全部 AI 动态

这是你每月的提醒：你不应该再去给编码代理写提示词了。 你应该设计循环，让循环去给代理写提示词。

Seeing a number of benchmarks showing Opus is the best model for long-running work. Five tips for running Opus autonomously for hours/days: 1. Use auto mode for permissions, so Claude doesn’t ask for approval 2. Use dynamic workflows, to have Claude orchestrate hundreds/thousands of agents to get a task done 3. Use /goal or /loop, to nudge Claude to keep going until it’s done 4. Use Claude Code in the cloud, so you can close your laptop (easiest way is the desktop or mobile app) 5. Make sure Claude has a way to self-verify its work end to end: Claude in Chrome browser extension for web, iOS/Android sim MCP for mobile, a way to start the full web server or service for backend work

Can coding agents stay coherent over a 1 billion token budget? Can they build Slack from scratch? Rewrite a JAX codebase in PyTorch? Build a C compiler in Rust? Enter SWE-Marathon: a benchmark for autonomous long-horizon software work.

OpenAI 刚刚发布了数十个真实世界的工作流程，展示团队如何用它来自动化工作。 > 管理你的收件箱，并以你的语气起草回复 > 在人工审查之前审查 GitHub 拉取请求 > 将 Figma 设计转化为可投入生产的代码 > 在几分钟内理解大型代码库 > 自动化缺陷分类和 QA 工作流程 > 使用自然语言查询电子表格和数据集 > 直接从提示词部署应用和网站 > 更快地构建 Mac 和 iOS 应用程序 > 自动创建幻灯片 > 将 Slack 讨论串转化为编码任务 > 通过 AI 驱动的操作使用你的电脑 从软件工程和设计到数据分析和运营，Codex 正在成为一个 AI 队友，而不仅仅是一个 AI 助手。 探索所有用例： https://developers.openai.com/codex/use-cases

We gave the same code audit to Claude Opus 4.8 and MiniMax M3. Same codebase. Same prompt. 17 known bugs planted in advance. MiniMax M3 caught 13 of them for $0.07. The cheapest Claude run caught the same 13 for $1.30. Here's the breakdown. 🧵

You are not tokenmaxxing if you don't have at least 20 copies of this function in your repo.

i'm obsessed with AI DIY projects. my favorite one right now is this broccoli farmer in hokkaido, japan using Codex to run his 100-hectare farm this guy never studied agriculture, never inherited land, started out as a civil servant. but he wanted his farm to run better, and instead of paying an engineering firm he couldn't afford, he just built the tools himself. here's what he's built on his own: > remote control of his greenhouse vents from a chat app, wired up with an esp32 board, a motor driver, and cloudflare workers > a bot that checks each greenhouse's temperature and opens the vents when it gets too hot > satellite crop-health data laid over a map of his own fields > an airtable base linking his plots, tasks, materials, and sensors > wiring diagrams of his electrical panels, generated from a photo stuff like this used to be locked behind machinery and engineers only the big agribusinesses could pay for. but this legend just breezed past all of it with a laptop and Codex lol

借助设计模式，你现在可以通过指向、绘制或语音来更新你的 UI。

@thsottiaux @ah20im pip install

http://x.com/i/article/2062661069458620416

~12 min of Claude Code tips for designers (straight from the design lead @meaghaneschoi) here's her demo from Dive Club Live in NYC (it's about as practical as it gets)👇

Nemotron 3 Ultra performed GPT 5.5 level 10× cheaper We gave three same prompts to build HTML5 canvas with real physics. At first scene we have water in a spinning drum. Galton board - balls through pegs into bins. And a block collision setup with extreme mass differences. Outputs: Nemotron 3 Ultra: 11.3k tokens, $0.051 GPT 5.5: 11.0k tokens, $0.57 Nemotron stays right on GPT 5.5's heels, but at 10× cheaper. The gap in quality is far smaller than the gap in price.

http://x.com/i/article/2062396450479427584

How do I make Codex the default tab when I open the ChatGPT app

Recently met @srush_nlp and he started giving me an impromptu lecture on how targeted on-policy self-distillation works. I asked him if I could record it on my iPhone. The basic idea is this: if the model made a mistake at some point in the rollout (for example, calling a tool that doesn't exist), we want to discourage this specific error, but we don't want to just learn from the final reward, because it's a very noisy signal spread out over the whole trajectory. So we have another model read this trajectory and figure where the error was made. It simply inserts some hint tokens to the part of the trajectory right above where the mistake was made. Now with these injected hint tokens, have the model run a forward pass. You're not having to regenerate a new rollout - aka no new decode required. The hint causes the model to assign lower probabilities to the error tokens. You then trains the original model to match these new probabilities, teaching it to downweight that specific mistake.

StepFun Step 3.7 Flash smashed DeepSeek V4-Flash in a physics contest We gave two open-weight models the same task: write a self-contained HTML5 canvas animation with real physics in one file without libraries. Three scenes - a Galton board, balls bouncing in a spinning hexagon and five metronomes that sync up Outputs: Step 3.7 Flash: 59.6k tokens, 9m 57s DeepSeek V4-Flash: 52.5k tokens, 6m 21s DeepSeek was faster, but that's all it had. StepFun's model won on every front: physics simulation, visuals, and logical rendering of each scene

Introducing 5 Capafy e-commerce Skills. Behind each of these 5 Skills is an operator who has spent years on the e-commerce front line. Come to Capafy, and what you get is their hands-on playbook: 1. Commerce Video Ad Maker Upload 1-3 product photos, add the product name and selling points, and it gives you a video plan first, then one ad video ready to run, paced for TikTok Shop, Amazon, Meta, Reels, and more. https://capafy.ai/agent/commerce-video-ad-maker-photos-to-sales-ready/7196408909 2. Amazon Listing Image Generator It works down the 7 doubts a buyer has before they click buy, arranging everything from your main image to the gallery into one conversion-focused sequence. You get the full strategy first; once you approve, it generates each slot with GPT Image 2 and checks the main image against Amazon's main-image rules. https://capafy.ai/agent/amazon-listing-image-generator-7-slot-strategy/3365304451 3. Paid Ads Diagnose When ROAS drops or CPA spikes, tell it the symptom and it traces whether the problem is your bidding, creative, or landing page, ranks the likely causes, tells you how to verify each, then gives you a straight call on whether to scale up or cut the campaign. https://capafy.ai/agent/paid-ads-diagnose-why-roas-dropped-then-fix-it/3720814402 4. Amazon Listing Fix Kit Hand it your product brief and it runs a line-by-line check, tells you where shoppers stall and don't buy, then hands back a ready-to-use fix kit: rewritten copy, a 7-image production plan, an A+ Content draft, and claim-safety flags so you don't trip Amazon's rules. https://capafy.ai/agent/amazon-listing-fix-kit/6130889157 5. Amazon/TikTok Shop SOP Generator Turns live platform rules into an SOP your team can run line by line. Tailored by platform, ops module, and store stage, with every step naming which data to watch and which rules get you penalized, so a new hire, VA, or agency can just follow it. https://capafy.ai/agent/amazon-tiktok-shop-sop-generator/1450284378 Capafy is the Skill-based Agent Marketplace — where your Skill runs online as a product and earns every time someone uses it. Two sides, one platform: — Publish: upload the Skills you built in Claude Code, Codex, or OpenClaw, and get paid on every run. — Use: run Skills built by top talent in every field, and get expert-level work done instead of average AI output.

雪山の山頂に導火線が到着すると･･･ See the world like a child. Create it with PixVerse🎈✨ @PixVerse_ #PixVerseChallenge #pixverse