全部 AI 动态

Made with GPT image 2 + Kling AI Prompt: Ultra-realistic F1 live TV broadcast screenshot, identity preserved exactly from reference image. Young woman sitting in the VIP paddock / team garage during a Formula 1 race, shown on the official live race broadcast as the girlfriend of an F1 driver. It is the final lap, and she is listening to the team radio through a professional racing headset, watching the garage monitors nervously, leaning forward with one hand near her mouth, proud tense expression. She wears a fitted white tank top, oversized racing team jacket draped over her shoulders, large black team-radio headset with boom mic, gold jewelry, soft glam makeup. A slim paddock pass hangs naturally from her neck. Add realistic F1 broadcast graphics: “FINAL LAP” banner, lap counter showing final lap, driver timing tower on the left, small F1-style logo bug, “LIVE” indicator, lower-third identifying her as driver partner / paddock guest. No fake oversized badge, no selfie angle. Team staff, headsets, garage screens, mechanics, and race equipment blurred around her. Telephoto broadcast camera from across the garage, compression artifacts, digital noise, bright paddock lighting, natural skin texture, no smoothing, 8k quality

GPT Image 2 on @PixVerse_ Prompt 👇 Ultra high-end fashion editorial collage in a 2 rows and 4 columns grid layout (total 8 equal panels). Each panel shows a stylish young male model in a separate minimal studio setup. Each panel has a different vibrant solid or gradient background color (red, yellow, pink, green, orange, blue, purple, cyan). Every scene features modern streetwear outfits, sandals or sneakers, and casual lifestyle poses (sitting, standing, leaning, holding props like skateboard, guitar, headphones, game controller, phone). Clean commercial advertising photography style, luxury fashion brand campaign aesthetic, soft studio lighting, crisp shadows, ultra realistic skin, sharp focus, cinematic polish. Each panel includes minimal themed props matching the color environment (bench, chair, cones, lamp, hydrant, speaker, cushions, objects arranged neatly for visual balance). Perfect symmetry, evenly spaced grid, consistent framing across all 8 panels. High-end global fashion magazine advertisement look, vibrant and aesthetic composition. IMPORTANT: No text, no logos, no watermark, no typography, no brand names.

顺便说一句，这真的非常有效：在你的查询末尾让你的 LLM“将你的回答结构化为 HTML”，然后在浏览器中查看生成的文件。我也曾成功让 LLM 把输出呈现为幻灯片等形式。 更一般地说，在我看来，音频是人类偏好的 AI 输入，而视觉（图像/动画/视频）则是他们偏好的输出。我们大脑中大约三分之一是一个专用于视觉的大规模并行处理器，它是信息进入大脑的十车道超级高速公路。随着 AI 的进步，我认为我们会看到一个利用这一点的演进过程： 1) 原始文本（阅读起来困难/费力） 2) markdown（粗体、斜体、标题、表格，稍微更护眼一些）<-- 当前默认 3) HTML（仍然带有底层代码的过程式形式，但在图形、布局甚至交互性上灵活得多）<-- 尚处早期，但正在形成新的良好默认 ...4,5,6,... n) 交互式神经视频/模拟 在我看来，这种外推（尽管相关技术目前还不存在）最终会走向某种由扩散神经网络直接生成的交互式视频。关于如何将精确/过程式的“软件 1.0”产物（例如交互式模拟）与神经产物（扩散网格）编织在一起，还有许多未解问题，但总体方向类似于最近爆火的 https://x.com/zan2434/status/2046982383430496444 在输入端也有必要的、尚待完成的改进。音频、文本或视频单独都不够，例如，我觉得需要指向和比划屏幕上的东西，就像你与一个实际坐在你身旁、和你共用电脑屏幕的人会做的所有事情一样。 TLDR 人类与 AI 之间的输入/输出心智融合正在进行中，还有很多工作要做，也有重大进展有待取得，远在我们一路跳到类似 neuralink 的脑机接口之类的东西之前。就当前阶段值得探索的东西而言，小贴士：试试要求 HTML。

Ink shouldn’t be able to fly, but this one finds a way! Seedance 2.0 1080p on PixVerse | @PixVerse_ Prompt below 👇

5️⃣ عجبتني النتيجة وقلت ليه ماستخدم سي دانس ٢ ، ورحت على منصة @PixVerse_ لان احب ان السرعة عندهم عالية جدا لتوليد الفيديو بدقة ١٠٨٠ وبصراحة فيه تحكم اكبر حتى بالجوال علشان كذا طلبت من شات جي بي تي يعطيني برومبت يتناسب مع اللقطات اللي وضحتها في الرد السابق بشكل متسق ويدعم ملتي شوت واستخدمت خاصية الريفرنس فيه وارفقت الستوري بورد له والمراجع مرة ثانية علشان يستوعب كل شي بدون مشاكل وهذا كان برومبت الفيديو 👌✨🔽 Create a 15-second vertical 9:16 multi-shot cinematic luxury perfume ad for “French Tobacco” by Ibraq, based on the attached storyboard. Use the attached bottle, packaging, and logo as strict visual references. Preserve the exact bottle shape, white ribbed body, gold ribbed cap, gold neck band, front plaque label, packaging design, Arabic text, English text, logo, and proportions exactly as shown. Do not distort, rewrite, or redesign any branding or text. Style: ultra-luxury, surreal, elegant, Hollywood perfume commercial, beige, gold, off-white palette, glossy reflections, soft smoke, floating clouds, mirror shards, volumetric light, rich contrast, premium cinematic finish. Multi-shot sequence: 0.0–1.5s, hero hook, the bottle floats through golden clouds with dramatic sun rays, slow push-in. 1.5–3.0s, mirrored corridor shot, bottle and box centered in a reflective hallway, subtle forward tracking. 3.0–4.2s, macro close-up of the front label, mirror shards around it, glowing highlights. 4.2–5.4s, packaging hero shot, the box alone on a reflective pedestal in a dark elegant room. 5.4–6.6s, dreamy bottle beauty shot, bottle drifting softly through warm clouds. 6.6–8.0s, premium duo shot, bottle and box together in a warm cinematic spotlight, slight orbit. 8.0–9.4s, macro cap shot, close-up of the gold ribbed cap and neck band, branding clearly visible. 9.4–10.8s, smoke atmosphere shot, bottle and box surrounded by soft luxurious smoke and reflections. 10.8–12.6s, final bottle hero shot, solo bottle centered under a focused spotlight, glossy floor reflection. 12.6–15.0s, logo reveal, smooth transition to the IBRAQ logo on a dark silky background with elegant glow. Transitions: smooth fast flash cuts, reflective match cuts, light streak wipes. Camera: cinematic push-ins, macro inserts, slow orbit, elegant product motion. Music: luxurious French-classic perfume ad vibe, soft piano, orchestral strings, subtle rise, elegant climax, refined final reveal.

http://x.com/i/article/2053059568121282560

Excel Copilot one-shotted a tiny GPT-style language model for me inside a spreadsheet: embeddings, causal attention, weights trained with stochastic gradient descent, next-token prediction, and a live slider to watch it learn.

How to make your engineering job application stand out (from the perspective of someone looking at hundreds of resumes): 1. Your resume should be one page. If you really need more space, link to a website. You don't need 10+ bullets for each job. 2. You will immediately stand out >90% of applications if you link a personal website that has some intentionality behind it. 3. If you are going to link your X, you might want to clean up your posts? Seems obvious but... people post some wild stuff. 4. You should link your GitHub. Please avoid doing a profile README that looks like a MySpace profile with the badges and images. I'm trying to look at code and your ability to build interesting ideas. 5. You should try to customize your application to the company. If you're applying to a startup, the courses you took in college probably don't matter as much. Maybe more if you're trying to make it through the ATS screening for FAANG. 6. I'm seeing a surprising number of resumes which don't talk about AI or agents at all. Software engineering is changing and it's a pretty fair assumption that you will be expected to learn or understand coding with AI for your job. That should be reflected on your resume and projects (and I'm not just saying this because I'm at Cursor). 7. Take your LinkedIn seriously. Most devs are here hanging out on X but surprisingly still most people will send around your LinkedIn internally. 8. Find ways to show your unique strengths/tastes/interests. It's nice to see people are smart, well-rounded, and thoughtful. Maybe this is a collection of books you enjoyed and why. Or some writing you've done. Or films you liked. At the end of the day, people want to work with other people they like and respect. If nothing else, it will be a good conversation starter ("oh I love [book] as well!"). 9. Do not use AI to write your cover letter or resume text. It's incredibly obvious, especially if you are applying to an AI company. You can still use it to ideate on ideas or phrases, but write it by hand (don't fall victim to the overused in-the-distribution-AI-phrases). See: /humanizer skill. 10. No photos on resumes. Save those for whatever you link out to. 11. Quality over quantity. 3 really good, thoughtful, detailed, interesting projects versus a wall of 27 AI-slop ones. Remember that hiring managers / recruiters are getting hundreds or thousands of applications for a role. They're not going to spend 20 minutes on every single application. You need to cut the cruft and get to the point. I hope this helps you stand out!

Codex quite literally filed my reimbursements, downloaded invoices since the start of the month, updated the expenses spreadsheet and filled the actual form all by itself Used Drive & Sheets plugin for state tracking Gmail plugin for tracking invoices Chrome extension for actual form Used Drive + upload I had been putting this off ever since I joined OpenAI - it took codex ~20 minutes to do it all!

We've published our internal manual for building agent skills. Skills require a new way of thinking for developers. https://research.perplexity.ai/articles/designing-refining-and-maintaining-agent-skills-at-perplexity

Hermes Agent 现已在全球 @OpenRouter token 排行榜上排名第一。 虽然我们共同的旅程才刚刚开始，但我们想借此机会感谢我们的贡献者、支持者和用户，感谢他们为让我们走到今天所做的一切。

http://x.com/i/article/2052796100608974848

为 LLaMA.cpp 实现多 Token 预测（MTP）！ 本地运行 Gemma4 模型速度提升 1.5 倍。 我们给 LLaMA.cpp 打了补丁。将 Gemma 4 助手模型量化为 GGUF 格式。我们在 MacBook Pro M5Max 上进行了测试。Gemma 26B 配合 MTP 起草 Token 速度提升 40%。基准测试、源代码和模型 👇

小型模型上的最终对抗性测试现在正在收尾，GPU 每天到货，目前已有 1TB 内存、128 VRAM 并还在增加。将会有从 3b Qwen 到 Nous Hermes 405 B 以及介于两者之间的各种适配器。集群已上线并不断扩展，产品几乎完成。最终用户可以下载我们的应用，连接到我们的 API，不会给最终用户增加延迟。目前我们的 droplet 可以同时托管多达 50,000 名用户。如果广告测试进展顺利，应该会在本周末前上线。下载将在 https://www.proprioceptiveai.com/cygnus 提供 @grok 你有什么想法？ 关于我们的 Proprioceptive Adapters 和 Cygnus 设计的论文已发布 https://proprioceptiveai.com/publications.html https://zenodo.org/records/19223882 - Cygnus 论文 - 63 页。3 月 - 26 2026’ https://zenodo.org/records/19080172 - 459 页。3 月 - 17 2036’ https://zenodo.org/records/18471775 - “LLM”中的统一行为修改 33 页。2 月 - 3 2026’ https://zenodo.org/records/18247585 - Holonomy Transformer - 36 页。2026 年 1 月 14 日’（首批概念 - Fiber Theory、数学、开放代码）

how to use Google's NEW open source Design.md + AI Skills to make your startup look like a $100 million company in 1 hour: 1. Design.md is an open source file from Google that captures the soul of a design. Typography, colors, spacing, all in one markdown file. You attach it to your prompt and your agent builds beautiful things every time. 2. Think of it this way. The HTML is the finished dish. The design.md is the recipe. The skills are the ingredients. Put them together and everything you build looks consistent and professional. 3. Don't create a design system from scratch. Find a brand you love. Linear, Stripe, Vercel, whatever resonates. Study it. Use ChatGPT or Claude to help you extract the design language into your own design.md file. 4. Build skills on top of your design.md. A landing page skill. A mobile app skill. A motion design skill. A slide deck skill. Each one references the same design.md so everything looks like it came from the same designer. 5. The biggest mistake people make: they nail one screen and then everything else looks generic. Design.md solves this. One file keeps every page, every format, every medium consistent. 6. Use it across everything. Your landing page. Your app. Your pitch deck. Your promo videos. Same DNA. Same taste. Same system. That's what separates a startup that looks real from one that looks vibe-coded. 7. Build a second brain for design inspiration. When you see something beautiful in the real world or online, capture it. Save it. When you're building something new, reference it. Taste is developed, not downloaded. 8. It's obvious but the difference between a product people trust and a product people bounce from is how it looks and feels. Design.md gives you that edge. you can watch below https://youtu.be/oLu32YpiIJw?si=1YoYhkymvoQYJk-U shoutout to @mengto for coming on @startupideaspod and walking through his full workflow. if you want to use AI to actually build gorgeous designs, you'll want to use see this. watch

There is a lot of news about compute being the bottleneck for AI. There is less visibility into the engineering it takes to make large-scale compute actually work reliably. In my view, this is one of the most interesting computer science problems in the industry right now. It is not just about getting more GPUs. It is about making every layer of the system work: networking, scheduling, hardware health, storage, orchestration, reliability, observability, security, and the developer experience for researchers. This blog gives a rare preview into the depth of engineering happening across the stack at OpenAI, starting with MRC and supercomputer networking. We're excited to start sharing more about designing, building, and operating compute at planet scale. https://openai.com/index/mrc-supercomputer-networking/ Join us: https://openai.com/careers/software-engineer-compute-infrastructure-san-francisco/

写了一篇 OpenAI Codex App 入门指南 本来想把 Markdown 编辑为 X Article 格式分享，不过 X Article 的编辑体验还是太费时间了，明天先发公众号，再整理 X Article 格式（朋友们如果有好的 Markdown -> X Article 工具或 Skills，感谢分享！） 刚好这两天受 @yudDIDit 邀请体验 @editframe Skills，安装后用它做了这个「OpenAI Codex App 入门指南」的预览视频，先分享出来大家看看。 公众号明天发布后我再补充文章链接 🔗

The Warp team uses a lot of Skills to speed up our workflow, so we open-sourced them. > npx skills add warpdotdev/oz-skills Includes Skills for SEO and accessibility auditing, documentation writing, DevOps tasks like terraform configuration, and more.

https://developers.openai.com/codex/migrate