{ "title": "2026-08-20 AI Daily | OpenAI Strengthens Data Boundaries, Replit Brings Cheaper Advanced Intelligence into Development Workflow", "url": "https://miaok.ong/en/ai-daily/ai-daily-2026-08-20/", "date": "2026-08-20T07:00:00+08:00", "lastmod": "2026-08-20T07:00:00+08:00", "type": "ai-daily", "kind": "page", "language": "en", "description": "OpenAI offers zero data retention for eligible frontier model customers and previews private, secure processing for multi-turn tasks; Replit is lowering the barrier to software creation with GPT-5.6 Luna. The signal today is clear: AI is evolving from merely \u0026ldquo;answering questions\u0026rdquo; to being controllable, auditable, and ready for production workflows. Simultaneously, research into clinical trial programming and runtime governance is raising the bar for real-world implementation.", "keywords": null, "tags": [], "categories": [], "author": "Mark (Miao) Kong", "image": "https://miaok.ong/images/avatar.jpg", "content": "\u003ch1 id=\"2026-08-20-ai-daily--openai-strengthens-data-boundaries-replit-brings-cheaper-advanced-intelligence-into-the-development-workflow\"\u003e\n 2026-08-20 AI Daily | OpenAI Strengthens Data Boundaries, Replit Brings Cheaper, Advanced Intelligence into the Development Workflow\n \u003ca class=\"heading-link\" href=\"#2026-08-20-ai-daily--openai-strengthens-data-boundaries-replit-brings-cheaper-advanced-intelligence-into-the-development-workflow\"\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 is offering zero data retention for eligible frontier model customers and previewing private, secure processing for multi-turn tasks, while Replit is lowering the barrier to software creation with GPT-5.6 Luna. The signal today is clear: AI is evolving from \u0026ldquo;being able to answer\u0026rdquo; to becoming \u0026ldquo;controllable, auditable, and production-ready,\u0026rdquo; as research in clinical trial programming and runtime governance simultaneously raises the standards for implementation.\u003c/p\u003e\n\u003c/blockquote\u003e\n\u003ch2 id=\"-this-issues-watch-list-a-deep-dive\"\u003e\n 馃摉 This Issue\u0026rsquo;s Watch List: A Deep Dive\n \u003ca class=\"heading-link\" href=\"#-this-issues-watch-list-a-deep-dive\"\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\u003eThere are three key threads worth reading today. First, model commercialization and platform boundaries: When you consider OpenAI\u0026rsquo;s zero data retention promise and Replit embedding cheaper, advanced intelligence into the software creation workflow with GPT-5.6 Luna, alongside DeepSeek\u0026rsquo;s ongoing challenge to the \u0026ldquo;closed-source, high-price\u0026rdquo; narrative, it\u0026rsquo;s clear that model capabilities are rapidly being commoditized. Second, security and controllability: Safe RAG, MLLM decision-making under uncertainty, and risk identification in multi-turn tasks all remind us that a truly production-ready system is not just a model that can answer questions, but one that knows when to refuse an answer and when to isolate evidence. Third, healthcare and compliant implementation: From clinical trial programming to PHI de-identification and ICD prediction, LLMs are entering high-risk workflows, but the demands for evaluation and auditing are rising in parallel.\u003c/p\u003e\n\u003ch2 id=\"-top-ai-news-on-x\"\u003e\n 馃寪 Top AI News on X\n \u003ca class=\"heading-link\" href=\"#-top-ai-news-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-1study-reveals-trade-offs-in-ai-vibe-coding-tools\"\u003e\n Topic 1:Study Reveals Trade-Offs in AI Vibe Coding Tools\n \u003ca class=\"heading-link\" href=\"#topic-1study-reveals-trade-offs-in-ai-vibe-coding-tools\"\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 路 News\u003c/li\u003e\n\u003cli\u003eSummary: Trending for: 23 hours ago, Related posts: 268\u003c/li\u003e\n\u003cli\u003eWhat happened: A study on AI \u0026ldquo;vibe coding\u0026rdquo; tools has drawn attention, pointing out that while these tools can rapidly generate applications and lower the barrier to coding, they come with trade-offs in controllability, stability, and consistency.\u003c/li\u003e\n\u003cli\u003eWhy it\u0026rsquo;s important: This is significant for the AI field as it directly concerns whether AI programming tools can transition from \u0026ldquo;demo-ready\u0026rdquo; to \u0026ldquo;production-ready for the long term,\u0026rdquo; impacting development efficiency, code quality, product security, and enterprise adoption decisions.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: Discussions on X are mainly divided: one side believes tools like Cursor, Claude, and Google AI Studio are significantly boosting development speed, even enabling non-engineers to deliver applications quickly; the other side worries this will amplify problems with code review, debugging, maintenance, and reliability, especially in complex projects and production environments.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-2stripe-acquires-openrouter-for-over-8-billion-in-ai-push\"\u003e\n Topic 2:Stripe Acquires OpenRouter for Over $8 Billion in AI Push\n \u003ca class=\"heading-link\" href=\"#topic-2stripe-acquires-openrouter-for-over-8-billion-in-ai-push\"\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 路 News\u003c/li\u003e\n\u003cli\u003eSummary: Trending for: 14 hours ago, Related posts: 6100\u003c/li\u003e\n\u003cli\u003eWhat happened: It\u0026rsquo;s trending on X that Stripe has acquired OpenRouter for over $8 billion, a move seen by outsiders as a significant step to bolster its AI business.\u003c/li\u003e\n\u003cli\u003eWhy it\u0026rsquo;s important: This is important because it could combine payment infrastructure with an AI model routing platform, affecting how developers access models, control costs, and the overall ecosystem distribution landscape.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: The discussion focuses on the authenticity of the news, whether the $8 billion valuation is reasonable, OpenRouter\u0026rsquo;s strategic value, and whether the acquisition will weaken its platform neutrality and impact model providers and developers.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-3truefoundry-open-sources-trueforge-for-cost-effective-ai-agents\"\u003e\n Topic 3:TrueFoundry Open Sources TrueForge for Cost-Effective AI Agents\n \u003ca class=\"heading-link\" href=\"#topic-3truefoundry-open-sources-trueforge-for-cost-effective-ai-agents\"\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 路 News\u003c/li\u003e\n\u003cli\u003eSummary: Trending for: , Related posts: 428\u003c/li\u003e\n\u003cli\u003eWhat happened: TrueFoundry announced it is open-sourcing TrueForge, a tool designed to help developers build, deploy, and manage AI Agents more cost-effectively.\u003c/li\u003e\n\u003cli\u003eWhy it\u0026rsquo;s important: As AI Agent applications grow, inference costs, infrastructure complexity, and observability are becoming bottlenecks for implementation. Open-source tools help lower the barrier to entry and promote enterprise-grade Agent engineering.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: Discussions on X are centered on whether TrueForge can significantly reduce Agent operational costs, its compatibility with existing frameworks and cloud platforms, and whether its open-source strategy can attract a developer ecosystem. Some also question whether its actual effectiveness still needs to be validated in real production environments.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-4etched-raises-700-million-at-21-billion-valuation-from-jane-street\"\u003e\n Topic 4:Etched Raises $700 Million at $21 Billion Valuation from Jane Street\n \u003ca class=\"heading-link\" href=\"#topic-4etched-raises-700-million-at-21-billion-valuation-from-jane-street\"\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 路 News\u003c/li\u003e\n\u003cli\u003eSummary: Trending for: 1 day ago, Related posts: 7200\u003c/li\u003e\n\u003cli\u003eWhat happened: AI inference chip startup Etched has raised $700 million at a $21 billion valuation, in a round led by Jane Street.\u003c/li\u003e\n\u003cli\u003eWhy it\u0026rsquo;s important: This reflects the market\u0026rsquo;s continued enthusiasm for specialized AI inference hardware and shows that alternative paths to Nvidia are gaining validation from both capital and actual customers, potentially impacting the AI infrastructure landscape.\u003c/li\u003e\n\u003cli\u003eDiscussion summary: The main topics of discussion on X are whether Etched\u0026rsquo;s valuation is overheated, whether Jane Street\u0026rsquo;s testing and purchasing of complete systems signifies commercial validation, and whether specialized inference chips can truly challenge existing chip giants.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"topic-5-tesla-owners-share-full-self-driving-real-world-wins-amid-cybercab-buzz\"\u003e\n Topic 5: Tesla Owners Share Full Self-Driving Real-World Wins Amid Cybercab Buzz\n \u003ca class=\"heading-link\" href=\"#topic-5-tesla-owners-share-full-self-driving-real-world-wins-amid-cybercab-buzz\"\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 路 News\u003c/li\u003e\n\u003cli\u003eOverview: Trending Time:, Related Posts: 303\u003c/li\u003e\n\u003cli\u003eWhat it is: Tesla owners are actively sharing their real-world experiences with FSD V14.2 on X, while the mass production and expansion plans for Cybercab and Robotaxi are drawing attention.\u003c/li\u003e\n\u003cli\u003eWhy it matters: This indicates that end-to-end autonomous driving systems are transitioning from demonstrations to larger-scale real-world testing and commercialization expectations. This is crucial for autonomous driving safety validation, in-car AI computing power, Robotaxi business models, and Tesla\u0026rsquo;s competitive position in embodied intelligence and mobility services.\u003c/li\u003e\n\u003cli\u003eDiscussion Overview: Discussions focus on whether FSD V14.2 has significantly improved its ability to handle complex road conditions, whether owner tests can represent general safety, the credibility of the Cybercab timeline, and whether European regulatory restrictions will slow down global deployment. Supporters highlight real-world cases and the speed of software iteration, while critics are concerned about regulation, liability attribution, and the risks of implementing unsupervised autonomous driving.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch4 id=\"summary-of-ai-public-opinion-on-x-today\"\u003e\n Summary of AI Public Opinion on X Today\n \u003ca class=\"heading-link\" href=\"#summary-of-ai-public-opinion-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\u003eThe main narrative on X today can be summarized as: AI is accelerating from \u0026ldquo;can demo\u0026rdquo; to \u0026ldquo;can deliver.\u0026rdquo; Discussions are centered on commercialization paths such as programming tools, Agent infrastructure, inference chips, and autonomous driving, with an overall enthusiastic sentiment. The broad consensus is that vibe coding, open-source Agent tools, and FSD real-world tests are all proving that AI\u0026rsquo;s efficiency and productization speed are indeed increasing, and capital continues to bet on underlying infrastructure and new entry points. The point of contention is whether these technologies can reliably enter production environments. Supporters emphasize the lower barrier to entry, increased speed, and real user feedback, while critics worry about code controllability, platform neutrality, potentially overheated valuations, and whether the optimism around autonomous driving and specialized chips is only temporary. The potential risks mainly stem from a disconnect between technological maturity and the business narrative, which could become particularly evident in issues related to code quality, system stability, supply chain/platform lock-in, regulatory compliance, and liability.\u003c/p\u003e\n\u003ch2 id=\"-influencer-insights\"\u003e\n 馃挕 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 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 馃摎 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\u003eTime Window: Last 3 days; Covering 22 sources; 34 updates in total\u003c/p\u003e\n\u003c/blockquote\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/apple-settles-with-e-u-u-s-app-store-fees-att-rules-in-germany/\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eApple Settles With E.U., U.S. App Store Fees, ATT Rules in Germany\u003c/a\u003e\u003c/strong\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-08-19 18:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary: - Apple\u0026rsquo;s App Store is finally facing the reality of lower fees, and the EU should be satisfied with its work; it\u0026rsquo;s ok it\u0026rsquo;s late.\n\u003cul\u003e\n\u003cli\u003e\u003cstrong\u003e$15\u003c/strong\u003e/month* or *\u003cstrong\u003e$150\u003c/strong\u003e/year.\u003c/li\u003e\n\u003cli\u003eSubstantive analysis of the day鈥檚 news via three weekly emails or a podcast.\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\u003eApple\u0026rsquo;s App Store is finally facing the reality of lower fees, and the EU should be satisfied with its work; it\u0026rsquo;s ok it\u0026rsquo;s late.\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\u003cstrong\u003e\u003ca href=\"https://openai.com/index/offering-zero-data-retention-for-frontier-models\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eOffering Zero Data Retention for frontier models\u003c/a\u003e\u003c/strong\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-08-20 03:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary: - Zero Data Retention provides a clear commitment to eligible API customers: OpenAI will not retain their prompts or model responses after processing a request.\n\u003cul\u003e\n\u003cli\u003eOpenAI personnel cannot view customer content\u003ca href=\"#itation-bottom-1\" \u003e1\u003c/a\u003e, and enterprise customer data is not used to train our models unless a customer explicitly opts in.\u003c/li\u003e\n\u003cli\u003eAs models take on longer and more complex tasks, some serious risks may only become apparent over multiple interactions.\u003c/li\u003e\n\u003cli\u003eExisting ZDR-compatible safety systems evaluate each interaction individually.\u003c/li\u003e\n\u003cli\u003eToday, we are previewing Private Safety Processing, which aims to identify patterns across related interactions without giving OpenAI personnel access to the underlying content.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003eOpenAI reaffirms Zero Data Retention for eligible API customers and previews Private Safety Processing for advanced AI safety without compromising data privacy.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003e\u003ca href=\"https://openai.com/index/replit\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eReplit expands access to software creation with GPT-5.6 Luna\u003c/a\u003e\u003c/strong\u003e\n\u003cul\u003e\n\u003cli\u003ePublished time: 2026-08-19 15:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary:\n\u003cul\u003e\n\u003cli\u003eAs models become more capable, their economics are rapidly shifting.\u003c/li\u003e\n\u003cli\u003eBetter price-performance makes advanced intelligence practical in more products, more workflows, and more moments.\u003c/li\u003e\n\u003cli\u003eFor software creation, this shift is narrowing the distance between having an idea and building something viable.\u003c/li\u003e\n\u003cli\u003eReplit was an early user of GPT-3, building with OpenAI models as natural language software development began to take shape.\u003c/li\u003e\n\u003cli\u003eNow, GPT-5.6 Luna is powering Replit Free Mode, demonstrating how the GPT-5.6 series\u0026rsquo; price-performance and recent OpenAI price drops can translate to broader access at scale.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN highlights:\n\u003cul\u003e\n\u003cli\u003eReplit introduces Free Mode, powered by GPT-5.6 Luna, so anyone can turn ideas into working software without worrying about token costs.\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=kyYepbhe1g8\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eDeepSeek Just Made Closed AI Look Ridiculous\u003c/a\u003e\u003c/strong\u003e\n\u003cul\u003e\n\u003cli\u003ePublished time: 2026-08-20 02:02 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary:\n\u003cul\u003e\n\u003cli\u003e鉂わ笍 Check out Lambda here and sign up for their GPU Cloud:.\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\u003eDeepSeek just made closed AI look ridiculous.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN highlights:\n\u003cul\u003e\n\u003cli\u003e鉂わ笍 Check out Lambda here and sign up for their GPU Cloud:\u003c/li\u003e\n\u003cli\u003eDeepSeek V4 Pro 0813:\u003c/li\u003e\n\u003cli\u003eDSpark full episode:\u003c/li\u003e\n\u003cli\u003e馃檹 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:\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/2608.16890\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eGxP-Agent: Process-DAG Topology for Reliable Clinical Trial Programming with LLM Agents\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eSummary:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.16890v1 Announce Type: new.\u003c/li\u003e\n\u003cli\u003eAbstract: Clinical trial programming鈥攖ranslating study protocols into analysis-ready datasets according to CDISC standards鈥攊s a bottleneck for regulatory submissions, yet LLM-based code generation catastrophically fails at this task: none of 11 single attempts using 5 cutting-edge models produced a valid subject-level analysis dataset.\u003c/li\u003e\n\u003cli\u003eWe introduce GxP-Agent, a multi-agent system that encodes regulatory process orchestration as a Directed Acyclic Graph (DAG), decomposing holistic dataset generation into 15 domain-specific nodes executed by worker agents with pharmaverse skill context, validation gates, and conditional retries.\u003c/li\u003e\n\u003cli\u003eOn CDISC-Bench, a new execution-based benchmark constructed from FDA pilot submission document CDISSCPilot01 (254 subjects, 49 true ADSL variables), GxP-Agent with Claude Sonnet 4.6 achieved 100% structural match (49/49 variables, 254 correct records) across three independent runs, compared to 59.2% for the best retrieval-augmented baseline and 0% for all baseline single-agent and flat multi-agent approaches.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.16890v1 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: Clinical trial programming \u0026ndash; transforming study protocols into analysis-ready datasets under CDISC standards \u0026ndash; is a bottleneck in regulatory submiss鈥u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe introduce GxP-Agent, a multi-agent system that encodes regulatory process ordering as a directed acyclic graph (DAG), decomposing monolithic dataset generati鈥u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eOn CDISC-Bench, a new execution-based benchmark built from the FDA pilot submission CDISCPilot01 (254 subjects, 49 ground-truth ADSL variables), GxP-Agent with鈥u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2608.16891\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eRuntime Governance for Agentic AI: Action-Boundary Control with Trusted Provenance and Fail-Closed Execution\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.16891v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Agentic AI systems request tool actions that can modify files, send messages, launch jobs, or change workflow state.\u003c/li\u003e\n\u003cli\u003eThis shifts the safety problem from harmful text generation to harmful operational side effects.\u003c/li\u003e\n\u003cli\u003ePrompt-level governance can shape model behavior, but it does not create an execution boundary.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.16891v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Agentic AI systems request tool actions that can modify files, send messages, launch jobs, or change workflow state\u003c/li\u003e\n\u003cli\u003eThis shifts the safety problem from harmful text generation to harmful operational side effects\u003c/li\u003e\n\u003cli\u003ePrompt-level governance can shape model behavior, but it does not create an execution boundary\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/2608.16956\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eThe Price of Thinking: Reasoning Effort as a Model-Specific API Contract\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.16956v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: API buyers purchase a dated contract, not just a model name: the contract includes the requested and served model, reasoning-effort terms or their omission, output tracks, service offerings, prompts, and a price list.\u003c/li\u003e\n\u003cli\u003eWe study the reasoning-effort term through a registered paired contrast of Sonnet 5 with explicit high effort against the same model with effort omitted, using 30 AIME 2026 problems and 5 calls per problem.\u003c/li\u003e\n\u003cli\u003eEach paid attempt is assigned a frozen terminal category and reasoning resamples the item while preserving its duplicate calls.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.16956v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: API buyers purchase a dated contract, not a model name alone: the contract includes the requested and served model, reasoning-effort term or its omiss鈥u003c/li\u003e\n\u003cli\u003eWe study the reasoning-effort term through a registered paired contrast of Sonnet 5 with explicit high effort against the same model with effort omitted, using鈥u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eEvery paid attempt was assigned one frozen terminal category, and inference resampled items while retaining their repeated calls\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2608.16971\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eFedPref: Federated Preference Learning for Structured Radiology Report Extraction\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.16971v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Radiology reports describe findings and locations in free text, but downstream search and analysis require these relations in a fixed schema.\u003c/li\u003e\n\u003cli\u003eLearning this extraction requires labels that are unevenly distributed across institutions: smaller hospitals have less local evidence, and pooling data may be infeasible.\u003c/li\u003e\n\u003cli\u003eWe introduce FedPref: a frozen public language model proposes alternative JSON extractions, local annotations rank them, and sites collaboratively train a compact Qwen3-8B adapter while only sharing model updates.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.16971v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Radiology reports describe findings and locations in free text, but downstream search and analysis require these relations in a fixed schema\u003c/li\u003e\n\u003cli\u003eLearning this extraction requires labels that are unevenly distributed across institutions: smaller hospitals have less local evidence, and pooling data may be鈥u003c/li\u003e\n\u003cli\u003eWe introduce FedPref: frozen public language models propose alternative JSON extractions, local annotations rank them, and sites collaboratively train compact Q鈥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/2608.16977\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eThe Problem Is the Problem: Towards Scalable Mathematical Discovery\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.16977v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: AI systems are increasingly capable of contributing to mathematical research.\u003c/li\u003e\n\u003cli\u003eIn research practice, frontier model inference resources are limited, and expert mathematical review is even more severely constrained.\u003c/li\u003e\n\u003cli\u003eTherefore, the proper allocation of these scarce resources is crucial for improving the efficiency of AI-assisted mathematical discovery.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.16977v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: AI systems are increasingly capable of contributing to mathematical research\u003c/li\u003e\n\u003cli\u003eIn research practice, frontier-model reasoning is a limited resource, and expert mathematical review is even more sharply constrained\u003c/li\u003e\n\u003cli\u003eAllocating these scarce resources well is therefore central to making AI-assisted mathematical discovery efficient\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/2608.17007\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eSkillEffect: Checked Lowering for Memory-Bounded Agent Tools\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17007v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Agent skills can specify procedures and resource obligations for tool use, which language models then instantiate into concrete programs.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eHowever, when models translate this guidance into code for existing tool interfaces, even a semantically correct program may load an entire input and exceed the memory available for a single tool call.\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe present SkillEffect, a checked-lowering runtime for computations, featuring a recoverable source relation, an audited bounded implementation, and registered output post-conditions.\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17007v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Agent Skills can specify procedural and resource obligations for tool use, and language models instantiate them as concrete programs\u003c/li\u003e\n\u003cli\u003eHowever, when models turn this guidance into code for existing tool interfaces, even a semantically correct program may load an entire input and exceed the memo鈥u003c/li\u003e\n\u003cli\u003eWe present SkillEffect, a checked-lowering runtime for computations with a recoverable source relation, an audited bounded implementation, and a registered outp鈥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/2608.17053\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eMemory Is Communication: The Frontier Between Remembering and Signaling\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17053v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: A bounded agent may obtain information for a decision from its own past, from peers, or from both sources.\u003c/li\u003e\n\u003cli\u003eRetaining task-relevant history can reduce later communication, while a peer message can supply what memory lacks.\u003c/li\u003e\n\u003cli\u003eUnder limits on both resources, how should an agent allocate its information budget?\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17053v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: A bounded agent may obtain information for a decision from its own past, from peers, or from both sources\u003c/li\u003e\n\u003cli\u003eRetaining task-relevant history can reduce later communication, while a peer message can supply what memory lacks\u003c/li\u003e\n\u003cli\u003eUnder limits on both resources, how should an agent allocate its information budget\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/2608.17067\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eDiSCO: Defending text-to-image generation through distribution-guided contrastive prompt optimization\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17067v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: With the advancements in text-to-image generation models, they have raised serious security concerns, particularly the generation of Not Safe For Work (NSFW) content such as violence and nudity, a problem further exacerbated by red-teaming adversarial attacks.\u003c/li\u003e\n\u003cli\u003eExisting defenses primarily operate under a white-box assumption, relying on text encoder optimization, weight editing, or inference-time interventions, and are fundamentally not scalable to proprietary models.\u003c/li\u003e\n\u003cli\u003eBlack-box alternatives based on LLM prompt rewriting offer broader applicability but fail in a key mechanism we identify as the \\textit{benign adversarial} problem: where prompts are linguistically safe yet still trigger harmful generations due to the data distributions learned by the model.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17067v1 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 text-to-image generative models advance, they raise critical safety concerns, particularly the generation of Not-Safe-For-Work (NSFW) content such鈥u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eExisting defenses predominantly operate under white-box assumptions, relying on text encoder optimization, weight editing, or inference-time intervention, and f鈥u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eBlack-box alternatives based on LLM prompt rewriting offer broader applicability, yet fail in a critical regime we identify as the \\textit{benign adversarial} p鈥u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2608.17071\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eKernelArc: A Multi-Agent Framework for GPU Kernel Optimization\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17071v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: We present KernelArc, a multi-agent framework for autonomous GPU kernel optimization across heterogeneous workloads.\u003c/li\u003e\n\u003cli\u003eStrategy-specialized agents run in parallel and coordinate through conclusions-only shared memory, a deterministic benchmark guard, and a read-only cross-agent state with smooth-trigger drafting.\u003c/li\u003e\n\u003cli\u003eWe evaluate \\kernelarc{} on NVIDIA H100 and B200 GPUs using category-representative SOL-ExecBench workloads.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17071v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: We present KernelArc, a multi-agent framework for autonomous GPU kernel optimization across heterogeneous workloads\u003c/li\u003e\n\u003cli\u003eStrategy-specialized agents run in parallel and coordinate through conclusions-only shared memory, a deterministic benchmark guard, and read-only cross-agent st鈥u003c/li\u003e\n\u003cli\u003eWe evaluate \\kernelarc{} on NVIDIA H100 and B200 GPUs using category-representative SOL-ExecBench workloads\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/2608.17124\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eA decodability criterion predicts when hidden-state selection beats majority voting in large language models\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17124v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Combining the answers a large language model (LLM) samples for a question into one decision is a test-time information fusion problem, usually solved by majority voting.\u003c/li\u003e\n\u003cli\u003eFor difficult problems, voting is unreliable when sampled answers have correlated errors, so an incorrect answer may win, and drawing more samples can make the decision worse.\u003c/li\u003e\n\u003cli\u003eSelecting candidates by reading a correctness signal from the model\u0026rsquo;s hidden states is a promising alternative, but its accuracy varies by model and task, and there is no measure to indicate when it can be trusted.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17124v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Combining the answers a large language model (LLM) samples for a question into one decision is a test-time information fusion problem, usually solved鈥u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eVoting is unreliable on difficult questions, where the sampled answers share correlated errors, so the wrong answer can win and drawing more samples makes the d鈥u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eSelecting a candidate by reading a correctness signal from the model\u0026rsquo;s hidden states is a promising alternative, but its accuracy varies across models and tasks鈥u003c/p\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/2608.16975\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eMargin-Regularized Structured Semantic Alignment for Brain-Language Correspondence\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.16975v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: With the rapid advancement of large language models, brain-language decoding has achieved remarkable progress.\u003c/li\u003e\n\u003cli\u003eHowever, it remains unclear whether decoded content genuinely reflects neural representations or is largely reconstructed by the language model itself.\u003c/li\u003e\n\u003cli\u003eThis ambiguity limits interpretability and hinders the investigation of intrinsic brain-language correspondence.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.16975v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: With the rapid advancement of large language models, brain-language decoding has achieved remarkable progress\u003c/li\u003e\n\u003cli\u003eHowever, it remains unclear whether decoded content genuinely reflects neural representations or is largely reconstructed by the language model itself\u003c/li\u003e\n\u003cli\u003eThis ambiguity limits interpretability and hinders the investigation of intrinsic brain-language correspondence\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/2608.17050\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eCross-Model Memory Transfer via Target-Side Reader Adaptation\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17050v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Methods for improving knowledge use in large language models typically fall into two regimes.\u003c/li\u003e\n\u003cli\u003eNon-parametric retrieval offers flexible access to external knowledge, but adds retrieval latency, context overhead, and only shallow integration with the backbone.\u003c/li\u003e\n\u003cli\u003eParametric adaptation is efficient at inference time, but entangles knowledge with model weights and can be hard to update, audit, or transfer.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17050v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Methods for improving knowledge use in large language models typically fall into two regimes\u003c/li\u003e\n\u003cli\u003eNon-parametric retrieval offers flexible access to external knowledge, but adds retrieval latency, context overhead, and only shallow integration with the backb鈥u003c/li\u003e\n\u003cli\u003eParametric adaptation is efficient at inference time, but entangles knowledge with model weights and can be hard to update, audit, or transfer\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/2608.17051\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eInstitution-Specific LLM Prompting Recovers PHI That De-identification Systems and Their Gold Standards Both Miss\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17051v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: The secondary use of electronic health records requires de-identification, yet existing systems miss \\emph{institution-specific} protected health information (PHI), such as hospital abbreviations, building names, and internal codes whose status is locally determined.\u003c/li\u003e\n\u003cli\u003eWe ask whether large language models (LLMs) with in-context learning (ICL) can close this gap and control the precision-recall trade-off.\u003c/li\u003e\n\u003cli\u003eOn 100 annotated pediatric oncology notes (5,322 PHI spans) from Texas Children\u0026rsquo;s Hospital, we benchmarked eight LLMs against two specialized systems (Stanford TiDE, OpenMed PII) and two pattern-based baselines.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17051v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Secondary use of electronic health records requires de-identification, yet existing systems miss \\emph{institutionally situated} protected health info鈥u003c/li\u003e\n\u003cli\u003eWe ask whether large language models (LLMs) with in-context learning (ICL) can close this gap and control the precision\u0026ndash;recall trade-off\u003c/li\u003e\n\u003cli\u003eOn 100 annotated pediatric oncology notes (5,322 PHI spans) from Texas Children\u0026rsquo;s Hospital, we benchmarked eight LLMs against two purpose-built systems (Stanfor鈥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/2608.17075\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eFoundation Agents Meet Agentic Deep Research: Evidence-Grounded Clinical Code Forecasting\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17075v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Next-encounter ICD forecasting predicts which standardized diagnosis codes will be documented at a future visit based on the previously available longitudinal record.\u003c/li\u003e\n\u003cli\u003eThe task is prospective and multi-label: the target annotation does not yet exist, and multiple codes may be correct.\u003c/li\u003e\n\u003cli\u003eStructured EHR foundation models capture recurrence and temporal progression, while language foundation models generate flexible diagnostic hypotheses.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17075v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Next-encounter ICD forecasting predicts which standardized diagnosis codes will be documented at a future visit from the longitudinal record available鈥u003c/li\u003e\n\u003cli\u003eThe task is prospective and multi-label: the target note does not yet exist, and several codes may be correct\u003c/li\u003e\n\u003cli\u003eStructured EHR foundation models capture recurrence and temporal progression, whereas language foundation models generate flexible diagnostic hypotheses\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/2608.17084\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eUncertainty-Aware Decision Making in Multimodal Large Language Models\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17084v1 Announcement Type: New.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eAbstract: Multimodal large language models (MLLMs) increasingly answer questions whose correctness depends on visual, textual, temporal, acoustic, document, chart, or concrete evidence.\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eTherefore, their failures are not just linguistic.\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eA fluent answer may conceal poor input quality, perceptual errors, weak grounding, conflicts between modalities, unstable reasoning, distribution shifts, or an inability to answer from the provided evidence.\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eEN Highlights:\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17084v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Multimodal large language models (MLLMs) increasingly answer questions whose correctness depends on visual, textual, temporal, acoustic, document, cha鈥u003c/li\u003e\n\u003cli\u003eTheir failures are therefore not only linguistic\u003c/li\u003e\n\u003cli\u003eA fluent answer may conceal poor input quality, a perceptual error, weak grounding, conflict between modalities, unstable reasoning, distribution shift, or a qu鈥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/2608.17088\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eThere is No Theoretical Curse of Multilinguality For Embedding Space Structure\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eRelease Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17088v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: A central goal of multilingual NLP is to achieve high monolingual performance per language and cross-lingual alignment through multilingual models to realize large-scale language coverage.\u003c/li\u003e\n\u003cli\u003eThe curse of multilinguality describes the phenomenon of performance degradation in multilingual models as we increase language coverage, posing a threat to the aforementioned goal.\u003c/li\u003e\n\u003cli\u003eThis paper asks whether multilingual embedding spaces are inherently incapable of achieving perfect multilinguality without a prohibitive increase in required capacity.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17088v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: A central goal of multilingual NLP is to achieve high monolingual performance per language and cross-lingual alignment for large-scale language covera鈥u003c/li\u003e\n\u003cli\u003eThe curse of multilinguality describes the phenomenon of degradation in multilingual model performance as we increase language coverage, posing a threat to the鈥u003c/li\u003e\n\u003cli\u003eThis paper asks whether multilingual embedding spaces are inherently incapable of achieving perfect multilinguality without a prohibitive increase in required c鈥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/2608.17096\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eA Glyph Is Not a Letter, a Token Is Not a Word, a Space Is Not a Space: What the Units of Voynichese Are Not\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eRelease Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17096v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: The Voynich manuscript (Beinecke MS 408) is commonly analyzed based on three unstated assumptions: its glyphs are letters, the strings between spaces are words, and every space is a word-space.\u003c/li\u003e\n\u003cli\u003eWe test all three against the Zandbergen-Landini transliteration using matched prose, cryptographic, and pseudotext controls, as well as quire-level resampling.\u003c/li\u003e\n\u003cli\u003eNone of them hold, and the failures have a common shape: order in Voynichese is located at the edges of the signs and at the graded boundaries between them, not in the continuity of the signs themselves.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17096v1 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: The Voynich manuscript (Beinecke MS 408) is usually analysed on three unstated assumptions: that its glyphs are letters, that the strings between blan鈥u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eWe test all three against the Zandbergen-Landini transliteration with matched prose, cipher, and pseudo-text controls and quire-level resampling\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eNone holds, and the failures share a shape: the order in Voynichese sits at the edges of tokens and at graded boundaries between them, not in the succession of鈥u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2608.17102\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eEmotion Across Speech and Faces: Shared Affective Mechanisms in Multimodal Foundation Models\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17102v1 Announcement Type: New.\n-Abstract: Modern multimodal foundation models (MFMs) have made rapid progress on tasks requiring integrated perception across speech, vision, and language, including emotion recognition.\n\u003cul\u003e\n\u003cli\u003eHowever, it remains unclear whether they recognize speech and facial emotion through shared affective functional units or modality-specific pathways.\u003c/li\u003e\n\u003cli\u003eWe explore emotion-sensitive neurons (ESNs), i.e., sparse decoder neurons selectively associated with emotion categories, in three MFMs: Gemma-4-12B-it, MiniCPM-o-4.5, and Qwen2.5-Omni-7B.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17102v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Modern multimodal foundation models (MFMs) have made rapid progress on tasks requiring integrated perception across speech, vision, and language, incl鈥u003c/li\u003e\n\u003cli\u003eHowever, it remains unclear whether they recognize speech and facial emotion through shared affective functional units or modality-specific pathways\u003c/li\u003e\n\u003cli\u003eWe explore emotion-sensitive neurons (ESNs), sparse decoder neurons selectively associated with emotion categories, in three MFMs: Gemma-4-12B-it, MiniCPM-o-4.5鈥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/2608.17120\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eChildren, but not language models, show accelerating returns in word learning\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17120v1 Announcement Type: New.\n\u003cul\u003e\n\u003cli\u003eAbstract: Children learn hundreds of words over the first years of their lives, a process that begins slowly but quickly accelerates.\u003c/li\u003e\n\u003cli\u003ePrevious models have described vocabulary growth as an accumulation of evidence over time.\u003c/li\u003e\n\u003cli\u003eHere, we show that this process is best characterized by accelerating accumulation: children learn more from each additional unit of language experience than from the previous one.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17120v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Children learn hundreds of words over the first years of their lives, in a process that begins slowly but quickly picks up speed\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003ePrior models describe vocabulary growth as evidence accumulation over time\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eHere we show that the process is best characterized as accelerating accumulation: children learn more from each additional unit of linguistic experience than th鈥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/2608.17153\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eTowards Safer RAG: Only Agents Capable of System 2 Thinking may Access Untrusted Documents\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.17153v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Retrieval-Augmented Generation (RAG) has significantly enhanced the performance of large language models (LLMs), yet these systems remain vulnerable to knowledge poisoning attacks, where misinformation in retrieved documents can compromise the model\u0026rsquo;s final output.\u003c/li\u003e\n\u003cli\u003eNotably, an LLM may correctly detect that a document contains incorrect information but still be influenced by it.\u003c/li\u003e\n\u003cli\u003ePrior work has addressed this vulnerability through the Cordon Principle, which prevents models responsible for final answer synthesis from directly accessing the original evidence.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.17153v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Retrieval-Augmented Generation (RAG) has significantly enhanced the performance of large language models (LLMs), yet these systems remain vulnerable t鈥u003c/li\u003e\n\u003cli\u003eNotably, an LLM may correctly detect that a document contains incorrect information while nevertheless being influenced by it\u003c/li\u003e\n\u003cli\u003ePrior work has addressed this vulnerability through the Cordon Principle, which prevents models responsible for final answer synthesis from directly accessing r鈥u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3 id=\"arxiv-cslg-b_introsearch\"\u003e\n ArXiv cs.LG (B_intro+search)\n \u003ca class=\"heading-link\" href=\"#arxiv-cslg-b_introsearch\"\u003e\n \u003ci class=\"fa-solid fa-link\" aria-hidden=\"true\" title=\"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/2608.14556\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eLearning Discrete Riemannian Metrics for Physical Fields with Cochain-Frame Equivarianc\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.14556v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Physical fields on meshes require a separation between topology and geometry: conservation laws are topological and should be exact, while geometry, material response, and anisotropic coupling must be learned from data.\u003c/li\u003e\n\u003cli\u003eExisting neural surrogates often mix these roles in unconstrained message passing.\u003c/li\u003e\n\u003cli\u003eWe introduce Riemannian Hodge Message Passing (RHMP), which turns this separation into an architectural principle.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.14556v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Physical fields on meshes require a separation between topology and geometry: conservation laws are topological and should be exact, while geometry, m鈥u003c/li\u003e\n\u003cli\u003eExisting neural surrogates often mix these roles inside unconstrained message passing\u003c/li\u003e\n\u003cli\u003eWe introduce Riemannian Hodge Message Passing (RHMP), which turns this separation into an architectural principle\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/2608.14563\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eForward Pass Domain Adaptation (Without Cross-Layer Backpropagation)\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.14563v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Forward-Pass-Only MLP training (FPO) adapts large language models without a backward pass through the model body, achieving 2.7\u0026ndash;3.2x the throughput of standard finetuning with roughly 40% reduction in peak training memory, while preserving out-of-domain benchmarks within the seed noise of a baseline, an attribute not reliably reproduced by full-network finetuning.\u003c/li\u003e\n\u003cli\u003eFPO rests on a single empirical observation: at late layers of a Transformer, the output-layer prediction error approximates the true gradient with cosine similarity 0.47\u0026ndash;0.59 across six public models we investigate.\u003c/li\u003e\n\u003cli\u003eWe introduce a two-minute diagnostic that quantifies this approximation per layer for any model, identifying where late-layer adaptation is viable.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.14563v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Forward-Pass-Only MLP training (FPO) adapts large language models without a backward pass through the model body, achieving 2.7\u0026ndash;3.2x the throughput o鈥u003c/li\u003e\n\u003cli\u003eFPO rests on a single empirical observation: at late layers of a transformer, the output-layer prediction error approximates the true gradient with cosine simil鈥u003c/li\u003e\n\u003cli\u003eWe introduce a two-minute diagnostic that quantifies this approximation per layer for any model, identifying where late-layer adaptation is viable\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/2608.14570\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eCoarse-to-Fine Multi-Resolution Diffusion Models for Trajectory Generation in Urban Systems\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.14570v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Understanding human mobility is critical for a wide range of urban applications, including traffic management, epidemic control, and urban planning.\u003c/li\u003e\n\u003cli\u003eHowever, due to privacy concerns, the availability of large-scale public trajectory data remains limited, posing challenges for downstream mobility analysis.\u003c/li\u003e\n\u003cli\u003eExisting methods for synthetic trajectory generation primarily focus on matching global distribution similarity, while often overlooking mobility patterns across different spatial and temporal resolutions, which are crucial for practical applications.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.14570v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Understanding human mobility is critical for a wide range of urban applications, including traffic management, epidemic control, and urban planning\u003c/li\u003e\n\u003cli\u003eHowever, due to privacy concerns, the availability of large-scale public trajectory data remains limited, posing challenges for downstream mobility analysis\u003c/li\u003e\n\u003cli\u003eExisting methods for synthetic trajectory generation primarily focus on matching global distribution similarity, while often overlooking mobility patterns acros鈥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/2608.14594\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eGeometry Is Not Robustness: A Trajectory-Level Study of PGD Evaluation\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eAbstract: - arXiv:2608.14594v1 Announce Type: new.\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eAbstract: Projected Gradient Descent (PGD) is widely used to evaluate adversarial robustness, typically through final adversarial accuracy, but it does not capture the model\u0026rsquo;s behavior throughout the attack process.\u003c/li\u003e\n\u003cli\u003eRecent work proposes trajectory-level diagnostics, such as loss evolution, gradient alignment, and steps-to-failure, for deeper insight into adversarial optimization dynamics.\u003c/li\u003e\n\u003cli\u003eHowever, it remains unclear whether these diagnostics reliably indicate robustness strength.\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:2608.14594v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Projected Gradient Descent (PGD) is widely used to evaluate adversarial robustness, typically via final adversarial accuracy, which does not capture m鈥u003c/li\u003e\n\u003cli\u003eRecent work proposes trajectory-level diagnostics, such as loss evolution, gradient alignment, and steps-to-failure, for deeper insight into adversarial optimis鈥u003c/li\u003e\n\u003cli\u003eHowever, whether these diagnostics reliably indicate robustness strength remains unclear\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/2608.14614\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eDumpsterCluster: From Dumpster Diving to Serving LLaMA-70B on $60 GPUs\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.14614v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: As AI data centers retire functional GPUs, large quantities of still-capable accelerators enter the secondary market.\u003c/li\u003e\n\u003cli\u003eThis paper investigates whether these retired GPUs can find a productive afterlife by forming a DumpsterCluster that can serve modern LLM inference, and under what conditions this repurposing is economically feasible and environmentally sustainable.\u003c/li\u003e\n\u003cli\u003eWe physically built a 128-GPU DumpsterCluster from scratch using only second-hand components and ran it for one year.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Highlights:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.14614v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: As AI datacenters retire functional GPUs, vast quantities of still capable accelerators enter secondary markets\u003c/li\u003e\n\u003cli\u003eThis paper investigates whether these retired GPUs can find a productive afterlife to form a DumpsterCluster that can serve modern LLM inference, and under what鈥u003c/li\u003e\n\u003cli\u003eWe physically built a 128-GPU DumpsterCluster from scratch using only second-hand components and ran it for one year\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/2608.14617\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eCalibrated Trust, Not Sharper Prediction: An Empirical Test of Uncertainty Fusion\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublished: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract: - arXiv:2608.14617v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: A recurring proposal in the field of legal AI is to improve case outcome prediction by fusing uncertainty tools (evidence graphs with belief propagation, sequential Bayesian odds updating, Dempster-Shafer combination, and conformal prediction) into a pipeline.\u003c/li\u003e\n\u003cli\u003eWe tested this on 1,000 real European Court of Human Rights cases from LexGLUE and FairLex, predicting whether the court found a violation of the Convention from the factual paragraphs of the cases.\u003c/li\u003e\n\u003cli\u003eWe compared three series from two leading LLMs (Claude Opus 4.8 and GPT-5.5) as factual evidence estimators: (A) the raw LLMs, (B) the LLMs routed through the fusion pipeline, and (C) a term-frequency baseline through the same pipeline.\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 瑕佺偣:\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003earXiv:2608.14617v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: A recurring proposal in legal AI is to improve case-outcome prediction by fusing uncertainty tools (evidence graphs with belief propagation, sequentia鈥u003c/li\u003e\n\u003cli\u003eWe test this on 1,000 real European Court of Human Rights cases from LexGLUE and FairLex, predicting whether the Court found a Convention violation from the cas鈥u003c/li\u003e\n\u003cli\u003eWe compare three families across two frontier LLMs (Claude Opus 4.8 and GPT-5.5) as per-fact evidence estimators: (A) the raw LLM, (B) the LLM routed through th鈥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/2608.14619\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003ePIKFNO: An Interpretable Neural Operator Based on Physics Informed Kernel Function\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublish Time:2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract:- arXiv:2608.14619v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: This work proposes a new interpretable neural operator framework, termed the Physics Informed Kernel Function Neural Operator (PIKFNO), which explicitly incorporates physics-informed kernel functions derived from governing equations into the neural operator architecture.\u003c/li\u003e\n\u003cli\u003eUnlike traditional neural operators such as DeepONet, which rely on deep networks to implicitly learn basis functions, PIKFNO constrains the trunk network through physics-informed kernel functions, thereby aligning its operator structure with kernel extensions used in mesh-free collocation methods.\u003c/li\u003e\n\u003cli\u003eTwo construction strategies are introduced: one learns kernel functions directly from data, where the learned kernel can be regarded as a non-singular fundamental solution, while the other constructs them through transformations of analytical fundamental solutions.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 瑕佺偣:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.14619v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: This work proposes a new interpretable neural operator framework, termed the Physics Informed Kernel Function Neural Operator (PIKFNO), which explicit鈥u003c/li\u003e\n\u003cli\u003eUnlike traditional neural operators such as DeepONet, which rely on deep networks to implicitly learn basis functions, PIKFNO constrains the trunk network throu鈥u003c/li\u003e\n\u003cli\u003eTwo construction strategies are introduced: one learns kernel functions directly from data, where the learned kernel can be regarded as a nonsingular fundamenta鈥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/2608.14620\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eExplaining Reinforcement Learning Decisions in Self-adaptive Systems\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublish Time:2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract:- arXiv:2608.14620v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Reinforcement Learning (RL) has been widely applied in autonomous and self-adaptive systems, but RL policies, especially deep RL policies relying on neural networks, lack transparency and are difficult to understand.\u003c/li\u003e\n\u003cli\u003eThis can lead to decreased user trust and make system verification more challenging.\u003c/li\u003e\n\u003cli\u003eTo address this challenge, this paper introduces Explanations using Reinforcement Learning Alternative Realities (EARL), a Python library for generating counterfactual explanations in RL settings.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN 瑕佺偣:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.14620v1 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: Reinforcement Learning (RL) has been extensively used in autonomous and self-* systems, but RL policies, especially deep RL ones relying on neural net鈥u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eThis can lead to diminished user trust, and makes for a more challenging verification of systems\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eTo address this challenge, this paper introduces Explanations using Alternative Realities for Reinforcement Learning (EARL), a Python library to produce counter鈥u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003e\u003ca href=\"https://arxiv.org/abs/2608.14634\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eMetaplasticity as adaptive gradient preconditioning for incremental learning\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract:- arXiv:2608.14634v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Biological intelligence naturally prevents catastrophic forgetting through Complementary Learning Systems (CLS) theory, a macroscopic consolidation process driven by synaptic metaplasticity at the local level: continuous, history-dependent neuromodulation of individual synapses.\u003c/li\u003e\n\u003cli\u003eWhile artificial neural networks struggle with the stability-plasticity dilemma in non-stationary environments, existing solutions often require task labels or incur significant memory overhead, contrary to biological reality.\u003c/li\u003e\n\u003cli\u003eRe-framing this localized neuromodulation as an optimization-driven process, we introduce $\\textbf{SynGAP}$: $\\textbf{Syn}$aptic $\\textbf{G}$eometric $\\textbf{A}$daptive $\\textbf{P}$reconditioning.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.14634v1 Announce Type: new\u003c/li\u003e\n\u003cli\u003eAbstract: Biological intelligence naturally prevents catastrophic forgetting through Complementary Learning Systems (CLS) theory, a macroscopic consolidation pr鈥u003c/li\u003e\n\u003cli\u003eWhile artificial neural networks struggle with the stability-plasticity dilemma in non-stationary environments, existing solutions often require task labels or鈥u003c/li\u003e\n\u003cli\u003eRe-framing this localized neuromodulation as an optimization-driven process, we introduce $\\textbf{SynGAP}$: $\\textbf{Syn}$aptic $\\textbf{G}$eometric $\\textbf{A鈥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/2608.14636\" class=\"external-link\" target=\"_blank\" rel=\"noopener\"\u003eFractional Optimizers Meet Fractal Activation Functions: An Empirical Study of Multi-Scale Optimization in Neural Network\u003c/a\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003ePublication Time: 2026-08-19 12:00 Beijing Time\u003c/li\u003e\n\u003cli\u003eAbstract:- arXiv:2608.14636v1 Announce Type: new.\n\u003cul\u003e\n\u003cli\u003eAbstract: Fractional optimization methods and fractal activation functions are two independent directions for improving neural network training.\u003c/li\u003e\n\u003cli\u003eFractional optimizers extend first-order optimization through fractional derivatives and memory effects, while fractal activations introduce multi-scale nonlinear representations based on self-similar Weierstrass and Blancmange-type functions.\u003c/li\u003e\n\u003cli\u003eHere, we investigate their interaction within a unified experimental framework.\u003c/li\u003e\n\u003c/ul\u003e\n\u003c/li\u003e\n\u003cli\u003eEN Key Points:\n\u003cul\u003e\n\u003cli\u003earXiv:2608.14636v1 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: Fractional optimization methods and fractal activation functions are two independent directions for improving neural network training\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eFractional optimizers extend first-order optimization through fractional derivatives and memory effects, whereas fractal activations introduce multi-scale nonli鈥u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003eHere, we investigate their interaction within a unified experimental framework\u003c/p\u003e\n\u003c/li\u003e\n\u003c/ul\u003e\n", "wordCount": 6448, "readingTime": 31, "tableOfContents": "\u003cnav id=\"TableOfContents\"\u003e\n \u003cul\u003e\n \u003cli\u003e\u003ca href=\"#-this-issues-watch-list-a-deep-dive\"\u003e馃摉 This Issue\u0026rsquo;s Watch List: A Deep Dive\u003c/a\u003e\u003c/li\u003e\n \u003cli\u003e\u003ca href=\"#-top-ai-news-on-x\"\u003e馃寪 Top AI News on X\u003c/a\u003e\n \u003cul\u003e\n \u003cli\u003e\u003ca href=\"#topic-1study-reveals-trade-offs-in-ai-vibe-coding-tools\"\u003eTopic 1:Study Reveals Trade-Offs in AI Vibe Coding Tools\u003c/a\u003e\u003c/li\u003e\n \u003cli\u003e\u003ca href=\"#topic-2stripe-acquires-openrouter-for-over-8-billion-in-ai-push\"\u003eTopic 2:Stripe Acquires OpenRouter for Over $8 Billion in AI Push\u003c/a\u003e\u003c/li\u003e\n \u003cli\u003e\u003ca href=\"#topic-3truefoundry-open-sources-trueforge-for-cost-effective-ai-agents\"\u003eTopic 3:TrueFoundry Open Sources TrueForge for Cost-Effective AI Agents\u003c/a\u003e\u003c/li\u003e\n \u003cli\u003e\u003ca href=\"#topic-4etched-raises-700-million-at-21-billion-valuation-from-jane-street\"\u003eTopic 4:Etched Raises $700 Million at $21 Billion Valuation from Jane Street\u003c/a\u003e\u003c/li\u003e\n \u003cli\u003e\u003ca href=\"#topic-5-tesla-owners-share-full-self-driving-real-world-wins-amid-cybercab-buzz\"\u003eTopic 5: Tesla Owners Share Full Self-Driving Real-World Wins Amid Cybercab Buzz\u003c/a\u003e\u003c/li\u003e\n \u003c/ul\u003e\n \u003c/li\u003e\n \u003cli\u003e\u003ca href=\"#-influencer-insights\"\u003e馃挕 Influencer Insights\u003c/a\u003e\u003c/li\u003e\n \u003cli\u003e\u003ca href=\"#-appendix-todays-watch-list-source-updates\"\u003e馃摎 Appendix: Today\u0026rsquo;s Watch List Source Updates\u003c/a\u003e\n \u003cul\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 }