System translated (Gemini)

🤖 AI 速览

Today’s core trend is the shift in enterprise agents from capability demonstrations toward cost, controllability, and explainability. OpenAI disclosed that Sophos threat investigation time was compressed by 96%, automating 52% of MDR; Asana’s browser agent reduced costs by 76x and …
📋 文章元数据
发布时间
2026-10-10
类型
ai-daily
字数
6563
阅读时长
31 min

2026-10-10 AI Daily | Enterprise Agents Move to Unit Economics Competition: OpenAI Case Study Speeds Up by 96%, Costs Cut by 76x Link to heading

Today’s core trend is that enterprise agents are shifting from capability demonstrations to cost, control, and interpretability. OpenAI disclosed that Sophos compressed threat investigation time by 96% and automated 52% of MDR cases; Asana reduced browser agent costs by 76x and boosted speed by 5x. Google released enterprise-grade Gemini agents, dedicated small models are making a comeback, and real-world deployments place greater value on being good enough and controllable.

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

Today, two enterprise case studies on OpenAI’s official blog are worth close reading by engineering teams: Sophos, with the help of Daybreak, compressed threat investigation time by 96% and automated 52% of MDR cases; Asana, in browser tests, reduced model costs by 76x and boosted speed by 5x. Together, they demonstrate that the competitive focus of agent deployment has shifted from capability demonstrations to cost and controllability. Meanwhile, a group of underlying threads from papers on arXiv—concerning 50M token long-term memory, sparse attention as matrix approximation, and lossy compression text autoencoders—discuss how long context can break free from computing power stacking. The third noteworthy trend is the comeback of specialized small models: fine-tuned small models outperformed prompted frontier models in grammatical annotation, and when paired with vertical research such as historical Polish corpora and transportation planning, real-world deployments are prioritizing “good enough, controllable, and interpretable.”

🌐 X Platform AI Hot News Link to heading

Topic 1: X Prepares Unified Subscription Bundling Grok and Cursor AI Tools Link to heading

  • Category: AI · News
  • Overview: Trending time: 1 day ago, Related posts: 20,000
  • What it is: X plans to launch a unified subscription service, bundling Grok and Cursor AI tools for users.
  • Why it matters: This shows X is extending from a social platform to an AI productivity platform, which may drive the trend of unified subscriptions for AI assistants and programming tools, and intensify competition with ecosystems like OpenAI and Microsoft Copilot.
  • Discussion overview: Discussions on X mainly focus on whether the package pricing is cost-effective, the actual experience of Grok and Cursor integration, and whether X can leverage this to become an AI super app; divergences center on the attractiveness of bundled subscriptions, data privacy, and vendor lock-in risks.

Topic 2: Grok Bot Gains Real-Time X Monitoring and Custom Email Features Link to heading

  • Category: AI · News
  • Overview: Trending time: 2 days ago, Related posts: 52,000
  • What it is: Grok Bot has added real-time monitoring of the X platform and custom email features, allowing it to handle subscriptions, replies, and multi-agent coordination like an always-on AI assistant.
  • Why it matters: This marks the shift of AI agents from one-off Q&A tools to persistent workflow nodes capable of long-term autonomous operation and cross-agent collaboration, lowering the barrier to automation and agent collaboration.
  • Discussion overview: The focus of discussions includes its practical use cases as an “always-on AI teammate,” its potential to collaborate with other agents (such as Hermes) via email, as well as the iOS experience and feature extensions; some users remain in a wait-and-see attitude regarding stability and reliability.

Topic 3: Google Unveils Enterprise Gemini Agent at NASA Hangar Event Link to heading

  • Category: AI · News
  • Overview: Trending time: 1 day ago, Related posts: 6,400
  • What it is: On October 8, Google unveiled its enterprise-grade Gemini agent at the Gemini at Work event held at a NASA hangar. The agent can handle Q&A, knowledge work, media creation, and coding, while connecting to company systems and maintaining context across devices.
  • Why it matters: This marks Google’s push of Gemini from a consumer-grade assistant to an enterprise-grade autonomous agent, directly competing with Microsoft Copilot and OpenAI for the enterprise AI market, potentially accelerating the entry of AI agents into core workflows.
  • Discussion overview: Discussions on X mainly focus on its capability comparison with Microsoft Copilot and OpenAI’s enterprise products, with attention to enterprise data security, permission control, and actual productivity gains. Some also believe the venue was chosen to reinforce Google’s enterprise-level ambitions.

Topic 4: Joma Tech Parodies GPT-6 Dishwashing with Math Puzzle Twist Link to heading

  • Category: AI · News
  • Overview: Trending time: 19 hours ago, Related posts: 2,100
  • What it is: Joma Tech released a parody video, spoofing GPT-6 for requiring a math puzzle to be solved before even washing dishes, satirizing AI’s over-reasoning.
  • Why it matters: It uses comedy to highlight the current issue where large language models overcomplicate simple tasks, showing an imbalance between reasoning capabilities and practical utility, which serves as a cautionary tale for AI product design.
  • Discussion overview: The focus of discussions on X is whether AI should consume massive computing resources in unnecessary scenarios, and whether this parody accurately reflects usability flaws ignored in real-world benchmarks; some see it as an exaggerated spoof, while others feel it hits the pain points of the reasoning-enhancement trend.

Topic 5: Austin Riders Embrace Tesla Cybercab Robotaxis in Rain and Night Link to heading

  • Category: AI · Entertainment
  • Overview: Trending time: 1 day ago, Number of related posts: 4700
  • What happened: Passengers in Austin used Tesla Cybercab robotaxis in rainy and nighttime conditions, receiving positive responses.
  • Why it matters: Rainy and nighttime conditions are critical challenges for autonomous driving. This event provides validation for the reliability of pure vision-based solutions under real-world, complex conditions, and also drives discussions on Robotaxi commercialization.
  • Discussion overview: Discussions on X focus on Cybercab’s safety performance in low-visibility conditions, the debate between Tesla’s vision-only approach and the LiDAR-based route, and whether it has the conditions for large-scale operations.

Summary of Today’s AI Public Sentiment on X Link to heading

Today, the main narrative in public opinion is that AI is accelerating its transition from one-off Q&A tools to long-running, cross-application collaborative agents and productivity platforms embedded in actual workflows. This trend is reflected in X’s integration of Grok with Cursor, Grok Bot’s persistent email collaboration, and Google’s release of enterprise-grade Gemini. The consensus is that the industry generally values AI’s practicality and large-scale implementation capabilities, but divergences are also prominent: one side is concerned with whether bundled subscriptions are cost-effective and whether they bring data privacy and vendor lock-in risks, while the other side debates whether enterprise-grade agents can truly boost productivity, and whether Tesla’s vision-only approach is reliable enough in complex scenarios like rain and night. Potential risks are concentrated in data security and permission control, agent stability and the imbalance in computing power consumption, as well as the still insufficient safety validation for the large-scale operation of autonomous driving under low-visibility conditions. Joma Tech’s spoof also perfectly points out the risk that over-reasoning might sacrifice the usability of simple tasks.

💡 Influencer Insights Link to heading

Today’s influencer insights are temporarily unavailable. Reading the in-depth content in the Watch List is recommended.

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

Time window: Recent 3 days; Covers 22 sources; 33 updates in total

Stratechery by Ben Thompson (A_full) Link to heading

  • 2026.41: It’s Not You, It’s Me
    • Publication Time: 2026-10-10 01:00 Beijing Time
    • Summary: [To be translated] - (Photo by Justin Sullivan/Getty Images).
      • Welcome back to This Week in Stratechery!
      • As a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone.
      • Additionally, you have complete control over what we send to you.
      • On that note, here were a few of our favorites this week.
    • EN Highlights:
      • (Photo by Justin Sullivan/Getty Images)
      • Welcome back to This Week in Stratechery
      • As a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone
      • Additionally, you have complete control over what we send to you

OpenAI Blog (A_full) Link to heading

  • Sophos cuts threat investigation time by 96% with OpenAI Daybreak

    • Publication Time: 2026-10-09 15:00 Beijing Time
    • Summary: [To be translated] - Discover how Sophos uses OpenAI’s Daybreak to cut cyber-threat investigation time by 96% and automate 52% of MDR cases while preserving human oversight.

  • This piece from OpenAI Blog explains how Sophos cuts threat investigation time by 96% with OpenAI Daybreak shapes the broader AI and infrastructure landscape.

    • It also surfaces practical implications for founders, operators, and investors following Sophos cuts threat investigation time by 96% with OpenAI Daybreak.
    • EN Key Points:
      • Discover how Sophos uses OpenAI’s Daybreak to cut cyber-threat investigation time by 96% and automate 52% of MDR cases while preserving human oversight.

  • Asana cuts model costs 76x in browser tests with GPT-6.1 Sol

    • Release Time: 2026-10-09 15:00 Beijing Time
    • Summary: - Using GPT-6 Astra in Codex, Asana made its browser agent 76x cheaper and 5x faster in tests to offer customers more capable models.
      • This piece from OpenAI Blog explains how Asana cuts model costs 76x in browser tests with GPT-6.1 Sol shapes the broader AI and infrastructure landscape.
      • It also surfaces practical implications for founders, operators, and investors following Asana cuts model costs 76x in browser tests with GPT-6.1 Sol.
    • EN Key Points:
      • Using GPT-6 Astra in Codex, Asana made its browser agent 76x cheaper and 5x faster in tests to offer customers more capable models.

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

  • An Explainable Header-Centric Framework for Large-Scale Semantic Table Interpretation and Data Quality Assessment

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: - arXiv:2610.10541v1 Announce Type: new.
      • Abstract: Knowledge Graph (KG) quality depends not only on downstream graph validation, but also on the quality of tabular metadata used before integration.
      • In metadata-only Semantic Table Interpretation (STI), where cell values are unavailable, noisy, or unsuitable, column headers become a critical source of semantic evidence for traceable KG preparation.
      • We present an explainable, header-centric framework for metadata-only Column Type Annotation (CTA) and Data Quality Assessment (DQA).

  • EN Highlights:

    • arXiv:2610.10541v1 Announce Type: new
    • Abstract: Knowledge Graph (KG) quality depends not only on downstream graph validation, but also on the quality of tabular metadata used before integration
    • In metadata-only Semantic Table Interpretation (STI), where cell values are unavailable, noisy, or unsuitable, column headers become a critical source of semant…
    • We present an explainable, header-centric framework for metadata-only Column Type Annotation (CTA) and Data Quality Assessment (DQA)

  • Synthesis Through Simulation: Generating Coherent Enterprise Data via Scalable Agent-System Interaction

    • Publication Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10549v1 Announce Type: new.
      • Abstract: Tool-calling agents have become central to enterprise AI, yet training and evaluating them at scale remains severely constrained due to business and legal restrictions on enterprise systems, data, and database schemas.
      • Tabular data synthesis offers a natural alternative, but its effectiveness is fundamentally limited by structural validity and schema availability, while procedure-based approaches yield the opposite weakness, typically lacking distributional fidelity without per-domain authoring.
      • We introduce Synthesis Through Simulation (STS), a schema–free data synthesis paradigm in which an LLM agent generates data by executing operations against policy-enforcing APIs within simulated enterprise environments.
    • EN Highlights:
      • arXiv:2610.10549v1 Announce Type: new
      • Abstract: Tool-calling agents have become central to enterprise AI, yet training and evaluating them at scale remains severely constrained due to business and l…
      • Tabular data synthesis offers a natural alternative, but its effectiveness is fundamentally limited by structural validity and schema availability, while proced…

  • We introduce Synthesis Through Simulation (STS), a schema–free data synthesis paradigm in which an LLM agent generates data by executing operations aga…

  • Agent-Controlled Forgetting for Tool-Using Agents: Reversible Context Curation in Practice

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10590v1 Announce Type: new.
      • Abstract: Tool-using agents repeatedly carry observations whose useful content can be much smaller than their original payload.
      • We study agent-controlled forgetting: the acting model selects previously observed tool results, replaces each with a short note at its original position, and retains the exact original in a recoverable archive.
      • A Python harness exposes batch archival and explicit recovery without task-specific model training, while protecting user instructions and assistant messages from these operations.
    • EN Key Points:
      • arXiv:2610.10590v1 Announce Type: new
      • Abstract: Tool-using agents repeatedly carry observations whose useful content can be much smaller than their original payload
      • We study agent-controlled forgetting: the acting model selects previously observed tool results, replaces each with a short note at its original position, and r…
      • A Python harness exposes batch archival and explicit recovery without task-specific model training, while protecting user instructions and assistant messages fr…

  • Verification and Self-Improvement in Agentic AI: Foundations and Limits

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10611v1 Announce Type: new.
      • Abstract: Agentic AI systems can improve by searching longer, receiving additional support, or modifying how they propose and verify outputs.
      • A performance score does not distinguish these mechanisms.
      • We compare these changes through bounded verification with hidden terminal randomness.
    • EN Key Points:
      • arXiv:2610.10611v1 Announce Type: new

  • Abstract: Agentic AI systems can improve by searching longer, receiving additional support, or modifying how they propose and verify outputs

    • A performance score does not distinguish these mechanisms
    • We compare these changes through bounded verification with hidden terminal randomness

  • The Harness as the Only Mutable Surface: Compliance-Bounded Self-Evolution of LLM Agents in Credit Pipelines, with a Measured Admission Gate

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10629v1 Announce Type: new.
      • Abstract: Self-improving LLM agents can adapt a credit pipeline to a changed rule, but an agent that rewrites itself destroys the artefact a supervisor reviews: a named change, a recorded test, an approval.
      • We argue that self-evolution is reviewable only if it is confined to the runtime harness (instruction text, tool-call logic and primitive composition) while model weights stay fixed, so that every adaptation is a diff with a cause and a test attached.
      • We give a dual-loop engine built on that bound, with one admission gate that writes a hash-chained record before deployment, and we measure the gate in simulation, with a simulated agent and a seeded-search proposer rather than language models.
    • EN Highlights:
      • arXiv:2610.10629v1 Announce Type: new
      • Abstract: Self-improving LLM agents can adapt a credit pipeline to a changed rule, but an agent that rewrites itself destroys the artefact a supervisor reviews:…
      • We argue that self-evolution is reviewable only if it is confined to the runtime harness (instruction text, tool-call logic and primitive composition) while mod…
      • We give a dual-loop engine built on that bound, with one admission gate that writes a hash-chained record before deployment, and we measure the gate in simulati…

  • Speaking the Navigator’s Language: Trajectory-Grounded Instruction Translation for Frozen Aerial VLN Agents

    • Release Time: 2026-10-09 12:00 Beijing Time

  • Summary: arXiv:2610.10635v1 Announce Type: new.

    • Abstract: Aerial vision-and-language navigation (VLN) agents are typically trained on detail-rich, trajectory-aligned commands, whereas users issue short, intent-driven instructions; on a frozen OpenFly navigator, this \emph{instruction gap} drops success rate (SR) from $31.03%$ to $11.33%$.
    • To scale translator training, we prompt a language model with human-written style examples to convert original commands into paired, intent-centered Weak commands, which yield $15.27%$ SR.
    • We introduce the \textbf{Trajectory-Grounded Instruction Translator (TGIT)}, a front-end that keeps the navigator frozen and translates Weak inputs into agent-executable commands by learning from its trajectory outcomes.
    • EN Key Points:
      • arXiv:2610.10635v1 Announce Type: new
      • Abstract: Aerial vision-and-language navigation (VLN) agents are typically trained on detail-rich, trajectory-aligned commands, whereas users issue short, inten…
      • To scale translator training, we prompt a language model with human-written style examples to convert original commands into paired, intent-centered Weak comman…
      • We introduce the \textbf{Trajectory-Grounded Instruction Translator (TGIT)}, a front-end that keeps the navigator frozen and translates Weak inputs into agent-e…

  • Plan-and-Patch: Diffusion Language Models for Agentic Planning

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Summary: arXiv:2610.10786v1 Announce Type: new.
      • Abstract: Planning is increasingly important for long-horizon agents, where successful execution requires coordinating subgoals, tool use, and intermediate outcomes over many steps.
      • Yet assumptions made during planning may be invalidated by the environment, tools may return unexpected results, or actions may fail.
      • Effective agents must therefore not only generate plans, but also revise them.
    • EN Key Points:
      • arXiv:2610.10786v1 Announce Type: new

  • Abstract: Planning is increasingly important for long-horizon agents, where successful execution requires coordinating subgoals, tool use, and intermediate outc…

    • Yet assumptions made during planning may be invalidated by the environment, tools may return unexpected results, or actions may fail
    • Effective agents must therefore not only generate plans, but also revise them

  • Whose Ground Truth? Embracing Ambiguity in Human-Centered AI

    • Published: 2026-10-09 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2610.10805v1 Announce Type: new.
      • Abstract: As AI systems increasingly interact with people and make decisions about them, understanding human interpretations becomes an important part of developing human-centered AI.
      • Conventional machine learning and AI systems are largely developed under the assumption that a single definitive ground truth exists, with variability in human annotations often resolved through aggregation or treated as noise.
      • However, for many human-centered tasks, human interpretation is inherently ambiguous, and multiple interpretations of the same input may be simultaneously reasonable and valid.
    • EN Key Points:
      • arXiv:2610.10805v1 Announce Type: new
      • Abstract: As AI systems increasingly interact with people and make decisions about them, understanding human interpretations becomes an important part of develo…
      • Conventional machine learning and AI systems are largely developed under the assumption that a single definitive ground truth exists, with variability in human…
      • However, for many human-centered tasks, human interpretation is inherently ambiguous, and multiple interpretations of the same input may be simultaneously reaso…

  • On the Clock: Towards Punctual and Productive Time-Budgeted AI Agents

    • Published: 2026-10-09 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2610.10833v1 Announce Type: new.

  • Abstract: We study whether small LLM agents can operate effectively under explicit wall-clock time budgets by both respecting the allocated runtime and using available time productively.

    • We evaluate Qwen3.6-27B on five competitions from MLE-Bench Lite and Qwen3-4B on Zork I (Jericho), two agentic benchmarks where additional computational time can meaningfully improve performance.
    • In the simplest setting, where the budget is stated only in the prompt, agents fail to translate the stated budget into controlled use of time.
    • EN Key Points:
      • arXiv:2610.10833v1 Announce Type: new
      • Abstract: We study whether small LLM agents can operate effectively under explicit wall-clock time budgets by both respecting the allocated runtime and using av…
      • We evaluate Qwen3.6-27B on five competitions from MLE-Bench Lite and Qwen3-4B on Zork I (Jericho), two agentic benchmarks where additional computational time ca…
      • In the simplest setting, where the budget is stated only in the prompt, agents fail to translate the stated budget into controlled use of time

  • Self-Supervised Keyframe Discovery for Horizon-Invariant Behavior Cloning

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2610.10857v1 Announce Type: new.
      • Abstract: Behavior cloning (BC) in non-Markovian environments is a challenging problem because policies have to reason over contextual information over long horizons.
      • Existing policy architectures rely on recurrent or attention-based mechanisms to capture long-term dependencies.
      • However, recurrent models suffer from hidden-state collapse and gradient instability under backpropagation through time, while attention-based models are fundamentally limited by context length.
    • EN Key Points:
      • arXiv:2610.10857v1 Announce Type: new
      • Abstract: Behavior cloning (BC) in non-Markovian environments is a challenging problem because policies have to reason over contextual information over long hor…

  • Existing policy architectures rely on recurrent or attention-based mechanisms to capture long-term dependencies

    • However, recurrent models suffer from hidden-state collapse and gradient instability under backpropagation through time, while attention-based models are fundam…

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

  • Diffu-LoRA: A Novel Low-Rank Adaptation for Personalized Diffusion Models

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2610.10550v1 Announce Type: new.
      • Abstract: Personalizing text-to-image diffusion models from a few reference images requires preserving subject identity while following prompts that describe new contexts.
      • Full-model fine-tuning is parameter-intensive, whereas low-rank adaptation (LoRA) reduces the number of trainable parameters but leaves open how adaptation capacity should be distributed across layers.
      • We introduce Diffu-LoRA, a parameter-efficient method that learns this allocation through gated low-rank adaptation.
    • EN Key Points:
      • arXiv:2610.10550v1 Announce Type: new
      • Abstract: Personalizing text-to-image diffusion models from a few reference images requires preserving subject identity while following prompts that describe ne…
      • Full-model fine-tuning is parameter-intensive, whereas low-rank adaptation (LoRA) reduces the number of trainable parameters but leaves open how adaptation capa…
      • We introduce Diffu-LoRA, a parameter-efficient method that learns this allocation through gated low-rank adaptation

  • Wieszcz-XIX: A 3.1-Billion-Word Corpus of Pre-1918 Polish and Temporally Bounded Language Models Trained From Scratch

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2610.10592v1 Announce Type: new.
      • Abstract: Historical Polish is well documented as a language but annotated in machine-readable form only to about a million words for the period this paper covers; the rest sits behind optical character recognition of variable quality.

  • We present Wieszcz-XIX, a corpus of 6.75 billion tokens (about 3.1 billion words) in 294,369 documents, most of them periodical issues, of Polish published from 1800 to 1918, assembled from Wolne Lektury and the Internet Archive by a pipeline that filters, deduplicates, audits for post-1918 leakage and splits at the document level.

    • It is over three orders of magnitude larger than the annotated corpus of the same period, and we quantify its defects: recognition corruption against a false-positive floor, near-identical duplication, which is removed, and post-1918 leakage, which is excluded from the training corpus itself down to a known residue of 0.04 to 0.38% of its bytes, found in the transcribed source, so the published corpus is the trained one document for document.
    • EN Highlights:
      • arXiv:2610.10592v1 Announce Type: new
      • Abstract: Historical Polish is well documented as a language but annotated in machine-readable form only to about a million words for the period this paper cove…
      • We present Wieszcz-XIX, a corpus of 6.75 billion tokens (about 3.1 billion words) in 294,369 documents, most of them periodical issues, of Polish published from…
      • It is over three orders of magnitude larger than the annotated corpus of the same period, and we quantify its defects: recognition corruption against a false-po…

  • Large Language Model-Assisted Preparation of Transportation Management Plans: A Case Study with WisDOT WisTMP System

    • Publication Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10650v1 Announce Type: new.
      • Abstract: Work zones are critical yet hazardous components of transportation infrastructure, requiring carefully designed Transportation Management Plans (TMPs) to ensure safety and mobility.
      • However, TMP preparation remains labor-intensive and heavily dependent on practitioner expertise.

  • This paper proposes a Large Language Model (LLM)-assisted framework to automate TMP content generation, leveraging the WisDOT WisTMP system as the application context.

    • EN Key Points:
      • arXiv:2610.10650v1 Announce Type: new
      • Abstract: Work zones are critical yet hazardous components of transportation infrastructure, requiring carefully designed Transportation Management Plans (TMPs)…
      • However, TMP preparation remains labor-intensive and heavily dependent on practitioner expertise
      • This paper proposes a Large Language Model (LLM)-assisted framework to automate TMP content generation, leveraging the WisDOT WisTMP system as the application c…

  • Cognitive Thermometers: Machine Learning and Logical Complexity

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10724v1 Announce Type: new.
      • Abstract: How does the human mind represent semantic categories?
      • Why do natural languages favor certain meanings over others?
      • Prior explanations have relied on logical definability and complexity, but these are highly sensitive to the choice of logical language, rendering some design choices unmotivated.
    • EN Key Points:
      • arXiv:2610.10724v1 Announce Type: new
      • Abstract: How does the human mind represent semantic categories
      • Why do natural languages favor certain meanings over others
      • Prior explanations have relied on logical definability and complexity, but these are highly sensitive to the choice of logical language, rendering some design c…

  • Lossy Compressive Text Autoencoders

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10738v1 Announce Type: new.
      • Abstract: Our work explores learning a compressed latent representation of text, at the intersection of data compression and representation learning.

  • We propose an autoencoder architecture that performs residual downscaling and upscaling of hidden representations along the time axis, with a residual low-dimension discrete bottleneck.

    • We analyze our approach for different quantization methods, training objectives, and datasets.
    • EN Key Points:
      • arXiv:2610.10738v1 Announce Type: new
      • Abstract: Our work explores learning a compressed latent representation of text, at the intersection of data compression and representation learning
      • We propose an autoencoder architecture that performs residual downscaling and upscaling of hidden representations along the time axis, with a residual low-dimen…
      • We analyze our approach for different quantization methods, training objectives, and datasets

  • Clarify, Then Focus: Statement Normalization for Conversation Analytics at Scale

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10758v1 Announce Type: new.
      • Abstract: Enterprise conversation analytics asks many questions of millions of interactions.
      • Each question can require reconstructing what people mean and identifying which information matters, repeating costly interpretive work across the same transcripts.
      • We propose a simple principle: clarify the text, then focus the reader.
    • EN Key Points:
      • arXiv:2610.10758v1 Announce Type: new
      • Abstract: Enterprise conversation analytics asks many questions of millions of interactions
      • Each question can require reconstructing what people mean and identifying which information matters, repeating costly interpretive work across the same transcri…
      • We propose a simple principle: clarify the text, then focus the reader

  • Grammar Concept Annotation at Scale: Deployed Fine-Tuned Small Language Models Outperform Prompted Frontier Models

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10827v1 Announce Type: new.

  • Abstract: Corrective feedback is among the best-evidenced drivers of second-language acquisition, yet corrections delivered during lessons rarely accumulate into an actionable view of grammar mastery.

    • Prompted frontier models can provide such a view from learner–tutor lesson transcripts, but they are costly at scale.
    • We close this gap by fine-tuning Qwen3.5 small language models (SLMs) on filtered and rebalanced teacher-generated supervision, then deploying an efficient 0.8B model in an end-to-end grammar mastery tracker for all English learners on our platform.
    • EN Key Points:
      • arXiv:2610.10827v1 Announce Type: new
      • Abstract: Corrective feedback is among the best-evidenced drivers of second-language acquisition, yet corrections delivered during lessons rarely accumulate int…
      • Prompted frontier models can provide such a view from learner–tutor lesson transcripts, but they are costly at scale
      • We close this gap by fine-tuning Qwen3.5 small language models (SLMs) on filtered and rebalanced teacher-generated supervision, then deploying an efficient 0.8B…

  • Real Long-Term Memory for AI: A 50-Million-Token Window That Is Faster and Cheaper Than Recompute

    • Publication Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2610.10845v1 Announce Type: new.
      • Abstract: A large language model can only use the text that fits in its context window, and it recomputes its internal key-value (KV) state for a prompt every time the prompt is sent.
      • We test a memory layer, the public package galahad-kv, that saves the KV state of each block of about 16,000 tokens to encrypted local NVMe disk and loads it back later, byte-exact, without recomputing it.
      • We ran it on 50,000,000 tokens of real public text, served through vLLM on one NVIDIA H100, with Gemma 4 12B and Gemma 4 31B.
    • EN Key Points:
      • arXiv:2610.10845v1 Announce Type: new

  • Abstract: A large language model can only use the text that fits in its context window, and it recomputes its internal key-value (KV) state for a prompt every t…

    • We test a memory layer, the public package galahad-kv, that saves the KV state of each block of about 16,000 tokens to encrypted local NVMe disk and loads it ba…
    • We ran it on 50,000,000 tokens of real public text, served through vLLM on one NVIDIA H100, with Gemma 4 12B and Gemma 4 31B

  • Disentangling Linguistic and Paralinguistic Information with Routed Sparse Autoencoders

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To Be Translated] - arXiv:2610.10865v1 Announce Type: new.
      • Abstract: Self-supervised speech encoders contain linguistic and paralinguistic information in a shared, entangled representation space.
      • We combine a TopK sparse autoencoder with route-specific supervision and cross-factor adversaries.
      • Across frozen SPEAR and WavLM encoders, independent probes show factor-specific retention and suppression: linguistic information remains stronger in the linguistic route, while paralinguistic factors, including speaker identity, emotion, and prosody, are retained in the paralinguistic route and substantially reduced in the linguistic route.
    • EN Key Points:
      • arXiv:2610.10865v1 Announce Type: new
      • Abstract: Self-supervised speech encoders contain linguistic and paralinguistic information in a shared, entangled representation space
      • We combine a TopK sparse autoencoder with route-specific supervision and cross-factor adversaries
      • Across frozen SPEAR and WavLM encoders, independent probes show factor-specific retention and suppression: linguistic information remains stronger in the lingui…

  • Sparse Attention Is Matrix Approximation, Not Choosing from a Bag of Values

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To Be Translated] - arXiv:2610.10871v1 Announce Type: new.

  • Abstract: Large Language Models (LLMs) achieve strong performance across many domains, but their efficiency is limited by the quadratic cost of attention with respect to prompt length.

    • Sparse attention reduces this cost by retaining only a small fraction of query-key interactions to approximate the full attention matrix.
    • However, existing methods are trapped in a mathematically wrong view: they simply keep large scalar entries or high-mass regions of the attention matrix.
    • EN Key Points:
      • arXiv:2610.10871v1 Announce Type: new
      • Abstract: Large Language Models (LLMs) achieve strong performance across many domains, but their efficiency is limited by the quadratic cost of attention with r…
      • Sparse attention reduces this cost by retaining only a small fraction of query-key interactions to approximate the full attention matrix
      • However, existing methods are trapped in a mathematically wrong view: they simply keep large scalar entries or high-mass regions of the attention matrix

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

  • Freeze the Decoder, Heal the Encoder: Parameter-Efficient Adaptation for SVD-Based KV-Cache Compression

    • Publication Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10552v1 Announce Type: new.
      • Abstract: Comparing parameter-efficient fine-tuning recipes under a single, shared learning rate is a common but flawed practice: when the arms being compared have very different trainable-parameter counts, a shared rate can simultaneously depress the larger arms’ means and inflate their variance, manufacturing a large, seemingly multi-seed-significant advantage for the smallest arm that is not a real effect.

  • We document this confound in a concrete setting: post-hoc SVD-based KV-cache compression, where an already-pretrained model is converted to a low-rank (multi-head-latent-attention-style) cache by factorizing its key/value weights into a down-projection (“encoder”) and an up-projection (“decoder”), after which a short fine-tune (“healing”) recovers the accuracy lost to truncation.

    • Under a shared learning rate, freezing the decoder and healing only the encoder looks like a clear win over healing the decoder or both factors; once every arm is given its own tuned learning rate, that apparent advantage disappears, and encoder-only healing instead reaches parity with the alternatives, at a real, measured saving of 3x fewer trainable parameters and 3x less optimizer-state memory.
    • EN Key Points:
      • arXiv:2610.10552v1 Announce Type: new
      • Abstract: Comparing parameter-efficient fine-tuning recipes under a single, shared learning rate is a common but flawed practice: when the arms being compared h…
      • We document this confound in a concrete setting: post-hoc SVD-based KV-cache compression, where an already-pretrained model is converted to a low-rank (multi-he…
      • Under a shared learning rate, freezing the decoder and healing only the encoder looks like a clear win over healing the decoder or both factors; once every arm…

  • SPERA: Spherical Prior EEG Foundation Model with Geometry- and Frequency-Aware Latent Prediction

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10571v1 Announce Type: new.
      • Abstract: Electroencephalography (EEG) provides a non-invasive measure of ongoing neural activity, but building general-purpose EEG models remains challenging due to the heterogeneity of subjects, devices, and electrode montages.
      • Existing EEG foundation models predominantly rely on reconstruction-based objectives defined on the observed signal, which contains both neural and non-neural components.

  • We introduce SPERA (Spherical Prior EEG Representation Architecture), an EEG foundation model that adopts the joint-embedding predictive architecture (JEPA) to predict in latent space.

    • EN Highlights:
      • arXiv:2610.10571v1 Announce Type: new
      • Abstract: Electroencephalography (EEG) provides a non-invasive measure of ongoing neural activity, but building general-purpose EEG models remains challenging d…
      • Existing EEG foundation models predominantly rely on reconstruction-based objectives defined on the observed signal, which contains both neural and non-neural c…
      • We introduce SPERA (Spherical Prior EEG Representation Architecture), an EEG foundation model that adopts the joint-embedding predictive architecture (JEPA) to…

  • Coverage, Not Difficulty, Sets How Much Synthetic Data an Activation Probe Needs

    • Release Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [Pending Translation] - arXiv:2610.10594v1 Announce Type: new.
      • Abstract: Activation probes that monitor deployed language models are trained on synthetic conversations, and how many a probe needs is open.
      • We trace learning curves over 10-590 synthetic samples for three monitoring concepts, high-stakes situations, replies harmful to a person, and replies that do not follow the user’s instruction, on fourteen held-out evaluation distributions and four probe models, varying the generator LLM and the prompt’s detail.
      • The need is set by what is monitored: probes for high-stakes and harmful are within a few hundredths of their plateau from 80 samples on Gemma-3-27B-IT, instruction probes need several times as many, and the ordering holds on three smaller probe models and on real samples (from dev set).
    • EN Highlights:
      • arXiv:2610.10594v1 Announce Type: new
      • Abstract: Activation probes that monitor deployed language models are trained on synthetic conversations, and how many a probe needs is open

  • We trace learning curves over 10-590 synthetic samples for three monitoring concepts, high-stakes situations, replies harmful to a person, and replies that do n…

    • The need is set by what is monitored: probes for high-stakes and harmful are within a few hundredths of their plateau from 80 samples on Gemma-3-27B-IT, instruc…

  • Temporal transformer CAN encoder with federated lightweight heads for anomaly detection

    • Published: 2026-10-09 12:00 Beijing Time
    • Summary: [To be translated] - arXiv:2610.10613v1 Announce Type: new.
      • Abstract: Modern vehicles rely on large numbers of Electronic Control Units (ECUs) that constantly exchange information over the Controller Area Network (CAN) bus.
      • Due to the rapidity, structure, and repetition of this communication, even slight variations in timing, payload values, or message patterns can point to unusual activity.
      • Whether due to errors, malfunctions, or deliberate interference, these anomalies are frequently subtle and challenging to identify with conventional methods that handle messages separately or rely on manually created rules.
    • EN Key Points:
      • arXiv:2610.10613v1 Announce Type: new
      • Abstract: Modern vehicles rely on large numbers of Electronic Control Units (ECUs) that constantly exchange information over the Controller Area Network (CAN) b…
      • Due to the rapidity, structure, and repetition of this communication, even slight variations in timing, payload values, or message patterns can point to unusual…
      • Whether due to errors, malfunctions, or deliberate interference, these anomalies are frequently subtle and challenging to identify with conventional methods tha…

  • When Routing Reveals Membership: Privacy Leakage from MoE Router Telemetry

    • Published: 2026-10-09 12:00 Beijing Time
    • Summary: [To be translated] - arXiv:2610.10616v1 Announce Type: new.

  • Abstract: Mixture-of-Experts (MoE) language models produce routing information during inference that may be logged or exposed for monitoring, debugging, load analysis, and safety auditing.

    • Unlike ordinary model outputs, this telemetry reveals a view of the model’s internal computation, raising a privacy question: can it reveal whether an example was used to fine-tune the deployed model?
    • We introduce a router-augmented membership inference attack that combines conventional output-side signals with aggregated routing features and applies a membership classifier learned from independently fine-tuned shadow models to the target model.
    • EN Highlights:
      • arXiv:2610.10616v1 Announce Type: new
      • Abstract: Mixture-of-Experts (MoE) language models produce routing information during inference that may be logged or exposed for monitoring, debugging, load an…
      • Unlike ordinary model outputs, this telemetry reveals a view of the model’s internal computation, raising a privacy question: can it reveal whether an example w…
      • We introduce a router-augmented membership inference attack that combines conventional output-side signals with aggregated routing features and applies a member…

  • Self-Organization from Constrained Geometric Radiation

    • Published: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10621v1 Announce Type: new.
      • Abstract: How does dynamic order emerge spontaneously in closed systems without external driving?
      • Existing paradigms all require external energy flows, temperature quenching, or slow driving.
      • Here we report constraint-induced self-organization via geometric radiation in coupled metric evolution systems.
    • EN Highlights:
      • arXiv:2610.10621v1 Announce Type: new
      • Abstract: How does dynamic order emerge spontaneously in closed systems without external driving
      • Existing paradigms all require external energy flows, temperature quenching, or slow driving

  • Here we report constraint-induced self-organization via geometric radiation in coupled metric evolution systems

  • Recurrent Self-Improvement: Dynamic Cross-Loop On-Policy Distillation for Looped Language Models

    • Publication Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10623v1 Announce Type: new.
      • Abstract: Looped Language Models (LoopLMs) offer a parameter efficient approach to scaling reasoning by reusing shared parameters across recurrent computation steps.
      • Despite their promise, effective post-training of LoopLMs remains challenging.
      • Existing approaches either provide reward based supervision that is sparse or costly to extend across loops, or rely on external teachers or privileged information, leading to limited teacher availability or teacher-student context mismatch.
    • EN Key Points:
      • arXiv:2610.10623v1 Announce Type: new
      • Abstract: Looped Language Models (LoopLMs) offer a parameter efficient approach to scaling reasoning by reusing shared parameters across recurrent computation s…
      • Despite their promise, effective post-training of LoopLMs remains challenging
      • Existing approaches either provide reward based supervision that is sparse or costly to extend across loops, or rely on external teachers or privileged informat…

  • Exact SO(3)-Equivariant Isotropic Kernels for Rotation-Robust Neural Dynamics

    • Publication Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10626v1 Announce Type: new.
      • Abstract: Neural surrogates for vector-valued partial differential equations can fit training data yet change their predictions when the same physical state is expressed in a rotated coordinate frame.
      • We study this failure on three-dimensional Navier–Stokes dynamics observed at irregularly placed points.

  • We introduce the Invariant-Conditioned Isotropic Kernel Neural Operator (IKNO), a compact graph model that builds local interactions from scalar quantities unchanged by rotation and vector directions that rotate with the data.

    • EN Key Points:
      • arXiv:2610.10626v1 Announce Type: new
      • Abstract: Neural surrogates for vector-valued partial differential equations can fit training data yet change their predictions when the same physical state is…
      • We study this failure on three-dimensional Navier–Stokes dynamics observed at irregularly placed points
      • We introduce the Invariant-Conditioned Isotropic Kernel Neural Operator (IKNO), a compact graph model that builds local interactions from scalar quantities unch…

  • Sample-Efficiency of Kolmogorov-Arnold Networks

    • Publish Time: 2026-10-09 12:00 Beijing Time
    • Abstract: [To be translated] - arXiv:2610.10627v1 Announce Type: new.
      • Abstract: Deep reinforcement learning has achieved substantial performance gains over classical control approaches.
      • Yet, a central challenge to learning in real-world applications is acquiring costly samples.
      • Kolmogorov-Arnold Networks are a recently proposed architecture that can learn physical relationships in control problems effectively, with significantly higher parameter efficiency and interpretability when compared to Multi-Layer-Perceptron architectures.
    • EN Key Points:
      • arXiv:2610.10627v1 Announce Type: new
      • Abstract: Deep reinforcement learning has achieved substantial performance gains over classical control approaches
      • Yet, a central challenge to learning in real-world applications is acquiring costly samples
      • Kolmogorov-Arnold Networks are a recently proposed architecture that can learn physical relationships in control problems effectively, with significantly higher…

  • Phase-HDC: Replacing Optimizer History with Gradient Thresholds in Discrete Phase Learning

    • Publish Time: 2026-10-09 12:00 Beijing Time

  • Summary: [To be translated] - arXiv:2610.10630v1 Announce Type: new.

    • Abstract: Training a compact model often needs far more memory than storing it, because the optimizer keeps its own records of past gradients.
    • For a hyperdimensional classifier whose learned parameters are low-bit angles, which we call a \emph{phase memory}, these records take several times more memory than the model itself.
    • We ask whether such a model can be trained while storing nothing but the model.
    • EN Key Points:
      • arXiv:2610.10630v1 Announce Type: new
      • Abstract: Training a compact model often needs far more memory than storing it, because the optimizer keeps its own records of past gradients
      • For a hyperdimensional classifier whose learned parameters are low-bit angles, which we call a \emph{phase memory}, these records take several times more memory…
      • We ask whether such a model can be trained while storing nothing but the model