🤖 AI 速览

今天的关键变化，不在于模型又刷新了多少分数，而在于 AI 是否能稳定、低成本地完成长期任务。新研究开始评估智能体记忆对实际执行的边际价值，并将生成评测转化为可操作的改进反馈；与此同时，OpenAI 邀请对齐研究者进入基金会安全治理体系，显示前沿模型的能力扩张正与责任机制同步推进。
📋 文章元数据
发布时间
2026-09-10
类型
ai-daily
字数
5172
阅读时长
25 min

2026-09-10 AI日更 | 当 AI 开始长期工作：记忆成本、过程反馈与安全治理浮出水面 链接到标题

今天的关键变化，不在于模型又刷新了多少分数，而在于 AI 是否能稳定、低成本地完成长期任务。新研究开始评估智能体记忆对实际执行的边际价值，并将生成评测转化为可操作的改进反馈；与此同时，OpenAI 邀请对齐研究者进入基金会安全治理体系，显示前沿模型的能力扩张正与责任机制同步推进。

📖 本期 Watch List 深度导读 链接到标题

今天最值得关注的主线，是智能体从“能力展示”走向“长期治理”。OpenAI 一边引入 Paul Christiano 参与基金会与安全事务，一边发布面向工作场景的 GPT-6 Astra；相关研究则进一步追问：谁来维护智能体技能、如何设计自主目标的治理机制，以及安全监控为何可能“高召回却不等于保护”。Meta 的个人智能体发布，也值得结合这些治理问题审视。

第二条主线是评测体系的细化。医疗语音识别、Tamil 语音端点检测和阿片类药物使用识别表明，真实场景更关心关键错误与可复现性，而非单一平均分。与此同时，RAG 查询重写、并行解码和 CriticGen 等工作，正在把评测从“判断对错”推进到“解释并改进生成过程”。 乐彩

🌐 X 平台 AI 热点快讯 链接到标题

话题 1:DeepSeek V4.1 Flash Delivers Top AI Performance at Low Cost 链接到标题

  • 分类:AI · News
  • 概况:热度时间:1 day ago,相关帖子数:14000
  • 是什么事:据标题，DeepSeek V4.1 Flash 被指以较低成本实现了领先的人工智能性能。
  • 为什么重要:这关系到高性能 AI 模型的推理成本、商业化可及性以及开源或低价模型对现有市场格局的影响。
  • 讨论概况:X 上的讨论重点可能集中在其性能评测是否可靠、成本优势的实际幅度、与其他主流模型的对比，以及这一结果能否持续；由于未提供代表性推文，具体分歧无法进一步确认。

话题 2:AI Researcher Quits Anthropic Over Superintelligence Risks 链接到标题

  • 分类:AI · News
  • 概况:热度时间:23 hours ago,相关帖子数:723000
  • 是什么事:Anthropic研究员Jacob Coxon因担忧自我改进型超级智能可能带来失控和人类灭绝风险而离职，并公开批评行业竞逐超级智能的做法。
  • 为什么重要:事件凸显前沿AI实验室在能力竞赛、对齐研究、风险披露和监管责任之间的矛盾，也使超级智能风险发生概率及现有安全机制是否足够再次成为公共议题。
  • 讨论概况:X上的讨论主要围绕Coxon的警告是否可信、Anthropic对未来AI造成灾难的概率评估、企业是否因竞争和游说而弱化安全监管展开；分歧集中在应立即放缓或加强监管，还是继续研发并依靠安全研究降低风险。

话题 3:Omarchy Linux Distro Gains Traction with AI Agents and Sleek Design 链接到标题

  • 分类:AI · News
  • 概况:热度时间:2 days ago,相关帖子数:11000
  • 是什么事:面向开发者的 Omarchy Linux 发行版因支持 AI 代理工作流和简洁的桌面设计在 X 平台获得关注。
  • 为什么重要:它反映出 AI 代理应用正在推动开发者重新关注可控、轻量且便于自动化的操作系统环境，并可能影响本地 AI 工具链和开发体验。
  • 讨论概况:讨论主要集中在 Omarchy 的界面设计、安装与配置体验、对 AI 代理的实际支持程度，以及硬件兼容性、系统稳定性和与现有 Linux 发行版相比的优势；部分用户认为其热度更多来自设计风格而非技术突破。

话题 4:Meme Juxtaposes AI Doomsday Warnings with Apple’s Camera Upgrades 链接到标题

  • 分类:AI · News
  • 概况:热度时间:2 hours ago,相关帖子数:1400
  • 是什么事:一则网络梗图将人工智能末日预警与苹果持续升级相机功能并置，形成反差。
  • 为什么重要:这反映出公众对 AI 安全风险的宏大讨论与消费电子产品渐进式创新之间的落差，有助于观察 AI 热潮如何影响科技行业议题和舆论注意力。
  • 讨论概况:X 上的讨论焦点主要是 AI 末日论是否被过度炒作、苹果相机升级是否缺乏创新，以及科技公司对高风险 AI 议题与实际产品体验的关注是否失衡。

话题 5:Meme Reimagines Job Interview as Robot Apocalypse Stand-Off 链接到标题

  • 分类:AI · Entertainment
  • 概况:热度时间:,相关帖子数:2700
  • 是什么事:一则网络迷因把求职面试戏剧化地重构为人类与机器人在末日场景中的对峙。
  • 为什么重要:它以娱乐化方式呈现人工智能进入工作场景后对就业、岗位竞争和人机关系的影响，反映AI从对话工具走向智能体和企业工作流所引发的社会想象。
  • 讨论概况:X上的讨论主要围绕这一设定是否是在调侃AI抢夺工作、面试焦虑和人类对自动化的担忧展开，也可能延伸至机器人拟人化表达，以及娱乐化叙事是否夸大了AI对就业市场的冲击。

话题 6:Teaser Trailer Drops for Guadagnino’s Sam Altman Drama Artificial 链接到标题

  • 分类:AI · Entertainment
  • 概况:热度时间:1 day ago,相关帖子数:52000
  • 是什么事:卢卡·瓜达尼诺执导、聚焦 Sam Altman 相关故事的剧情片《Artificial》发布首支预告片，引发 X 平台关注。
  • 为什么重要:该片把 OpenAI 权力更迭、AI 商业化与科技领袖形象带入主流影视叙事，显示生成式 AI 已从技术议题扩展为大众文化和公共治理话题。
  • 讨论概况:X 上讨论集中在影片是否会准确呈现 OpenAI 内部危机、Sam Altman 的公众形象是否被戏剧化处理，以及好莱坞是否过早将仍在发展的 AI 行业事件改编成娱乐作品。

今日 X 上的 AI 舆情小结 链接到标题

今天的舆论主线围绕“AI能力扩张与社会承受力之间的张力”展开：一方面，DeepSeek V4.1 Flash、Omarchy Linux等话题强化了低成本高性能模型、本地化工具链和AI代理工作流正在加速普及的共识；另一方面，Anthropic研究员离职、求职与机器人迷因以及相关影视作品，则集中放大了超级智能失控、就业冲击、企业治理和科技权力的担忧。分歧主要在于，行业应继续推进能力竞赛并依靠安全研究解决风险，还是应立即放缓研发、加强监管，同时公众也在争论AI末日叙事究竟是必要警醒还是过度炒作。潜在风险包括模型性能和成本宣传缺乏可复现验证、企业在商业竞争中弱化安全责任，以及娱乐化和迷因化叙事放大焦虑、模糊真实风险与技术炒作之间的界限。

💡 大佬观点(Influencer Insights) 链接到标题

今日大佬观点暂缺,推荐阅读 Watch List 深度内容。

📚 附录:今日 Watch List 更新源列表 链接到标题

时间窗口:最近 3 天;覆盖 22 个源;共 33 条更新

Stratechery by Ben Thompson (A_full) 链接到标题

  • OpenAI Does Math, Reward-Hacking, Meta Launches Personal Agent
    • 发布时间:2026-09-09 18:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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) 链接到标题

  • Paul Christiano joins OpenAI Foundation Board

    • 发布时间:2026-09-10 01:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 19:00 北京时间
    • 摘要:【待翻译】- 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. 链接到标题

      • 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 要点:
      • 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) 链接到标题

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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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) 链接到标题

  • When Agent Governance Helps

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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.
    • EN 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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) 链接到标题

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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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

    • 发布时间:2026-09-09 12:00 北京时间
    • 摘要:【待翻译】- 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 要点:
      • 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…