The new Kimi K2 1T model (4-bit quant) runs on 2 512GB M3 Ultras with mlx-lm and mx.distributed. 1 trillion params, at a speed that's actually quite usable:

We are creating a multi-agent AI software company @xAI, where @Grok spawns hundreds of specialized coding and image/video generation/understanding agents all working together and then emulates humans interacting with the software in virtual machines until the result is excellent.

Sam Altman

Every robot you see is a data firehose generating terabytes of chaos. This hidden crisis is the #1 reason robots fail, and it's costing the industry billions. You see hardware, but not the data swamp drowning engineers. In 2025, a quiet revolution is fixing it. Here’s how. 🧵

🫣 how innovative

Pretty cool. I think 2025-2026 will be a stronger focus on these in open source tooling. I.e. having LLMs delegate knowledge-based queries to search, which in turn frees up model capacity to improve reasoning capabilities and tool use.

Marc Andreessen’s career advice for the AI era:

Grok Imagine prompt: A funny and sassy KFC advertisement video featuring Elon Musk as a delightfully quirky showman, dressed in a dazzling bright yellow chicken costume with a large, exaggerated beaked headpiece and a vibrant red comb, wearing a clean white t-shirt, mode...

1/ Qwen cli is really under valued compare to gemini cli. Similar code base but model is much better solving hard issue. Gemini kept writing unconventional strange case. If u use vscode or insider, chances are current GitHub chat is not gonna fix complex project so as gemini chat

2/ haven't test their mcp skills. But github has really bad integration. Agent mode for Gemini is still Preview feature.

3/grok is better explaining number thoery than gpt5 although it inference twice and think long.

save this photo and ask grok to explain this problem in details. Even the answer is given. grok make it bit complicated. It derails itself early on. So human still need to revert back to previous steps. However, gpt5 free not only rate limit on image but also hallucinated more

土狗栈 汪汪汪🐶

🧸Turn your favorite memes into real-world collectibles — one click with Qwen-Image-Edit. 🎁Try it now: chat.qwen.ai/?inputFeature=… ✨Prompt: Create a 1/7 scale commercialized figure of the character in this image, in a realistic style and real-life setting. The figure is placed on a

罗素厉害之处便是几十年前便认识到过度竞争的结果是 - 不能繁衍.....

this is one of the best blog posts of 2025 by the openai researcher Shunyu Yao. "we're at AI's halftime," it's a playbook of what will matter the most in AI research and the startup ecosystem, and how to prepare best for it. for decades, AI research focused on algorithms and

just made an ad film for @mkbhd's chevron sweater experimenting with nano banana + kling 2.1 + seedance pro @mkbhd hope this is cool with you 🙏🏼 happy to take it down if otherwise

🍌Banana 应用实践： 1️⃣教学展示 提示词：绘制一幅展示汽车发动机工作原理的3D立体剖面设计图，以高度还原的方式精细呈现其内部结构。每个零部件被拆解并有序排列，各部分均配有清晰英文标注，注明结构名称与功能说明，整体布局兼具专业性与视觉逻辑性，呈现出清晰、整洁且极具科技感的解析示意图。

可能 Apple 的 AI 有救了? 一定要关注下刚发布的 Apple-FastVLM-0.5B. 我想简单测试下，于是把摄像头怼着手机刷了会抖音，来看效果，识别率惊人。要知道这可 是0.5B 的 ONNX 模型, 跑在浏览器里. 苹果还发布了 1.5B 和 7B 版本 更多详细测试稍后为大家带来 模型地址: huggingface.co/collections/ap…