模型发布/更新
-
xAI releases Grok 4.6, strengthening long-running agent capabilities
xAI today released Grok 4.6, which builds on Grok 4.5 with a focus on strengthening long-running agents and more…
-
阿里开放 Qwen3.8-2.4T-A95B 模型权重:2.4T MoE、激活 95B、原生 256K 上下文
阿里 Qwen 团队正式开放 Qwen3.8-2.4T-A95B 模型权重,这是 Qwen-Max 级别模型首次开源。模型总参数 2.4T,每个 Token 激活 95B,原生支持 262,144 Token 上下文并可扩展至 1,010…
-
LTX-2.5 model debuts: AI generates 10-second 720P video in just 6.8 seconds, natively integrated with ComfyUI
LTX launched the LTX-2.5 model, natively integrated with ComfyUI, which generates a 10-second 720P video in just 6.8…
-
Microsoft debuts its first self-developed reasoning model, MAI-Thinking-1
Our first reasoning model, MAI-Thinking-1, was built from scratch and is now available in Microsoft Foundry. Kudos to…
产品发布/更新
-
OpenRouter launches real-time web search benchmark: how to choose engine, depth and model for agents
OpenRouter released a real-time leaderboard that systematically evaluates combinations of four configuration types:…
-
Claude in Chrome sidebar upgraded to Claude Cowork sessions
The sidebar of the Claude in Chrome browser extension has been upgraded to Claude Cowork sessions. Conversations are…
-
SGLang 与 Miles 为 Qwen3.8-2.4T-A95B 提供 Day-0 支持
SGLang 与 Miles 在发布首日即支持 Qwen3.8-2.4T-A95B,这是 Qwen 最大的开源模型,总参数 2.4T,每 token 激活 95B,采用混合注意力架构。
-
WhatsApp 如何用端到端加密与可验证性构建 Scam Alert 诈骗提醒功能
WhatsApp 推出可选功能 Scam Alert,通过端到端加密保护下在设备端运行机器学习模型,识别潜在诈骗消息,且消息内容不会离开设备或自动上报。该功能遵循仅设备端处理、无自动上报、用户控制三大原则,模型权重公开供独立验证,遥测数据…
-
Claude Code v2.1.229 发布:新增远程会话恢复、插件市场命令源及多项修复
Claude Code v2.1.229 发布,新增远程控制会话恢复、自托管 runner 的服务器端 hook 支持,以及插件市场命令源。修复了长响应流式输出丢失、窄终端渲染崩溃、Windows 扩展路径崩溃等问题。改进工作流扇出以复用…
行业动态
-
RingCentral 如何用 ChatGPT Work 和 Codex 构建 AI 原生工作流
RingCentral 通过全员发放 ChatGPT Work 和 Codex,推动从工程到运营的 AI 原生开发,其 AI-Native Challenge 让数千名员工(含非技术人员)交付了可运行项目。
论文研究
-
Empty shelves or lost keys? Google study: Recall is the bottleneck of parametric factuality
Google Research proposes a knowledge profiling framework and finds that frontier LLMs (such as Gemini3 and GPT-5) are…
-
Anthropic 联合独立研究者发布工人再培训项目证据综述
Anthropic 与独立研究者 David Roodman 合作发布报告,基于 56 项美国随机研究和欧洲实验证据,评估工人再培训项目应对 AI 劳动力市场冲击的效果。
技巧与观点
-
A 12-step hands-on workflow for absolute beginners to get started with AI in half a day
The article presents a 12-step hands-on workflow for absolute beginners to get started with AI in half a day: prepare…
-
DeepSeek V4 Pro and Grok 4.6 launch same day, both approaching Claude Fable 5 experience
The official DeepSeek V4 Pro and Grok 4.6 were released within 2 hours of each other, both 1.6T/1.5T parameter models,…
-
How AutoGPT uses AGENTS.md and skill gating to manage AI-generated pull requests
AutoGPT maintainers found that AI agents do not proactively read documentation, so they placed instructions in…
-
I wrote an AI textbook—how long until AI can write a better one?
After finishing an RLHF textbook, the author reflects that LLMs have stalled on long-form nonfiction writing, with…
-
OpenRouter 工具调用指南:一次编写循环,切换模型字符串即可跨模型运行
OpenRouter 发布工具调用指南,展示如何用同一套代码在 Claude、GPT 和开源权重模型间切换,仅需更改模型字符串。指南涵盖定义工具、发送请求、读取 tool_calls 响应、执行函数并返回结果的完整循环,支持 OpenAI…
-
LangChain 详解:什么是 AI 智能体?
AI 智能体是在大语言模型循环中自主运行的系统,通过反复调用模型、观察结果并调整下一步行动来完成复杂任务。工作流(workflow)则是对固定步骤的预编排,两者互补:工作流保证确定性,智能体提供灵活性。理解二者差异是构建可靠、可投入生产的…
-
OpenAI 研究:企业如何用 ChatGPT 和 Codex 落地智能体 AI
OpenAI 研究揭示企业采用智能体 AI 的方式,以及前沿企业如何在 AI 应用上拉开差距。企业正通过 ChatGPT 和 Codex 将 AI 从辅助转向执行,头部公司已率先将智能体投入实际业务流程。