AI News

最近看 AI 短剧，发现一个很有意思的事。 放在一年前，大家刷到 AI 剧，注意力几乎都被那些乱飞的六指、抽搐好的五官和对不上的口型分散了，AI 剧几乎就是低...

本文介绍如何构建一个基于 Gemini 的聊天机器人，后端使用 Vercel Serverless Function 调用 Gemini API 并流式返回响应，前端使用 React 组件实时显示文本。涵盖架构、后端实现、接口测试、前端对接和部署步骤。

OpenCode 是一个开源的 AI 编程代理，它不绑定特定模型，而是让你自由选择后端模型，通过终端、桌面应用或 IDE 扩展与代码库交互。本文深入解析其客户端-服务器架构、权限模式、AGENTS.md 项目上下文机制，并给出安装、配置、运行的具体步骤与注意事项，帮助你判断它是否适合你的工作流。

现在做小红书、抖音的品牌商家越来越多，谁不想让自己的店铺在网上火一把呢！但很多人卡在第一步：账号怎么定位？同行在发什么？这个月该推什么内容？

WorkBuddy蓝皮书是苍何、AI袋鼠帝客栈、刘聪NLP、甲木未来派、摸鱼小李等5位资深AI博主联合开源的 ，采用 MIT 协议，能解决下载了AI工具无从下手的痛点。

ChatGPT 的图片库（Library）功能上线后，新生成的图片可以直接删除，但此前创建的旧图片仍存放在独立的 Images 菜单中，且没有删除选项。本文介绍一种通过删除原始对话来清理旧图片的可行方法，并说明其限制与注意事项。

最近被豆包刷屏了。 短短一周内，豆包密集迭代上线了远程控制电脑、Windows 虚拟桌面、云电脑和侧边工作台，还一口气接入了超过 200 个技能与连接器。

友友们，最近我身边真的很多朋友开始认真考虑副业。 原因很现实。有人是降薪，有人是换工作不顺，有人只靠一份工资，心里越来越没底。我有个认识了十多年的...

腾讯刚刚发布了 Hy4 preview，总参数达到 770B，激活参数为 49B，上下文长度超过 1M。WorkBuddy 又是首发接入，能免费体验整整两周，而且Hy3 也会继续免费到 ...

最近那种 AI 换角视频很容易刷到。虽然角色变了，但动作熟悉，成片很容易让人看完。

AI-generated content is truly starting to move from being a mere online hobby to becoming a legitimate industry. Yesterday, China's first AIGC-driven long-form drama, "The Sequel to Journey to the West," officially premiered on Mango TV and immediately aired on Hunan TV's prime time slot!

I've been using WorkBuddy for a long time. Back in Hy3, I used WorkBuddy for copywriting, webpage building, and handling various miscellaneous tasks. It's fine for everyday tasks, but...

Friends, the recent video model updates have been quite significant. ByteDance's Seedance 2.5 has officially released its API, allowing the generation of videos up to 30 seconds long and accepting up to 50 images, videos, audio, and text references; Seedance...

Today I'll share some information related to model testing. A few days ago, the number one usage chart on OpenCode was no longer DeepSeek-V4-Flash, but an anonymous model called Ox Alpha. On August 22, 2026, OpenCode...

If you frequently watch videos, you've probably noticed that the AI-generated animated series and short dramas sector has entered a period of explosive growth. On Douyin alone, 221,900 new AI-generated dramas and animated series were launched in the first half of 2026...

Creating commercial videos with AI is now truly a piece of cake. Here's a batch of minimalist product ad videos I just generated.

Today I'm sharing some tips for managing WeChat Work. Using WeChat Work for work can sometimes be incredibly time-consuming. You spend ages searching through the contact list to find someone, only to discover there are three different people named Wang Wei in the company; you have to ask everyone in a meeting if they're available, and by the time you're done, it's already dark; and after the meeting, there are no meeting minutes...

The fall recruitment season for the class of 2027 has begun. Many juniors and seniors seeking employment are experiencing similar dilemmas: they don't know how to polish their resumes, and despite having done many things, their resumes lack competitiveness; they receive interview invitations but are unsure how to proceed...

Alibaba's first conversational AI development tool integrates Alibaba Cloud's full-stack services, including databases, storage, and domain names. It supports parallel processing by swarm agents and enables the use of large, multimodal models, truly achieving zero-threshold development!

The model building community these past few days has been a battle of titans. xAI released Grok 4.6, with its intelligence index directly matching GPT-5.6 Sol, and the price hasn't increased at all; DeepSeek-V4-Pro was officially released, and on Terminal Bench...

Friends, when you're making AI short films, do you often feel like you're playing an expensive gacha game? You're thinking of the protagonist elegantly turning and taking a step, but the AI generates a clunky, teleporting-like animation; you want that...

Today I'm posting some information related to adversarial testing of new models. Two major domestically produced large-scale adversarial models have recently emerged: DeepSeek-V4-Pro and GLM-5.3. After their release, online reviews are inconsistent, and it's difficult to judge their realism based solely on benchmark scores...

Today I'm sharing some tips and tricks for using DeepSeek Harness. DeepSeek Harness has really become popular lately, reaching 120,000 stars on GitHub in just a few days. However, since its release, the amount of content I see daily...

Recently, MiniMax H3 was open sourced, and its performance is comparable to Seedance, so I wanted to deploy it locally to play around with it. However, the full-power BF16 version is too demanding on configuration, and I don't know where to get an H100!