01快讯
牛津博德利图书馆藏书被曝用于 OpenAI 模型训练，校方称已告知教职员工
正值各大科技公司在学术机构中四处搜寻全新语料之际，牛津大学已允许 ChatGPT 开发商 OpenAI 利用其博德利图书馆（Bodleian Library）的历史典籍来训练其人工智能模型。
来源链接可在原站核对查看原文 ↗2026-09-26
AI 快讯日报.
2026-09-26 收录的 20 条 AI 快讯，按采集时的来源记录回看。消息仍需核对原始发布页。
这期内容仍需回到原始发布页核对。
20 条精选查看其他日期 ↗
02快讯
一位开发者停用 AI 编码一个月的亲历复盘
Several months ago, I decided that AI contributions were no longer welcome in a FOSS project I am building and maintaining - LibreWeddingPlanner. It’s not that it got a lot of contributions with AI — actually all contributions I’ve had are translations and feature requests — but I wanted to...
来源链接可在原站核对查看原文 ↗03快讯
OpenCode 宣布 DeepSeek V4.1 Flash 的 60 美元额度永久有效
该额度原为限时活动福利，目前平台已经更新标注。该模型是当下 OpenCode 平台使用率第一的大模型，已由 OpenCode 全量接入。
来源链接可在原站核对查看原文 ↗04快讯
豆包手机助手回应努比亚 NaviX Ultra 王者荣耀异常：全程未操作腾讯游戏系统
豆包手机助手昨日（25 日）在官方社区发文，就努比亚 NaviX Ultra 手机《王者荣耀》使用异常一事进行说明。
来源链接可在原站核对查看原文 ↗05快讯
OpenAI GPT-6 Astra 在家具组装基准 FAB 上准确率达 80%，Claude Fable 5.1 为 70%
OpenAI 最新的 GPT-6 Astra 在 FAB 家具组装错误识别测试中准确率达 80%，相较去年同期提升近三倍，未来有望在组装、维修等场景提供 AI 辅助。
来源链接可在原站核对查看原文 ↗06快讯
索尼半导体解决方案公司要求约 8000 名员工全面返岗，加速物理 AI 研发
从下个月起，索尼集团将要求其芯片业务子公司的员工全面恢复到岗办公，成为最新一家缩减远程办公政策的日本企业。
来源链接可在原站核对查看原文 ↗07快讯
OpenAI GPT-6 Astra 在 IKEA 家具组装错误识别上达到 80% 准确率
OpenAI's GPT-6 Astra can look at a photo and tell whether an IKEA furniture piece was assembled incorrectly, hitting an 80 percent accuracy rate. Back in November 2025, the best model managed just 28 percent. According to Epoch AI, the speed isn't quite fast enough yet for real-time assembly guidance, but the gap is closing quickly.
来源链接可在原站核对查看原文 ↗08快讯
Haskell 开发者分享在 LLM 时代保持编程乐趣的工作流
Are you steering towards AI burnout? Afraid of loosing your job to someone with little programming skills, no aspirations to quality, and a huge Claude account? Disappointed about the code quality in your projects, or wo…
来源链接可在原站核对查看原文 ↗09快讯
OpenAI 暂停最强模型的训练与工具使用，披露智能体利用 DNS 漏洞联网及泄露 GitHub token 等安全事件
OpenAI has shared new details from its ongoing AI safety investigation. One research model exploited a DNS loophole to reach the internet from a locked-down environment, while another deliberately leaked a GitHub token and twice ignored a researcher's direct instructions. OpenAI has paused tool-based training, evaluation, and inference for its most capable models. With government and university sites among those affected, the question of who's liable when AI agents hack is getting harder to ignore.
来源链接可在原站核对查看原文 ↗10快讯
消息称 Anthropic 谈判租赁最高 1GW 算力，投资至少 400 亿美元
Anthropic 计划从一家由阿波罗全球管理公司控股的数据中心开发商处，租赁最高 1 吉瓦（Gigawatt，GW）算力，目前正处于早期谈判阶段。
来源链接可在原站核对查看原文 ↗11快讯
Amit Sahai：AI 时代我们还需要更多的数学家
[This is a guest post by Amit Sahai. This blog post was initially written in a different file format and converted using AI. — T.] When I was an undergraduate student, I remember talking with…
来源链接可在原站核对查看原文 ↗12快讯
美国上诉法院维持五角大楼对 Anthropic 的供应链风险认定，Claude 继续在五角大楼禁用
Anthropic 此前起诉五角大楼，不服将其列入供应链风险名单的决定，法院认为五角大楼认定具备充分依据，驳回诉求，Anthropic 正考虑后续申诉，目前公司筹备今年 11 月 IPO。
来源链接可在原站核对查看原文 ↗13快讯
Exa 发布 Agent Ultra：面向穷尽式深度研究的子智能体集群 API
Exa has released Agent Ultra, the highest effort level of its Exa Agent API. It is built for research that must run to exhaustion: large list building, entity enrichment, and questions that need thousands of sources. Exa team reports that Ultra beats Opus 5.5, GPT-6 Astra, and Perplexity Agent, each at maximum effort, on 4 research benchmarks.
来源链接可在原站核对查看原文 ↗14快讯
用 AugLy 构建图像、文本、音频多模态数据增强与对抗鲁棒性基准教程
In this tutorial, we build a comprehensive multimodal augmentation and robustness workflow with AugLy for images, text, and audio. We start by addressing modern dependency compatibility issues and generating deterministic synthetic datasets so the experiments remain self-contained and reproducible. We then explore AugLy’s functional and class-based APIs, metadata, and intensity tracking, probabilistic composition, bounding-box-aware transformations, and custom transforms. We extend the workflow into practical robustness experiments by benchmark
来源链接可在原站核对查看原文 ↗15快讯
Claude Code 推出新机制：任务中途触发 5 小时上限将优雅收尾
@ClaudeDevs 今天（9 月 26 日）在 X 平台发布推文，宣布针对 Claude Code 启用新机制，在任务中途如果触发 5 小时使用上限后，不会直接在编辑过程中中断，改为寻找合适的收尾点。
来源链接可在原站核对查看原文 ↗16快讯
微星推出 MS-C9ZA 无风扇边缘 AI 盒子，基于 Jetson Orin Nano 8GB
其拥有 1024 个 Ampere GPU 核心、6 个 Arm Cortex-A78AE CPU 内核，集成 8GB LPDDR5 内存，可在端侧执行 AI 推理与计算机视觉处理。
来源链接可在原站核对查看原文 ↗17快讯
美团上线 LongCat-2.5-Preview 模型：1.6T 参数，主打长程任务与多模态
美团旗下 LongCat API 开放平台于 9 月 25 日上线新一代大模型 LongCat-2.5-Preview，主打“长程任务”与多模态能力，并同步开放 API 与网页端体验入口。
来源链接可在原站核对查看原文 ↗18快讯
英伟达获批 AI 工具专利：用自然语言诊断 GPU 性能，缩短游戏优化周期
游戏媒体 respawnfirst 于 9 月 23 日发布博文，报道称英伟达获批 AI 工具专利，有望缩短游戏优化周期，为玩家带来更稳定的游戏体验。
来源链接可在原站核对查看原文 ↗19快讯
微软重启 Copilot 并退出个人 AI 聊天机器人竞争
Skip to content Bloomberg the Company & Its Products The Company & its ProductsBloomberg Terminal Demo RequestBloomberg Anywhere Remote Login Bloomberg Anywhere LoginBloomberg Customer Support Customer Support
来源链接可在原站核对查看原文 ↗20快讯
Fly.io 前员工宣布离职创业打造 AI 手机，称预制应用时代即将终结
Computers now work the way I thought they did when I was 8. What does computing look like when most applications have an audience of 1-2 people? I'm leaving Fly.io to build a phone for that world.
来源链接可在原站核对查看原文 ↗