全部 AI 动态

GPT-Realtime 2 是操作系统的未来。 我已经试验了它几周，不得不说，它真是相当不可思议。 打开应用、搜索网页，甚至在 Premiere 里剪辑。全都只用我的声音完成。 而且只需几个提示就能设置好。 在这个视频里，我会向你展示具体怎么做。 0:00 开场 1:46 什么是 GPT-Realtime 2？ 4:46 设置它（无需编程） 7:53 修复常开麦克风（按键说话） 9:41 演示：搜索网页 11:38 演示：通过 MCP 连接应用（Obsidian） 14:13 演示：控制 Premiere Pro（无障碍树） 18:00 诚实的注意事项 19:00 结尾

http://x.com/i/article/2066964446086676480

http://x.com/i/article/2066964446086676480

Had early access to GPT-5.2. Its an impressive model. Here is GPT 5.2 Pro's version of "create a visually interesting shader that can run in twigl-dot-app make it like an infinite city of neo-gothic towers partially drowned in a stormy ocean with large waves," single shot.

forked clicky into a tiny Mac top-bar app that reviews my website designs, talks back, and patches the code itself. the loop: “what’s wrong with this?” → @MiniMax_AI M3 reads the screen + points at the weak parts “fix it” → it edits the actual files on disk @togethercompute served the full model stack for the demo: Parakeet for STT, MiniMax Speech 2.8 for voice, and MiniMax M3 for reasoning. Even saw a reduction in latency from 546ms to 277ms when switching STT to Together AI demo below

I built a @threejs game director skill system for Codex & Claude Code to help agents create more polished playable browser games. It guides gameplay loops, graphics, HUD/UI, debugging, QA, and optional @tripoai @ElevenLabs @NanoBanana 3D/image/audio assets with API keys. Open source, get the skills in this thread 👇

so close to 90k followers... what codex articles do i need to write to get there...

🤖 made this entire 15-second UGC ad with AI zero camera, zero studio, zero real product shoot. Here's the quick breakdown: ✅ ChatGPT in @PixVerse_ for the first images ✅ Nana Banana Pro 2 (Flow AI) for the rest ✅ Kling 3.0 in @PixVerse_ for the video animation ✅ edited on TikTok for final polish this is what brands are paying for now. drop "BLEND" if you want the full prompt breakdown 👇

Here are my 7 phases of AI-powered development. I've been thinking that the pre-PRD phase needs more structure. You need to figure out the shape of the design tree first, before then walking down it with higher-fidelity prototypes. In other words, /grill-with-docs needs to change again IMO

Open-weight MiniMax M3 filled out a US customs form from a driver's license photo For this test we deployed MiniMax M3 Q4 using MLX-VLM on a Mac Studio M3 Ultra 512GB RAM. The model was tasked with reading a scanned document and an ID card photo, then completing a declaration form Output: 736 tokens · Input: 1,847 tokens · Time: ~31s The model analyzed both inputs, streamed its reasoning, and then called three tools: write_field for text fields, mark for Yes/No checkboxes, and sign for the signature and date. It extracted the required information, mapped it to the correct fields and completed the form without any manual input

tips for codex goals sure you can use /goal but it also has a set_goal() function its almost better to prompt the model to set its own goal, it will likely write a better prompt than you

http://x.com/i/article/2066286219416469504

http://x.com/i/article/2065692454490103808

我现在基本上再也不自己写 /goal 了。 我会让 Codex 给它自己写一个，再给它生成的每个 agent 各写一个。 就像这样 👇

Bro committed grand theft diamond with ZERO exit strategy 😭💎 Created using @PixVerse_ Check the process below of how to create this -

Created with @PixVerse_ The stadium suddenly goes dark. A single beam of golden light falls from the sky onto the center of the pitch. The ground cracks with glowing energy as a legendary FIFA World Cup player rises from beneath a crystal platform. Time slows down. Millions of floating holographic flags swirl around him while the crowd's heartbeat echoes through the stadium. As he takes his first step, massive waves of light race across the field, igniting fireworks around the entire arena. The camera flies from the sky, through the roaring crowd, and locks onto the player's face. Epic cinematic atmosphere, god-like entrance, ultra-realistic visuals, dramatic lighting, volumetric effects, blockbuster sports movie quality, 4K, masterpiece @PixVerse_ #pixversecpp #fifa

pi + DeepSeek 画的，才发现这个技能不需要生图模型，是通过 LLM 将自然语言描述转为结构化 JSON → Node.js 渲染器用纯几何算法生成 SVG → 注入自包含 HTML。 https://github.com/tt-a1i/archify

信息卡制作完整方法我做成了 Skill「infocard-skills」，开源在这里了： https://github.com/shaom/infocard-skills 支持 16/9、4/3、1/1、3/4、9/16 几种常见比例信息卡，和 2.35/1、3/1 等常见比例封面的适配性针对设计。 输入需要生成的内容和比例要求（默认 4/3），AI Agent 使用 这个 Skill 生成 HTML 并截图输出 PNG。 下面是这个 Skill 生成的样例 🔽 大家可以在使用中调整成最适合你的样式风格，并沉淀更新到 Skill 里。

MiniMax M3 support added to mlx-vlm with MSA implementation! 🚀 Tested on M3 Ultra 512GB running at 24 tps with peak memory ~240GB. Now working on optimizing performance and adding ton of tests 💪 Model is here: https://huggingface.co/mlx-community/MiniMax-M3-4bit PR is here: https://github.com/Blaizzy/mlx-vlm/pull/1374

http://x.com/i/article/2065439304785039360

Claude Fable 5 scores very well on FrontierMath: Tiers 1–4 (v2), reaching 87% on Tiers 1–3 and 88% on Tier 4. This continues a streak of Anthropic models improving rapidly at math.

An ancient estate. Teenagers forced to devour the nightmares of the powerful. One defective recruit who drags the darkness back into the waking world. THE DREAM EATERS — by award-winning AI filmmaker @Shanzyin_ai . Built on PixVerse Canvas. Project file included. RT+Follow+Reply "DREAM" = 150 Creds & Workflow in DMs (72H ONLY)