System translated (Gemini)

🤖 AI 速览

On one hand, OpenAI is establishing a framework for tracking, investigating, and disclosing model mismatches, pushing anomalous behavior into an auditable process; on the other hand, it measures AI investment by usage, cost, and business outcomes. Meanwhile, intelligent agents are gradually …
📋 文章元数据
发布时间
2026-09-17
类型
ai-daily
字数
6744
阅读时长
32 min

2026-09-17 AI Daily | From Mismatch Disclosure to Input-Output: AI Begins to Be Integrated into Organizational Management Ledgers Link to heading

On one hand, OpenAI is establishing a framework for tracking, investigating, and disclosing model mismatches, moving abnormal behaviors into an auditable process. On the other hand, it measures AI investment through usage, cost, and business outcomes. Meanwhile, agents are gradually replacing traditional interfaces, becoming the new gateway for enterprises to connect data and workflows.

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

The first major theme to watch today is the shift of AI from “can it work” to “can it be audited.” OpenAI has released a model mismatch reporting framework and disclosed six case studies of abnormal behavior, which are essential reading for security, evaluation, and governance teams. At the same time, several papers point out that few-shot prompts, spelling errors, and bias audits can significantly affect evaluation results, reminding us not to treat a single score as a sufficient representation of a model’s capabilities or risks.

The second theme focuses on the organizational implementation of AI. Through ChatGPT Work and Codex analysis tools, OpenAI is connecting usage, costs, training needs, and business outcomes. Economic research, in turn, shows how AI is giving rise to new ways of working. Combined with Salesforce’s discussion on “Agents as UI” and the headless trend, we can see that the focus of corporate competition is shifting from interfaces to agents, data, and process orchestration.

Additionally, OpenAI’s collaboration with AARP to promote AI education for seniors, as well as Sponsored Agents for advertisers, are also worth observing as examples of social applications and commercialization boundaries.

🌐 AI Hot Topics on X Link to heading

Topic 1: Justin Sun Launches Math Prize for Proofs and Machine Verification Link to heading

  • Category: AI · News
  • Summary: Hot for: 14 hours ago, Related posts: 9000
  • Abstract: Justin Sun Launches Math Prize for Proofs and Machine Verification: 🧠 Justin Sun launches a math prize The Justin Sun Prize rewards solutions to complex math problems, with prizes paid on-chain. Both humans and AI can compete — and rewards go to researchers who solve problems and formalize proofs f…

Topic 2: Xi Jinping Stroke Rumors Follow BRICS Summit; Ramaphosa Officially Ill Link to heading

  • Category: AI · News
  • Summary: Hot for: 22 hours ago, Related posts: 63000
  • Abstract: Xi Jinping Stroke Rumors Follow BRICS Summit; Ramaphosa Officially Ill:

Topic 3: BNB Chain’s Walter Announces Departure After Four Years Link to heading

  • Category: AI · News
  • Summary: Hot for: 12 hours ago, Related posts: 158
  • Abstract: BNB Chain’s Walter Announces Departure After Four Years:

Topic 4: Anthropic Merges Claude Chat and Cowork into Unified Experience Link to heading

  • Category: AI · News
  • Summary: Hot for: 7 hours ago, Related posts: 5400
  • What it is: Anthropic is integrating features like Claude Chat and Cowork into a unified experience, allowing users to handle conversations, documents, slides, and design tasks within the same workflow.
  • Why it’s important: This marks a shift for AI products from providing Q&A services to undertaking and completing continuous workflows, which helps enhance task automation capabilities and product stickiness.
  • Discussion summary: Discussions on X are mainly focused on whether a unified workspace can reduce tool switching and improve the efficiency of complex tasks, as well as the impact on user control, reliability, and product positioning as the lines between Chat and Cowork blur.

Topic 5: Xiaomi Livestreams Massive AI Model Training with Full Transparency Link to heading

  • Category: AI · News
  • Summary: Hot for: 3 hours ago, Related posts: 1200
  • What it is: Xiaomi is publicly showcasing the training process of its large-scale AI model via livestream, emphasizing transparency in the training phase.
  • Why it’s important: This could push AI model training from result disclosure to process disclosure, increasing public understanding of computing power usage, training progress, and R&D standards, and providing a reference for industry transparency and reproducibility.
  • Discussion Summary: Discussions on the X platform primarily center on whether the live stream constitutes genuine technical transparency, whether the public content can be independently verified, and disagreements regarding data privacy, model security, computational costs, and whether the move is promotional in nature.

Topic 6: AI Coding Harnesses Drive Costs More Than Success, Study Shows Link to heading

  • Category: AI · News
  • Summary: Hotness Time: 5 hours ago, Related Posts: 588
  • Abstract: AI Coding Harnesses Drive Costs More Than Success, Study Shows: Designing Sandboxes That Unlock an Agent’s Full Potential What happened in AI between January 2024 & September 2026, and what we are building toward.

Topic 7: Grok Bot Turns Bookmarks into Working Software Demos Overnight Link to heading

  • Category: AI · News
  • Summary: Hotness Time: 2 days ago, Related Posts: 8300
  • What Happened: Grok Bot can allegedly automatically organize users’ bookmarked content and transform it into runnable software demo prototypes, with some tasks being completed overnight.
  • Why It Matters: This demonstrates that generative AI is evolving beyond answering questions and generating code to possessing end-to-end software construction capabilities that integrate requirements, programming, and deployment.
  • Discussion Summary: Discussions on X primarily focus on its practical usability, the quality and reliability of the generated prototypes, whether it truly achieves automated development, and issues concerning intellectual property, code security, and the impact on developer jobs.

Topic 8: Clearlake Capital Takes Full Ownership of Chelsea FC Link to heading

  • Category: AI · Sports
  • Summary: Hotness Time:, Related Posts: 7300
  • Abstract: Clearlake Capital Takes Full Ownership of Chelsea FC: 🚨 BREAKING: Chelsea FC have announced that affiliates of Clearlake Capital Group will acquire Todd Boehly’s ownership stake, as well as Mark Walter’s stake, giving Clearlake full control of the club. Hansjörg Wyss will remain a key stakeholder a…

Topic 9: MLB Bettors Share Home Run Parlay Picks for September 16 Slate Link to heading

  • Category: AI · Other
  • Summary: Hotness Time:, Related Posts: 448
  • Abstract: MLB Bettors Share Home Run Parlay Picks for September 16 Slate:

Topic 10: Arch Manning Apologizes for Laughing at AI Video of Coach Slapping Reporter Link to heading

  • Category: AI · Other
  • Summary: Hotness Time: 2 days ago, Related Posts: 198000
  • Abstract: Arch Manning Apologizes for Laughing at AI Video of Coach Slapping Reporter: Arch Manning Apologizes For Laughing At AI Video Of Coach Slapping Woman Reporter, Social Media Debates https://t.co/jO2FLGsQxN Arch Manning Apologizes For Laughing At AI Video Of Coach Slapping Woman Reporter, Social Medi…

Topic 11: G2 Esports Eyes Leviatán Stars for 2027 Valorant Overhaul Before Champions Link to heading

  • Category: AI · Sports
  • Summary: Hotness Time: 4 hours ago, Related Posts: 3700
  • Abstract: G2 Esports Eyes Leviatán Stars for 2027 Valorant Overhaul Before Champions:

Topic 12:Lando Norris and Oscar Piastri Shine at McLaren Fashion Show Link to heading

  • Category: AI · Sports
  • Overview: Trending for: 3 hours ago, Related posts: 4500
  • Summary: Lando Norris and Oscar Piastri Shine at McLaren Fashion Show:

Topic 13:SpaceXAI Engineers Build Company in Three Days Using Grok Bots Link to heading

  • Category: AI · News
  • Overview: Trending for: 15 hours ago, Related posts: 555
  • What happened: AI engineers at SpaceX reportedly built a new company in three days using Grok bots, sparking attention on X.
  • Why it matters: This highlights the potential of generative AI agents to accelerate code development, product building, and the startup process. It also sparks discussion about the future scale of software companies, the role of engineers, and the boundaries of AI automation.
  • Discussion summary: The discussion focuses on the authenticity of the “three-day startup” claim, the actual workload handled by Grok, and the extent of human contribution. Some users connect it to the competition with companies like Anthropic, NVIDIA, and Intel, and the huge investments in the industry. Others believe the narrative is exaggerated and not yet sufficient proof that AI can independently create a startup.

Topic 14:Looksmaxxing Enters Merriam-Webster Dictionary After Streamer Spotlight Link to heading

  • Category: AI · Entertainment
  • Overview: Trending for:, Related posts: 119
  • Summary: Looksmaxxing Enters Merriam-Webster Dictionary After Streamer Spotlight:

Topic 15:Players Debate Addon Ban in World of Warcraft: Forever Link to heading

  • Category: AI · Entertainment
  • Overview: Trending for:, Related posts: 80
  • Summary: Players Debate Addon Ban in World of Warcraft: Forever:

💡 Influencer Insights Link to heading

No influencer tweet updates were retrieved today, or the bird CLI fetch failed.

📚 Appendix: Today’s Watch List Source Updates Link to heading

Timeframe: Last 3 days; 22 sources covered; 36 updates in total

Stratechery by Ben Thompson (A_full) Link to heading

  • Salesforce AI Force, Agents as UI, The Race to Headless
    • Published: 2026-09-16 18:18 Beijing Time
    • Summary: [Translation Pending] - Salesforce is abandoning UI as a moat, which is a very smart move because it’s disappearing for everyone.
      • $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:
      • Salesforce is abandoning UI as a moat, which is a very smart move because it’s disappearing for everyone.

OpenAI Blog (A_full) Link to heading

  • Our framework for reporting model misalignment

    • Published: 2026-09-17 01:00 Beijing Time

  • Summary: OpenAI shares a framework for tracking, investigating, and disclosing model misalignment, alongside six reports of unexpected or concerning model behavior.

    • This piece from OpenAI Blog explains how Our framework for reporting model misalignment shapes the broader AI and infrastructure landscape.
    • It also surfaces practical implications for founders, operators, and investors following Our framework for reporting model misalignment.
    • Key Points:
      • OpenAI shares a framework for tracking, investigating, and disclosing model misalignment, alongside six reports of unexpected or concerning model behavior.

  • Helping older adults use AI in everyday life

    • Published: 2026-09-17 00:00 Beijing Time
    • Summary: OpenAI and AARP are bringing free, hands-on ChatGPT workshops to 1,000 older adults across 10 U.S. cities to build practical AI skills safely.
      • This piece from OpenAI Blog explains how Helping older adults use AI in everyday life shapes the broader AI and infrastructure landscape.
    • Key Points:
      • OpenAI and AARP are bringing free, hands-on ChatGPT workshops to 1,000 older adults across 10 U.S. cities to build practical AI skills safely.

  • Reimagining advertising with AI

    • Published: 2026-09-16 21:00 Beijing Time
    • Summary: Explore new AI-powered advertising experiences from OpenAI, including Sponsored Agents, tools for marketers, and integrations with HubSpot and Shopify.
      • This piece from OpenAI Blog explains how Reimagining advertising with AI shapes the broader AI and infrastructure landscape.
      • It also surfaces practical implications for founders, operators, and investors following Reimagining advertising with AI.
    • Key Points:
      • Explore new AI-powered advertising experiences from OpenAI, including Sponsored Agents, tools for marketers, and integrations with HubSpot and Shopify.

  • How to connect AI usage to business value

    • Published: 2026-09-16 20:00 Beijing Time

  • Summary: Learn how ChatGPT Work and Codex analytics help teams understand AI usage and spend, identify training needs, and connect adoption to business outcomes.

    • This piece from OpenAI Blog explains how How to connect AI usage to business value shapes the broader AI and infrastructure landscape.
    • It also surfaces practical implications for founders, operators, and investors following How to connect AI usage to business value.
    • EN Key Points:
      • Learn how ChatGPT Work and Codex analytics help teams understand AI usage and spend, identify training needs, and connect adoption to business outcomes.

  • How workers are unlocking new ways of working

    • Published: 2026-09-16 17:00 Beijing Time
    • Summary: New OpenAI Economic Research shows how workers use AI beyond traditional roles and which new activities become recurring parts of their work.
      • This piece from OpenAI Blog explains how How workers are unlocking new ways of working shapes the broader AI and infrastructure landscape.
      • It also surfaces practical implications for founders, operators, and investors following How workers are unlocking new ways of working.
    • EN Key Points:
      • New OpenAI Economic Research shows how workers use AI beyond traditional roles and which new activities become recurring parts of their work.

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

  • ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search

    • Published: 2026-09-16 12:00 Beijing Time
    • Summary: arXiv:2609.13356v1 Announce Type: new.
      • Abstract: In this work, we present ZGCM-1, a fully open 7B dense foundation model trained from scratch with extreme data, system, and algorithmic efficiency.
      • ZGCM-1 is founded on a core premise: compact models cannot passively memorize the open web, but can overcome parametric capacity limits by coupling deliberate internal thinking with active external tool use.

  • To support this paradigm across a 256K context, we develop an end-to-end, high-efficiency open training recipe: Architecture & System Co-design: interleaved gated sliding-window and full attention, and a stable FP8 Muon optimizer; Progressive Curriculum & MDP Mid-Training: context scaling across 16K, 64K, and 256K, and the reformulation of interaction traces into Markov Decision Processes.

    • Key English Points:
      • arXiv:2609.13356v1 Announce Type: new
      • Abstract: In this work, we present ZGCM-1, a fully open 7B dense foundation model trained from scratch with extreme data, system, and algorithmic efficiency
      • ZGCM-1 is founded on a core premise: compact models cannot passively memorize the open web, but can overcome parametric capacity limits by coupling deliberate i…
      • To support this paradigm across a 256K context, we develop an end-to-end, high-efficiency open training recipe: Architecture & System Co-design: interleaved gat…

  • Converge Then Diversify: Decoupling Convergence and Diversity in Multi-Objective Bayesian Optimisation

    • Release Time: 2026-09-16 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.13396v1 Announce Type: new.
      • Abstract: Multi-objective Bayesian optimisation (MOBO) is a sample-efficient approach for optimising expensive black-box functions with multiple objectives.
      • In MOBO, the goal is to adequately approximate the Pareto front; that is, to obtain a high-quality solution set with 1) good convergence (closeness to the Pareto front) and 2) good diversity (spread across the Pareto front).
      • Existing MOBO methods typically aim to accomplish these two tasks simultaneously, i.e., driving the search towards the Pareto front while maintaining a diverse set of nondominated solutions, such that the solutions, ideally, can gradually approach the entire front.
    • Key English Points:
      • arXiv:2609.13396v1 Announce Type: new

  • Abstract: Multi-objective Bayesian optimisation (MOBO) is a sample-efficient approach for optimising expensive black-box functions with multiple objectives

  • In MOBO, the goal is to adequately approximate the Pareto front; that is, to obtain a high-quality solution set with 1) good convergence (closeness to the Paret…

  • Existing MOBO methods typically aim to accomplish these two tasks simultaneously, i.e., driving the search towards the Pareto front while maintaining a diverse…

  • Generalized Agent Iteration: One Formal Framework for Iterative Policy Improvement and Recursive Self-Improvement

    • Published: 2026-09-16 12:00 Beijing Time
    • Abstract: [Translation pending] - arXiv:2609.13406v1 Announce Type: new.
      • Abstract: When we speak of recursive self-improvement (RSI), are we speaking of a phenomenon, a mechanism, or a prospect?
      • Towards autonomous and evolving intelligence, RSI is being claimed at many scales, while no single framework that formally describes these emerging instances exists.
      • Its counterpart in the classical realm, iterative policy improvement, is characterized by generalized policy iteration (GPI), a framework of broad applicability with well-understood theoretical properties, but only where the update principle and the evaluation base lie outside the agent.
    • EN Highlights:
      • arXiv:2609.13406v1 Announce Type: new
      • Abstract: When we speak of recursive self-improvement (RSI), are we speaking of a phenomenon, a mechanism, or a prospect
      • Towards autonomous and evolving intelligence, RSI is being claimed at many scales, while no single framework that formally describes these emerging instances ex…
      • Its counterpart in the classical realm, iterative policy improvement, is characterized by generalized policy iteration (GPI), a framework of broad applicability…

  • Vibe Patenting: Evaluating LLM Judges for Professional Patent-Drafting Agents

    • Published: 2026-09-16 12:00 Beijing Time

  • Abstract: [To be translated] - arXiv:2609.13422v1 Announce Type: new.

    • Abstract: LLM judges are increasingly used to evaluate and improve AI-generated outputs, yet their reliability for complex professional work remains unclear.
    • We study this problem through Vibe Patenting, an end-to-end patent-drafting testbed for AI-agent evaluation.
    • A separately-invoked LLM judge evaluates generated patent drafts and provides structured feedback for iterative revision.
    • EN Key Points:
      • arXiv:2609.13422v1 Announce Type: new
      • Abstract: LLM judges are increasingly used to evaluate and improve AI-generated outputs, yet their reliability for complex professional work remains unclear
      • We study this problem through Vibe Patenting, an end-to-end patent-drafting testbed for AI-agent evaluation
      • A separately-invoked LLM judge evaluates generated patent drafts and provides structured feedback for iterative revision

  • Toward Self-Adaptive Physical AI: Can LLM Agents Manage Long-Horizon Physical Tasks?

    • Published: 2026-09-16 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.13436v1 Announce Type: new.
      • Abstract: Large Language Model (LLM) agents offer a promising path toward autonomously managing long-term physical tasks without human intervention.
      • However, physical tasks require agents to continuously observe the environment, make consequential actions, and remain effective as the environment changes.
      • Existing approaches either require substantial data and retraining, or primarily focus on agents operating in the virtual world.
    • EN Key Points:
      • arXiv:2609.13436v1 Announce Type: new
      • Abstract: Large Language Model (LLM) agents offer a promising path toward autonomously managing long-term physical tasks without human intervention
      • However, physical tasks require agents to continuously observe the environment, make consequential actions, and remain effective as the environment changes

  • Existing approaches either require substantial data and retraining, or primarily focus on agents operating in the virtual world

  • LabAgent: Customize Any Research Hubs for Scientific Discoveries Using AI Agents

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: Scientific research is a continuous process that emphasizes inheritance.
      • Methods developed by predecessors are often expanded upon by new researchers to explore more novel and in-depth scientific questions.
      • However, the change of lab staff, such as student graduation, leads to a lack of personnel capable of replicating methods.
    • Key Points (EN):
      • arXiv:2609.13437v1 Announce Type: new
      • Abstract: Scientific research is a continuous process that emphasizes inheritance
      • Methods developed by predecessors are often expanded upon by new researchers to explore more novel and in-depth scientific questions
      • However, the change of lab staff, such as student graduation, leads to a lack of personnel capable of replicating methods

  • TimeThink: Eliciting Compositional Reasoning in Timeseries Large Language Models

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: Timeseries multimodal large language models (TS-MLLMs) have recently begun leveraging the reasoning capabilities of large language models (LLMs) for question-answering tasks.
      • However, these models often fail to capture dynamic temporal patterns, providing only implicit reasoning that lacks the underlying explanations critical for high-stakes applications like healthcare.
      • While reinforcement learning (RL)-based timeseries language models aim to address this, they often fall short because they are trained on narrow, in-distribution data and struggle with out-of-distribution compositional questions.
    • Key Points (EN):
      • arXiv:2609.13457v1 Announce Type: new

  • Abstract: Timeseries multimodal large language models (TS-MLLMs) have recently begun leveraging the reasoning capabilities of large language models (LLMs) for q…

  • However, these models often fail to capture dynamic temporal patterns, providing only implicit reasoning that lacks the underlying explanations critical for hig…

  • While reinforcement learning (RL)-based timeseries language models aim to address this, they often fall short because they are trained on narrow, in-distributio…

  • Root-Cause Attribution Is a Search Problem: Continual Search for Long-Horizon Agent Failures

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.13463v1 Announce Type: new.
      • Abstract: The increasing deployment of AI agents in long-horizon tasks yields massive execution logs.
      • Diagnosing failures within these records is crucial for reliability, as it transforms outcome-level signals into actionable interventions.
      • The sheer scale of the data renders human review impractical, driving the need for automated root-cause attribution (RCA).
    • EN Key Points:
      • arXiv:2609.13463v1 Announce Type: new
      • Abstract: The increasing deployment of AI agents in long-horizon tasks yields massive execution logs
      • Diagnosing failures within these records is crucial for reliability, as it transforms outcome-level signals into actionable interventions
      • The sheer scale of the data renders human review impractical, driving the need for automated root-cause attribution (RCA)

  • Governing at Machine Speed: An Adaptive Intelligence Architecture for Real-Time AI Policy Enforcement

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.13466v1 Announce Type: new.
      • Abstract: Enterprise AI adoption has reached 78% of organizations globally, yet the infrastructure to govern that adoption has not kept pace.

  • This paper identifies and characterizes the attestation deficit, a structural condition in which organizations maintain governance policies but cannot produce auditable, tamper-evident evidence of enforcement within regulatory timelines.

  • Drawing on empirical data from the Stanford 2026 AI Index Report (362 documented incidents), the IBM/Ponemon 2026 Cost of a Data Breach study (USD 4.99M average cost, 92% lacking access controls), and the EY/AIUC-1 Consortium survey (38% end-to-end monitoring, 17% agent-to-agent coverage), this paper demonstrates that the governance failure is organizational and architectural rather than technical.

  • EN Key Points:

    • arXiv:2609.13466v1 Announce Type: new
    • Abstract: Enterprise AI adoption has reached 78% of organizations globally, yet the infrastructure to govern that adoption has not kept pace
    • This paper identifies and characterizes the attestation deficit, a structural condition in which organizations maintain governance policies but cannot produce a…
    • Drawing on empirical data from the Stanford 2026 AI Index Report (362 documented incidents), the IBM/Ponemon 2026 Cost of a Data Breach study (USD 4.99M average…

  • OrchSLM: Probing the Dynamics of Small Language Model Orchestration

    • Published: 2026-09-16 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.13470v1 Announce Type: new.
      • Abstract: Although large language models (LLMs) have demonstrated remarkable capabilities, their reliance on cloud-scale infrastructure poses fundamental challenges for deployment in agentic pipelines, including latency, privacy, connectivity, and substantial computational cost.
      • Small language models (SLMs) offer a compelling alternative: recent studies suggest that many repetitive and narrowly scoped subtasks in agentic workloads may be better served by specialized SLMs than by monolithic LLMs.

  • However, the limited capacity and context windows of SLMs can constrain long-horizon reasoning and interaction-heavy orchestration strategies such as iterative verification and debate.

    • EN Key Points:
      • arXiv:2609.13470v1 Announce Type: new
      • Abstract: Although large language models (LLMs) have demonstrated remarkable capabilities, their reliance on cloud-scale infrastructure poses fundamental challe…
      • Small language models (SLMs) offer a compelling alternative: recent studies suggest that many repetitive and narrowly scoped subtasks in agentic workloads may b…
      • However, the limited capacity and context windows of SLMs can constrain long-horizon reasoning and interaction-heavy orchestration strategies such as iterative…

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

  • Few-Shot Degradation Is Not What It Seems: Behavioral Evidence, Representation Analysis, and a Random-Text Control Across 12 Models, 2 Tasks, and 2 Architectures

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: arXiv:2609.15990v1 Announce Type: new.
      • Abstract: Few-shot prompting sometimes degrades language models instead of helping them, but why this happens is unknown.
      • We evaluate 12 open-weight models on two Ukrainian tasks news classification and legal case outcome prediction and find that the effect is strongly task-dependent: the same models that gain +24 pp on news show only +3.4 pp on legal text, with two models degrading.
      • To understand why, we look inside the models.
    • EN Key Points:
      • arXiv:2609.15990v1 Announce Type: new
      • Abstract: Few-shot prompting sometimes degrades language models instead of helping them, but why this happens is unknown
      • We evaluate 12 open-weight models on two Ukrainian tasks news classification and legal case outcome prediction and find that the effect is strongly task-depende…
      • To understand why, we look inside the models

  • The Functionalizer: Lossless Functional Decomposition for Subword Tokenization

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Summary: [To be translated] - arXiv:2609.15991v1 Announce Type: new.
      • Abstract: Standard subword tokenizers either treat every orthographic variation of a word (such as hello, Hello, HELLO, and H'ello) as unrelated vocabulary entries, which fragments the embedding space, or discard this variation through lossy normalization.
      • We present the Functionalizer, a lossless pre-tokenizer framework that factors orthographic and structural variations into a compositional opcode/operand prefix stream before tokenization: a canonical base token (operand) prefixed by parametric transformation operators (opcodes) encoded in the Unicode Private Use Area.
      • We introduce operators covering casing (CAPITALIZE), diacritics (13 dedicated opcodes), and character repetition (REPEAT, MULTIREPEAT), which are fully reversible.
    • EN Highlights:
      • arXiv:2609.15991v1 Announce Type: new
      • Abstract: Standard subword tokenizers either treat every orthographic variation of a word (such as hello, Hello, HELLO, and H'ello) as unrelated vocabulary ent…
      • We present the Functionalizer, a lossless pre-tokenizer framework that factors orthographic and structural variations into a compositional opcode/operand prefix…
      • We introduce operators covering casing (CAPITALIZE), diacritics (13 dedicated opcodes), and character repetition (REPEAT, MULTIREPEAT), which are fully reversib…

  • Optimal Model Activation Policies for Inference Networks of Large Language Models

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Summary: [To be translated] - arXiv:2609.15992v1 Announce Type: new.
      • Abstract: Recent advances in large language models (LLMs) have rendered them necessary for Natural Language Processing (NLP) tasks, and their high inference cost motivates the study of cost-performance trade-offs.

  • In practice, several expert LLMs are used in synergy for inference, either in an ensemble mode or in series, yet without a principled approach on how to best use the available models.

    • An adaptive approach can route simple queries to cheaper LLMs and complex ones to more capable, costly models.
    • EN Key Points:
      • arXiv:2609.15992v1 Announce Type: new
      • Abstract: Recent advances in large language models (LLMs) have rendered them necessary for Natural Language Processing (NLP) tasks, and their high inference cos…
      • In practice, several expert LLMs are used in synergy for inference, either in an ensemble mode or in series, yet without a principled approach on how to best us…
      • An adaptive approach can route simple queries to cheaper LLMs and complex ones to more capable, costly models

  • Single Document Extractive Summarization using Domination in Hypergraph

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.15993v1 Announce Type: new.
      • Abstract: Automatic Text Summarization (ATS) in Natural Language Processing has been an important task in Information Retrieval.
      • It compresses a document to create a summary that captures all the relevant and important information conveyed in the document.
      • This study explores Hypergraph for extractive text summarization of single documents.
    • EN Key Points:
      • arXiv:2609.15993v1 Announce Type: new
      • Abstract: Automatic Text Summarization (ATS) in Natural Language Processing has been an important task in Information Retrieval
      • It compresses a document to create a summary that captures all the relevant and important information conveyed in the document
      • This study explores Hypergraph for extractive text summarization of single documents

  • Latent Undertow: How Ordinary Typos Break Probes

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.15994v1 Announce Type: new.

  • Abstract: LLMs handle ordinary typing variation fluently: a typo or missing punctuation leaves both user intent and the model’s response substantively unchanged.

  • Yet probes that detect malicious prompts by reading the model’s hidden states tell a different story: the same edit rotates the readout vector by 43–56 at the perturbed token, decaying below 15% within ~10 downstream tokens.

  • Stacking ~3 common typos per message cuts a single-position prompt-injection probe’s TPR @FPR$=1% by 12.0pp, a gap recalibration alone cannot close.

    • EN Key Points:
      • arXiv:2609.15994v1 Announce Type: new
      • Abstract: LLMs handle ordinary typing variation fluently: a typo or missing punctuation leaves both user intent and the model’s response substantively unchanged
      • Yet probes that detect malicious prompts by reading the model’s hidden states tell a different story: the same edit rotates the readout vector by 43–56 at the…
      • Stacking ~3 common typos per message cuts a single-position prompt-injection probe’s TPR @FPR$=1% by 12.0pp, a gap recalibration alone cannot close

  • Bias Audits Detect Bias but Disagree on Ranking: Evidence from Ten Instruments and Ten Frontier Models

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.15995v1 Announce Type: new.
      • Abstract: Emerging AI regulation mandates bias audits of high-risk systems, and audit scores are beginning to be used to rank models.
      • Both uses assume different audit tools measure the same thing well enough to compare.
      • We test that assumption directly, running ten extrinsic audit instruments over a shared panel of ten frontier models through one pooled inference gateway, first on occupational gender bias, then on age and socioeconomic status.
    • EN Key Points:
      • arXiv:2609.15995v1 Announce Type: new
      • Abstract: Emerging AI regulation mandates bias audits of high-risk systems, and audit scores are beginning to be used to rank models

  • Both uses assume different audit tools measure the same thing well enough to compare

  • We test that assumption directly, running ten extrinsic audit instruments over a shared panel of ten frontier models through one pooled inference gateway, first…

  • Comment on arXiv:2607.01233: Survivorship Bias in Published-Paper Baselines for Research-Idea Distributions

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.15996v1 Announce Type: new.
      • Abstract: Chen, Zhao, and Cohan introduce a valuable distributional evaluation of LLM-generated research ideas.
      • This comment raises a narrower identification concern: their human baseline consists of published papers, whereas the LLM baseline consists of one-shot proposals.
      • If bridge-like or synthesis-like ideas are relatively easy to generate but relatively unlikely to survive publication, then the published human baseline will understate their prevalence in the unseen human idea pool.
    • EN Key Points:
      • arXiv:2609.15996v1 Announce Type: new
      • Abstract: Chen, Zhao, and Cohan introduce a valuable distributional evaluation of LLM-generated research ideas
      • This comment raises a narrower identification concern: their human baseline consists of published papers, whereas the LLM baseline consists of one-shot proposal…
      • If bridge-like or synthesis-like ideas are relatively easy to generate but relatively unlikely to survive publication, then the published human baseline will un…

  • Crash Narrative-Guided Countermeasure Recommendation Using Large Language Models: A Retrieval-Augmented Generation Framework for Intersection Safety

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2609.15997v1 Announce Type: new.
      • Abstract: Improving safety at intersections requires identifying crash mechanisms and recommending appropriate countermeasures.

  • However, this process traditionally relies on expert judgment, making it labor-intensive, difficult to scale, and dependent on the availability of experienced traffic safety engineers.

    • Although crash narratives contain rich description of crash mechanisms, this unstructured information remains largely underutilized in safety analyses.
    • EN Key Points:
      • arXiv:2609.15997v1 Announce Type: new
      • Abstract: Improving safety at intersections requires identifying crash mechanisms and recommending appropriate countermeasures
      • However, this process traditionally relies on expert judgment, making it labor-intensive, difficult to scale, and dependent on the availability of experienced t…
      • Although crash narratives contain rich description of crash mechanisms, this unstructured information remains largely underutilized in safety analyses

  • Self-reported archetypes and behavioral failures in Large Language Models

    • Published: 2026-09-16 12:00 Beijing Time
    • Abstract: - arXiv:2609.15998v1 Announce Type: new.
      • Abstract: Every large language model (LLM) has behavioral traits and moral preferences that comprise its character.
      • Whether by design or as an emergent property of training, these systems exhibit persistent dispositions that shape how they interact, comply, resist, and err, yet the structure of LLM character remains poorly understood.
      • We map the self-reported personality archetypes of 22 LLMs spanning closed-source frontier systems (GPT-4.0-5.2, Grok-3/4, Gemini 2.5 Pro/Flash, Claude Sonnet 4.5/4.6) and open-source models (Llama, DeepSeek, OLMo, and Qwen series).
    • EN Key Points:
      • arXiv:2609.15998v1 Announce Type: new
      • Abstract: Every large language model (LLM) has behavioral traits and moral preferences that comprise its character
      • Whether by design or as an emergent property of training, these systems exhibit persistent dispositions that shape how they interact, comply, resist, and err, y…

  • We map the self-reported personality archetypes of 22 LLMs spanning closed-source frontier systems (GPT-4.0-5.2, Grok-3/4, Gemini 2.5 Pro/Flash, Claude Sonnet 4…

  • NepKANUN: A RAG-Based Nepali Legal Assistant

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Summary: [Translation Pending] - arXiv:2609.15999v1 Announce Type: new.
      • Abstract: Accessing legal information in Nepal is difficult due to complex terminology, limited resources, and misinformation.
      • We introduce an AI-powered legal assistant that is tailored for Nepali legal texts and is built on a fine-tuned large language model.
      • The technology provides precise, streamlined answers to natural language legal inquiries when integrated into a Retrieval-Augmented Generation (RAG) framework.
    • EN Key Points:
      • arXiv:2609.15999v1 Announce Type: new
      • Abstract: Accessing legal information in Nepal is difficult due to complex terminology, limited resources, and misinformation
      • We introduce an AI-powered legal assistant that is tailored for Nepali legal texts and is built on a fine-tuned large language model
      • The technology provides precise, streamlined answers to natural language legal inquiries when integrated into a Retrieval-Augmented Generation (RAG) framework

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

  • Causal neural set filtering for online multi-target tracking

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Summary: [Translation Pending] - arXiv:2609.16054v1 Announce Type: new.
      • Abstract: Transformer-based multi-target tracking (MTT) jointly learns data association and state estimation, but MT3/Track-MT3-style trackers repeatedly re-encode measurement windows, incurring redundant computation.
      • We propose Causal Neural Set Filtering (CNSF)\footnote{\href{ a neural set filter that encodes only current measurements while carrying past evidence in a structured recursive track state.

  • CNSF combines exclusive Sinkhorn association, association-conditioned Kalman-shaped updates with moment matching, and recurrent Bernoulli lifecycle modeling with measurement-driven birth.

    • EN Highlights:
      • arXiv:2609.16054v1 Announce Type: new
      • Abstract: Transformer-based multi-target tracking (MTT) jointly learns data association and state estimation, but MT3/Track-MT3-style trackers repeatedly re-enc…
      • We propose Causal Neural Set Filtering (CNSF)\footnote{\href{ a neural set filter that encodes only current measurements while carrying past evidence in a struc…
      • CNSF combines exclusive Sinkhorn association, association-conditioned Kalman-shaped updates with moment matching, and recurrent Bernoulli lifecycle modeling wit…

  • Managing Action Preconditions in Neuro-Symbolic RL: Three Placement Strategies for Embodied Agents

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: [Translation Pending] - arXiv:2609.16056v1 Announce Type: new.
      • Abstract: Humans carry behaviour knowledge of how to act in familiar situations into every new task rather than relearning it from scratch.
      • There is no reason a Reinforcement Learning (RL) agent shouldn’t do the same: known behaviour patterns need not be learned, only applied.
      • Neuro-symbolic RL bridges prior knowledge and RL by injecting symbolic knowledge alongside a learned policy.
    • EN Highlights:
      • arXiv:2609.16056v1 Announce Type: new
      • Abstract: Humans carry behaviour knowledge of how to act in familiar situations into every new task rather than relearning it from scratch
      • There is no reason a Reinforcement Learning (RL) agent shouldn’t do the same: known behaviour patterns need not be learned, only applied
      • Neuro-symbolic RL bridges prior knowledge and RL by injecting symbolic knowledge alongside a learned policy

  • OmniHarness: Harnessing Generalizable Visual Generation via Symbolic Policy Learning

    • Publication Time: 2026-09-16 12:00 Beijing Time

  • Summary: [To be translated] - arXiv:2609.16057v1 Announce Type: new.

    • Abstract: Unified multimodal large language models (MLLMs) and multi-agent systems have advanced visual generation.
    • However, three limitations remain.
    • (1) Existing methods often distill task-specific experience with limited generalizability.
    • EN Key Points:
      • arXiv:2609.16057v1 Announce Type: new
      • Abstract: Unified multimodal large language models (MLLMs) and multi-agent systems have advanced visual generation
      • However, three limitations remain
      • (1) Existing methods often distill task-specific experience with limited generalizability

  • Driver Behavior Estimation at Signalized Intersections Using a Physics-Constrained Decision-Conditioned Autoregressive Transformer

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Summary: [To be translated] - arXiv:2609.16058v1 Announce Type: new.
      • Abstract: Red-light violations and harsh braking at signalized intersections are major contributors to traffic accidents.
      • This paper analyzes and predicts human driver decision-making and longitudinal trajectory behavior during traffic light signal transitions.
      • We collected a diverse real-world dataset comprising 449 approach runs under varying speed and distance conditions.
    • EN Key Points:
      • arXiv:2609.16058v1 Announce Type: new
      • Abstract: Red-light violations and harsh braking at signalized intersections are major contributors to traffic accidents
      • This paper analyzes and predicts human driver decision-making and longitudinal trajectory behavior during traffic light signal transitions
      • We collected a diverse real-world dataset comprising 449 approach runs under varying speed and distance conditions

  • HintMiner: Automatic Question Hints Mining From Q&A Web Posts with Language Model via Self-Supervised Learning

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Summary: [To be translated] - arXiv:2609.16060v1 Announce Type: new.

  • Abstract: Users often need ask questions and seek answers online.

  • The Question - Answering (QA) forums such as Stack Overflow cannot always respond to the questions timely and properly.

  • In this paper, we propose HintMiner, a novel automatic question hints mining tool for users to help them find answers.

  • EN Key Points:

    • arXiv:2609.16060v1 Announce Type: new
    • Abstract: Users often need ask questions and seek answers online
    • The Question - Answering (QA) forums such as Stack Overflow cannot always respond to the questions timely and properly
    • In this paper, we propose HintMiner, a novel automatic question hints mining tool for users to help them find answers

  • POSPAN: Position-Constrained Span Masking for Language Model Pre-training

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: [Awaiting Translation] - arXiv:2609.16061v1 Announce Type: new.
      • Abstract: Span-level masked language modeling (MLM) has shown to be advantageous to pre-trained language models over the original single-token MLM, as entities/phrases and their dependencies are critical to language understanding.
      • Previous works only consider span length with some discrete distributions, while the dependencies among spans are ignored, i.e., assuming that the positions of masked spans are uniformly distributed.
      • In this paper, we present POSPAN, a general framework to allow diverse position-constrained span masking strategies via the combination of span length distribution and position constraint distribution, which unifies all existing span-level masking methods.
    • EN Key Points:
      • arXiv:2609.16061v1 Announce Type: new
      • Abstract: Span-level masked language modeling (MLM) has shown to be advantageous to pre-trained language models over the original single-token MLM, as entities/…
      • Previous works only consider span length with some discrete distributions, while the dependencies among spans are ignored, i.e., assuming that the positions of…

  • In this paper, we present POSPAN, a general framework to allow diverse position-constrained span masking strategies via the combination of span length distribut…

  • Signed p-adic Residual Encodings of Finite-Domain All-Different Systems with a Sudoku Case Study

    • Release Date: 2026-09-16 12:00 Beijing Time
    • Abstract: [[OC_PH_TODO_TRANSLATE]] - arXiv:2609.16063v1 Announce Type: new.
      • Abstract: We study signed, weighted affine $p$-adic residual objectives as native encodings of finite-domain constraints.
      • For primes that separate the finite alphabet, sufficiently weighted positive unary rows pin each coefficient to its allowed set, while negative rows reward unequal endpoints or clause satisfaction.
      • A coordinatewise domination theorem places every global minimiser in the finite domain; there the loss is, up to an additive constant, the all-different conflict count or the negative number of satisfied CNF clauses.
    • EN Key Points:
      • arXiv:2609.16063v1 Announce Type: new
      • Abstract: We study signed, weighted affine $p$-adic residual objectives as native encodings of finite-domain constraints
      • For primes that separate the finite alphabet, sufficiently weighted positive unary rows pin each coefficient to its allowed set, while negative rows reward uneq…
      • A coordinatewise domination theorem places every global minimiser in the finite domain; there the loss is, up to an additive constant, the all-different conflic…

  • You Don’t Need To Train: Agentic Heuristic Learning Studio for Executable Human Activity Recognition

    • Release Date: 2026-09-16 12:00 Beijing Time
    • Abstract: [[OC_PH_TODO_TRANSLATE]] - arXiv:2609.16065v1 Announce Type: new.
      • Abstract: Human activity recognition (HAR) is usually framed as gradient-based training of neural networks.

  • Agentic Heuristic Learning (AHL) Studio explores a complementary view inspired by human cognitive learning: people learn activities by remembering examples, forming rules, and repairing mistakes, not by backpropagating.

  • This proposed tool implements AHL for HAR: a learning-time agent reasons over sensor protocols, proposes executable heuristic policies, records repair traces, and exports an LLM-free policy for edge deployment.

    • EN Key Points:
      • arXiv:2609.16065v1 Announce Type: new
      • Abstract: Human activity recognition (HAR) is usually framed as gradient-based training of neural networks
      • Agentic Heuristic Learning (AHL) Studio explores a complementary view inspired by human cognitive learning: people learn activities by remembering examples, for…
      • This proposed tool implements AHL for HAR: a learning-time agent reasons over sensor protocols, proposes executable heuristic policies, records repair traces, a…

  • A panoramic aerodynamic performance prediction method for turbomachinery cascades using transformer-enhanced neural operator

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: [Awaiting Translation] - arXiv:2609.16066v1 Announce Type: new.
      • Abstract: To enable flexible and rapid aerodynamic performance evaluation in turbomachinery design, this paper proposes a panoramic performance prediction framework.
      • Unlike most previous prediction models that directly predict the objective functions of interest, our approach first predicts the basic parameters of the Navier-Stokes equations, such as temperature, pressure, and density.
      • Utilizing these basic physical quantities, it subsequently predicts key performance parameters of the turbine stage meridian plane.
    • EN Key Points:
      • arXiv:2609.16066v1 Announce Type: new
      • Abstract: To enable flexible and rapid aerodynamic performance evaluation in turbomachinery design, this paper proposes a panoramic performance prediction frame…

  • Unlike most previous prediction models that directly predict the objective functions of interest, our approach first predicts the basic parameters of the Navier…

  • Utilizing these basic physical quantities, it subsequently predicts key performance parameters of the turbine stage meridian plane

  • A Dynamic Aggregation Strategy Enhanced Efficient Global Optimization Algorithm for Solving High-Dimensional Turbomachinery Design Problems

    • Publication Time: 2026-09-16 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2609.16067v1 Announce Type: new.
      • Abstract: In order to solve the high-dimensional ($d \geq 30$) expensive black-box problems within budget, an efficient global optimization (EGO) algorithm with a dynamic aggregation strategy is proposed, labeled as DA-EGO.
      • Specifically, the DA-EGO decomposes the original high-dimensional design space into a set of low-dimensional subspaces for efficient surrogate-based optimization search, and the optimal solutions of subspaces are combined as an elite point for the global search.
      • Most importantly, the subspaces are not fixed.
    • EN Key Points:
      • arXiv:2609.16067v1 Announce Type: new
      • Abstract: In order to solve the high-dimensional ($d \geq 30$) expensive black-box problems within budget, an efficient global optimization (EGO) algorithm with…
      • Specifically, the DA-EGO decomposes the original high-dimensional design space into a set of low-dimensional subspaces for efficient surrogate-based optimizatio…
      • Most importantly, the subspaces are not fixed