EasyChen retweeted
：Python 社区迎来了一个新的包管理器
：（看着十几种包管理器方案）这有什么奇怪的
：它是 pnpm
experimental: pnpm v12.4 can install Rust and Python dependencies too. So in a multi-language repository you get js packages, crates, and pypi packages downloaded concurrently.
pnpm.io/blog/releases/12.4
EasyChen retweeted
我的合伙人 @xdanger 二十多年前就是受 DHH 的 Ruby on Rails 介绍视频的影响，从 Windows 平台换到的 Mac。他今天捐了 100 万美元给到 Omacom 基金会，用实际支持来让这个世界有更多改变。
今天的 Mac OS 已经失去了当年那种独特和前卫的气质。开发个软件还要每年交 99 刀，上架 App Store 还有各种苛刻的限制和 30% 的苹果税，这一切都在 AI 时代显得格格不入。期待有更多的革新者出现，打破沉闷无聊的旧世界。
Omacom Foundation has raised another $2 million from @xdanger and @brian_armstrong for now a combined TWELVE MILLION DOLLARS from our Founding Patrons. I'm incredibly grateful for this continued flow of support. WE WILL FIX EVERYTHING! 🙏 omarchy.org/news/2026/08/oma…
EasyChen retweeted
誰でも簡単にチップチューン（ピコピコ音楽）が作れる【EASY 8BIT EDITOR】を公開しました。自動作曲機能がついているので、テイストを調整してSHUFFLEボタンを押すだけでもOK！ミックスも楽しめます。ゲームやYouTube・WebサイトのBGMに！
amix-design.com/tl/tool-s-8b…
EasyChen retweeted
Some of you guessed right. 👀
Owl Alpha on @OpenRouter — that's us.
Since going live, it has reached Top 3 globally by daily volume — and #1 on Hermes Agent, #2 on Claude Code, #3 on OpenClaw by monthly volume.
Thank you to everyone who tested and used Owl Alpha during stealth — you helped shape what's coming next.
Owl Alpha will be retiring soon. But this isn't an ending — stay tuned!
EasyChen retweeted
🚀 DeepSeek-V4 Preview is officially live & open-sourced! Welcome to the era of cost-effective 1M context length.
🔹 DeepSeek-V4-Pro: 1.6T total / 49B active params. Performance rivaling the world's top closed-source models.
🔹 DeepSeek-V4-Flash: 284B total / 13B active params. Your fast, efficient, and economical choice.
Try it now at chat.deepseek.com via Expert Mode / Instant Mode. API is updated & available today!
📄 Tech Report: huggingface.co/deepseek-ai/D…
🤗 Open Weights: huggingface.co/collections/d…
1/n
EasyChen retweeted
We ran Qwen3.6-35-A3B GGUF performance benchmarks to help you choose the best quant.
Unsloth ranks first in 21 of 22 model sizes on mean KL divergence, making them SOTA.
GGUFs: huggingface.co/unsloth/Qwen3…
EasyChen retweeted
We're delighted to announce that MiniMax M2.7 is now officially open source.
With SOTA performance in SWE-Pro (56.22%) and Terminal Bench 2 (57.0%).
You can find it on Hugging Face now. Enjoy!🤗
huggingface：huggingface.co/MiniMaxAI/Min…
Blog: minimax.io/news/minimax-m27-…
MiniMax API: platform.minimax.io/
Readers added context they thought people might want to know
The MiniMax M2.7 model weights are publicly available, but under a license prohibiting commercial use without authorization. This does not meet the Open Source Initiative's definition of open source, which requires allowing commercial use.
License: huggingface.co/MiniMaxAI/Mini…
OSI: opensource.org/osd
EasyChen retweeted
‼️Do not npm install or deploy anything right now
Supply chain attack on axios 1.14.1 - even if you don’t use axios it may be a nested dep.
Pin versions or wait until this is resolved
@npmjs @GHSecurityLab there is an active supply chain attack on [email protected] which pulls in a malicious package published today - [email protected] - someone took over a maintainer account for Axios
EasyChen retweeted
The Next.js Adapter API is now stable - the result of over a year of collaboration with Netlify, OpenNext, and other platform partners.
We are deeply committed to developers running Next.js on Cloudflare. Looking forward to building our official adapter on this new foundation.
Next.js 16.2 introduces a stable Adapter API, built with Netlify, Cloudflare, OpenNext, AWS, and Google Cloud. But the API is only part of the story.
Next.js is used by millions of developers across every major cloud, and making it work well everywhere is on us. Here are our commitments.
nextjs.org/nextjs-across-pla…
EasyChen retweeted
🔥 Introducing LongCat-Next: A Discrete Native Autoregressive Multimodal Model
LongCat-Next integrates language, vision, and audio into a unified discrete autoregressive model, extending Next-Token Prediction to native multimodality and delivering industrial-strength performance across diverse multimodal domains.
🔑 Key Features:
⚙️ 68.5B total params, 3B active, LongCat-Flash-Lite MoE backbone, excels at seeing, painting, and speaking in a unified discrete autoregressive framework.
🧩 Discrete Native Autoregression Paradigm (DiNA): We introduce DiNA, a unified paradigm that extends next-token prediction from language to native multimodality, internalizing diverse modalities into a shared discrete token space.
🌐 Discrete Native Any‑Resolution Vision Transformer (dNaViT): A unified visual tokenizer and de-tokenizer that encodes images into discrete IDs with semantic completeness, enabling both understanding and generation at any resolution. This approach overcomes the performance ceiling of discrete vision modeling in understanding tasks and enables to reconcile the conflict between understanding and generation.
👀 Visual Understanding: Fine-grained visual perception for complex tasks such as OCR, Charts, GUI interpretation, and document analysis, and advanced STEM reasoning capabilities.
🎨 Visual Generation: Generation under 28x compression ratio at arbitary resolution with competitive performance, especially in text rendering.
🎧 Speech: Strong audio comprehension capabilities, low-latency and intelligent audio-to-audio interaction, as well as speech synthesis featuring customizable voice cloning.
📄 Paper: github.com/meituan-longcat/L…
🔗 GitHub: github.com/meituan-longcat/L…
😊 HuggingFace: huggingface.co/meituan-longc…
💻 demo: longcat.chat/longcat-next
📖 blog: longcat.chat/longcat-next/in…
EasyChen retweeted
Building agent skills from scratch is a grind.
We've been there. So we built the fix — and open-sourced all of it under MIT.
PDF, Excel, PPTX, DOCX, and more.
Use it, fork it, PR it.
Intelligence with everyone.
Github Repo → github.com/MiniMax-AI/skills
Try online → agent.minimax.io
EasyChen retweeted
🚀 Qwen3.5-397B-A17B is here: The first open-weight model in the Qwen3.5 series.
🖼️Native multimodal. Trained for real-world agents.
✨Powered by hybrid linear attention + sparse MoE and large-scale RL environment scaling.
⚡8.6x–19.0x decoding throughput vs Qwen3-Max
🌍201 languages & dialects
📜Apache2.0 licensed
🔗Dive in:
GitHub: github.com/QwenLM/Qwen3.5
Chat: chat.qwen.ai
API：modelstudio.console.alibabac…
Qwen Code: github.com/QwenLM/qwen-code
Hugging Face: huggingface.co/collections/Q…
ModelScope: modelscope.cn/collections/Qw…
blog: qwen.ai/blog?id=qwen3.5
EasyChen retweeted
Andrej Karpathy：
相比起“提示工程”（prompt engineering），我更倾向使用“上下文工程”（context engineering）这个词。
为什么呢？
因为大多数人一想到“提示”，就联想到我们日常跟大语言模型（LLM）互动时，输入的那些短小的任务描述或问题。而在工业级的 LLM 应用场景中，“上下文工程”指的是一种精妙而复杂的技术：你要精准地将上下文窗口填充上恰到好处的信息，让模型能准确地迈出下一步。
这是一门科学，也是门艺术。
说它是科学，因为你要把任务描述、说明、少量样例（few-shot examples）、检索增强生成（RAG）、各种相关数据（甚至可能是多模态数据）、工具、状态、历史信息等全部巧妙地组合在一起，同时还要考虑如何压缩信息。这就像烹饪一道精致的菜肴，配料太少或搭配不对，模型无法获得足够的信息，性能会变差；配料太多或毫无关联，则会增加成本甚至降低表现。要做好这件事，需要的不仅仅是简单堆叠，更是高度专业化的技巧。
说它是艺术，则是因为操作者还要掌握一种近似“心理学”的直觉，敏锐地洞察 LLM 和人类用户心理之间的微妙互动。
除了上下文工程之外，一个工业级 LLM 应用还要：
* 将复杂问题巧妙地拆解成合理的控制流程；
* 精细地管理每次调用模型时的上下文窗口；
* 根据具体需求分派不同类型、能力等级的模型；
* 妥善处理生成与验证的交互式用户界面和用户体验；
* 以及更多其他任务——如安全防护、性能评估、并行计算、预取数据等等……
所以说，“上下文工程”其实仅仅是众多工作中的一环。围绕 LLM 应用，还在逐渐兴起一层厚重、复杂的软件基础设施，它的任务是协调、优化各种 LLM 调用和相关操作，组合出真正有效、有价值的 LLM 应用。
因此，以前常听到的“ChatGPT 套壳”（wrapper）这样的说法，现在看来已经不仅陈旧，更是非常不恰当了。
+1 for "context engineering" over "prompt engineering".
People associate prompts with short task descriptions you'd give an LLM in your day-to-day use. When in every industrial-strength LLM app, context engineering is the delicate art and science of filling the context window with just the right information for the next step. Science because doing this right involves task descriptions and explanations, few shot examples, RAG, related (possibly multimodal) data, tools, state and history, compacting... Too little or of the wrong form and the LLM doesn't have the right context for optimal performance. Too much or too irrelevant and the LLM costs might go up and performance might come down. Doing this well is highly non-trivial. And art because of the guiding intuition around LLM psychology of people spirits.
On top of context engineering itself, an LLM app has to:
- break up problems just right into control flows
- pack the context windows just right
- dispatch calls to LLMs of the right kind and capability
- handle generation-verification UIUX flows
- a lot more - guardrails, security, evals, parallelism, prefetching, ...
So context engineering is just one small piece of an emerging thick layer of non-trivial software that coordinates individual LLM calls (and a lot more) into full LLM apps. The term "ChatGPT wrapper" is tired and really, really wrong.