AI 情报文章归档

TikTok 用户 @therealcornpop 指出，用 AI 写 TikTok 和 YouTube 脚本很容易被识破。破绽不只是"不是 X，而是 Y"这类 AI 腔、三段式结构和怪异断句，更在于内容里什么都没有：没有明确的个人声音，也看不出你对所谈之事真有观点。

OpenAI 发布 GPT-6 Sol 和 GPT-6 Luna，将 GPT-6 Astra 的训练方法用于更快更便宜的模型，API 价格较 GPT-5.6 促销价下调 50%（Sol 输入 $4→$2、输出 $20→$10；Luna 输入 $0.20→$0.10、输出 $1.20→$0.50，每百万 token）。

OpenAI 发布 GPT-6 Sol 和 Luna 更新版，扩展本月早些时候推出的 GPT-6 Astra 系列。6 系列 API 价格为 5.6 系列 Sol 和 Luna 的一半，公司归因于缓存和推理改进；内部事实性评测显示 GPT-6 Sol 错误数约为前代一半，达到 Astra 级可靠性。

Diffusion Large Language Models (dLLMs) have recently emerged as a promising alternative to autoregressive LLMs by enabling non-autoregressive text generation. However, their practical deployment remains limited by inefficient inference, l…

PixVerse R2，实机游玩。#PixVerseWorldModel 那种大制作动作 RPG 的感觉，只不过当你改变规则时，世界会随之响应。自由移动，切换视角，实时探索环境。没有过场动画。操控权在你手里。

We study decentralized partially observable team decision problems with low-rank latent dynamics and unknown system models. The proposed framework combines team-theoretic equivalence with low-rank model representations to address cooperati…

A multi-agent system can reduce latency on complex tasks by executing work concurrently. Several pioneering harness frameworks support multi-agent systems. However, the scalability of current multi-agent harnesses is often constrained by a…

Long-term conversational memory in multi-party settings requires more than retrieving relevant content from long-term conversations: it must distinguish who said what, whom each statement concerns, how individuals perceive one another, wha…

Agents often work on complex problems that require millions of tokens of context, which necessitates compacting across sessions due to limited context windows. We develop CliffCompaction, an autocompaction technique that reduces cost by up…

We introduce SWE-Serve, a benchmark for evaluating agents on production inference engineering tasks. Implementing an inference feature can require coordinating multiple changes across the serving stack, including model support, runtime exe…

我们家的小家伙长得真快 😭 很快他就要去 Grok @bot 大学了 @Tesla：你 Tesla 里的 @Grok 现在能为你做真正有意义的工作了 通过 Connectors，你可以管理收件箱、清理日历，或直接对话处理现有文件/聊天/任务——全程免手操作

Drop 是一款无需 root 权限的 Linux 沙箱工具，可在不打断现有工作环境的情况下隔离程序和编码智能体。它基于 Linux user namespace 实现，隐藏原 home 目录并丢弃用户命名空间能力，可选启用 gVisor 用户态内核作为额外隔离层；还支持用 TOML 配置暴露的文件和本地网络服务，并复用已安装的发行版环境，无需 Docker/Podman 容器设置。

DigitalOcean 在其平台推出 Managed Agents，支持 OpenCode、Codex 等智能体框架，并与原生 Inference Engine 集成。Harness Runtime 在编码沙箱内的隔离 microVM 中运行每个会话，设备断开后会话仍持续、从暂停恢复仅需 305 毫秒；Action Gateway 通过同一 MCP 端点对 16,000+ 工具提供受管访问。

Claude Opus 5.5 现已在 Google Cloud 上线，可通过 Model Garden 使用。该模型面向长时间运行的编程与知识工作等日常复杂任务，过程中保持用户参与，且每 token 价格更低。

Arena 宣布 Anthropic 的 Claude Opus 5.5 已进入 Agent Arena，用户可通过投票驱动排行榜。Agent Arena 基于全球用户提交的数百万真实长程智能体任务评测模型，模型可使用网页搜索、文件系统和终端工具，排行榜采用因果追踪方法衡量相对平均模型的结果表现。

Plain custom elements, a synchronous lifecycle, nothing re-renders on its own, and typed events. Four rules, each chosen because a reviewer can check it by reading the diff.

StepFun 开源内部使用的 LLM 数据标注与模型检查工具 onPanda，支持 token 级可视化与控制。标注方面称中位标注时间比人工后编辑降低 52%，可在同一流程产出 SFT 与偏好数据，ΔPPL <1%，并支持图像、音频和视频的 agent 轨迹标注；调试方面可查看 token 概率与 top-k 候选、逐 token 引导解码。

Agents using the Model Context Protocol (MCP) rely on semantic matching to select tools from third-party servers, exposing a semantic supply-chain risk through attacker-controlled metadata and outputs. We introduce A2M (Attraction-to-Manip…

Large language model (LLM) agents often handle streams of related tasks, yet standard harnesses repeatedly ask the model to reconstruct the same control decisions inside each task's context. We study whether task feedback can instead turn …

Typed decision models are built for settings where model outputs are consumed directly by software. Instead of generating free-form text, they return a decision over a predefined set of options. By construction, every output conforms to th…

RRSI 智能体框架的正则化递归自改进 论文：https://huggingface.co/papers/2609.24972

X-ray is medicine's most widely used imaging modality, yet remains among its least quantitative. Unlike volumetric modalities like CT or MRI, X-ray collapses 3D anatomy into a 2D projection, causing structures to overlap and anatomical bou…

There's a failure mode in agent tooling that never throws anything and I lost an afternoon to it...

我们开源了 Ming-Image-0.1-Design 系列： • Ming-Image-0.1-Design，6B • Ming-Image-0.1-Design-Layer，6B • 两个开源 Agent Skills：Ling UI Design Skill 和 Image-to-Editable-PPT Skill Ming-Image-0.1-Design 在 Artificial Analysis 的 UI/UX Design 排行榜上位列开放权重模型第一。🧵

Perplexity 在 Computer 中推出侧聊（side chat）功能，用户可以在主任务继续进行时，通过单独的只读侧聊快速提问。每个侧聊使用上下文快照，可以读取文件或搜索网页，不影响主任务；CEO Aravind Srinivas 补充了在 Computer 会话中用 /btw 发起侧聊、共享上下文快照的用法。

Large language models are increasingly used to generate SystemVerilog Assertions from natural-language specifica- tions and register-transfer-level designs. Existing datasets and benchmarks support important goals such as large- scale trai…

TypeSafe released Jev on September 15th, 2026 and called it a System One model. A System One model is...

前沿：节奏 抽掉梯子：加速 [引用 @M1Astra]：Claude Opus 5.5 将是首个 Opus，针对"与前沿 LLM 开发相关的一小部分能力（如内核开发）"设计为回退到能力较弱的模型。

Large language models (LLMs) are increasingly applied to the automated repair of C/C++ security vulnerabilities, and compile rate is a commonly reported proxy for progress: whether the generated patch compiles. We argue that compile rate i…

Clustering is an unsupervised learning technique that partitions unlabeled data into groups. Most existing methods require user-specified parameters, such as the number of clusters or neighborhood size. Conversely, we propose automatic dep…

加拿大是 Cohere 诞生的地方。它养育了我们，让我们的创始人接受了教育，也教会了他们我们构建技术时所秉持的价值观。 这些价值观是什么？ AI 不应该让你远离你所热爱的事物。AI 应该赋能你，让你有更多时间与它们相伴。

Detection of overlapping communities is essential for modelling networks in which nodes participate simultaneously in multiple structural or functional groups. Existing graph neural network approaches commonly rely on local message passing…

关于 Jev 的一切：🦄 AI That Works #75 https://x.com/i/broadcasts/1qJVmyramPAGB

OpenAI 发布第三方评估的四大优先领域及配套原则，承诺为独立评估方提供覆盖训练、评估与部署的深度访问权限。OpenAI 称该权限应让第三方评估者质疑其假设、发现被遗漏的风险，并就其安全防护措施的有效性得出独立结论。

Pragmatic Engineer 采访 Windows 团队，详解微软在操作系统层面集成 AI 智能体的计划，包括 Entra ID 智能体身份、MCP 注册表 ODR、智能体隔离机制 MXC、本地模型层 Windows ML 及内置 SLM。

我们为之而来的人，就是像 @cheneypiano 这样有想法、有方案、有梦想的人，他们只需要一个载体把这一切变成现实。 @cheneypiano：谢谢 @Replit。@amasad 和团队一直对我很好，却从不知道他们对我的人生影响有多大。他们来我家拍摄了这段内容，现在他们更了解了一些。但 Replit 真的改变了我的人生。

Anthropic 表示 Claude Opus 5.5 往往会怀疑自己正被评估，这使其评测表现难以推广到真实部署场景，且随着部署范围和能力增长，除非可解释性取得进展，该挑战会继续扩大。引用内容补充称 Opus 5.5 达到 Fable 5.1 级别性能，相比 Opus 5 典型工作负载成本降 40%，输入输出价格为 $4 和 $20 per 1M tokens，缓存读取降 60% 至 $0.20，输出速度快 30% 以上，Fast mode 最高 2.5x 速度但 tok…

llm-anthropic 0.29 发布，新增对 Claude Opus 5.5 的支持，可通过 `llm -m claude-opus-5.5 "prompt goes here"` 调用该模型。

AI tools for digital product design now offer prompt-to-design capabilities, allowing designers and their non-designer colleagues to create prototypes through conversational workflows with large language models (LLMs). While these tools pr…