Tweets

Today we release the transformers version 5 RC! 🤗

With this, we enable e2e interoperability with our friends in ecosystem, ease up adding new models and simplify the library 🙌🏻

Read our blog to learn morhuggingface.co/blog/transform…RgR

🚀 LangSmith Agent Builder is now available in Public Beta

Now anyone can create production ready agents without writing code, just chat.

Agent Builder guides you from initial idea to deployed agent, creating detailed prompts, selecting required tools, and even creating subagents.

Our Beta release also includes:
🧰 Bring your own tools via MCP server
🧩 Browse, copy, and customize agents across a shared workspace
🧠 Use your preferred model

Try Agent Builder frlangsmith.com/?utm_medium=so…LDt1
Read more on the launblog.langchain.com/langsmith-agen…M3l8

✨ Build Deep Agents in our LangChain Academy course ✨

Many agents today follow the same simple pattern: run in a loop, call tools. That architecture works well enough, but it breaks down as tasks get more complex.

Today, companies of all sizes – from startups to large enterprises – are building their own Deep Agents.

These agents dive deeper. They’re able to plan complex tasks and carry them out over longer time horizons.

There are four key features that set Deep Agents apart from regular agents:
1. Planning – keeps agents on track
2. File system – allows agents to offload context
3. Sub-agents – act as focused specialists
4. Prompting – provides agents with detailed instructions

Our LangChain Academy course, Deep Agents, shows you how to combine these pieces. By the end, you’ll have the skills to design, implement, and deploy your own Deep Agent.

Enroll for free ➡️ bit.ly/48owKfL

我们利用 GitHub 平台来构建 GitHub 平台。 🛠️

我们使用提供给各位的相同工具，包括 GitHub Copilot。它已不仅仅是自动补全工具。在我们的核心代码库中，它现在是一位积极的贡献者，可以：

🧑‍💻 被分配问题
🙌 打开拉取请求
⚡️ 处理繁琐的任务

想了解它的运作方式吗？请看这里。 ⬇️
github.blog/ai-and-ml/gith…

"I didn't take any notes this time. Everything's going so well."

— Jessica after a YC Board meeting

😍Really loved this framing from @eugeneyan on how to build product evals (for ai applications) in three simple steps!

We built LangSmith to make it easier to build evals - so I recorded a quick video showing how you can do each of these three steps:youtu.be/mz7mAo4zIC8b

谢谢 @LipuAIX 整理的翻译提示词，分享一些我用 AI 翻译文章的心得。虽然我认为这提示词太复杂了一些，但其中思想是没问题的，我以前也分享过类似的观点：
1. 最好的翻译就是重写
2. 好的翻译效果要分成几步来做

但也要分场景，普通翻译场景，重写一次就足够了，以现在大语言模型的能力，尤其是 Gemini 3 Pro 这样的，一次重写质量已经相当高了。

如果真要做专业翻译，第一遍重写之后，再让 AI 去校对润色是有必要的。

但是校对和润色不要放在一个提示词里面来做，除非内容很短。

主要原因就是我昨天提到的：大模型可以输入很长，但是输出太长就会偷工减料，幻觉严重。

想象一下，如果你翻译一篇 2 千字的文章，按照提示词这样三遍下来，输出就要 5-6 千字了，到后面输出质量就不高了。

所以翻译、校对、润色最好是独立做。

先说翻译，翻译的提示词不需要原推那么复杂，简单的让它“重写”即可，重点是说明：
- 文章风格
- 常用的翻译词汇对照表
- 目标读者（可选）

校对，校对需要原文、和译文，检查遗漏、错误的地方。如果对翻译精度要求不高，这一步也可以省略。

润色则不需要英文原文，只需要译文就够了，这时候只需要让模型去检查翻译后的结果语句是否通顺，是不是符合中文表达习惯。已经不需要英文原文了。

另外，文章如果太长，还是需要分块，分块最好按照自然章节和段落，通常段落一级就够了。

至于每一块和上一块怎么衔接，一个简单有效的办法就是把上一块的原文、翻译结果加到上下文中，这样翻译下一块就可以参考上一块的内容和风格。

至于保留多少块的历史，取决于模型和分块大小。通常上一块的原文和翻译就足够了，不用太多，甚至不加历史结果也没问题。

这个过程可以写个程序用 API，也可以手动去模型做。比如我个人就是用 Gemini，不同的提示词做成不同的 Gem，需要用就把内容贴到 Gem 就好。

Crowds aren’t wise - markets are wise, and mobs are dumb.

Mistral 3 is now available on Ollama v0.13.1 (currently in pre-release on GitHub). 

14B: 
ollama run ministral-3:14b

8B: 
ollama run ministral-3:8b

3B: 
ollama run ministral-3:3b

Please update to the latest Ollama.

If you don’t have Wabi access yet, you can edit this prompt yourself: 

{
"subject": {
"description": "A woman posing for a photo"
"age": "25"
"expression": "realistic"
"ethnicity": "white",
"eye color": "blue",
"hair": {
"color": "brown"
"style":  "half up half down"
},
"clothing": {
"top": {
"type": "oversized t-shirt",
"color": "charcoal gray",
"details": "white Google lettering"
},
"bottom": {
"type": leggings
"color": black
"details": n/a
}
},
"face": {
"preserve_original": true,
"makeup": light
}
},
"accessories": big headphones
},
"prop": {
"type": "none
},
"device": {
"type": macbook laptop on counter
}
},
"photography": {
"camera_style": "direct candid photograph, not a selfie",
"angle": "eye-level direct shot",
"shot_type": "waist-up composition, subject positioned on the right side of the frame",
"aspect_ratio": "9:16 vertical",
"texture": "sharp focus, natural indoor lighting, social media realism, clean details"
},
"background": {
"setting": "nice but small kitchen"
"wall_color": white
"elements":
],
"atmosphere": "casual"
"lighting": "imperfect"
}