AAI Search
AI 资讯日报 · 2026-09-17

2026-09-17 · AI 资讯日报

今日新收录 146 条公开资讯，按模型 / 产品 / 行业 / 论文 / 观点 自动归类汇编（非 AI 生成，点击可溯源原文）。

今日精选6 条

  1. 1.
    论文研究Agentic Societies Need a Social Harness— arXiv

    An agentic society is a collection of AI agents that coordinate autonomously across trust boundaries, on behalf of different principals whose objectives may only partially align. We show experimentall…

  2. 2.
    论文研究When Should LLMs Abstain? Chain-of-Self-Questioning for Selective Risk Control— arXiv

    Large language models can produce fluent answers when their factual support is weak. This paper introduces Chain-of-Self-Questioning (CoSQ), a prompt-only framework that makes answer commitment condit…

  3. 3.
    论文研究What Breaks Under Pruning in Smart Homes, and When? Evaluating LLM Degradation Across Architectures and Task Complexity— arXiv

    Pruning can reduce the deployment cost of large language models (LLMs), but its impact on context-grounded tool calling remains poorly understood. We systematically study pruning-induced degradation i…

  4. 4.
    论文研究LACE: Layer-Wise Compression for Dynamic Frame Rate Codecs— arXiv

    Neural audio codecs are a key component in speech language modeling. However, their high frame rates lead to long sequence lengths, increasing computational costs. Dynamic frame rate codecs mitigate t…

  5. 5.
    论文研究ENCP: Episode-Normalized Conformal Prediction for Vision-and-Language Navigation— arXiv

    Uncertainty estimation for Vision-Language-Navigation (VLN) models is a critical task since it can help identify ambiguous and unreliable predictions, enabling agents to make safer navigation decision…

  6. 6.
    论文研究Verifiable Social Reasoning for LLM Assistants— arXiv

    LLM assistants are widely used for daily social advice, yet evaluating their social reasoning in such consultation settings remains challenging since (i) it requires setups where the assistant learns…

模型发布/更新6 条

  1. 1.
    Our framework for reporting model misalignment— OpenAI

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

  2. 2.
    Helping older adults use AI in everyday life— OpenAI

    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.

  3. 3.
    Reimagining advertising with AI— OpenAI

    Explore new AI-powered advertising experiences from OpenAI, including Sponsored Agents, tools for marketers, and integrations with HubSpot and Shopify.

  4. 4.
    How workers are unlocking new ways of working— OpenAI

    New OpenAI Economic Research shows how workers use AI beyond traditional roles and which new activities become recurring parts of their work.

  5. 5.
    Your AI agents can now control your Google Home devices— TechCrunch

    Google is launching early access to a new MCP server for Google Home, allowing AI agents like Claude, ChatGPT, and others to control connected devices, review camera summaries, and…

  6. 6.
    NVIDIA Vera Rubin NVL72 Delivers Leading Performance in MLPerf Inference v6.1 Debut— NVIDIA

    System performance, efficient infrastructure scaling and continuous software optimization are key levers that determine AI inference economics. Higher system performance means more…

产品发布/更新6 条

  1. 7.
    Amazon launches Alexa+ in India with Hindi support— TechCrunch

    Amazon is letting all customers use Alexa+ assistant in early access period.

  2. 8.
    Snap is launching a new Specs AI tool, and it’s coming to iOS and Mac— The Verge

    AI 导读 · Snap将AI助手与硬件眼镜联动，试图在可穿戴赛道抢占先机，但生态整合能力才是成败关键。

  3. 9.
    Apple is reportedly building an enterprise AI server with its own M8 Ultra chips— The Decoder

    AI 导读 · 苹果自研服务器芯片剑指AI推理，意在摆脱英伟达依赖，但2029年才可能落地，节奏明显偏慢。

  4. 10.
    Apple reportedly building server packed with M-series Ultra chips for AI— Ars Technica

    AI 导读 · 苹果自研AI服务器芯片，标志其从终端算力向云端基础设施的战略延伸。

  5. 11.
    modelscope/ms-cookbook— GitHub

    魔搭紫皮书｜ModelScope Cookbook：面向开发者的开源模型应用实战指南，覆盖模型选型、推理、微调、评测、RAG、Agent 与 AIGC，从跑通第一个模型到构建实际应用。

  6. 12.
    Eurekaleo/awesome-ai-for-games— GitHub

    A curated, searchable collection of research on AI and foundation models across the game lifecycle.

行业动态6 条

  1. 13.
    Nvidia announces native GPU programming in Rust— Hacker News
  2. 14.
    OpenSpec – A lightweight and configurable AI spec framework— Hacker News

    AI 导读 · 轻量可配置的AI规范框架，有望降低智能体开发门槛，让规格定义像写配置一样简单。

  3. 15.
    Anthropic and OpenAI want to embed safety evaluators. Will they really be independent?— TechCrunch

    AI 导读 · 把安全评估员请进自家实验室，独立性能否守住，是AI治理的关键试金石。

  4. 16.
    After accusations of selling ‘perv glasses,’ Meta prepares to sell a pair without a camera— TechCrunch

    AI 导读 · 去掉摄像头只是规避争议，Meta仍未回应隐私担忧的核心。

  5. 17.
    AI labs want in-house auditors — but maybe they should shut the front door first— TechCrunch

    AI 导读 · AI实验室想设内部审计，却忽视更简单的入口管控，值得关注。

  6. 18.
    Anthropic merges Claude Chat, Cowork, and more into a single product— The Decoder

    Anthropic is merging Claude Chat and Cowork into a single product. Instead of users picking between interfaces, Claude now decides on its own whether a task needs a quick answer or…

论文研究6 条

  1. 19.
    Agentic Societies Need a Social Harness— arXiv

    An agentic society is a collection of AI agents that coordinate autonomously across trust boundaries, on behalf of different principals whose objectives may only partially align. We show experimentall…

  2. 20.
    When Should LLMs Abstain? Chain-of-Self-Questioning for Selective Risk Control— arXiv

    Large language models can produce fluent answers when their factual support is weak. This paper introduces Chain-of-Self-Questioning (CoSQ), a prompt-only framework that makes answer commitment condit…

  3. 21.
    What Breaks Under Pruning in Smart Homes, and When? Evaluating LLM Degradation Across Architectures and Task Complexity— arXiv

    Pruning can reduce the deployment cost of large language models (LLMs), but its impact on context-grounded tool calling remains poorly understood. We systematically study pruning-induced degradation i…

  4. 22.
    LACE: Layer-Wise Compression for Dynamic Frame Rate Codecs— arXiv

    Neural audio codecs are a key component in speech language modeling. However, their high frame rates lead to long sequence lengths, increasing computational costs. Dynamic frame rate codecs mitigate t…

  5. 23.
    ENCP: Episode-Normalized Conformal Prediction for Vision-and-Language Navigation— arXiv

    Uncertainty estimation for Vision-Language-Navigation (VLN) models is a critical task since it can help identify ambiguous and unreliable predictions, enabling agents to make safer navigation decision…

  6. 24.
    Verifiable Social Reasoning for LLM Assistants— arXiv

    LLM assistants are widely used for daily social advice, yet evaluating their social reasoning in such consultation settings remains challenging since (i) it requires setups where the assistant learns…

技巧与观点6 条

  1. 25.
    How to connect AI usage to business value— OpenAI

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

  2. 26.
    Optimizing agent system prompts with Amazon Bedrock AgentCore— AWS ML

    AgentCore optimization turns production traces into proposed configuration changes, then validates them before promotion. This technical companion to the launch post explains how t…

  3. 27.
    Build a serverless PII redaction pipeline with Amazon Bedrock Data Automation— AWS ML

    Learn how to automate end-to-end PII detection and redaction from scanned documents at scale using Amazon Bedrock Data Automation with a custom blueprint, AWS Step Functions, and A…

  4. 28.
    Improving HCLS AI reasoning with open-source agent skills— AWS ML

    AI 导读 · 通用模型懂指南却用错，开源技能包正补上医疗决策落地的最后一环。

  5. 29.
    Fault tolerant distributed training on Amazon EKS using NVRx— AWS ML

    AI 导读 · 大模型训练最怕GPU故障中断，秒级恢复加检查点与训练重叠，直接拉高集群可用性。

  6. 30.
    Claude Cowork and chat are now one Claude— Simon Willison

    AI 导读 · 产品线从三个入口收敛为单一Claude，降低了用户认知负担，也意味着Anthropic正把协作与编码

快讯

日报生成时间：2026-09-28
← 2026-09-16最新日报2026-09-18 →