AI 情报文章归档

TL;DR Code is being written three times faster than anyone can review it: GitHub logged...

One harness, same conditions, four dimensions, and the raw receipts of every call — this round, seven...

Google 宣布把 Guided Learning 功能引入 Google Search 学生账号和 Google Classroom，引导 AI 时代下的学习方式。

《魔兽世界：永恒》周三上线前，相关 subreddit 已被大量 AI 生成插件占据，这些插件抄袭长期开发者的作品。Warcraft Logs 母公司 Archon.gg 的 CEO Sildri Silentcrest 表示，因 AI 劣质插件涌入，公司正在调整其免费开放的 API。该 API 过去主要服务业余开发者，如今商业项目越来越多，本应申请许可以避免违反服务条款或无法扩展。

If you've ever tried to parse an LLM response in code, you know the pain: the format changes slightly...

Modem 向所有用户开放应用内反馈功能，可在 Settings > Integrations > In-app feedback 配置，或把起始提示词粘贴给编码智能体自动接入。

At MCP Dev Summit Toronto, Accenture's Michael Forrester told a story about an employee who wanted...

What racking wine teaches about agent memory The mead's lees on October 21, with the racking cane...

数学家们现在因为害怕AI而隐藏自己的研究想法 😀 ~ Terence Tao “现在，数学家们都在隐藏自己的问题。过去他们会非常公开地分享所有有前景的研究方向，而现在他们担心，一旦这样做，某个拥有AI的人就会把他们的研究想法输入前沿模型，赶在他们完成工作之前抢先发表。” ---- 完整视频见“SAIR”YouTube频道，（链接在评论中）

Back in September I published Stop bloating your AGENTS.md: reference your conventions instead of...

When an application sends requests to more than one AI model, keeping those integrations organized...

If you give an AI a vague prompt, it will give you a confident answer. And that confidence will cost...

The most common complaint about AI photo editing is not quality. It's "that's not me anymore". You...

Background I have a primary LINE Bot that I use heavily every day:...

Recap In the previous post "Comprehensive Analysis of Managed Agents API", I created an...

据 FT 报道，Nvidia 要求云与数据中心客户将至少 51% 的服务交给非中国企业，仅适用于新合同。此举针对中国企业通过海外云租赁远程使用先进芯片训练模型的路径，因为美国出口管制一般不覆盖远程访问；云厂商还被要求上报芯片最终用户身份。

据 The Information 报道，中国经济规划部门表示将密切监测 AI 对就业的影响，把 AI 对就业的冲击列为监测风险。此举在此前 10 月国务院发布的指导意见基础上，要求对可能造成重大就业扰动的 AI 项目进行风险评估。

On September 15 at 10:27 UTC, an agent working for me ran a dry-run seven times, once per language of...

OpenClaw 发布 v2026.10.10 版本，带来三项更新：Chat Details 可在宽聊天旁显示、Advisor 建议课程修正、Browser Talk 支持分屏语音。该版本包含 3,028 个源代码 PR 和 454 位贡献者，完整发布说明见 https://docs.openclaw.ai/releases/2026.10.10

Anthropic 在 Claude 平台以公开测试版形式推出 Message Threads。此前每次 API 请求都需重发完整对话历史，新机制只传输自上一条消息以来的变化，据 Anthropic 称在 50 万至 100 万 token 的请求上可将首字时间缩短约 35%。

作者向英国在线超市 Ocado 提交日历邀请格式错误的技术反馈，收到署名“Lulu”的回复，因谄媚式附和、破折号用法等特征判断其为 AI 生成，随后在满意度调查中打 1 星并投诉。几天后 Ocado 人工回复致歉，确认反馈已转交技术团队。作者认为用 AI 回复真实投诉是不尊重用户，主张公开指出糟糕的 AI 回复并要求真人处理。

Large language models (LLMs) on personal devices keep user data local, but limited memory restricts their general capability. Since most user-specific tasks do not require the breadth of frontier models, a practical solution is to adapt a …

Expert offloading enables large mixture-of-experts (MoE) language models to run locally, but conventional batching stalls the GPU while missing experts are fetched. A popular remedy, known as prefetching, is to predict what experts will be…

Reinforcement learning objectives that maximize expected return can concentrate probability mass on a small subset of high-reward outcomes, limiting diversity when multiple distinct solutions are desirable. We introduce Density-Aware Rewar…

Reinforcement learning with verifiable rewards (RLVR) has become an effective approach for improving mathematical reasoning in large language models. However, a fundamental question remains: when a model fails to solve a problem, does it t…

Large language models are candidates for collecting and analyzing records in criminal investigations and mass surveillance. Their capability and accuracy in this work cannot currently be measured. Real deployments lack the ground truth nee…

据 The Information 与 Financial Times 报道，Nvidia 已建议部分云客户（包括其投资的 neocloud）减少来自中国公司的业务，指引不具强制性、只适用于未来合同，并通过销售与区域团队口头传达，云客户还需披露访问芯片的终端用户身份。

SAE-based feature-level interventions to suppress PII leakage in LLMs are primarily evaluated by log-probability reduction. We show that this criterion cannot distinguish feature specific suppression from a magnitude artifact: a direction-…

In single-pixel (ghost) imaging, known illumination patterns probe an object and a detector with no spatial resolution returns one number per pattern. Each measurement costs acquisition time and photons and, for ionizing probes, contribute…

Split Inference (SI) addresses the hardware resource limitations of Internet of Things (IoT) devices by distributing tiny Machine Learning (TinyML) model segments across multiple IoT devices. However, a device failure can interrupt the inf…

We study cooperative multi-agent reinforcement learning with decoupled dynamics and a monotone submodular reward function over joint trajectories. Due to the non-additive reward, an optimal policy may depend on the history of all agents. T…

A deployed LLM coaching agent answers requests whose correct response depends on player state and game version, so advice that is right for one player can be wrong for the next. We describe TactiNavi, the serving system behind the LLM coac…

Probabilistic time-series forecasting requires models that simultaneously capture structured temporal dynamics, expressive uncertainty, and efficient inference, yet existing approaches fall short of this goal: latent dynamical models impos…

An activation steering edit should depend on the input exactly when the destination varies \emph{inside} the class a query requests, and that is measurable before any rule is fitted or applied. Contrastive Activation Addition (CAA) adds on…

作者实测OiiOii新上线的全能复刻功能，复刻了品牌UI动效、MG动画、单品广告、游戏实机、口播和多维度批量裂变6类视频，全程无需写代码或用AE。其原理是镜头光影交给Seedance 2.5生成，识别到文字排版和UI时调用Opus 5.5写代码并在云端浏览器渲染录制，文字边缘比原片更清晰；复刻时留存中间工程文件，同一创意反复修改速度可快4到5倍、费用省约80%。

Your next customer doesn't scroll Retailers have a habit called a store walk. The manager walks the...

Hybrid models let neural networks supply the unresolved parameters of a process-based simulator, and are fitted to observations that constrain aggregate fluxes and storages but not the internal partitioning that produces them. Such a fit i…

Graph neural network explanations are often evaluated via deletion fidelity, i.e., by deleting their top-ranked edges and measuring the resulting change in model output. Yet, in a message-passing GNN, "deleting an edge" does not specify a …

Transaction event sequences support diverse prediction tasks in online commerce and digital financial services. The same transaction records can yield prediction targets with different semantics, including values, changes, and relations to…

A 2025 emoji shifted Claude Code's TUI one cell until a width-table update. How wcwidth, VS16, ZWJ and mode 2027 decide where your agent's borders land.