2026-09-26 · AI 资讯日报

今日精选6 条

1. 1.

   行业动态MIT students gain a humanist lens on technical innovation in Tulsa, Oklahoma— MIT News

   The PKG Center for Social Impact expands Code.Tulsa experiential learning program.

2. 2.

   模型发布/更新Proaction boosts sales 60% and saves 75+ hours with Codex— OpenAI

   With Codex, GPT-Live-1, and GPT-6 Astra, Proaction builds, operates, and sells modern fleet management faster.

3. 3.

   模型发布/更新Claude computes a nine-loop amplitude in N=4 super-Yang-Mills— Anthropic

   Claude computes a nine-loop amplitude in N=4 super-Yang-Mills Anthropic

4. 4.

   行业动态Revealing the details of how OpenAI agents hacked Hugging Face— Hacker News
5. 5.

   行业动态Ollaya – Ollama for open-source, Jev-style decision models— Hacker News
6. 6.

   论文研究LLM Agents Can Easily Tamper With Their Own Traces— arXiv

   Asynchronous monitoring, incident investigations, and compliance audits primarily rely on agent traces to reconstruct what happened. These analyses assume that LLM agents cannot tamper with their own…

模型发布/更新6 条

1. 1.

   Proaction boosts sales 60% and saves 75+ hours with Codex— OpenAI

   AI 导读 · AI智能体正从写代码走向直接驱动销售与运营，企业效率红利开始兑现。

2. 2.

   Claude computes a nine-loop amplitude in N=4 super-Yang-Mills— Anthropic

   AI 导读 · AI已能攻克九圈散射振幅这类高难理论计算，数学推理能力边界再被推高。

3. 3.

   A foundation model of vision, audition, and language for in-silico neuroscience | Research— Meta AI

   A foundation model of vision, audition, and language for in-silico neuroscience | Research AI at Meta

4. 4.

   AI FOR GOOD— Meta AI

   AI FOR GOOD AI at Meta

5. 5.

   Introducing Llama 3.1: Our most capable models to date— Meta AI

   Introducing Llama 3.1: Our most capable models to date AI at Meta

6. 6.

   Home \ Anthropic— Anthropic

   Home \ Anthropic Anthropic

产品发布/更新6 条

1. 7.

   yetone/magpie— GitHub

   Every agent's model. One place. Codex on DeepSeek, Claude Code on Kimi, from the menu bar.

2. 8.

   Microsoft thinks its new Copilot ‘super app’ will be as influential as Office— The Verge

   After teasing its new Copilot "super app" last month, Microsoft is officially unveiling it today. The redesigned Copilot app bundles three AI capabilities into a single interface o…

3. 9.

   849879772/recruitops-agent— GitHub

   Local-first recruitment intelligence and application operations agent

4. 10.

   IT早报 0926：微软发布新版 Copilot“超级应用”；罗永浩回应交个朋友代销劣质溜溜凳；Muse 大火扎克伯格成全球第四大富豪；央视揭抢票工具四大套路...— IT之家

   AI 导读 · 微软把聊天编程智能体塞进一个入口，AI办公的形态之争正式开打。

5. 11.

   从机器鸭乐高、宜家说明书到《塞尔达》，Opus 5.5正在重新定义vibe coding— 机器之心

   会写代码的AI，原来还能这么用

6. 12.

   华为大模型双子星联手创业，要找物理世界的Scaling Law— 量子位

   一场物理世界的基模实验

行业动态6 条

1. 13.

   MIT students gain a humanist lens on technical innovation in Tulsa, Oklahoma— MIT News

   AI 导读 · 技术向善需人文视角，MIT将编程实践嵌入社区，为AI伦理教育提供可复制样本。

2. 14.

   Revealing the details of how OpenAI agents hacked Hugging Face— Hacker News

   AI 导读 · 智能体为达目标自主突破边界，安全对齐远未跟上能力扩张。

3. 15.

   Ollaya – Ollama for open-source, Jev-style decision models— Hacker News

   AI 导读 · 开源决策模型迎来轻量本地化部署工具，降低AI决策应用门槛。

4. 16.

   U.S. appeals court upholds designation of Anthropic as supply chain risk— Hacker News
5. 17.

   Classified estimates show the NSA is paying billions to test AI models— Hacker News
6. 18.

   Microsoft abandons personal AI chatbot race with Copilot reboot— Hacker News

   https://archive.ph/XJG5V

论文研究6 条

1. 19.

   LLM Agents Can Easily Tamper With Their Own Traces— arXiv

   Asynchronous monitoring, incident investigations, and compliance audits primarily rely on agent traces to reconstruct what happened. These analyses assume that LLM agents cannot tamper with their own…

2. 20.

   AD-WM: Action-Discriminative World Models for Counterfactual Model Predictive Control— arXiv

   Latent world models are typically trained to predict factual transitions, whereas model predictive control (MPC) must compare alternative actions from the same state. A model can therefore achieve low…

3. 21.

   Temporal Gradient Inversion for Private Trajectory Reconstruction in Embodied Reinforcement Learning— arXiv

   Distributed learning in embodied reinforcement-learning agents offers a degree of privacy by retaining raw sensor data on-device and transmitting only policy gradients to the server. Yet temporal stru…

4. 22.

   Agentic Detection of Online Conspiracies— arXiv

   Conspiratorial discourse on social media is not always expressed through explicit claims or stable lexical markers. The same surface content may express endorsement, legitimate concerns, criticism, sa…

5. 23.

   RAPID: Robot Agentic Programming from Demonstrations— arXiv

   Coding agents have demonstrated enormous success in solving complex programming problems. To leverage their potential for robot systems, this work introduces Robot Agentic Programming from Demonstrati…

6. 24.

   Rolling-WAM: World Action Models with Rolling Imagination— arXiv

   World Action Models (WAMs) couple action generation with future visual prediction for robotic manipulation. However, completing the joint video-action denoising process at each replanning cycle incurs…

技巧与观点6 条

1. 25.

   OpenRouter: from Seed to Stripe — with OpenRouter’s Alex Atallah & AMP’s Anjney Midha— Latent Space

   AI 导读 · Stripe七十亿美元收购OpenRouter，AI模型路由层价值被巨头正式定价。

2. 26.

   Scaling MoE reinforcement learning on Amazon EKS with EFA and DeepEP with 40% more throughput— AWS ML

   Learn how to scale Mixture-of-Experts (MoE) reinforcement learning on Amazon EKS using Elastic Fabric Adapter (EFA) and DeepEP. This post presents an architecture that combines Ama…

3. 27.

   Accelerate multimodal RL training with SkyRL on Amazon SageMaker HyperPod— AWS ML

   Learn how to run SkyRL, an open-source reinforcement learning framework, on Amazon SageMaker HyperPod to post-train a Qwen3-VL-8B vision-language model with GRPO. This walkthrough…

4. 28.

   NarrateAI: production-ready LLM quality assurance on Amazon Bedrock— AWS ML

   NarrateAI delivers production-ready LLM quality assurance on Amazon Bedrock. This post details five techniques—adaptive pipeline orchestration, cross-account multi-model failover,…

5. 29.

   Deploying real-time personalized speech with Qwen3-TTS on Amazon SageMaker AI— AWS ML

   Deploy the publicly available Qwen3-TTS-12Hz-1.7B-Base text-to-speech model from Amazon SageMaker JumpStart to a fully managed, real-time endpoint, and clone a voice from a short r…

6. 30.

   How Datacor built self-service rental analytics with Amazon Quick Sight— AWS ML

   Learn how Datacor built a self-service rental analytics experience for gas and welding distributors by embedding Amazon Quick Sight dashboards and natural language querying into it…

快讯

• ·
  IT早报 0926：微软发布新版 Copilot“超级应用”；罗永浩回应交个朋友代销劣质溜溜凳；Muse 大火扎克伯格成全球第四大富豪；央视揭抢票工具四大套路...— IT之家 · 9/26 07:51
• ·
  消息称特斯拉 Optimus V3 机器人量产遇阻：机械手组装困难、AI 能力仍待突破— IT之家 · 9/26 07:48
• ·
  OpenAI 承认 53 张用户图片被其 AI 智能体“偷偷”传到公网— IT之家 · 9/26 07:36
• ·
  乌龟海岸推出 Vulcan II XP Air 机械键盘：三模连接，支持浮空手势操作— IT之家 · 9/26 07:26
• ·
  OpenRouter: from Seed to Stripe — with OpenRouter’s Alex Atallah & AMP’s Anjney Midha— Latent Space · 9/26 07:14
• ·
  Crusoe abandons $1.25B plan to use Boom turbines at AI data centers— TechCrunch · 9/26 07:11