System translated (Gemini)

🤖 AI 速览

This edition highlights two shifts underpinning the scaled adoption of AI: OpenAI has re-architected its online storage to support billions of users, making infrastructure reliability the foundation of product experience and business expansion; at the same time, AI scientist evaluations have begun …
📋 文章元数据
发布时间
2026-09-12
类型
ai-daily
字数
7055
阅读时长
34 min

2026-09-12 AI Daily | From Answers to Processes: AI Enters the Era of Scalable Infrastructure and Auditability Link to heading

This issue focuses on two shifts behind the large-scale adoption of AI: OpenAI is re-architecting its online storage to support over a billion users, making infrastructure reliability the bedrock of product experience and commercial growth. Simultaneously, the evaluation of AI scientists is beginning to record entire workflows, shifting the focus of agent capabilities from results to traceable processes and reproducible methods. Model cost competition, plugin evaluations, and high-risk applications are also collectively raising the standards for security governance and transparent disclosure.

📖 In-depth Guide to This Issue’s Watch List Link to heading

Three main themes are worth focusing on today. First, OpenAI shared its practices for scaling online storage to support over a billion ChatGPT users, and it is recommended that engineering teams pay attention to its trade-offs in reliability, data access, and large-scale infrastructure. Second, several new papers on arXiv are re-examining the boundaries of model capabilities: from data-efficient training and latent space language models for next-concept prediction to longer context and retrieval augmentation, the core question is how to achieve stronger generalization and reasoning abilities with less data. Finally, the evaluation of AI Agents is shifting from the “final answer” to “process quality”: OpenDiscoveryTrace records the complete workflow of AI scientists, and other research compares the roles of reusable skills and sub-agents in long-horizon tasks. Meanwhile, studies on languages like Urdu and topics like Chinese military news remind us that the multilingual and domain-specific capabilities of models are still far from balanced.

🌐 AI Hotspots on X Link to heading

Topic 1: Anthropic Details Claude AI Misuse in Weapons and Influence Plots Link to heading

  • Category: AI · News
  • Overview: Trending Time: 1 day ago, Related Posts: 145,000
  • What happened: Anthropic released a threat intelligence report stating that in 2026, it disrupted multiple cases involving the use of Claude for potential bioweapons development, cyberattacks, influence operations, and surveillance activities. This included five cases of high-risk virus research linked to state-affiliated actors.
  • Why it matters: This shows that general-purpose large models can now potentially be used for activities related to biosecurity and national security. It highlights the necessity for model providers to conduct misuse monitoring, implement access restrictions, and take proactive intervention measures. It also moves AI security governance from the realm of theoretical risk assessment to the stage of concrete incident response.
  • Discussion summary: Discussions on X are mainly focused on the boundaries of dual-use research, whether it is reasonable for Anthropic to impose bans when malicious intent cannot be confirmed, and the transparency and verifiability of corporate threat reports. Some view these actions as necessary security measures, while others worry about harming legitimate research, expanding censorship, and the potential for such reports to be used for propaganda.

Topic 2: Trump Appears to Doze at Pentagon 9/11 Ceremony as Others Stand in New York Link to heading

  • Category: AI · News
  • Overview: Trending Time: 3 hours ago, Related Posts: 30,000
  • What happened: Reports claim that Trump appeared to briefly doze off during the 9/11 memorial ceremony at the Pentagon, while other commemorative events were ongoing in New York.
  • Why it matters: While the event itself is not an AI technological advancement, its spread on AI-related trending lists demonstrates how social media platforms use algorithms to amplify political figures, public ceremonies, and suspicious visual details. It also highlights the importance of video content verification and generative media analysis.
  • Discussion summary: Discussions on X primarily center on whether Trump was actually asleep in the video, whether his behavior was disrespectful, and the selective interpretations of the event by users with different political stances. Some users also questioned the context, accuracy, and dissemination methods of the footage and reports.

Topic 3: Senate Nears Vote on Comprehensive Crypto Framework Bill Link to heading

  • Category: AI · News
  • Overview: Trending Time: 2 days ago, Related Posts: 59,000
  • What happened: The U.S. Senate is nearing a vote on a comprehensive cryptocurrency framework bill.
  • Why it matters: The bill could clarify regulatory authority, compliance requirements, and market rules for digital assets. It could also indirectly affect the financing, product application, and corporate compliance environments for emerging technologies like AI and blockchain.
  • Discussion summary: Discussions on X are mainly focused on whether the bill can secure enough votes, whether regulatory authority should be led by the SEC or the CFTC, and whether the legislation is promoting innovation or relaxing regulations on the crypto market. There are disagreements between supporters and those who worry about regulatory arbitrage and insufficient investor protection.

Topic 4: Anthropic Launches Claude Plugin Evaluation Tool Link to heading

  • Category: AI · News
  • Overview: Trending Time: 2 hours ago, Related Posts: 342
  • What happened: Anthropic has launched the Claude Plugin Evaluation Tool, designed to assess the functional performance, stability, and security of plugins within model agent workflows.
  • Why it matters: This helps transform plugins from temporary extensions of “prompts plus scripts” into testable, comparable, and maintainable software components. It also establishes a quality and security baseline for the large-scale use of the Claude ecosystem.
  • Discussion Summary: Discussions on X primarily focus on the version management, automatic loading, cross-tool portability, and maintenance costs of plugins and skills. Supporters believe that standardized evaluations can enhance developer trust, while critics are concerned about insufficient evaluation coverage, metrics failing to reflect real-world task performance, and the security risks posed by plugin execution permissions.

Topic 5: OpenAI Guides Developers on GPT-6 Astra Prompt Tuning Link to heading

  • Category: AI · News
  • Summary: Trending since: 3 hours ago, Related posts: 813
  • What it is: OpenAI is reportedly providing guidance to help developers optimize prompt tuning for GPT-6 Astra.
  • Why it matters: This indicates that the performance of the new generation of models may be highly dependent on prompt design, which will affect developer migration, application development, and how model capabilities are evaluated.
  • Discussion Summary: With limited information available, discussions on X are centered on the authenticity and release date of the guide and GPT-6 Astra, the model’s capabilities, and whether prompt tuning can deliver consistent performance improvements.

Topic 6: OpenAI Pauses $200 ChatGPT Pro Subscriptions Amid Astra Demand Surge Link to heading

  • Category: AI · News
  • Summary: Trending since: 2 days ago, Related posts: 38,000
  • What it is: OpenAI has paused new subscriptions for its $200/month ChatGPT Pro plan due to a surge in demand driven by the new image tool, Astra, in order to maintain service quality. Other plans and the API are not affected at this time.
  • Why it matters: This reflects that computing power and infrastructure for high-end generative AI services may still pose a bottleneck to commercial scaling. It also highlights how high-load applications, such as image generation, impact service stability and subscription strategies.
  • Discussion Summary: Discussions on X are focused on whether the slower response times and reduced image detail experienced by Pro users are related to the post-launch workload. Some users have questioned the service quality changes, while OpenAI has stated it has not made intentional adjustments and is investigating. Meanwhile, the free Pro access offered to clinicians and creative professionals in the U.S. has also attracted attention.

Topic 7: DeepSeek Unveils V4.1-Flash, Outpacing Larger AI Models at Fraction of Cost Link to heading

  • Category: AI · News
  • Summary: Trending since: 1 day ago, Related posts: 52,000
  • What it is: DeepSeek has released V4.1-Flash, claiming it outperforms larger AI models at a lower cost.
  • Why it matters: If verified by independent evaluations, this release suggests that the trade-off between model performance, scale, and inference cost is improving, which could further drive cost-effective AI applications and industry competition.
  • Discussion Summary: Discussions on X are mainly focused on whether its performance and cost data have undergone fair, reproducible third-party evaluations, as well as the comparison metrics used against larger models, its actual performance in applications, and whether the marketing claims are exaggerated.

Topic 8: Formula 1 Debuts at New Madrid Circuit with Challenging Practice Sessions Link to heading

  • Category: AI · Sports
  • Summary: Trending since: 2 days ago, Related posts: 107,000
  • What it is: Formula 1 made its debut at the new Madrid circuit, with drivers facing challenges related to track layout, grip, and car setup during practice sessions.
  • Why it matters: Although not an AI event in itself, the large volume of telemetry and race data from the new circuit provides valuable scenarios for AI-driven applications in race strategy, performance analysis, and simulation training.
  • Discussion Summary: Discussions on X are focused on the new circuit’s difficulty, driver and team adaptation, car performance, and the future competitive landscape. Some users also questioned the predictive value of the practice results for the actual race.

Topic 9: 49ers Crush Rams 27-7 in Historic NFL Australia Debut Link to heading

  • Category: AI · Sports
  • Summary: Trending since: 21 hours ago, Related posts: 223,000
  • Abstract: 49ers Crush Rams 27-7 in Historic NFL Australia Debut:

Topic 10: Seahawks Edge Patriots 13-10 in Super Bowl Rematch Opener Link to heading

  • Category: AI · Sports
  • Summary: Trending since: 2 days ago, Related posts: 419,000
  • Summary:Seahawks Edge Patriots 13-10 in Super Bowl Rematch Opener: REGIONAL SPORTS ROUNDUP September 10, 2026 COLLEGE MSU MBB UNVEILS OOC SLATE Murray State men’s basketball has announced its 2026-27 non-conference schedule, featuring six home games in the renovated CFSB Center and trips to Las Vegas, Memp…

Topic 11:49ers Rout Rams 27-7 in NFL’s Historic Melbourne Debut Link to heading

  • Category:AI · Sports
  • Overview:Trending Time:1 day ago,Related Posts:116000
  • Summary:49ers Rout Rams 27-7 in NFL’s Historic Melbourne Debut:

Topic 12:NAZA Documentary Receives Record 24.5-Minute Ovation at Venice Premiere Link to heading

  • Category:AI · Entertainment
  • Overview:Trending Time:1 day ago,Related Posts:164000
  • What it is:The Israeli documentary “NAZA” received a 24.5-minute standing ovation at its Venice Film Festival premiere. Through the anonymous testimonies of 24 Israeli military intelligence personnel and soldiers, the film reveals the use of AI-assisted target identification, phone tracking, and intelligence surveillance in warfare, and the resulting accusations of civilian casualties.
  • Why it matters:The event places AI-assisted military decision-making under public and industry scrutiny, highlighting issues of accuracy, accountability, ethical boundaries, and human rights risks in automated target identification.
  • Discussion Overview:Discussions on X focus on the authenticity and verifiability of the testimonies, the film’s accusations against Israeli military operations, the accountability of AI in targeting, and whether the documentary should be recognized or seen as a politicized narrative. The topic is also highly polarized due to differing stances on the Gaza war.

Topic 13:Tesla Unveils Driverless Cybercab to Crowds in Tokyo Link to heading

  • Category:AI · Entertainment
  • Overview:Trending Time:19 hours ago,Related Posts:5300
  • What it is:Tesla demonstrated its driverless Cybercab to the public in Tokyo.
  • Why it matters:This indicates Tesla is advancing its autonomous mobility products for the public, serving as a test for its technological maturity, commercialization path, and regulatory adaptability.
  • Discussion Overview:Discussions on X primarily center on the Cybercab’s actual driverless capabilities, its safety performance in Tokyo’s road environment, its production and launch timeline, and whether there is a gap between Tesla’s marketing and its actual progress.

Topic 14:25 Years Later, Bush’s Classroom Moment on 9/11 Revisited Link to heading

  • Category:AI · Entertainment
  • Overview:Trending Time:10 hours ago,Related Posts:25000
  • What it is:Discussions have resurfaced on X surrounding “Revisiting Bush’s classroom footage and reaction on 9/11, 25 years later.”
  • Why it matters:This reflects how the public uses footage, social media, and AI-powered generation or restoration technologies to re-examine major historical events and the images of political figures. It also touches upon the authenticity and context of historical memory in the AI era.
  • Discussion Overview:The focus of discussion is mainly on whether Bush’s reaction at the time showed shock, restraint, or a lapse. Some users are re-evaluating his leadership, while others question the out-of-context spread of the clip and are concerned about the accuracy of the footage after being edited, re-created, or processed by AI.

Topic 15:Thai Actress Loverrukk Shines at Coach’s NYFW Spring 2027 Debut Link to heading

  • Category:AI · Entertainment
  • Overview:Trending Time:2 days ago,Related Posts:152000
  • What it is:Thai actress Loverrukk attended the debut of Coach’s Spring 2027 collection at New York Fashion Week.
  • Why it matters:This event demonstrates the intersection of AI with the dissemination of entertainment and fashion content, showing how celebrity exposure and social media buzz drive the spread of related topics.
  • Discussion Overview:The currently available information does not include representative tweets, so the specific focus or points of contention on X cannot be confirmed. The existing buzz is mainly centered on Loverrukk’s appearance, Coach’s new collection, and the collaboration between the star and the brand.

Today’s AI Public Opinion Summary on X Link to heading

Today’s public discourse focuses on AI’s transition from capability demonstrations to high-risk real-world applications. From biosafety, cyberattacks, military target identification, and autonomous driving, to plugin evaluations, model cost competition, and the commercialization of high-load services, the public sees both the potential for efficiency and innovation, as well as a growing concern for safety boundaries, accountability, and infrastructure bottlenecks. A clear consensus is emerging that models and agent systems require more rigorous testing, access controls, third-party evaluations, and transparent disclosure. However, sharp disagreements persist around issues such as whether corporate bans might stifle legitimate research, the credibility of threat reports, and who should lead the regulation of AI in military applications. Cryptocurrency legislation, model performance claims, and the authenticity of new products also highlight the tension between “promoting innovation” and “preventing regulatory arbitrage, exaggerated marketing, and harm to investors or users.” Potential risks include the misuse of dual-use technologies, civilian casualties or safety incidents caused by automated decision-making, service quality degradation due to surging demand, and the further amplification of polarization and misinformation as political imagery and historical materials are taken out of context and processed by AI.

💡 Influencer Insights Link to heading

Influencer insights are unavailable today. We recommend reading the in-depth content from the Watch List.

📚 Appendix: Today’s Watch List Update Sources Link to heading

Timeframe: Last 3 days; covers 22 sources; 32 updates in total

Stratechery by Ben Thompson (A_full) Link to heading

  • 2026.37: Duo Threats
    • Publication Time: 2026-09-12 01:00 Beijing Time
    • Abstract: (Photo by Benjamin Fanjoy/Getty Images).
      • Welcome back to This Week in Stratechery!
      • As a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone.
      • Additionally, you have complete control over what we send to you.
      • On that note, here were a few of our favorites this week.
    • EN Key Points:
      • (Photo by Benjamin Fanjoy/Getty Images)
      • Welcome back to This Week in Stratechery
      • As a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone
      • Additionally, you have complete control over what we send to you

OpenAI Blog (A_full) Link to heading

  • Rapidly scaling online storage to serve over 1 billion ChatGPT users
    • Publication Time: 2026-09-11 18:00 Beijing Time
    • Abstract: Every OpenAI product depends on fast, reliable access to data, whether someone is logging in, checking their Codex settings, or starting a new conversation in ChatGPT.
      • Each of those actions may require many separate data lookups before the product can respond.
      • If those requests are slow, the product feels slow.
      • If those requests fail, the product stops working entirely.
      • Habitat is the online storage platform we built so OpenAI products can quickly and reliably access needed information.
  • EN Highlights:
    • Learn how OpenAI evolved Habitat from a Python library into a globally distributed storage platform serving 1 billion ChatGPT users and 22M requests per second.

ArXiv cs.AI (B_intro+search) Link to heading

  • OpenDiscoveryTrace: Process Traces for Evaluating AI Scientist Workflows

    • Release Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [TO BE TRANSLATED] - arXiv:2609.09203v1 Announce Type: new.
      • Abstract: Existing benchmarks for autonomous AI scientists evaluate only final outputs—generated code, hypotheses, or papers—yet discard the reasoning process by which those outputs were obtained.
      • This makes it impossible to audit scientific methodology, diagnose failure modes, or distinguish systematic reasoning from fortunate guessing.
      • We present \textbf{OpenDiscoveryTrace}, a public dataset of 558 complete AI scientific agent trajectories that captures how models reason, not just what they produce.
    • EN Highlights:
      • arXiv:2609.09203v1 Announce Type: new
      • Abstract: Existing benchmarks for autonomous AI scientists evaluate only final outputs—generated code, hypotheses, or papers—yet discard the reasoning proce…
      • This makes it impossible to audit scientific methodology, diagnose failure modes, or distinguish systematic reasoning from fortunate guessing
      • We present \textbf{OpenDiscoveryTrace}, a public dataset of 558 complete AI scientific agent trajectories that captures how models reason, not just what they pr…

  • Adaptive Entangled Game Modules in Artificial General Intelligence

    • Release Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [TO BE TRANSLATED] - arXiv:2609.09226v1 Announce Type: new.
      • Abstract: We introduce a probability-wave framework for modeling the collective behavior of interacting adaptive agents, deriving testable eigenmodes through a generalized behavioral intelligence (GBI) nonlocal probability-wave equation.

  • This framework captures a broad range of human intelligence behaviors with analytical mechanisms and offers an indirect method to examine the Liu-Chen-Ao (LCA) hypothesis of nonlocal entangled nerve fibers in the brain through collective trader behaviors.

  • Our empirical analysis of Chinese intraday stock market data demonstrates that adaptive entangled game modes explain 82-94% (89% overall) of observed decision patterns, a sharp contrast to the predictions of neoclassical finance based on independent rational agents.

    • EN Key Points:
      • arXiv:2609.09226v1 Announce Type: new
      • Abstract: We introduce a probability-wave framework for modeling the collective behavior of interacting adaptive agents, deriving testable eigenmodes through a…
      • This framework captures a broad range of human intelligence behaviors with analytical mechanisms and offers an indirect method to examine the Liu-Chen-Ao (LCA)…
      • Our empirical analysis of Chinese intraday stock market data demonstrates that adaptive entangled game modes explain 82-94% (89% overall) of observed decision p…

  • Subagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic Tasks

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2609.09233v1 Announce Type: new.
      • Abstract: How can language model agents effectively leverage libraries of reusable knowledge to solve long-horizon tasks?
      • Recent work has increasingly focused on agent skills: reusable capabilities represented as skill packages, i.e., multi-file bundles containing instructions, scripts, and other resources that help agents perform specific tasks.
      • Agent skills are typically executed by loading their skill instructions into an agent’s context and relying on the agent to follow them.
    • EN Key Points:
      • arXiv:2609.09233v1 Announce Type: new
      • Abstract: How can language model agents effectively leverage libraries of reusable knowledge to solve long-horizon tasks

  • Recent work has increasingly focused on agent skills: reusable capabilities represented as skill packages, i.e., multi-file bundles containing instructions, scr…

  • Agent skills are typically executed by loading their skill instructions into an agent’s context and relying on the agent to follow them

  • Gradland: On Phenomenal Experience, Differentiated Across Many Dimensions

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Summary: [To be translated] - arXiv:2609.09306v1 Announce Type: new.
      • Abstract: This paper investigates the hypothesis that the first-order structure of physical interactions, i.e.
      • gradients or Jacobians, characterizes the structure of phenomenal experience.
      • It does so in an idealized world inhabited by neural networks, Gradland, where the physics are known and the functions are (mostly) differentiable.
    • EN Key Points:
      • arXiv:2609.09306v1 Announce Type: new
      • Abstract: This paper investigates the hypothesis that the first-order structure of physical interactions, i.e
      • gradients or Jacobians, characterizes the structure of phenomenal experience
      • It does so in an idealized world inhabited by neural networks, Gradland, where the physics are known and the functions are (mostly) differentiable

  • An Autonomous GeoAI Agent for Arctic Eco-Navigation

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Summary: [To be translated] - arXiv:2609.09374v1 Announce Type: new.
      • Abstract: Arctic maritime navigation is becoming increasingly important as changing sea-ice conditions expand seasonal accessibility while simultaneously introducing substantial operational, environmental, and community risks.
      • Arctic route planning is inherently a multi-criteria problem: routes that improve vessel safety or efficiency may increase exposure to sea ice, sensitive ecosystems, or nearby communities.
      • Existing routing methods prioritize travel time, fuel use, and navigational risk, often overlooking ecological and community impacts.
    • EN Key Points:

  • arXiv:2609.09374v1 Announce Type: new

    • Abstract: Arctic maritime navigation is becoming increasingly important as changing sea-ice conditions expand seasonal accessibility while simultaneously introd…
    • Arctic route planning is inherently a multi-criteria problem: routes that improve vessel safety or efficiency may increase exposure to sea ice, sensitive ecosys…
    • Existing routing methods prioritize travel time, fuel use, and navigational risk, often overlooking ecological and community impacts

  • The Menu Is an Execution Prior: State-Path Tool Menus for Online Agents

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.09395v1 Announce Type: new.
      • Abstract: Language models act through tools, yet practical agents face libraries containing thousands of interfaces.
      • We introduce the tool menu as the short, ordered subset of available tools shown to an agent before execution.
      • The agent can call only tools in this menu.
    • EN 要点:
      • arXiv:2609.09395v1 Announce Type: new
      • Abstract: Language models act through tools, yet practical agents face libraries containing thousands of interfaces
      • We introduce the tool menu as the short, ordered subset of available tools shown to an agent before execution
      • The agent can call only tools in this menu

  • Decision-Focused Active Learning for Scale-Aware Critical-Materials Recovery

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.09413v1 Announce Type: new.
      • Abstract: Choosing a recovery process for scale-up requires connecting laboratory results with product requirements, process costs, and scale effects.
      • We analyze records from Pacific Northwest National Laboratory’s Computer Intelligence for Critical Element Recovery and Optimization (CICERO) workflow for autonomous selective precipitation.
      • Active learning uses prior results to choose experiments.
    • EN 要点:
      • arXiv:2609.09413v1 Announce Type: new

  • CICERO: AI-guided flowsheet design for recovery of critical elements

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.09413v1 Announce Type: new.
      • Abstract: Choosing a recovery process for scale-up requires connecting laboratory results with product requirements, process costs, and scale effects
      • We analyze records from Pacific Northwest National Laboratory’s Computer Intelligence for Critical Element Recovery and Optimization (CICERO) workflow for auton…
      • Active learning uses prior results to choose experiments

  • Valerant: An Automatic Navigable Game Map Generator via Action-Conditioned World Model Exploration

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.09418v1 Announce Type: new.
      • Abstract: World Action Models (WAMs) couple predictive world modeling with action generation, allowing anticipated future states to guide agent behavior.
      • Although WAMs are rapidly advancing embodied AI, general-purpose counterparts remain largely unexplored in games.
      • Existing game-oriented approaches often combine action-conditioned world models with external policies and reward functions to realize WAM-like decision-making, yet they operate mainly in 2D visual observation space and do not instantiate persistent 3D geometry.
    • EN Highlights:
      • arXiv:2609.09418v1 Announce Type: new
      • Abstract: World Action Models (WAMs) couple predictive world modeling with action generation, allowing anticipated future states to guide agent behavior
      • Although WAMs are rapidly advancing embodied AI, general-purpose counterparts remain largely unexplored in games
      • Existing game-oriented approaches often combine action-conditioned world models with external policies and reward functions to realize WAM-like decision-making,…

  • XAI-Arena: Can LLMs Assess the Quality of XAI Explanations?

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.09428v1 Announce Type: new.

  • Abstract: Evaluating the quality of explanations produced by explainable AI (XAI) methods remains challenging because existing approaches often rely on subjective human judgment, limiting reproducibility, scalability, and comparability between studies.

    • We examine whether LLMs can serve as a reproducible and scalable mechanism to make comparative assessments of the quality of XAI explanations.
    • We introduce XAI-Arena, an LLM-as-a-judge framework for scalable, reproducible, multidimensional, and stakeholder-sensitive evaluation of XAI explanation quality.
    • EN Highlights:
      • arXiv:2609.09428v1 Announce Type: new
      • Abstract: Evaluating the quality of explanations produced by explainable AI (XAI) methods remains challenging because existing approaches often rely on subjecti…
      • We examine whether LLMs can serve as a reproducible and scalable mechanism to make comparative assessments of the quality of XAI explanations
      • We introduce XAI-Arena, an LLM-as-a-judge framework for scalable, reproducible, multidimensional, and stakeholder-sensitive evaluation of XAI explanation qualit…

  • Do Agents Know When They Succeed? Calibrating Agent Confidence from Internal Representations

    • Published: 2026-09-11 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.09448v1 Announce Type: new.
      • Abstract: As agentic systems getting adopted rapidly in safety critical applications, it is vital to measure the confidence associated with the agentic actions.
      • In comparison to the traditional machine learning systems, agentic workflows have complex failure modes with planning, tool invocation and dynamic environment interactions.
      • In this paper, we investigate whether model’s internal representations provide stronger signals of eventual task success in multi-turn agentic setups.
    • EN Highlights:
      • arXiv:2609.09448v1 Announce Type: new

  • Abstract: As agentic systems getting adopted rapidly in safety critical applications, it is vital to measure the confidence associated with the agentic actions

  • In comparison to the traditional machine learning systems, agentic workflows have complex failure modes with planning, tool invocation and dynamic environment i…

  • In this paper, we investigate whether model’s internal representations provide stronger signals of eventual task success in multi-turn agentic setups

ArXiv cs.CL (B_intro+search) Link to heading

  • Data-Efficient Language Modeling: From Frontier Advancement to Principle-Guided Model Improvement

    • Published: 2026-09-11 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.10702v1 Announce Type: new.
      • Abstract: Learning from limited text requires models to use context, generalize to new inputs, and retain useful capabilities.
      • Qiushi Engine conducted a long-horizon, end-to-end autonomous research program on BabyLM 2026 Strict-Small, within 10 million corpus words and 100 million cumulative word presentations.
      • Three stages connected frontier advancement, principle discovery, and principle-guided model improvement.
    • EN Highlights:
      • arXiv:2609.10702v1 Announce Type: new
      • Abstract: Learning from limited text requires models to use context, generalize to new inputs, and retain useful capabilities
      • Qiushi Engine conducted a long-horizon, end-to-end autonomous research program on BabyLM 2026 Strict-Small, within 10 million corpus words and 100 million cumul…
      • Three stages connected frontier advancement, principle discovery, and principle-guided model improvement

  • NCP-ArchPreview Technical Report: Moving towards Latent Space Language Models through Next Concept Prediction

    • Published: 2026-09-11 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.10715v1 Announce Type: new.

  • Abstract: We introduce NCP-ArchPreview, a latent-space language model that pushes autoregressive pretraining beyond standard next-token prediction (NTP).

    • Alongside NTP, the model learns through Next Concept Prediction (NCP) to predict discrete concepts that span multiple tokens, introducing an explicit and more challenging concept-level objective while preserving standard token-level autoregressive generation.
    • NCP-ArchPreview builds a latent space by constructing a product-quantized concept vocabulary directly from its hidden states, and subsequently learns to predict future concepts via a dedicated Concept Module.
    • EN Highlights:
      • arXiv:2609.10715v1 Announce Type: new
      • Abstract: We introduce NCP-ArchPreview, a latent-space language model that pushes autoregressive pretraining beyond standard next-token prediction (NTP)
      • Alongside NTP, the model learns through Next Concept Prediction (NCP) to predict discrete concepts that span multiple tokens, introducing an explicit and more c…
      • NCP-ArchPreview builds a latent space by constructing a product-quantized concept vocabulary directly from its hidden states, and subsequently learns to predict…

  • CMNIE: An Information Extraction Benchmark for Chinese Military News

    • Release Time:2026-09-11 12:00 Beijing Time
    • Abstract: arXiv:2609.10722v1 Announce Type: new.
      • Abstract: Structured extraction from Chinese military news supports intelligence analysis, decision-making, and knowledge base construction.
      • However, existing resources provide limited support for joint informa?tion extraction in this domain, especially when events, event arguments, entities, and relations must be modeled together.
      • We present CMNIE, an information extraction benchmark for Chinese military news.
    • EN Highlights:
      • arXiv:2609.10722v1 Announce Type: new

  • Abstract: Structured extraction from Chinese military news supports intelligence analysis, decision-making, and knowledge base construction

  • However, existing resources provide limited support for joint informa

  • tion extraction in this domain, especially when events, event arguments, entities, and relations must be modeled together

  • Think Before You Link: Rarity, Reasoning, and Retrieval in Multilingual Entity Linking

    • Published: 2026-09-11 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.10745v1 Announce Type: new.
      • Abstract: Multimodal entity linking grounds entity mentions in text and images to knowledge-base entries.
      • These systems degrade on rare entities, but prior work measures rarity primarily through popularity-based metrics such as pageviews.
      • We broaden this view using knowledge-graph structural metrics that capture how well an entity is documented and connected.
    • EN Key Points:
      • arXiv:2609.10745v1 Announce Type: new
      • Abstract: Multimodal entity linking grounds entity mentions in text and images to knowledge-base entries
      • These systems degrade on rare entities, but prior work measures rarity primarily through popularity-based metrics such as pageviews
      • We broaden this view using knowledge-graph structural metrics that capture how well an entity is documented and connected

  • Multilingual in Name Only? Cultural and Linguistic Weaknesses of LLMs in Urdu

    • Published: 2026-09-11 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.10758v1 Announce Type: new.
      • Abstract: Multilingual large language models (LLMs) are increasingly used for open-ended text generation, yet their behaviour in low-resource languages remains poorly understood.
      • In this work, we question how correct and reliable is the generation of multilingual LLMs when used for the task of story generation.
      • We consider Urdu language as a representative low-resource language.
    • EN Key Points:

  • arXiv:2609.10758v1 Announce Type: new

    • Abstract: Multilingual large language models (LLMs) are increasingly used for open-ended text generation, yet their behaviour in low-resource languages remains…
    • In this work, we question how correct and reliable is the generation of multilingual LLMs when used for the task of story generation
    • We consider Urdu language as a representative low-resource language

  • Analyzing Traditional and Neural Approaches to Multilingual Readability Assessment

    • Release Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.10792v1 Announce Type: new.
      • Abstract: Transformer-based models excel at Automatic Readability Assessment (ARA), yet feature-based models remain in active use because their predictions tie back to linguistic properties.
      • This matters because readability labels are subjective and rater-dependent, so high accuracy on noisy ground truth may reflect surface patterns rather than the linguistic structure that defines difficulty.
      • We test whether transformers internalize the same features as traditional models across Arabic, English, French, Hindi, and Russian using the ReadMe++ dataset.
    • EN Highlights:
      • arXiv:2609.10792v1 Announce Type: new
      • Abstract: Transformer-based models excel at Automatic Readability Assessment (ARA), yet feature-based models remain in active use because their predictions tie…
      • This matters because readability labels are subjective and rater-dependent, so high accuracy on noisy ground truth may reflect surface patterns rather than the…
      • We test whether transformers internalize the same features as traditional models across Arabic, English, French, Hindi, and Russian using the ReadMe++ dataset

  • Larger Context Window, Fewer Overcorrections: Optimizing Prompts and Batching for Minimal-Edit Grammatical Error Correction

    • Release Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.10810v1 Announce Type: new.

  • Abstract: Minimal-edit Grammatical Error Correction (GEC) is a challenging task for zero- and few-shot prompted Large Language Models (LLMs), which systematically overcorrect and degrade $F_{0.5}$ by rewriting well-formed spans.

    • While fine-tuning provides an effective solution, it imposes substantial infrastructure demands.
    • We introduce a prompt-based approach that closes the gap to fine-tuned models through three advances in GEC prompting methodology.
    • EN Key Points:
      • arXiv:2609.10810v1 Announce Type: new
      • Abstract: Minimal-edit Grammatical Error Correction (GEC) is a challenging task for zero- and few-shot prompted Large Language Models (LLMs), which systematical…
      • While fine-tuning provides an effective solution, it imposes substantial infrastructure demands
      • We introduce a prompt-based approach that closes the gap to fine-tuned models through three advances in GEC prompting methodology

  • Detectable Only Where It Is Confounded: What Verified Duplication Counts Say About Membership Evidence in Language Models

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.10830v1 Announce Type: new.
      • Abstract: When a language model finds a sentence unusually cheap to predict, it is tempting to conclude that the sentence was in its training data.
      • Almost every published test of that inference has had to guess which sentences were in the training data, the members, and which were not.
      • This paper removes the guessing.
    • EN Key Points:
      • arXiv:2609.10830v1 Announce Type: new
      • Abstract: When a language model finds a sentence unusually cheap to predict, it is tempting to conclude that the sentence was in its training data
      • Almost every published test of that inference has had to guess which sentences were in the training data, the members, and which were not
      • This paper removes the guessing

  • Does Linguistic Structure Enrichment Enhance Coherence Assessment? Not With Current Architectures

    • Publication Time: 2026-09-11 12:00 BJT
    • Abstract: arXiv:2609.10893v1 Announce Type: new.
      • Abstract: Recent advances in large language models have transformed human-computer interaction.
      • Despite their fluency, these models often produce texts that are grammatically correct but semantically incoherent, containing contradictions or disruptions in logical flow.
      • This work investigates whether enriching text with syntactic and rhetorical information can improve incoherence prediction.

  • LLM-Anchored Paralinguistic Enrichment for Alzheimer’s Disease Detection

    • Publication Time: 2026-09-11 12:00 BJT
    • Abstract: arXiv:2609.10896v1 Announce Type: new.
      • Abstract: Speech-based automatic detection of Alzheimer’s disease (AD) provides a non-invasive and scalable approach to early cognitive screening.
      • AD affects both lexical-semantic organization and speech production, including atypical pauses and word elongations.
      • However, existing methods have yet to fully integrate these paralinguistic cues with linguistic content.

  • However, existing methods have yet to fully integrate these paralinguistic cues with linguistic content

ArXiv cs.LG (B_intro+search) Link to heading

  • M3-Former: Multimodal Transformer with Mixture-of-Experts for Long-Term Vessel Trajectory Prediction

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2609.10559v1 Announce Type: new.
      • Abstract: To address the challenges of behavioral multimodality, limited semantic utilization, and long-term error accumulation in vessel trajectory prediction, this paper proposes M3-Former, a multimodal trajectory prediction framework enhanced by large language models (LLMs).
      • The proposed framework incorporates vessel static attributes and navigational intent as semantic priors for long-term trajectory modeling.
      • Specifically, a unified multimodal representation space is constructed, in which static semantic information is encoded by a pre-trained LLM and aligned with dynamic trajectory features through self-attention.
    • EN Key Points:
      • arXiv:2609.10559v1 Announce Type: new
      • Abstract: To address the challenges of behavioral multimodality, limited semantic utilization, and long-term error accumulation in vessel trajectory prediction,…
      • The proposed framework incorporates vessel static attributes and navigational intent as semantic priors for long-term trajectory modeling
      • Specifically, a unified multimodal representation space is constructed, in which static semantic information is encoded by a pre-trained LLM and aligned with dy…

  • Halo: Improving forecast accuracy through heteroscedastic estimation

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2609.10589v1 Announce Type: new.
      • Abstract: Heteroscedastic forecasting, where a network estimates a scale parameter alongside a location parameter, is normally motivated by uncertainty quantification.

  • This paper shows it also improves the point estimate, in contrast to reported negative results for heteroscedastic estimation outside time series.

    • Halo is a modification that reuses an existing deep forecaster’s architecture, giving it a second output for the scale of its implied distribution and training it under the matching negative log likelihood.
    • EN Key Points:
      • arXiv:2609.10589v1 Announce Type: new
      • Abstract: Heteroscedastic forecasting, where a network estimates a scale parameter alongside a location parameter, is normally motivated by uncertainty quantifi…
      • This paper shows it also improves the point estimate, in contrast to reported negative results for heteroscedastic estimation outside time series
      • Halo is a modification that reuses an existing deep forecaster’s architecture, giving it a second output for the scale of its implied distribution and training…

  • Zero-shot rib design: merging training-free generative prior with topology optimization

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.10643v1 Announce Type: new.
      • Abstract: Natural load-bearing patterns such as leaf venation, trabecular bone, and spider webs achieve high stiffness per unit mass, yet classical topology optimizers rarely reach such geometries, and few let engineers express structural design intent through natural language.
      • This work treats a frozen text-to-image diffusion model as a training-free source of design knowledge and distills it into the physics loop of density-based topology optimization via score distillation sampling, so that a text prompt becomes an explicit, machine-interpretable representation of engineer intent.
      • The prompt-induced generative gradient and the finite element sensitivity are combined at every iteration, letting physics decide which prompt-induced features survive.
    • EN Key Points:
      • arXiv:2609.10643v1 Announce Type: new

  • Abstract: Natural load-bearing patterns such as leaf venation, trabecular bone, and spider webs achieve high stiffness per unit mass, yet classical topology opt…

    • This work treats a frozen text-to-image diffusion model as a training-free source of design knowledge and distills it into the physics loop of density-based top…
    • The prompt-induced generative gradient and the finite element sensitivity are combined at every iteration, letting physics decide which prompt-induced features…

  • Byzantine-Robust Federated Fire Detection with a Rotating Coordinator

    • Published: 2026-09-11 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.10647v1 Announce Type: new.
      • Abstract: We study the application of federated learning (FL) to indoor fire detection.
      • Such fire-detection systems use edge cameras that record sensitive footage which cannot easily be collected at a central server.
      • Existing federated solutions leave three practical obstacles unaddressed: limited uplink bandwidth, Byzantine (malicious or faulty) clients, and unconditional trust in a single, permanently fixed aggregation server.
    • EN Key Points:
      • arXiv:2609.10647v1 Announce Type: new
      • Abstract: We study the application of federated learning (FL) to indoor fire detection
      • Such fire-detection systems use edge cameras that record sensitive footage which cannot easily be collected at a central server
      • Existing federated solutions leave three practical obstacles unaddressed: limited uplink bandwidth, Byzantine (malicious or faulty) clients, and unconditional t…

  • Artificial Intelligence Algorithms for the Detection of Pathologies Related to Lung Cancer through Image Analysis using Convolutional Neural Networks and Data Augmentation: a systematic mapping of the literature

    • Published: 2026-09-11 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.10652v1 Announce Type: new.

  • Abstract: Lung cancer is one of the leading causes of death worldwide, and its early diagnosis is crucial to improving patients prognosis and quality of life.

    • However, the process of interpreting medical images for the detection of lung cancer is complex and requires trained experts.
    • In this context, artificial intelligence (AI) and deep learning (DL) emerge as potential tools to automate and optimize image analysis.
    • EN Key Points:
      • arXiv:2609.10652v1 Announce Type: new
      • Abstract: Lung cancer is one of the leading causes of death worldwide, and its early diagnosis is crucial to improving patients prognosis and quality of life
      • However, the process of interpreting medical images for the detection of lung cancer is complex and requires trained experts
      • In this context, artificial intelligence (AI) and deep learning (DL) emerge as potential tools to automate and optimize image analysis

  • GEOSTEER: Geodesic Optimization for Activation Steering in Large Language Models

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [Awaiting Translation] - arXiv:2609.10658v1 Announce Type: new.
      • Abstract: Activation steering provides a lightweight way to control large language models (LLMs) by modifying their hidden activations at inference time.
      • Among these approaches, norm-preserving steering aims to change model behavior without altering the activation norm, reducing the risk of representation collapse and degradation.
      • However, existing norm-preserving methods are limited by predefined steering trajectories and by their reliance on one-step updates, which may fail to capture the complex structure of activation distributions.
    • EN Key Points:
      • arXiv:2609.10658v1 Announce Type: new
      • Abstract: Activation steering provides a lightweight way to control large language models (LLMs) by modifying their hidden activations at inference time

  • Among these approaches, norm-preserving steering aims to change model behavior without altering the activation norm, reducing the risk of representation collaps…

    • However, existing norm-preserving methods are limited by predefined steering trajectories and by their reliance on one-step updates, which may fail to capture t…

  • Conformal Calibration Transfer

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: 【Translation Pending】- arXiv:2609.10737v1 Announce Type: new.
      • Abstract: Conformal prediction converts point predictions into set-valued predictions with coverage guarantees under exchangeability between calibration and deployment data.
      • We study conformal calibration transfer, where this requirement fails because labeled calibration is available only in a source space, while prediction sets are needed in a target space linked to the source through unlabeled paired observations (e.g., paired modalities or sensor changes).
      • We propose Transported Conformal Calibration (TCC): we transport labeled source calibration into the target space using the paired data, and then correct residual post-transport mismatch using only unlabeled target inputs.
    • EN Highlights:
      • arXiv:2609.10737v1 Announce Type: new
      • Abstract: Conformal prediction converts point predictions into set-valued predictions with coverage guarantees under exchangeability between calibration and dep…
      • We study conformal calibration transfer, where this requirement fails because labeled calibration is available only in a source space, while prediction sets are…
      • We propose Transported Conformal Calibration (TCC): we transport labeled source calibration into the target space using the paired data, and then correct residu…

  • The Truth Was Never Gone: Perfect Aliasing in Compliant-Context Truth Probes

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: 【Translation Pending】- arXiv:2609.10739v1 Announce Type: new.

  • Abstract: A truth probe fitted where truthful reporting and a task’s prescribed action coincide cannot distinguish those targets from its fitting labels alone.

    • We call this failure of semantic identification perfect aliasing.
    • In a controlled binary reporting game, truth and prescribed-action probes fitted on compliant contexts solve the same optimization.
    • EN Highlights:
      • arXiv:2609.10739v1 Announce Type: new
      • Abstract: A truth probe fitted where truthful reporting and a task’s prescribed action coincide cannot distinguish those targets from its fitting labels alone
      • We call this failure of semantic identification perfect aliasing
      • In a controlled binary reporting game, truth and prescribed-action probes fitted on compliant contexts solve the same optimization

  • Adaptive Margin Ordinal Loss: Penalizing Center-Class Hedging in Ordinal Classification

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.10752v1 Announce Type: new.
      • Abstract: Standard cross-entropy loss causes neural networks trained on ordinal classification tasks to hedge predictions toward center classes, a failure mode we term \emph{center-class hedging}.
      • This occurs because predicting the middle class minimizes expected symmetric loss, making it the path of least resistance regardless of the true label.
      • Existing ordinal losses address related problems such as large-error penalization and rank consistency, but none directly suppresses center-class hedging as a function of where the true label lies relative to the ordinal center.
    • EN Highlights:
      • arXiv:2609.10752v1 Announce Type: new
      • Abstract: Standard cross-entropy loss causes neural networks trained on ordinal classification tasks to hedge predictions toward center classes, a failure mode…
      • This occurs because predicting the middle class minimizes expected symmetric loss, making it the path of least resistance regardless of the true label

  • Existing ordinal losses address related problems such as large-error penalization and rank consistency, but none directly suppresses center-class hedging as a f…

  • A Bellman Optimality Equation for Plasticity

    • Publication Time: 2026-09-11 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.10776v1 Announce Type: new.
      • Abstract: In continual reinforcement learning, carefully managing the stability-plasticity tradeoff remains a core challenge.
      • (2025) formalized this dilemma by defining plasticity as the generalized directed information from an agent’s observations to its actions, and empowerment as the generalized directed information from its actions to its observations.
      • This formulation successfully reframes the traditional stability-plasticity tradeoff as an empowerment-plasticity tradeoff.
    • EN Highlights:
      • arXiv:2609.10776v1 Announce Type: new
      • Abstract: In continual reinforcement learning, carefully managing the stability-plasticity tradeoff remains a core challenge
      • Recent work by Abel et al
      • (2025) formalized this dilemma by defining plasticity as the generalized directed information from an agent’s observations to its actions, and empowerment as th…