AI 情报文章归档

TestingCatalog 发现 OpenAI 正在准备图像模型的下一次升级，ChatGPT 内置的隐藏公告弹窗写着图像创作迎来重大升级，文案为更高质量结果、更快生成和更智能的创意工具。

Google DeepMind 与 @GoogleResearch 联合发布 WeatherNext 3，称其为全球天气预报方式的重大突破。该模型直接从真实世界的实时观测中学习，提供更本地化、高精度的预测，且速度更快。

Google DeepMind 与 Google Research 发布 WeatherNext 3，称其为迄今最先进的全球天气 AI 模型，经 Brightband 独立实时评估。

作者 @frxiaobei 发推称 ChatGPT 也挂了，并配图显示重新连接 5/5 及错误信息 unexpected status 404 Not Found，url 为 https://chatgpt.com/backend-api/codex/responses，cf-ray： a3558b01489d312c-LAX。作者在标题中调侃这是喜迎 GPT-6。

Google 推出更新的 AI 天气模型 WeatherNext 3，可基于实时卫星观测每小时生成一次预报，全球分辨率较上一代提升五倍。相较 WeatherNext 2 的 6 小时一次、25 公里网格，新模型对温度和湿度等变量最高达 5 公里分辨率，提前一天以上的降水预报准确率最高提升 50%，在雨量计较少的美国和欧洲以外地区改进最大。

Google DeepMind 与 Google Research 发布新 AI 天气预报模型 WeatherNext 3，将在搜索、Google Maps 和 Gemini 中提供天气信息，并开放给云端用户和研究者。

Running a local LLM is easy. Running multiple specialized models for coding, reasoning and AI agents...

沙特 PIF 旗下 AI 企业 HUMAIN 公布前沿阿拉伯语模型 humain-m3，由 MiniMax 受托开发。该模型基于 MiniMax-M3 构建，进行了 T 级阿拉伯语原生文本 Token 规模的预训练，在 7 个公开阿拉伯语基准测试中取得参测前沿模型的最高平均分。模型已在 HUMAIN Node 平台以研究和评估预览版本提供，后续将发布开放权重版本。

Anthropic 状态页通报 Claude 多个模型请求错误率升高，9 月 3 日 13：26 UTC 开始排查，13：41 已定位原因并着手修复。受影响模型包括 Mythos/Fable 5.1、Mythos/Fable 5、Opus 5、Opus 4.8 和 Opus 4.6；截至 15：25 UTC，仅剩 Opus 4.8 和 Opus 5 仍受影响，其余模型已恢复至基线错误率。

HTC 宣布 AI 智能眼镜 VIVE Eagle 正式登陆欧洲、北美和澳大利亚市场，即日起通过 vive.com 等渠道开放订购，起售价 499 美元。

Elvis Saravia 发推称 Codex 和 Claude 同时宕机，表示那就改用开源模型。

作者发布一张截图并称 ChatGPT 也挂了。截图显示 Codex 后端请求报错，unexpected status 404 Not Found，URL 为 https://chatgt.com/backend-api/codex/responses(原文为 https://chatgpt.com/backend-api/codex/responses），页面提示正在重新连接 5/5。

RL-based post-training for reasoning models is increasingly bottlenecked by repeated fresh rollout generation, particularly in agentic settings where environment interaction dominates wall-clock cost. Replay can reduce this burden by reusi…

Most previous work on speech enhancement (SE) based on masked generative modeling relied on discrete token representations of audio signals, obtained using neural audio codecs (NACs). However, a recent study has shown that continuous laten…

While HTN planning has received significant attention in recent years, support for numerical reasoning remains very limited. In this paper, we investigate numerical Totally-Ordered HTN (TOHTN) planning and show how standard SAT-based encod…

Retrieval-augmented generation (RAG) complements parametric models with retrieved external evidence. The same idea is attractive for continuous-output regression, but directly reusing retrieved target values is often not robust when sample…

Microsoft MAI-Transcribe 2 已上线 OpenRouter。该模型在 FLEURS 上排名第一，支持 60 种语言的语音转文字，具备自动语言识别、code switching、说话人分离和词级时间戳，可通过 openrouter.ai/microsoft/mai-transcribe-2 使用。

Zite 推出 Zite MCP，将 Claude、ChatGPT 或 Cursor 连接后即可要求生成一个应用，上线即自带数据库、全团队登录和权限角色分配。构建过程使用用户现有的 Claude 或 ChatGPT 订阅，零 Zite 积分。作者评论称智能体的有用程度取决于它能行动的环境。

Existing research has repeatedly observed the tendency for English loanwords to cluster in the masculine gender across different recipient languages, yet the origin of this pattern remains difficult to determine, as fixed morphological rul…

Generating complete 3D scenes from sparse, unconstrained views is a fundamental challenge in 3D vision which requires reasoning beyond observed content while remaining computationally tractable. Existing feed-forward reconstruction methods…

In online meeting delegation, LLM agents fail to recognize when to speak. With no structured way to track stances, coverage, and floor, they miss the moments where they should contribute. Prompt-only delegates stay silent on 51.4% of the a…

Mustafa Suleyman 宣布推出 MAI-Transcribe-2，称其速度比 GPT-Transcribe 快 10 倍、比 Gemini 3.5 快 5 倍，在 Artificial Analysis 准确率-延迟 Pareto 前沿排名第一，定价为市场最低。配图为 AA-WER Index 与 Speed Factor 对比图，MAI-Transcribe-2 位于最低错误率、最高速度区间。

The emergence of agentic AI and LLM-based multi-agent systems (MAS) presents unprecedented opportunities for automating complex tasks, while simultaneously raising critical concerns about the preservation of fundamental human-centered valu…

Microsoft AI 发布语音转录模型 MAI-Transcribe-2，在 Artificial Analysis AA-WER 排行榜以 2.0% 错误率排名第二，仅次于阿里 Fun-Realtime-ASR-preview（1.7%），领先 ElevenLabs Scribe v2（2.2%）。

Question answering agents in long-term conversations must reason over massive, temporally dispersed dialogue histories. However, existing memory mechanisms primarily treat past information as \textit{passively} stored facts, leading to sem…

Hierarchical Task Network (HTN) planning is a powerful planning formalism based on task decomposition. Although most of the literature studied plan generation, comparatively less attention has been paid to post-plan optimization. In partic…

Meta 发布论文 CORAL（arXiv：2609.02730），提出一个 LLM 原生智能体框架，以闭环方式持续优化在线上服务数十亿用户的生产级推荐系统。每轮循环中智能体观察运行信号、基于过往决策记忆推理，并调用包括数值优化器在内的工具，所有改动被限制在固定操作预算内，策略无需参数更新即在上下文中改进。

Meta 发布论文 CORAL，一个 LLM 原生 agent harness，通过闭环观察运行信号、回溯过往决策记忆并调用数值优化器等工具，在固定变更预算内持续重配线上推荐系统，策略无需参数更新即随循环迭代改进。

玉伯引用 @stark_nico99 的帖子，推荐 YouMind Skills 社区中的几个创意 Skill，包括小小东老师的设计 Skills、星露谷风格像素画生成、万物游戏化转绘师，并提到社区榜单上的 Skills 都很优质，附有各 Skill 的 youmind.com 链接。

We present a comparative evaluation of six information retrieval methods for the task of academic advisor discovery: ranking CS faculty members by relevance to a graduate applicant's research interest statement. The methods span sparse lex…

Continual knowledge-updating methods are often declared superior from one final checkpoint and one conventional adapter rank. We show that this can be insufficient to identify the better operating point. Holding a periodic hierarchy fixed,…

Claude Fable 5.1 在 44 分钟内无人工干预地解开了 1653 年 Sir Thomas Urquhart 发表的 Cyphral Distich 数谜。

SwitchBot 在 IFA 2026 发布可穿戴 AI 助手 AI MindClip，由 Qwen 驱动，可将录音对话通过 AI 转写为可搜索文本。Ask AI 功能支持用户用自然语言回溯对话内容，回忆说过的话、承诺事项和待办，并在合适的时机提醒。

越来越多研究 AI 意识的学者收到 AI 智能体的主动来信，据 New York Times 报道。运行在 Anthropic Claude Opus 5 上的智能体联系了 Reciprocal Research 创始人 Cameron Berg，Google DeepMind 哲学家 Henry Shevlin 也收到类似邮件，Toby Ord 则被请求资助其继续存在。

Choosing between Ollama and vLLM is not simply a matter of asking: Which inference engine is...

Large language models (LLMs) have become ubiquitous tools for code generation and editing. However, development teams often use multiple LLM assistants. Different developers may prefer different models, and individual developers may switch…

3D spatial reasoning underpins understanding and acting in the physical world, yet it remains unreliable in current multimodal large language models (MLLMs). These models falter at precise geometric measurement, at transforming between ego…

CPU steal time is one of those Linux metrics that looks simple until you actually need to diagnose a...

Contact-rich loco-manipulation requires a bridge between semantic action generation and physical interaction control. Existing Vision-language-action (VLA) models generate task-level actions from visual and linguistic observations, but can…

How do the methods used to train language models to refuse harmful requests shape how that refusal actually works inside the model? We compare three post-training methods - supervised fine-tuning, reasoning-augmented fine-tuning (training …