luna, it also has practical implications for founders, operators, and investors.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003eMeet GPT-6 Sol and Luna, two models that bring frontier intelligence to everyday work with different balances of capability and 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\u003e\u003cstrong\u003e\u003ca href=\"https://openai.com/index/parallel-cuts-time-and-cost-with-astra\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eParallel cut research time and cost in half with GPT‑6 Astra\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 20:00 Beijing Time\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eSummary: - GPT-6 Astra allowed Parallel\u0026rsquo;s agents to research and synthesize labor-market data in half the time and at half the cost.\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eCompared to previous models, GPT-6 Astra allowed Parallel\u0026rsquo;s agents to research and synthesize labor market data in half the time and at half the cost.\u003c/li\u003e\n\u003cli\u003eUsing GPT-6 Astra cut research time and cost for Parallel in half.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eEN Highlights:\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eGPT-6 Astra allowed Parallel’s agents to research and synthesize labor-market data in half the time and at half the cost vs\u003c/li\u003e\n\u003cli\u003eprior models.\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/priorities-principles-third-party-assessments\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003ePriorities and principles for effective third party assessments\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-09-22 08:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary: - OpenAI outlines priorities and principles for rigorous, secure, and independent third-party AI safety assessments of frontier models and safeguards.\n\u003cul\u003e\n\u003cli\u003eThis article from the OpenAI blog explains how the priorities and principles for effective third-party assessments shape the broader AI and infrastructure landscape.\u003c/li\u003e\n\u003cli\u003eIt also provides founders, operators, and investors with the practical implications of following the priorities and principles for effective third-party assessments.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003eOpenAI outlines priorities and principles for rigorous, secure, and independent third-party AI safety assessments of frontier models and safeguards.\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=\"two-minute-papers-b_introsearch\"\u003e\n  Two Minute Papers (B_intro+search)\n  \u003ca class=\"heading-link\" href=\"#two-minute-papers-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\u003cstrong\u003e\u003ca href=\"https://www.youtube.com/watch?v=qBBRRsH0rQc\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eYes, Jev Is Insane, But There\u0026rsquo;s A Catch\u003c/a\u003e\u003c/strong\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-09-22 17:32 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary: - ❤️ Check out Lambda here and sign up for their GPU Cloud:.\n\u003cul\u003e\n\u003cli\u003ePrevious papers, implementations, models:.\u003c/li\u003e\n\u003cli\u003eAdam Bridges, B Shang, Carlos Galarza, Christian Ahlin, Eric Tyson, Juan Benet, Lukas Biewald, Michael Tedder, Owen Skarpness, Ryan Stankye, Shawn Becker, Steef, Taras Bobrovytsky, Tazaur Sagenclaw, Tybie Fitzhugh, Ueli Gallizzi.\u003c/li\u003e\n\u003cli\u003eYes, Jev is insane, but there\u0026rsquo;s a catch.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003e❤️ Check out Lambda here and sign up for their GPU Cloud:\u003c/li\u003e\n\u003cli\u003ePrevious papers, implementations, models:\u003c/li\u003e\n\u003cli\u003e🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:\u003c/li\u003e\n\u003cli\u003eAdam Bridges, B Shang, Carlos Galarza, Christian Ahlin, Eric Tyson, Juan Benet, Lukas Biewald, Michael Tedder, Owen Skarpness, Ryan Stankye, Shawn Becker, Steef…\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/2609.20971\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eRBS-Attention: Radius-Bounded Sparse Prefill for Long-Context Large Language Models\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary: - arXiv:2609.20971v1 Announcement type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Long-context large language model inference is increasingly constrained by prefill, where dense self-attention processes the entire prompt before generation begins.\u003c/li\u003e\n\u003cli\u003eSparse block selection can reduce this cost, but block centroids can hide highly relevant tokens among many irrelevant ones.\u003c/li\u003e\n\u003cli\u003eWe call this failure mode \u0026ldquo;mean dilution\u0026rdquo; and propose RBS-Attention, a training-free sparse prefill method with two complementary selection branches.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eEN Highlights:\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003earXiv:2609.20971v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Long-context large language model inference is increasingly limited by prefill, where dense self-attention processes the entire prompt before generati…\u003c/li\u003e\n\u003cli\u003eSparse block selection can reduce this cost, but a block centroid may hide a highly relevant token among many irrelevant ones\u003c/li\u003e\n\u003cli\u003eWe call this failure mode mean dilution and propose RBS-Attention, a training-free sparse-prefill method with two complementary selection branches\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/2609.20974\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eAttention-Aware Routing: Coupling Routing and Attention in MoEs\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.20974v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: In Mixture-of-Experts language models, the router typically utilizes limited contextual information to select and weight experts based on the token\u0026rsquo;s hidden state.\u003c/li\u003e\n\u003cli\u003eWe propose Attention-Aware Routing (AAR), which augments the router with temporal and spectral features extracted from a sliding window of attention weights that represent a summary of the model\u0026rsquo;s contextual state, separate from the hidden state.\u003c/li\u003e\n\u003cli\u003eKeeping the base transformer entirely frozen, we train only the routing parameters, isolating routing as the sole variable.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.20974v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: In Mixture-of-Experts language models, the router typically selects and weights experts based on the token\u0026rsquo;s hidden state, utilizing limited contextua…\u003c/li\u003e\n\u003cli\u003eWe propose Attention-Aware Routing (AAR), which augments the router with temporal and spectral features extracted from a sliding window of attention weights tha…\u003c/li\u003e\n\u003cli\u003eKeeping the base transformer entirely frozen, we train only the routing parameters, isolating routing as the sole variable\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/2609.20981\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eCaLR: Causal Latent Revision for Robust Diffusion Reasoning\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.20981v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Autoregressive (AR) models suffer from the local greediness problem, while diffusion language models (DLMs) often lack the strict causal structure required for reasoning.\u003c/li\u003e\n\u003cli\u003eTo combine the advantages of both and overcome their shortcomings, we propose Causal Latent Revision (CaLR), a framework that reformulates reasoning as constrained latent optimization.\u003c/li\u003e\n\u003cli\u003eBy employing a causal topological matrix (CTM) from an expert model and implicit differentiation, CaLR performs gradient-guided \u0026ldquo;thought revision\u0026rdquo; to enforce logical consistency, thereby enabling dynamic self-correction of intermediate steps during the parallel generation process.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.20981v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Autoregressive (AR) models suffer from local greediness, while diffusion language models (DLMs) often lack the strict causal structure required for re…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eTo combine the advantages and overcome the drawbacks of the dual, we propose Causal Latent Revision (CaLR), a framework that reformulates reasoning as constrain…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eBy adopting a causal topology matrix (CTM) from an expert model and implicit differentiation, CaLR performs gradient-guided ``thought revision\u0026quot; to enforce logic…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.21061\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eLoRA Enhanced Contrastive Learning with SAS Vision Transformers\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.21061v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Automatic target recognition (ATR) with synthetic aperture sonar (SAS) supports advanced naval capabilities, but deep learning is constrained by the scarcity of target images, background clutter, and human-in-the-loop evaluation.\u003c/li\u003e\n\u003cli\u003eWe adapt DINOv3 Vision Transformer (ViT) models to underwater SAS ATR using a three-stage parameter-efficient framework.\u003c/li\u003e\n\u003cli\u003eStage 1 uses Low-Rank Adaptation (LoRA) while freezing the ViT backbone, bridging the gap between natural-image pretraining and underwater acoustic propagation.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.21061v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Automatic target recognition (ATR) with synthetic aperture sonar (SAS) supports advanced naval capabilities, but deep learning is constrained by scarc…\u003c/li\u003e\n\u003cli\u003eWe adapt DINOv3 Vision Transformer (ViT) models to underwater SAS ATR using a three-stage parameter-efficient framework\u003c/li\u003e\n\u003cli\u003eStage 1 uses Low-Rank Adaptation (LoRA) while freezing the ViT backbone, bridging the gap between natural-image pretraining and underwater acoustic propagation\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/2609.21096\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eDetecting Hallucination in LLMs: Tracing the Topological Signatures of Impaired Context Sharing\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.21096v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: In this work, we examine the topology of information flow patterns within attention graphs to effectively distinguish between hallucinated and non-hallucinated responses.\u003c/li\u003e\n\u003cli\u003eWe analyze the Forman-Ricci curvature to identify structural patterns that indicate information bottlenecks in attention graphs.\u003c/li\u003e\n\u003cli\u003eWe then introduce a method to capture semi-local and global information flow features of attention heads associated with hallucinatory responses.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.21096v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: In this work, we examine the topology of information flow patterns within attention graphs to effectively distinguish hallucinated from non-hallucinat…\u003c/li\u003e\n\u003cli\u003eWe analyze the Forman-Ricci curvature to identify structural patterns indicating information bottlenecks in attention graphs\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 then introduce a method that captures both semi-local and global information-flow characteristics of attention heads associated with hallucinated responses\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.21113\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eDecoupling Internal Representational Changes and Causal Importance in Fine-Tuned Large Language Models\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eRelease Date: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.21113v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Fine-tuning has emerged as a widely adopted approach for adapting LLMs to a variety of downstream tasks.\u003c/li\u003e\n\u003cli\u003eHowever, how it reshapes their internal mechanisms remains poorly understood.\u003c/li\u003e\n\u003cli\u003eTo address this, we investigate how fine-tuning alters internal representations in LLMs, including attention patterns and layer-wise activations, and examine whether these changes are related to EAP-identified task-relevant components (e.g., attention heads and logic-level activations), thereby driving task performance.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.21113v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Fine-tuning has emerged as a widely adopted approach for adapting LLMs to a variety of downstream tasks\u003c/li\u003e\n\u003cli\u003eHowever, how it reshapes their internal mechanisms remains poorly understood\u003c/li\u003e\n\u003cli\u003eTo address this, we investigate how fine-tuning alters internal representations in LLMs, including attention patterns and layer-wise activations, and examine wh…\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/2609.21139\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eTinyCeNN-LM: Quality-Gated Conversion of Pretrained Attention with CeNN-Inspired Cellular-Recurrent Layers\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eRelease Date: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.21139v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Replacing attention in a pretrained language model is a compatibility problem: a plausible substitute may alter representations expected by later layers.\u003c/li\u003e\n\u003cli\u003eTinyCeNN-LM introduces a \\emph{quality-gated post-training conversion} framework using CeNN-inspired cellular-recurrent layers with bounded local processing, compact recurrent memory, routing, fusion, and acceptance or rollback validation.\u003c/li\u003e\n\u003cli\u003eThree implementations are studied: Integrated Memory, MemoryFusion, and PDelta3-GDN2-CLVR+Local32.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.21139v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Replacing attention in a pretrained language model is a compatibility problem: a plausible substitute may alter representations expected by later laye…\u003c/li\u003e\n\u003cli\u003eTinyCeNN-LM introduces a \\emph{quality-gated post-training conversion} framework using CeNN-inspired cellular-recurrent layers with bounded local processing, co…\u003c/li\u003e\n\u003cli\u003eThree implementations are studied: Integrated Memory, MemoryFusion, and PDelta3-GDN2-CLVR+Local32\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/2609.21149\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eClinician-Grounded Quality Assurance for AI-Assisted Psychiatric Intake\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eAbstract: - arXiv:2609.21149v1 Announcement Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Before patients can use AI-assisted psychiatric intake systems, health systems need practical methods to regularly evaluate these tools against their quality-assured clinical standards.\u003c/li\u003e\n\u003cli\u003eBecause clinicians may use different intake styles, the evaluation for this task must (1) support comparison across different interview methods, (2) minimize the burden on clinicians, and (3) measure clinically relevant performance for the health systems deploying these technologies.\u003c/li\u003e\n\u003cli\u003eWe propose a clinician-grounded evaluation platform built around a memory-augmented patient simulator for open-ended AI interviewing, InterviewPlayground.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.21149v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Before patients can use AI-assisted psychiatric intake systems, health systems need practical ways to routinely evaluate these tools against their cli…\u003c/li\u003e\n\u003cli\u003eBecause clinicians may use different intake styles, evaluation for this task must (1) support comparison across interviewing approaches, (2) minimize clinician…\u003c/li\u003e\n\u003cli\u003eWe present a clinician-grounded evaluation platform built around a memory-augmented patient simulator for open-ended AI interviewing, InterviewPlayground\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/2609.21157\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eCan Agents Design Better Chips with a Higher Level Abstraction?\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.21157v1 Announcement Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Large Language Model (LLM) agents are increasingly being explored for chip design, but most existing methods operate directly on RTL.\u003c/li\u003e\n\u003cli\u003eWe ask whether agents can design better chips by leveraging higher-level abstractions.\u003c/li\u003e\n\u003cli\u003eWe compare direct RTL design, agent-based HLS design, post-compiler HLS refinement, and post-HLS RTL refinement, and combine agent-based HLS design with post-HLS RTL refinement as Agent-based HLS with RTL Refinement (AHRR).\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.21157v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Large Language Model (LLM) agents are increasingly being explored for chip design, but most existing approaches operate directly at RTL\u003c/li\u003e\n\u003cli\u003eWe ask whether agents can design better chips by leveraging higher-level abstractions\u003c/li\u003e\n\u003cli\u003eWe compare Direct RTL Design, Agent-based HLS Design, Post-Compiler HLS Refinement, and Post-HLS RTL Refinement, and combine Agent-based HLS Design with Post-HL…\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/2609.21165\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eSpecOpt: Contact-Diff Reasoning for Agentic Molecule Optimization Toward Binding Specificity\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.21165v1 Announcement Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Off-target protein binding is a major source of adverse reactions for small molecule drugs, but most structure-based molecular design methods focus on de novo generation of selective compounds rather than improving the selectivity of existing, well-characterized drugs.\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 introduce specificity optimization (SpecOpt), a molecular design task that seeks to make limited structural modifications to existing compounds to increase their binding preference for the intended target over known off-targets, while preserving their structural and drug-like properties.\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eTo enable systematic evaluation, we constructed a ChEMBL-derived benchmark based on compound-target interaction data, identifying intended targets through curated drug mechanism-of-action annotations and off-targets through measured activity.\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.21165v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Off-target protein binding is a major source of adverse effects for small-molecule drugs, yet most structure-based molecular design methods focus on g…\u003c/li\u003e\n\u003cli\u003eWe introduce specificity optimization (SpecOpt), a molecular design task that seeks constrained structural modifications to an existing compound that increase i…\u003c/li\u003e\n\u003cli\u003eTo enable systematic evaluation, we construct a ChEMBL-derived benchmark from compound-target interaction data, identifying intended targets through curated dru…\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/2609.22090\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eRecognition, Simulation, and Refusal: A Contamination-Aware Study of Classic Psychological Effects in LLM Agents\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22090v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: An LLM that produces response patterns associated with human psychological effects is different from an LLM that possesses such biases.\u003c/li\u003e\n\u003cli\u003eWe present PsyAgentBench, a benchmark that re-runs classic psychology experiments on LLM agents, constructed to separate these experiments: each paradigm is run either with the paradigm explicitly labeled in the prompt (named) or as a regular task (blind), and on either a literal textbook version of the task (canonical) or a structurally matched variant to reduce lexical and scenario overlap with possible training data (counterfactual), crossed with role-playing manipulations.\u003c/li\u003e\n\u003cli\u003eAcross five completed paradigms, evaluated on up to three open-weight model families with 41,904 trials released, apparently human-like effects arise through qualitatively different pathways rather than a single sensitivity: gating by explicitly-covered paradigm labels (Asch conformity, 0% blind to 83.3% named on gpt-oss-120B), knowledge-dependent signal-dependence (anchoring, with near-zero precision on fact-based vs. near-total on invented quantities, a pattern equally consistent with rational use of the only available signal), amplification of new content under a label (framing), robust absence (sunk cost), and safety-mediated choices where refusal is itself the main finding (minimal group assignment).\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22090v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: An LLM producing the response pattern associated with a human psychological effect is not the same claim as the LLM possessing that bias\u003c/li\u003e\n\u003cli\u003eWe present PsyAgentBench, a benchmark that re-runs classic psychology experiments on LLM agents under a factorial design built to separate these: each paradigm…\u003c/li\u003e\n\u003cli\u003eAcross five completed paradigms, evaluated on up to three open-weight model families with 41,904 trials released, apparently human-like effects arise through qu…\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/2609.22091\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eMemory That Looks Forward: A Zero-Inference Prospective Term for Personal Memory Retrieval\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22091v1 Announcement Type: new.\u003c/li\u003e\n\u003cli\u003eAbstract: Retrieval over a personal memory store is retrospective: it surfaces content resembling the query and is blind to actions the user has committed to perform.\u003c/li\u003e\n\u003cli\u003eWe describe a prospective term for memory retrieval that requires no inference at query time.\u003c/li\u003e\n\u003cli\u003eCommitments are held in an explicit ledger as dated or trigger-conditioned entries; memory items linked to a triggered entry receive a salience boost and are multiplicatively blended into embedding-based retrieval so that relevance remains sovereign.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22091v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Retrieval over a personal memory store is retrospective: it surfaces what resembles the query, and it is blind to what the user has committed to do\u003c/li\u003e\n\u003cli\u003eWe describe a prospective term for memory retrieval that costs no inference at query time\u003c/li\u003e\n\u003cli\u003eCommitments are held in an explicit ledger as dated or trigger-conditioned entries; memory items linked to a firing entry receive a salience boost, blended mult…\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/2609.22094\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eSummarize, Judge, Refine: Decoupled Content Understanding and Policy Learning for Multimodal Content Moderation\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22094v1 Announcement Type: new.\u003c/li\u003e\n\u003cli\u003eAbstract: Content moderation systems traditionally entangle multimodal understanding with policy-specific classification, requiring full pipeline retraining for every policy change and suffering from label scarcity due to the inability to meaningfully augment multimedia.\u003c/li\u003e\n\u003cli\u003eWe propose Summarize-Judge-Refine (SJR), a two-model architecture that decouples these concerns via a natural language interface: a multimodal content model generates structured text summaries, and a text-only policy model classifies them according to policy definitions.\u003c/li\u003e\n\u003cli\u003eAn iterative co-training loop refines the content model via GRPO to generate policy-relevant summaries, while text-space augmentation generates adversarial summary variants (an augmentation path not achievable on raw multimedia), enabling few-shot policy guidance.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22094v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Content moderation systems traditionally entangle multimodal understanding with policy-specific classification, requiring full pipeline retraining for…\u003c/li\u003e\n\u003cli\u003eWe propose Summarize-Judge-Refine (SJR), a two-model architecture that decouples these concerns via a natural language interface: a multimodal Content Model pro…\u003c/li\u003e\n\u003cli\u003eAn iterative co-training loop refines the Content Model via GRPO to produce policy-relevant summaries, while text-space augmentation generates adversarial summa…\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/2609.22096\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eAI-inferred expressed well-being and collective-action discourse in climate-change campaigns on X\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22096v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Climate campaigns are often evaluated through attention and mobilization, but less is known about the well-being language that accompanies them.\u003c/li\u003e\n\u003cli\u003eWhether campaign periods alter positive affect and hope, and whether happiness aligns with action language, remains unresolved.\u003c/li\u003e\n\u003cli\u003eWe analysed 364,118 public Twitter/X posts from Earth Day, Earth Hour, Global Climate Action Day and World Environment Day in 19 occurrence-years, using 30-day pre-event, event, and post-event windows.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22096v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Climate campaigns are often evaluated through attention and mobilization, but less is known about the well-being language that accompanies them\u003c/li\u003e\n\u003cli\u003eWhether campaign periods alter positive affect and hope, and whether happiness aligns with action language, remains unresolved\u003c/li\u003e\n\u003cli\u003eWe analysed 364,118 public Twitter/X posts from Earth Day, Earth Hour, Global Climate Action Day and World Environment Day in 19 occurrence-years, using 30-day…\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/2609.22097\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eToken Signatures of Code: Comparing Coding Behaviors Across Large Language Models\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22097v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: The evaluation of large language models (LLMs) on coding tasks has primarily focused on performance metrics such as pass@k.\u003c/li\u003e\n\u003cli\u003eAs LLMs continue to advance, many models now meet baseline performance requirements, reducing the discriminative power of performance-based evaluation alone.\u003c/li\u003e\n\u003cli\u003eYet a key question remains largely unexplored: how do LLMs differ in their coding behavior?\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22097v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: The evaluation of large language models (LLMs) on coding tasks has primarily focused on performance metrics such as pass @projects/huanxi-teams-ultra/requirements/BACKLOG.json\u003c/li\u003e\n\u003cli\u003eAs LLMs continue to advance, many models now meet baseline performance requirements, reducing the discriminative power of performance-based evaluation alone\u003c/li\u003e\n\u003cli\u003eYet a key question remains largely unexplored: how do LLMs differ in their coding behavior\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/2609.22098\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eTreeSpark: Calibrated, Load-Adaptive Draft Trees for Semi-Autoregressive Speculative Decoding\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22098v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Speculative decoding accelerates language model inference by having an inexpensive drafter propose tokens that the target model validates in parallel.\u003c/li\u003e\n\u003cli\u003eRecent block drafters make drafting nearly free: a single backbone pass emits an entire block of draft tokens.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eDraft trees promise further gains—verifying several alternative continuations in one target—but existing structures rank candidates by per-position edges, ignoring the parents from which candidates extend. Consequently, in semi-autoregressive drafters, wider trees mostly add misranked nodes, and fixed-size trees ignore how much speculation can be supported per decoding round and per serving load.\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22098v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Speculative decoding accelerates language-model inference by letting a cheap drafter propose tokens that the target model verifies in parallel\u003c/li\u003e\n\u003cli\u003eRecent block drafters make drafting nearly free: a single backbone pass emits an entire block of draft tokens\u003c/li\u003e\n\u003cli\u003eDraft trees promise a further gain \u0026ndash; several alternative continuations verified in one target forward \u0026ndash; but existing constructions rank candidates by per-posi…\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/2609.22099\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eA framework for recipe data structure with applications for culinary and nutritional insights\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eRelease Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22099v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Cooking is a complex process that transforms raw ingredients into delicious and nutritious dishes, yet the recipes that encode this process remain largely free-text; readable by people but not directly computable.\u003c/li\u003e\n\u003cli\u003eExisting recipe collections capture fragments of this information, but no shared representation links a recipe\u0026rsquo;s structured ingredient composition, its geo-cultural origin, and its nutritional profile in a single queryable schema.\u003c/li\u003e\n\u003cli\u003eWe address this representation gap by formalizing a framework for recipe data structure that decomposes each recipe into typed ingredient entities, grounds those entities in a reference nutritional database, and annotates them with geo-cultural and dietary context.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22099v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Cooking is a complex process that transforms raw ingredients into delicious and nutritious dishes, yet the recipes that encode this process remain lar…\u003c/li\u003e\n\u003cli\u003eExisting recipe collections capture fragments of this information, but no shared representation links a recipe\u0026rsquo;s structured ingredient composition, its geo-cult…\u003c/li\u003e\n\u003cli\u003eWe address this representation gap by formalizing a framework for recipe data structure that decomposes each recipe into typed ingredient entities, grounds thos…\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/2609.22100\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eAdaMem: Adaptive Memory Token Allocation for Soft Compression in Retrieval-Augmented Generation\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eRelease Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22100v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Retrieval-Augmented Generation (RAG) improves language models with retrieved evidence, but processing many long passages is costly and can introduce distracting information.\u003c/li\u003e\n\u003cli\u003eSoft compression addresses this challenge by encoding passages into a compact sequence of continuous memory embeddings prior to generation.\u003c/li\u003e\n\u003cli\u003eHowever, existing methods typically allocate the same number of memory embeddings for each retained passage, without considering its query-specific relevance.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003earXiv:2609.22100v1 Announce Type: new\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eAbstract: Retrieval-augmented generation (RAG) improves language models with retrieved evidence, but processing many long passages is costly and can introduce d…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eSoft compression addresses this challenge by encoding passages as compact sequences of continuous memory embeddings before generation\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eHowever, existing methods typically assign each retained passage an identical number of memory embeddings, irrespective of its query-specific relevance\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.22101\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eContext Poisoning as Extreme-Value Attention Interference in Long-Context Language Models\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eRelease Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22101v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Large language models can process increasingly long prompts, but their ability to locate and use decisive evidence may be diminished as irrelevant or confusing context is added.\u003c/li\u003e\n\u003cli\u003eWe formulate this phenomenon, which we call context poisoning, as extreme-value interference in attention: the score for decisive evidence is upper-bounded, while the maximum score of effective distractors grows with their number.\u003c/li\u003e\n\u003cli\u003eUnder a softmax retrieval abstraction, we derive a finite-sample upper bound showing that maintaining a fixed accuracy target above the base rate requires the evidence margin to scale as $\\Omega(\\sqrt{\\log N})$, where N represents the effective distractor count, not necessarily the original context length.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22101v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Large language models can process increasingly long prompts, yet their ability to locate and use decisive evidence may degrade as irrelevant or confus…\u003c/li\u003e\n\u003cli\u003eWe formulate this phenomenon, which we call context poisoning, as extreme-value interference in attention: the decisive-evidence score is upper-bounded, while t…\u003c/li\u003e\n\u003cli\u003eUnder a softmax retrieval abstraction, we derive a finite-sample upper bound showing that maintaining a fixed accuracy target above base rate requires the evide…\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/2609.22104\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eDeepInstructor: An Agentic AI Instructor for Experience-Driven Idea Evaluation\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eRelease Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22104v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: With advancements in automated scientific discovery, Large Language Models (LLMs) can now generate research ideas at an unprecedented scale, shifting the bottleneck from idea generation to idea evaluation.\u003c/li\u003e\n\u003cli\u003eExisting evaluators primarily rely on parameterized LLM knowledge or unstructured retrieval, producing judgments that lack the experience-based reasoning used by human instructors.\u003c/li\u003e\n\u003cli\u003eTo address this, we propose DeepInstructor, an agentic framework that formulates idea evaluation as reasoning over structured academic experience.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22104v1 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: As automated scientific discovery advances, Large Language Models (LLMs) can now generate research ideas at an unprecedented scale, shifting the bottl…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eExisting evaluators mainly rely on parametric LLM knowledge or unstructured retrieval, producing judgments that lack the experience-grounded reasoning used by h…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eTo address this, we propose DeepInstructor, an agentic framework that formulates idea evaluation as reasoning over structured scholarly experience\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/2609.22106\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003ePRQuant: Permutation Residual Quantization for Low-Overhead Inference\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22106v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: The accuracy of low-bit quantization for linear layers is often determined by a small number of outliers.\u003c/li\u003e\n\u003cli\u003eWhile existing methods (e.g., smoothing, rotation, or residual-based approaches) can mitigate this issue, they often introduce new accuracy bottlenecks for the weights.\u003c/li\u003e\n\u003cli\u003eFurthermore, most of these techniques are implemented as online methods, which can lead to significant execution overhead.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22106v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Accuracy of Low-bit quantization of linear layers is often dominated by a small number of outliers\u003c/li\u003e\n\u003cli\u003eAlthough existing methods, such as smoothing, rotation, or residual-based approaches, may mitigate this problem, they often introduce new accuracy bottlenecks t…\u003c/li\u003e\n\u003cli\u003eBesides, most of these techniques are implemented as online approaches, which can result in heavy execution overheads\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/2609.22107\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eGeneralized Multimodal Foundation Model\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22107v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Utilizing multimodal data for prediction is widely applied in various scenarios.\u003c/li\u003e\n\u003cli\u003eExisting multimodal fusion models, once deployed, can only handle predefined modalities (e.g., vision, text, and audio) and single tasks, making it difficult to quickly adapt to new downstream applications.\u003c/li\u003e\n\u003cli\u003eThis raises a natural yet rather radical question: does a universal multimodal fusion model exist that can be applied to any combination of modalities and any prediction task.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22107v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Making prediction with multimodal data is widely used in diverse scenarios\u003c/li\u003e\n\u003cli\u003eExisting multimodal fusion models, once deployed, can only handle predefined modalities (e.g., vision, text and audio) and single tasks, making it difficult to…\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eTherefore, a natural yet rather aggressive question arises, whether there exists a general multimodal fusion model that can be applied to arbitrary modality com…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.22108\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eCorrecting Learning-based Perception for Safety\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22108v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Learning-enabled perception is important in many autonomous systems.\u003c/li\u003e\n\u003cli\u003eUnlike traditional sensors, the boundary where machine learning perception works or does not work is difficult to characterize.\u003c/li\u003e\n\u003cli\u003eIncorrect perception can lead to unsafe or overly conservative downstream control actions.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22108v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Learning-enabled perception is important in many autonomous systems\u003c/li\u003e\n\u003cli\u003eUnlike traditional sensors, the boundary where ML perception does or does not work is poorly characterized\u003c/li\u003e\n\u003cli\u003eIncorrect perception can lead to unsafe or overtly conservative downstream control actions\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/2609.22109\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eA Shared Learning Rate Is Not a Neutral Control in Selective On-Policy Distillation\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22109v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Selective on-policy distillation trains a student only at the token positions where the selector scores highest, and the literature compares selectors under a single shared learning rate (chosen as a neutral control).\u003c/li\u003e\n\u003cli\u003eUnder LoRA on GSM8K (Qwen2.5-1.5B student, 7B teacher), across an 8x learning rate grid, dense supervision is statistically flat (swing 1.8 pp, p=0.26), whereas each selective arm\u0026rsquo;s performance varies: a random 5% subset by 5.4 pp, the total variation selector by 6.7 pp, and the teachability selector by up to 17.7 pp.\u003c/li\u003e\n\u003cli\u003eTherefore, the verdict between dense and selective supervision is 10.1 pp at lr=1e-4 but 5.1 pp at 5e-5, a 2.0x difference determined by a parameter the protocol treats as incidental, and two of the six pairwise significance calls between selectors flip between adjacent rates without any rank reversal.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22109v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Selective on-policy distillation trains a student only at the token positions a selector scores highest, and the literature compares selectors under a…\u003c/li\u003e\n\u003cli\u003eWe show it is not\u003c/li\u003e\n\u003cli\u003eUnder LoRA on GSM8K (Qwen2.5-1.5B student, 7B teacher), across an 8x learning-rate grid, dense supervision is statistically flat (swing 1.8 pp, p=0.26) while ev…\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/2609.22113\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eToward Fairness in Machine Learning Models for Predicting Treatment Retention and Premature Discontinuation in Medication for Opioid Use Disorder\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22113v1 Announcement Type: New.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eAbstract: Persistent low retention and completion rates in medications for opioid use disorder (MOUD) have driven the use of machine learning (ML) models to predict retention and identify patients at risk for premature discontinuation.\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eHowever, the fairness of these models across patient populations remains largely unexplored, raising concerns about their application in treatment decision support.\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eThis study systematically assesses algorithmic fairness in ML models for predicting MOUD retention and premature discontinuation and investigates the effectiveness of bias mitigation techniques.\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eEN Highlights:\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22113v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Persistent low retention and completion rates in medications for opioid use disorder (MOUD) have driven the use of machine learning (ML) models to pre…\u003c/li\u003e\n\u003cli\u003eHowever, the fairness of these models across patient populations remains largely unexplored, raising concerns about their application in treatment decision supp…\u003c/li\u003e\n\u003cli\u003eThis study systematically assesses algorithmic fairness in ML models for predicting MOUD retention and premature discontinuation and investigates the effectiven…\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/2609.22115\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eZoAQ: Adaptive Zeroth-Order Querying via Query-Reuse Coupling\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22115v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Zeroth-order optimization (ZOO) estimates updates from function evaluations, making perturbation queries a primary cost.\u003c/li\u003e\n\u003cli\u003eFixed budgets spend the same number of queries at every step, while adaptive controllers may offset their savings by using additional oracle calls to test the reliability of the estimator.\u003c/li\u003e\n\u003cli\u003eWe introduce ZoAQ, an adaptive ZOO method built around query reuse.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22115v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Zeroth-order optimization (ZOO) estimates updates from function evaluations, making perturbation queries a primary cost\u003c/li\u003e\n\u003cli\u003eFixed budgets spend the same number of queries at every step, while adaptive controllers may offset their savings by using additional oracle calls to test estim…\u003c/li\u003e\n\u003cli\u003eWe introduce ZoAQ, an adaptive ZOO method built around query reuse\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/2609.22117\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eLE4Mob: Towards Inductive, Distance-Aware and General-Purpose Location Embedding for Human Mobility Modelling\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22117v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Location representations provide mobility models with essential information about spatial locations, functional characteristics, and the relationships between places.\u003c/li\u003e\n\u003cli\u003eHowever, existing embeddings often rely on mobility observations, are unable to represent unseen locations, and have weak constraints for preserving geographical distance.\u003c/li\u003e\n\u003cli\u003eThis limits their reuse across datasets and mobility tasks.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22117v1 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: Location representations provide mobility models with fundamental information about the spatial position, functional characteristics, and relationship…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eHowever, existing embeddings are often dependent on mobility observations, unable to represent unseen locations, and weakly constrained to retain geographic dis…\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eThis limits their reuse across datasets and mobility tasks\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2609.22120\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eSuccess Leaves Detours: Learning Executable Walkthroughs for Long-Horizon Agents\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublish Date: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22120v1 Announce Type: new.\u003c/li\u003e\n\u003cli\u003eAbstract: Test-time self-evolving agents improve by reusing past experience, yet sparse-reward trajectories contain failures, loops, and detours, while summaries often ignore the state conditions and operational dependencies required for execution.\u003c/li\u003e\n\u003cli\u003eWe study executable walkthrough induction from sparse-reward trajectories: extracting compact, state-conditioned, and verifiable procedures.\u003c/li\u003e\n\u003cli\u003eOur main observation is that delayed credit identifies actions associated with progress but cannot determine whether they produce facts required by subsequent actions.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22120v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Test-time self-evolving agents improve by reusing past experience, yet sparse-reward trajectories contain failures, loops, and detours, while summarie…\u003c/li\u003e\n\u003cli\u003eWe study executable Walkthrough induction from sparse-reward trajectories: extracting compact, state-conditioned, and verifiable procedures\u003c/li\u003e\n\u003cli\u003eOur key observation is that delayed credit identifies actions associated with progress but cannot determine whether they produce facts required by later actions\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/2609.22121\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eModelling daily activity patterns from mobile phone location data via deep representation learning\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublish Date: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22121v1 Announce Type: new.\u003c/li\u003e\n\u003cli\u003eAbstract: Passively collected mobile phone location data provide large-scale, longitudinal observations of human mobility but do not directly reveal activity purpose.\u003c/li\u003e\n\u003cli\u003eFunctional characteristics of visited locations provide useful contextual information, but their relationship to activity purpose remains uncertain, especially in mixed-use urban environments.\u003c/li\u003e\n\u003cli\u003eWe conceptualize activity pattern mining as an integrated process of representation, clustering, and explanation, and propose the Activity Chain Encoder (ACE) for the representation phase.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22121v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Passively collected mobile phone location data provide large-scale, longitudinal observations of human mobility but do not directly reveal activity pu…\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/2609.22122\"  class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eRank Portability Does Not Imply Feasibility Portability: Target-Specific Evaluation of Joint Hardware Constraints\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-09-22 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2609.22122v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Cross-device hardware evaluation often assumes that if architecture rankings transfer across devices, a proxy device can support target-side model selection.\u003c/li\u003e\n\u003cli\u003eWe stress-test this assumption to validate the joint latency-energy feasibility for two public architecture families.\u003c/li\u003e\n\u003cli\u003eOn NAS-Bench-201, cross-device rank correlations are moderate, while target-comparable feasible-set overlap remains incomplete.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2609.22122v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Cross-device hardware evaluation often assumes that if architecture rankings transfer across devices, a proxy device can support target-side model sel…\u003c/li\u003e\n\u003cli\u003eWe stress-test this assumption for joint latency-energy feasibility across two public architecture families\u003c/li\u003e\n\u003cli\u003eOn NAS-Bench-201, cross-device rank correlations are moderate, while target-comparable feasible-set overlap remains incomplete\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": 7199,
  "readingTime": 34,
  "tableOfContents": "\u003cnav id=\"TableOfContents\"\u003e\n  \u003cul\u003e\n    \u003cli\u003e\u003ca href=\"#-in-depth-guide-to-this-issues-watch-list\"\u003e📖 In-depth Guide to This Issue\u0026rsquo;s Watch List\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-copilotkit-launches-openmuse-open-source-personal-ai-assistant\"\u003eTopic 1: CopilotKit Launches OpenMuse Open-Source Personal AI Assistant\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-2-unreal-agent-open-source-harness-cuts-ai-coding-costs-by-39\"\u003eTopic 2: Unreal Agent Open-Source Harness Cuts AI Coding Costs by 39%\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-3-metas-muse-ai-agent-tops-app-charts-and-boosts-stock\"\u003eTopic 3: Meta\u0026rsquo;s Muse AI Agent Tops App Charts and Boosts Stock\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-4-white-circle-open-sources-halo-for-faster-model-training-on-hugging-face\"\u003eTopic 4: White Circle Open-Sources Halo for Faster Model Training on Hugging Face\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-5-ai-guide-makes-product-launch-videos-easy-for-beginners\"\u003eTopic 5: AI Guide Makes Product Launch Videos Easy for Beginners\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-6-openai-launches-faster-cheaper-gpt-6-sol-and-luna-models\"\u003eTopic 6: OpenAI Launches Faster, Cheaper GPT-6 Sol and Luna Models\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-7-pexos-mark-to-fix-tool-transforms-ai-video-creation\"\u003eTopic 7: Pexo\u0026rsquo;s Mark to Fix Tool Transforms AI Video Creation\u003c/a\u003e\u003c/li\u003e\n        \u003cli\u003e\u003ca href=\"#topic-8-meta-ai-chief-jokes-about-muse-logo-dodging-butthole-comparisons\"\u003eTopic 8: Meta AI Chief Jokes About Muse Logo Dodging Butthole Comparisons\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-source-updates\"\u003e📚 Appendix: Today\u0026rsquo;s Watch List Source Updates\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=\"#two-minute-papers-b_introsearch\"\u003eTwo Minute Papers (B_intro+search)\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
}