System translated (Gemini)

🤖 AI 速览

The key development today is not about new benchmark scores, but whether AI can stably and cost-effectively complete long-term tasks. New research is beginning to evaluate the marginal value of agent memory on practical execution and transforming generative evaluation into actionable feedback for …
📋 文章元数据
发布时间
2026-09-10
类型
ai-daily
字数
6456
阅读时长
31 min

2026-09-10 AI Daily | As AI Starts Long-Term Work: Memory Costs, Process Feedback, and Safety Governance Emerge Link to heading

The key shift today isn’t about models breaking new score records, but whether AI can stably and cost-effectively complete long-term tasks. New research is beginning to assess the marginal value of agent memory on actual execution and transform generation evaluations into actionable improvement feedback. Meanwhile, OpenAI’s invitation for alignment researchers to join its foundation’s safety governance system shows that the capability expansion of frontier models is advancing in step with accountability mechanisms.

📖 Deep Dive: This Issue’s Watch List Link to heading

The most important theme to watch today is the shift of AI agents from “capability showcases” to “long-term governance.” OpenAI is bringing Paul Christiano into its foundation and safety affairs while releasing GPT-6 Astra for workplace scenarios. Related research further asks: who will maintain agent skills, how to design governance mechanisms for autonomous goals, and why safety monitoring with “high recall doesn’t equal protection.” Meta’s release of its personal agent also warrants examination in light of these governance issues.

The second major theme is the refinement of evaluation systems. Medical speech recognition, Tamil speech endpoint detection, and opioid use identification show that real-world scenarios care more about critical errors and reproducibility than a single average score. At the same time, work on RAG query rewriting, parallel decoding, and CriticGen is advancing evaluation from simply “judging right or wrong” to “explaining and improving the generation process.”

🌐 AI Hotspots on X Link to heading

Topic 1: DeepSeek V4.1 Flash Delivers Top AI Performance at Low Cost Link to heading

  • Category: AI · News
  • Overview: Trending 1 day ago, 14,000 related posts
  • What it is: According to the headline, DeepSeek V4.1 Flash is reported to have achieved leading AI performance at a low cost.
  • Why it matters: This pertains to the inference costs of high-performance AI models, their commercial accessibility, and the impact of open-source or low-cost models on the existing market landscape.
  • Discussion summary: Discussions on X likely focus on the reliability of its performance benchmarks, the actual extent of its cost advantage, comparisons with other mainstream models, and whether this result is sustainable. Specific points of disagreement cannot be further confirmed as no representative posts were provided.

Topic 2: AI Researcher Quits Anthropic Over Superintelligence Risks Link to heading

  • Category: AI · News
  • Overview: Trending 23 hours ago, 723,000 related posts
  • What it is: Anthropic researcher Jacob Coxon resigned over concerns that self-improving superintelligence could lead to loss of control and human extinction, publicly criticizing the industry’s race to build it.
  • Why it matters: The event highlights the conflict within frontier AI labs between capability competition, alignment research, risk disclosure, and regulatory responsibility. It also brings the probability of superintelligence risks and the adequacy of existing safety mechanisms back into the public debate.
  • Discussion summary: Discussions on X primarily revolve around whether Coxon’s warning is credible, Anthropic’s probability assessment of future AI-induced disasters, and whether companies are weakening safety regulations due to competition and lobbying. The main point of disagreement is whether to immediately slow down or strengthen regulation, or to continue development while relying on safety research to mitigate risks.

Topic 3: Omarchy Linux Distro Gains Traction with AI Agents and Sleek Design Link to heading

  • Category: AI · News
  • Overview: Trending 2 days ago, 11,000 related posts
  • What it is: Omarchy Linux, a distribution aimed at developers, has gained attention on X for its support of AI agent workflows and its clean desktop design.
  • Why it matters: It reflects a trend where AI agent applications are driving developers to refocus on controllable, lightweight, and easily automatable operating system environments, potentially influencing local AI toolchains and the developer experience.
  • Discussion summary: Discussions mainly focus on Omarchy’s interface design, installation and configuration experience, the practical extent of its support for AI agents, hardware compatibility, system stability, and its advantages compared to existing Linux distributions. Some users believe its popularity stems more from its design aesthetic than from technical breakthroughs.

Topic 4: Meme Juxtaposes AI Doomsday Warnings with Apple’s Camera Upgrades Link to heading

  • Category: AI · News
  • Overview: Trending 2 hours ago, 1,400 related posts
  • What it is: A meme juxtaposes AI doomsday warnings with Apple’s continuous camera upgrades, creating a stark contrast.
  • Why it matters: This reflects the gap between the grand public discourse on AI safety risks and the incremental innovation in consumer electronics. It helps in observing how the AI boom affects tech industry topics and public attention.
  • Discussion summary: The focus of discussion on X is mainly on whether AI doomsday theories are overhyped, whether Apple’s camera upgrades lack innovation, and whether there is an imbalance in tech companies’ focus on high-risk AI issues versus actual product experience.

Topic 5: Meme Reimagines Job Interview as Robot Apocalypse Stand-Off Link to heading

  • Category: AI · Entertainment
  • Overview: Trending Time: , Related Posts: 2700
  • What it is: An internet meme dramatically reframes a job interview as a standoff between humans and robots in an apocalyptic scenario.
  • Why it matters: It presents the impact of AI entering the workplace on employment, job competition, and human-machine relationships in an entertaining way, reflecting the social imagination sparked by AI’s evolution from a conversational tool to an intelligent agent and enterprise workflow component.
  • Discussion Summary: Discussions on X primarily revolve around whether this setup is a commentary on AI taking jobs, interview anxiety, and human fears of automation. It also extends to the anthropomorphic expression of robots and whether the entertaining narrative exaggerates AI’s impact on the job market.

Topic 6: Teaser Trailer Drops for Guadagnino’s Sam Altman Drama Artificial Link to heading

  • Category: AI · Entertainment
  • Overview: Trending Time: 1 day ago, Related Posts: 52000
  • What it is: The first teaser trailer for “Artificial,” a drama directed by Luca Guadagnino focusing on the story of Sam Altman, has been released, drawing attention on the X platform.
  • Why it matters: The film brings the OpenAI power dynamics, AI commercialization, and the image of tech leaders into mainstream cinematic narrative, showing that generative AI has expanded from a technical issue to a topic of popular culture and public governance.
  • Discussion Summary: Discussions on X are focused on whether the film will accurately portray the internal crisis at OpenAI, whether Sam Altman’s public image has been overly dramatized, and whether Hollywood is prematurely adapting events from the still-developing AI industry into entertainment.

AI Public Opinion Summary on X Today Link to heading

Today’s main narrative revolves around the tension between the expansion of AI capabilities and society’s capacity to adapt. On one hand, topics like DeepSeek V4.1 Flash and Omarchy Linux reinforce the consensus that low-cost, high-performance models, localized toolchains, and AI agent workflows are rapidly becoming widespread. On the other hand, the departure of a researcher from Anthropic, the job interview vs. robot meme, and related film projects have amplified concerns about superintelligence getting out of control, job displacement, corporate governance, and the power of tech. The main point of contention is whether the industry should continue to push the capabilities race while relying on safety research to mitigate risks, or if it should immediately slow down R&D and strengthen regulation. Meanwhile, the public is also debating whether the “AI apocalypse” narrative is a necessary wake-up call or excessive hype. Potential risks include the lack of reproducible verification for model performance and cost claims, companies downplaying safety responsibilities amidst commercial competition, and the tendency for entertainment and memes to amplify anxiety, blurring the line between real risks and technological hype.

💡 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 Source List Link to heading

Timeframe: Last 3 days; 22 sources covered; 33 updates in total

Stratechery by Ben Thompson (A_full) Link to heading

  • OpenAI Does Math, Reward-Hacking, Meta Launches Personal Agent
    • Publication Time: 2026-09-09 18:00 Beijing Time
    • Summary: [TO BE TRANSLATED] - OpenAI solving one of the most famous math problems is extremely impressive, and of little impact to most people’s lives; Meta’s Muse agent launch has the potential to be the exact opposite.
      • $15 / month or $150 / year.
      • Substantial analysis of the news of the day delivered via three weekly emails or podcasts.
      • Stratechery Interviews.
      • Interviews with leading public CEOs, private company founders, and discussions with fellow analysts.
    • EN Key Points:
      • OpenAI solving one of the most famous math problems is extremely impressive, and of little impact to most people’s lives; Meta’s Muse agent launch has the poten…

OpenAI Blog (A_full) Link to heading

  • Paul Christiano joins OpenAI Foundation Board

    • Publish Time: 2026-09-10 01:00 Beijing Time
    • Summary: We’re announcing the appointment of Paul Christiano to the OpenAI Foundation Board.
      • He will be a non-voting observer on the OpenAI Group PBC Board.
      • Paul will also join the Safety and Security Committee (SSC) of the Foundation Board, working alongside its chair, Zico Kolter.
      • The SSC provides governance over safety and security practices across all of OpenAI, including OpenAI Group PBC.
      • Department of Commerce, where he is a Senior Tech Advisor1.
    • EN Key Points:
      • Paul Christiano joins the OpenAI Foundation Board and its Safety and Security Committee, bringing experience in AI alignment, safety, and standards.

  • GPT-6 Astra: The next generation in intelligence for work

    • Publish Time: 2026-09-09 19:00 Beijing Time
    • Summary: Last week we introduced GPT‑6 Astra, the world’s most intelligent and aligned model, now available in ChatGPT Work, Codex, and the API.
      • Astra is state-of-the-art on computer use, browsing, professional work, software engineering, cybersecurity, and science, so teams can take on the most demanding professional work with unmatched speed, accuracy, and judgment.

      • The world’s best model for complex work. Link to heading

      • Most AI systems require businesses to prepare their data, redesign workflows, and build custom integrations before they can deliver value.
      • In ChatGPT Work and Codex, it can write code and work through the same applications people use every day—even when those applications don’t have an API.
    • EN Key Points:
      • Meet GPT-6 Astra, OpenAI’s most capable model for business, with advanced reasoning, computer use, and stronger writing and design judgment.

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

  • Beyond Right and Wrong: Evaluating Second-order Social Reasoning in Large Language Models

    • Publish Time: 2026-09-09 12:00 Beijing Time

  • Summary: [To be translated] - arXiv:2609.05437v1 Announce Type: new.

    • Abstract: Previous AI alignment efforts have focused primarily on first-order social norms – teaching models what is socially acceptable or unacceptable (e.g., `do not steal’).
    • However, social intelligence depends not only on norm recognition, but also on anticipating who will enforce it and how (e.g., public shame or even imprisonment).
    • These second-order expectations, known as metanorms, govern how people respond when social rules are broken.
    • EN Key Points:
      • arXiv:2609.05437v1 Announce Type: new
      • Abstract: Previous AI alignment efforts have focused primarily on first-order social norms – teaching models what is socially acceptable or unacceptable (e.g.,…
      • However, social intelligence depends not only on norm recognition, but also on anticipating who will enforce it and how (e.g., public shame or even imprisonment…
      • These second-order expectations, known as metanorms, govern how people respond when social rules are broken

  • CriticGen: Generation-Aware Evaluation as Actionable Feedback

    • Publication Time: 2026-09-09 12:00 Beijing Time
    • Summary: [To be translated] - arXiv:2609.05439v1 Announce Type: new.
      • Abstract: Current evaluation methods for large language models are coarse-grained and decoupled from generation, producing generic explanations that fail to provide actionable feedback for model improvement.
      • We propose CriticGen, a fine-grained, generation-aware evaluation framework that turns evaluation into actionable control for answer improvement.
      • CriticGen first generates sample-specific evaluation dimensions and scoring criteria under high-level categories such as subjective, objective, and self-derived constraints.
    • EN Key Points:
      • arXiv:2609.05439v1 Announce Type: new
      • Abstract: Current evaluation methods for large language models are coarse-grained and decoupled from generation, producing generic explanations that fail to pro…

  • We propose CriticGen, a fine-grained, generation-aware evaluation framework that turns evaluation into actionable control for answer improvement

  • CriticGen first generates sample-specific evaluation dimensions and scoring criteria under high-level categories such as subjective, objective, and self-derived…

  • When Does Memory Help? A Cost-Aware Evaluation of Long-Term Memory in Tool-Using LLM Agents

    • Publication Time: 2026-09-09 12:00 Beijing Time
    • Summary: [Translation Pending] - arXiv:2609.05441v1 Announce Type: new.
      • Abstract: Long-term memory for LLM agents is evaluated today by conversational recall benchmarks (LoCoMo, LongMemEval), which measure question answering over dialogue history, not whether remembered facts change what a tool-using agent does.
      • We present MERIT (Memory Evaluation for Realistic Instrumented Tasks), a benchmark and harness that measures the marginal utility of memory for task-executing agents under explicit cost accounting.
      • MERIT provides episodic tool-use tasks in three domains whose dependence on earlier-episode facts is verified by an automated leak check; a difficulty ladder ending in updated-fact recall; controlled memory corruption; and full token and dollar metering of every memory operation.
    • EN Key Points:
      • arXiv:2609.05441v1 Announce Type: new
      • Abstract: Long-term memory for LLM agents is evaluated today by conversational recall benchmarks (LoCoMo, LongMemEval), which measure question answering over di…
      • We present MERIT (Memory Evaluation for Realistic Instrumented Tasks), a benchmark and harness that measures the marginal utility of memory for task-executing a…
      • MERIT provides episodic tool-use tasks in three domains whose dependence on earlier-episode facts is verified by an automated leak check; a difficulty ladder en…

  • AutoFyn Technical Report: Non-Parametric Expert Iteration for Long-Horizon Agents

    • Publication Time: 2026-09-09 12:00 Beijing Time

  • Abstract: [Translation Pending] - arXiv:2609.05446v1 Announce Type: new.

    • Abstract: We introduce AutoFyn, an agent harness inspired by the Expert Iteration algorithm, adapting a frozen model across many rounds by updating persistent state from verified reward signals rather than model weights.
    • Each round begins from a fresh model session, and durable information is reintroduced only through explicit interfaces such as persistent memory files, reports, and repository state.
    • Within a round, an orchestrator explores, plans and builds many alternative approaches with specialized agents, while a task-grounded verifier verifies the work and supplies an objective reward for measuring progress.
    • EN Highlights:
      • arXiv:2609.05446v1 Announce Type: new
      • Abstract: We introduce AutoFyn, an agent harness inspired by the Expert Iteration algorithm, adapting a frozen model across many rounds by updating persistent s…
      • Each round begins from a fresh model session, and durable information is reintroduced only through explicit interfaces such as persistent memory files, reports,…
      • Within a round, an orchestrator explores, plans and builds many alternative approaches with specialized agents, while a task-grounded verifier verifies the work…

  • Damage-Aware Bandit Pruning for Vision and Language Transformers

    • Publication Time: 2026-09-09 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.05448v1 Announce Type: new.
      • Abstract: Structured post-training pruning of transformers requires selecting complete functional units whose suppression causes limited degradation.
      • We formulate structured-unit selection for language and vision transformers as a damage-aware multi-armed bandit problem under a fixed candidate-evaluation budget.
      • Attention heads and MLP channel groups are temporarily masked on calibration batches.
    • EN Highlights:
      • arXiv:2609.05448v1 Announce Type: new

  • Abstract: Structured post-training pruning of transformers requires selecting complete functional units whose suppression causes limited degradation

  • We formulate structured-unit selection for language and vision transformers as a damage-aware multi-armed bandit problem under a fixed candidate-evaluation budg…

  • Attention heads and MLP channel groups are temporarily masked on calibration batches

  • Compiling VGDL into Causal Models

    • Published: 2026-09-09 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.05459v1 Announce Type: new.
      • Abstract: Reinforcement learning and large language models often struggle to accurately capture the causal mechanics of game environments.
      • Standard reinforcement learning agents tend to rely on spurious correlations, while large language models are prone to hallucinating game rules.
      • Although causal reinforcement learning improves interpretability, there is currently no formal methodology to map complex game mechanics directly into causal models.
    • EN Key Points:
      • arXiv:2609.05459v1 Announce Type: new
      • Abstract: Reinforcement learning and large language models often struggle to accurately capture the causal mechanics of game environments
      • Standard reinforcement learning agents tend to rely on spurious correlations, while large language models are prone to hallucinating game rules
      • Although causal reinforcement learning improves interpretability, there is currently no formal methodology to map complex game mechanics directly into causal mo…

  • ARC-Bench: Closed-Loop Replanning Masks Broken Action Ranking in Frozen JEPA World Models

    • Published: 2026-09-09 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.05461v1 Announce Type: new.
      • Abstract: Reward-free latent world models plan by scoring candidate actions with distances in a frozen latent space: an action is preferred if its predicted future embedding lands closer to the goal embedding.

  • This silently assumes that latent closeness is action-rankable, i.e., that ordering candidates by latent distance agrees with ordering them by true cost.

    • We audit this assumption directly.
    • EN Key Points:
      • arXiv:2609.05461v1 Announce Type: new
      • Abstract: Reward-free latent world models plan by scoring candidate actions with distances in a frozen latent space: an action is preferred if its predicted fut…
      • This silently assumes that latent closeness is action-rankable, i.e., that ordering candidates by latent distance agrees with ordering them by true cost
      • We audit this assumption directly

  • RAPID: Reliability-Aware Pair Importance Distillation

    • Release Time: 2026-09-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.05481v1 Announce Type: new.
      • Abstract: Inter example relational distillation transfers a teacher’s representation geometry by matching relations among examples within a mini batch.
      • Computing all pairs has quadratic complexity in the batch size, whereas uniform subsampling may use a limited relation budget inefficiently.
      • We introduce Reliability Aware Pair Importance Distillation, or RAPID, which separates a reliability gated relational target from a full support adaptive pair proposal.
    • EN Key Points:
      • arXiv:2609.05481v1 Announce Type: new
      • Abstract: Inter example relational distillation transfers a teacher’s representation geometry by matching relations among examples within a mini batch
      • Computing all pairs has quadratic complexity in the batch size, whereas uniform subsampling may use a limited relation budget inefficiently
      • We introduce Reliability Aware Pair Importance Distillation, or RAPID, which separates a reliability gated relational target from a full support adaptive pair p…

  • PGP-Clinical-TimeKAN: Prior-Guided Joint Probabilistic Forecasting of Clinical Trajectories

    • Release Time: 2026-09-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.05488v1 Announce Type: new.

  • Abstract: Clinical deterioration unfolds through coupled, partially observed trajectories, not a single diagnostic label.

  • We introduce PGP-Clinical-TimeKAN, a trajectory-first framework for joint probabilistic forecasting of multivariate physiology.

  • It combines missingness-aware temporal encoders, a soft organ-system prior, patient-specific relations, nonlinear Kolmogorov-Arnold messages, and a low-rank multivariate Student-t head.

    • EN Highlights:
      • arXiv:2609.05488v1 Announce Type: new
      • Abstract: Clinical deterioration unfolds through coupled, partially observed trajectories, not a single diagnostic label
      • We introduce PGP-Clinical-TimeKAN, a trajectory-first framework for joint probabilistic forecasting of multivariate physiology
      • It combines missingness-aware temporal encoders, a soft organ-system prior, patient-specific relations, nonlinear Kolmogorov-Arnold messages, and a low-rank mul…

  • SciLitBench: Benchmark and Design Principles for LLM-Powered Systematic Literature Reviews

    • Published: 2026-09-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.05505v1 Announce Type: new.
      • Abstract: Systematic reviews require sustained human judgment across thousands of records, yet existing evaluations of large language models (LLMs) typically examine review stages in isolation.
      • We introduce SciLitBench, a multi-stage benchmark spanning title and abstract screening, full-text screening, and schema-guided data extraction, with 42,981 retrieved records, 1,012 full texts, and annotations for 888 included papers.
      • Across 22 open-weight LLMs from six model families, explicit inclusion and exclusion criteria improve title and abstract screening $F_2$ by 28.8%, while researcher-authored rationales improve full-text screening by 15%.
    • EN Highlights:
      • arXiv:2609.05505v1 Announce Type: new

  • Abstract: Systematic reviews require sustained human judgment across thousands of records, yet existing evaluations of large language models (LLMs) typically ex…

  • We introduce SciLitBench, a multi-stage benchmark spanning title and abstract screening, full-text screening, and schema-guided data extraction, with 42,981 ret…

  • Across 22 open-weight LLMs from six model families, explicit inclusion and exclusion criteria improve title and abstract screening $F_2$ by 28.8%, while resear…

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

  • When Agent Governance Helps

    • Published: 2026-09-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.05531v1 Announce Type: new.
      • Abstract: No specification says how a governed autotelic AI agent organization, where agents pursue self-generated goals inside guardrails, should be designed and evaluated.
      • First, we synthesize the Governed Autotelic Multi-Agent Product Organization (GAMPO) framework from a document-based qualitative evidence synthesis of 321 sources, integrating agency, agile, platform, and governance theory into a runnable specification.
      • Second, we probe a prompt-layer instantiation of GAMPO on CHI-Bench, a long-horizon healthcare benchmark, across open and frontier models.
    • EN Key Points:
      • arXiv:2609.05531v1 Announce Type: new
      • Abstract: No specification says how a governed autotelic AI agent organization, where agents pursue self-generated goals inside guardrails, should be designed a…
      • We answer in two parts
      • First, we synthesize the Governed Autotelic Multi-Agent Product Organization (GAMPO) framework from a document-based qualitative evidence synthesis of 321 sourc…

  • TamilEOT: A Dataset and Model for Semantic End-of-Turn Detection in Tamil Telephone Speech

    • Published: 2026-09-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.05631v1 Announce Type: new.
      • Abstract: A voice agent has to decide, at every pause, whether the user has finished speaking.

  • Without a model of the language that decision falls back to a fixed silence timeout: set it short and the agent interrupts, set it long and every turn pays the full wait.

    • Open semantic end-of-turn detectors exist, but to our knowledge none covers a South Indian language.
    • EN Highlights:
      • arXiv:2609.05631v1 Announce Type: new
      • Abstract: A voice agent has to decide, at every pause, whether the user has finished speaking
      • Without a model of the language that decision falls back to a fixed silence timeout: set it short and the agent interrupts, set it long and every turn pays the…
      • Open semantic end-of-turn detectors exist, but to our knowledge none covers a South Indian language

  • Better Together: Complementary Query Rewriting Under a Strong RAG Baseline

    • Published: 2026-09-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.05637v1 Announce Type: new.
      • Abstract: A popular way to improve Retrieval-Augmented Generation (RAG) is to rewrite the user’s question into several variants and search with all of them.
      • We test whether this actually helps once the underlying search is already strong.
      • Under one fixed, competitive pipeline (BGE dense retrieval, cross-encoder reranking, and MMR diversification), we compare four query-rewriting strategies (S1-S4) against two strong LLM baselines (HyDE, Query2Doc) on three datasets (HotpotQA, AmbigNQ, and the 512K-document EnterpriseRAG-Bench) over three seeds with paired-bootstrap significance tests.
    • EN Highlights:
      • arXiv:2609.05637v1 Announce Type: new
      • Abstract: A popular way to improve Retrieval-Augmented Generation (RAG) is to rewrite the user’s question into several variants and search with all of them
      • We test whether this actually helps once the underlying search is already strong
      • Under one fixed, competitive pipeline (BGE dense retrieval, cross-encoder reranking, and MMR diversification), we compare four query-rewriting strategies (S1-S4…

  • Who Maintains Agent Skills? A Longitudinal Study of Human-Governed, AI-Assisted Skill Maintenance

    • Published: 2026-09-09 12:00 Beijing Time
    • Summary: [Translation Pending] - arXiv:2609.05677v1 Announce Type: new.
      • Abstract: Lifelong LLM agents increasingly rely on external skill artifacts as one element for preserving and reusing capabilities over time.
      • These skills (usually portable Markdown files such as SKILL.md) describe when and how to apply a capability and must be corrected, expanded, and consolidated as tools and usage patterns shift over deployment.
      • Recent work seeks to automate skill curation, but it largely evaluates against automated baselines and treats human maintenance as an unmeasured bottleneck.
    • EN Highlights:
      • arXiv:2609.05677v1 Announce Type: new
      • Abstract: Lifelong LLM agents increasingly rely on external skill artifacts as one element for preserving and reusing capabilities over time
      • These skills (usually portable Markdown files such as SKILL.md) describe when and how to apply a capability and must be corrected, expanded, and consolidated as…
      • Recent work seeks to automate skill curation, but it largely evaluates against automated baselines and treats human maintenance as an unmeasured bottleneck

  • A Rubric-Guided Large Language Model Solution for Opioid Use Disorder Computable Phenotyping

    • Published: 2026-09-09 12:00 Beijing Time
    • Summary: [Translation Pending] - arXiv:2609.05682v1 Announce Type: new.
      • Abstract: Opioid use disorder (OUD) remains a public health crisis in the United States, yet it is difficult to identify from electronic health records (EHRs) because missing diagnosis codes and supporting evidence are buried in clinical narratives.
      • Accurate OUD identification is critical to support interventions and improve health outcomes.
      • This study developed a rubric-guided large language model (LLM) that incorporated Optimization by PROmpting (OPRO) for OUD computable phenotyping (CP).
    • EN Highlights:

  • arXiv:2609.05682v1 Announce Type: new

  • Abstract: Opioid use disorder (OUD) remains a public health crisis in the United States, yet it is difficult to identify from electronic health records (EHRs) b…

  • Accurate OUD identification is critical to support interventions and improve health outcomes

  • This study developed a rubric-guided large language model (LLM) that incorporated Optimization by PROmpting (OPRO) for OUD computable phenotyping (CP)

  • Intra-Prompt Parallel Decoding for Common-Context Question Answering

    • Published: 2026-09-09 12:00 Beijing Time
    • Abstract: - arXiv:2609.05707v1 Announce Type: new.
      • Abstract: In common-context question answering (CCQA) tasks, multiple input questions share a common context to base their answers from.
      • However, Large Language Models typically generate each answer using an independent prompt.
      • While existing batching and caching techniques help improve parallelism and reduce repeated computations, the separation of questions across prompts limits the achievable speedup, as modern GPUs are underutilized due to a memory bottleneck during attention.
    • Key Points:
      • arXiv:2609.05707v1 Announce Type: new
      • Abstract: In common-context question answering (CCQA) tasks, multiple input questions share a common context to base their answers from
      • However, Large Language Models typically generate each answer using an independent prompt
      • While existing batching and caching techniques help improve parallelism and reduce repeated computations, the separation of questions across prompts limits the…

  • MedWER: A Reproducible, Model-Free Evaluation Protocol for Medical Speech Recognition

    • Published: 2026-09-09 12:00 Beijing Time
    • Abstract: - arXiv:2609.05728v1 Announce Type: new.
      • Abstract: Overall word error rate hides clinically critical errors: a transcript can be 95% correct and still swap one drug for another.

  • The usual fix weights errors on medical entities, and almost always depends on an evaluation-time named-entity recognition (NER) model or cloud API, which makes the metric’s denominator a versioned black box.

  • We present MedWER, an evaluation protocol and open-source tool for medical ASR whose denominator is a fixed, license-clean term list: 19,373 drug, diagnosis, symptom, and injury-mechanism entries projected from public sources.

  • EN Highlights:

    • arXiv:2609.05728v1 Announce Type: new
    • Abstract: Overall word error rate hides clinically critical errors: a transcript can be 95% correct and still swap one drug for another
    • The usual fix weights errors on medical entities, and almost always depends on an evaluation-time named-entity recognition (NER) model or cloud API, which makes…
    • We present MedWER, an evaluation protocol and open-source tool for medical ASR whose denominator is a fixed, license-clean term list: 19,373 drug, diagnosis, sy…

  • Some Tokens Behave like Magnets: Revealing Linguistic Organization in the Layers of Language Models

    • Published: 2026-09-09 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.05743v1 Announce Type: new.
      • Abstract: We identify a special group of token vectors inside large language models (LLMs), which we term magnetic vectors, that organize the surrounding tokens by either attracting or repelling them.
      • Particularly, tokens pointing the same way as an attracting magnet are elongated; tokens pointing the same way as a repelling magnet are compressed.
      • Just as physical magnets pull or push away the iron filings around them, these vectors organize their surroundings through two opposing polarities.
    • EN Highlights:
      • arXiv:2609.05743v1 Announce Type: new
      • Abstract: We identify a special group of token vectors inside large language models (LLMs), which we term magnetic vectors, that organize the surrounding tokens…

  • Particularly, tokens pointing the same way as an attracting magnet are elongated; tokens pointing the same way as a repelling magnet are compressed

    • Just as physical magnets pull or push away the iron filings around them, these vectors organize their surroundings through two opposing polarities

  • CrisisKD: Five-Stage Knowledge Distillation for Aspect-Level Sentiment and Emotion Analysis in Crisis Discourse

    • Release Time: 2026-09-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.05757v1 Announce Type: new.
      • Abstract: Identifying the target of emotional words or phrases in crisis situations, especially health-related ones, is important for understanding public concerns across cultural and linguistic contexts.
      • We propose CrisisKD, a five-stage teacher–student knowledge distillation framework for aspect-level sentiment and emotion analysis on unannotated social media data.
      • A teacher LLM generates aspect-level labels and reasoning traces that supervise a smaller student model across aspect extraction, syntactic parsing, opinion extraction, sentiment classification, and emotion classification.
    • EN Key Points:
      • arXiv:2609.05757v1 Announce Type: new
      • Abstract: Identifying the target of emotional words or phrases in crisis situations, especially health-related ones, is important for understanding public conce…
      • We propose CrisisKD, a five-stage teacher–student knowledge distillation framework for aspect-level sentiment and emotion analysis on unannotated social media…
      • A teacher LLM generates aspect-level labels and reasoning traces that supervise a smaller student model across aspect extraction, syntactic parsing, opinion ext…

  • Recall Is Not Protection: Evaluating Safety Monitors Against Model Compliance

    • Release Time: 2026-09-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.05797v1 Announce Type: new.

  • Abstract: Safety monitors screen prompts sent to deployed language models, flagging harmful requests so they are never answered.

    • They are evaluated by recall against harmfulness labels, but a catch only prevents harm if the model would otherwise have complied.
    • We measure the difference directly: we sample repeated responses from the target model, call a harmful prompt \emph{elicitable} if the model complies at least once, and report monitor recall separately on elicitable and non-elicitable prompts.
    • EN Highlights:
      • arXiv:2609.05797v1 Announce Type: new
      • Abstract: Safety monitors screen prompts sent to deployed language models, flagging harmful requests so they are never answered
      • They are evaluated by recall against harmfulness labels, but a catch only prevents harm if the model would otherwise have complied
      • We measure the difference directly: we sample repeated responses from the target model, call a harmful prompt \emph{elicitable} if the model complies at least o…

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

  • AhaBench: Do Agents Learn from Prior Experience? A Benchmark for Long-Horizon Continual Learning

    • Release Time: 2026-09-09 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.05435v1 Announce Type: new.
      • Abstract: Modern language agents are expected to operate over long horizons: they ask follow-up questions, reuse worked examples, handle tool feedback, and adapt to delayed consequences.
      • Most evaluations still reset the agent after a prompt or score only the final state of one trajectory.
      • AhaBench asks a more operational question: when a fixed model receives useful experience, does its later behavior improve under a related evaluation condition where the obvious support has been removed, changed, or delayed?
    • EN Highlights:
      • arXiv:2609.05435v1 Announce Type: new

  • Abstract: Modern language agents are expected to operate over long horizons: they ask follow-up questions, reuse worked examples, handle tool feedback, and adap…

  • Most evaluations still reset the agent after a prompt or score only the final state of one trajectory

  • AhaBench asks a more operational question: when a fixed model receives useful experience, does its later behavior improve under a related evaluation condition w…

  • When Do Options Help? Policy Necrosis and Redundant Coverage in Option-Critic

    • Published: 2026-09-09 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2609.05508v1 Announce Type: new.
      • Abstract: Option-critic learns options: sub-policies together with a learned rule for when each one hands control back.
      • Its headline result is that performance improves as options are added.
      • We explain that result, with theory and experiment.
    • EN Highlights:
      • arXiv:2609.05508v1 Announce Type: new
      • Abstract: Option-critic learns options: sub-policies together with a learned rule for when each one hands control back
      • Its headline result is that performance improves as options are added
      • We explain that result, with theory and experiment

  • Multi-granularity Adaptive Hypergraph Representation Learning via Granular-ball

    • Published: 2026-09-09 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2609.05574v1 Announce Type: new.
      • Abstract: Hypergraph representation learning aims to capture high-order information in graphs by constructing hyperedges that simultaneously connect multiple nodes.
      • These hyperedges adapt to the graph’s topological features, facilitating the extraction of high-order relationships at multiple granularities.
      • Most prior work relies on predefined definitions to generate hyperedges, overlooking the diversity in graph topological structures and the multi-granularity characteristics of hyperedges.
    • EN Highlights:
      • arXiv:2609.05574v1 Announce Type: new

  • Abstract: Hypergraph representation learning aims to capture high-order information in graphs by constructing hyperedges that simultaneously connect multiple no…

  • These hyperedges adapt to the graph’s topological features, facilitating the extraction of high-order relationships at multiple granularities

  • Most prior work relies on predefined definitions to generate hyperedges, overlooking the diversity in graph topological structures and the multi-granularity cha…

  • Capsule Lens: Locating and Tracking Concept Geometry in Model Representations

    • Publication Time: 2026-09-09 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2609.05575v1 Announce Type: new.
      • Abstract: Understanding how concepts are encoded in the internal representations of machine learning models is a central problem in mechanistic interpretability, essential both for the science of deep learning and for the trustworthy deployment of increasingly capable models.
      • Existing approaches to interpret model representations mainly map representations onto more interpretable spaces and do not directly characterize how concepts occupy representation space; various hypotheses have been proposed, but often lack of rigorous validation and largely focus on static representations.
      • In this work, we introduce Capsule Lens, a framework that matches the region a concept occupies with a simple, trackable geometric form, a capsule, defined by several interpretable parameters, fitted in closed form to each concept’s geometry and validated on held-out samples.
    • EN Key Points:
      • arXiv:2609.05575v1 Announce Type: new
      • Abstract: Understanding how concepts are encoded in the internal representations of machine learning models is a central problem in mechanistic interpretability…
      • Existing approaches to interpret model representations mainly map representations onto more interpretable spaces and do not directly characterize how concepts o…

  • In this work, we introduce Capsule Lens, a framework that matches the region a concept occupies with a simple, trackable geometric form, a capsule, defined by s…

  • HB-PVI: A Hierarchical Bayesian Personalization and Value-of-Information Framework for Complex Activity Recognition

    • Publication Time: 2026-09-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.05582v1 Announce Type: new.
      • Abstract: Personalization can improve activity-recognition performance, but participant-specific gains are heterogeneous, and every additional calibration label has an acquisition cost.
      • This study presents HB-PVI, a hierarchical Bayesian personalization and value-of-information framework jointly modeling participant heterogeneity, the benefit and harm of four personalization mechanisms, and the economic value of an additional label, for the 47-participant MUSIC-CAR complex-activity cohort.
      • A leakage-safe, leave-one-participant-out evaluation combines a sequential-Monte-Carlo participant-effect updater with a Student-$t$ hierarchical gain model and a one-step expected-value-of-sample-information (EVSI) stopping rule.
    • EN Key Points:
      • arXiv:2609.05582v1 Announce Type: new
      • Abstract: Personalization can improve activity-recognition performance, but participant-specific gains are heterogeneous, and every additional calibration label…
      • This study presents HB-PVI, a hierarchical Bayesian personalization and value-of-information framework jointly modeling participant heterogeneity, the benefit a…
      • A leakage-safe, leave-one-participant-out evaluation combines a sequential-Monte-Carlo participant-effect updater with a Student-$t$ hierarchical gain model and…

  • Endogenous Exploration in Reinforcement Learning with Intrinsic Curiosity

    • Publication Time: 2026-09-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.05650v1 Announce Type: new.

  • Abstract: We propose a reinforcement learning framework in which exploration is driven by intrinsic curiosity, designed for scenarios where environments are non-stationary and rewards are sparse, delayed, uninformative, or absent.

  • In our model, action selection is guided by a combination of external rewards and an epistemic motivation mechanism that biases the agent toward structured exploratory directions.

  • The central hypothesis is that effective exploration emerges at intermediate levels of incoherence, while performance degrades under both overly rigid and overly disordered dynamics.

  • EN Key Points:

    • arXiv:2609.05650v1 Announce Type: new
    • Abstract: We propose a reinforcement learning framework in which exploration is driven by intrinsic curiosity, designed for scenarios where environments are non…
    • In our model, action selection is guided by a combination of external rewards and an epistemic motivation mechanism that biases the agent toward structured expl…
    • The central hypothesis is that effective exploration emerges at intermediate levels of incoherence, while performance degrades under both overly rigid and overl…

  • Robustness of LLM-Generated SystemVerilog Assertions to Semantics-Preserving RTL Transformations

    • Publication Time: 2026-09-09 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.05658v1 Announce Type: new.
      • Abstract: Large language models (LLMs) are increasingly being explored for automating SystemVerilog Assertion (SVA) generation, yet most evaluations report correctness on a single syntactic representation of an input.
      • Such point accuracy does not reveal whether a model’s correct output is stable when the same RTL behavior is written differently.
      • This paper presents a controlled metamorphic evaluation of LLM-based SVA generation under semantics-preserving RTL transformations.
    • EN Key Points:
      • arXiv:2609.05658v1 Announce Type: new

  • Abstract: Large language models (LLMs) are increasingly being explored for automating SystemVerilog Assertion (SVA) generation, yet most evaluations report corr…

  • Such point accuracy does not reveal whether a model’s correct output is stable when the same RTL behavior is written differently

  • This paper presents a controlled metamorphic evaluation of LLM-based SVA generation under semantics-preserving RTL transformations

  • PAC-Private Autoregressive Generation: Calibrating Noise to Ensemble Disagreement

    • Publication Time: 2026-09-09 12:00 Beijing Time
    • Abstract: [Translation pending] - arXiv:2609.05676v1 Announce Type: new.
      • Abstract: Language models adapted on private text are often served through APIs, so privacy leakage occurs through generated outputs rather than exposed weights.
      • Private prediction protects these releases.
      • Methods such as PMixED incur privacy cost at each release and increasingly rely on the public model over long horizons.
    • EN Key Points:
      • arXiv:2609.05676v1 Announce Type: new
      • Abstract: Language models adapted on private text are often served through APIs, so privacy leakage occurs through generated outputs rather than exposed weights
      • Private prediction protects these releases
      • Methods such as PMixED incur privacy cost at each release and increasingly rely on the public model over long horizons

  • Connecting Score Matching, Maximum Likelihood, and Expectation-Maximization in Mixed Linear Regression

    • Publication Time: 2026-09-09 12:00 Beijing Time
    • Abstract: [Translation pending] - arXiv:2609.05688v1 Announce Type: new.
      • Abstract: We study variance-preserving diffusion of the response in mixed linear regression (MLR) with unknown mixing weights.
      • Our analysis separates the statistical guarantees of score matching from the loss geometry and optimization signal at a fixed diffusion noise level.

  • The KL divergence links the denoising score matching objective integrated over the diffusion path with the likelihood and a terminal discrepancy.

    • EN Key Points:
      • arXiv:2609.05688v1 Announce Type: new
      • Abstract: We study variance-preserving diffusion of the response in mixed linear regression (MLR) with unknown mixing weights
      • Our analysis separates the statistical guarantees of score matching from the loss geometry and optimization signal at a fixed diffusion noise level
      • The KL divergence links the denoising score matching objective integrated over the diffusion path with the likelihood and a terminal discrepancy

  • GraphNOSE: A Graph Transformer in Olfaction

    • Publication Time: 2026-09-09 12:00 Beijing Time
    • Summary: [Translation Pending] - arXiv:2609.05694v1 Announce Type: new.
      • Abstract: Predicting olfactory qualities from molecular structure is an open problem in chemoinformatics.
      • Although linear models can link molecular features to odor descriptors, they often fail when extrapolating to novel chemical scaffolds, extreme molecular weights, or complex odor mixtures.
      • To address this, we introduce GraphNOSE, an open-source graph transformer framework that predicts multi-label odor descriptors from simplified molecular-input line-entry system (SMILES) strings for single molecules and binary mixtures.
    • EN Key Points:
      • arXiv:2609.05694v1 Announce Type: new
      • Abstract: Predicting olfactory qualities from molecular structure is an open problem in chemoinformatics
      • Although linear models can link molecular features to odor descriptors, they often fail when extrapolating to novel chemical scaffolds, extreme molecular weight…
      • To address this, we introduce GraphNOSE, an open-source graph transformer framework that predicts multi-label odor descriptors from simplified molecular-input l…