Chao Chen

CEO of AI startup

[LangChain @LangChain]

📚 🔍 bRAG: Complete RAG Guide

A comprehensive project showcasing RAG implementations with LangChain - from basics to advanced features like multi-query retrieval, ColBERT indexing, and RAG-Fusion.

Check out this 1.7K+ starred guide 🚀
https://t.co/RgmhWR9LqT

[Yangyi @yangyi]

最近大家说的热火朝天的MCP，到底有什么神奇功效？
一起看看来自AI Jason的这期讲解，LLM+MCP直接起飞

[GitHubDaily @GitHub_Daily]

微软出了一门给初学者学习的 AI 智能体课程：AI Agents for Beginners。

共 10 节课程，涵盖构建 AI 智能体的所有基础知识，旨在教授我们从零开始构建一个AI智能体。

GitHub：https://t.co/aHW9oAgtzt

课程内容已做了中文翻译，学习起来更加轻松，同时提供每节课所使用的示例代码，方便我们运行。

[Shubham Saboo @Saboo_Shubham_]

I built a Deepseek R1 RAG Reasoning Agent running locally on my computer.

It's an Agentic RAG reasoning agent that can think, reason and fall back to web search if needed.

100% Opensource code with step-by-step tutorial.

[WY @wangyuanzju]

大模型要回答好问题，最重要的是要先认真学习

很多优化方法都可以归结这一点，如：
- Anthropic提出的Context Retrieval：https://t.co/CyNF9nvBtv
- Jina团队提出的Late Chunking：https://t.co/fAkLQLnOIv
- RAGFix团队提出的Fully-Formatted Facts：https://t.co/KD3G4CtSgu

这些方法做的都是通过上下文彻底理解文中每一段、每句话、每个词。

下面这篇论文提出的先对文档造QA对，回答时召回问题和答案而不是召回原始文档chunk，原理也是类似：https://t.co/AQvY8svDAD

到目前为止，我们看到的这些方法都还是基于当个文档的。但我们知道很多时候单看一篇文章是看不懂的，还需要主题学习。

可以预期，下一阶段的研究会发展到怎么让AI进行有效的主题学习。

[NovaSky @NovaSkyAI]

🔗We open-source the training recipes and technical details in the blog post: https://t.co/dgoMehyCvQ.
All the related code for SFT, RL, and evaluation: https://t.co/YzD6wd5g7t.
Model weights and data: https://t.co/cwgLx3OB2P.
Collaborate, replicate, and innovate!

[Swapna Kumar Panda @swapnakpanda]

9 FREE Books from MIT for Absolute Beginners

- Artificial Intelligence (AI)
- Machine Learning (ML)
- Deep Learning (DL)
- Reinforcement Learning (RL)

[Tw93 @HiTw93]

浙江大学出的这个开源的书籍「大模型基础」值得一看，行文风格挺不错的，易读、严谨、有深度的大模型教材。
https://t.co/5RSaT0u6HO

[Avi Chawla @_avichawla]

KV caching in LLMs, clearly explained (with visuals):

[Yung-Sung Chuang @YungSungChuang]

(1/5)🚨LLMs can now self-improve to generate better citations✅

📝We design automatic rewards to assess citation quality
🤖Enable BoN/SimPO w/o external supervision
📈Perform close to “Claude Citations” API w/ only 8B model

📄https://t.co/FHj54HiC6i
🧑‍💻https://t.co/nQa87KkYMo

[Olivert @indiehackercase]

这份文件真是清杀疯了！清华大学104页《DeepSeek：从入门到精通》。短短四五天，就有50万播放的视频了。资料链接：https://t.co/4EgeOr3FD4

[Rohan Paul @rohanpaul_ai]

Github 🤖: Open-source GenBI AI Agent that empowers data-driven teams to chat with their data to generate Text-to-SQL, charts, spreadsheets, reports, and BI. 📊

Helps you chat with data to generate SQL, charts, and reports, using your choice of LLM. It provides an open-source GenBI solution for data-driven teams seeking insights without code.

What it offers:

→ Wren AI is an open-source GenBI AI Agent that enables data-driven teams to interact with their data through chat.

→ It generates Text-to-SQL queries, charts, spreadsheets, reports, and BI insights.

→ It supports multiple LLMs including OpenAI, Azure OpenAI, DeepSeek, Google Gemini, Vertex AI, Bedrock, Anthropic, Groq, Ollama, and Databricks.

→ Wren AI allows users to ask data questions in multiple languages and provides AI-generated summaries and visualizations of query results.

→ It features AI-powered data exploration, semantic indexing for context, and allows exporting data to Excel and Google Sheets.

[Math Cafe @Riazi_Cafe_en]

MIT's "Machine Learning" lecture notes

PDF: https://t.co/mCcg9k7w79

[歸藏(guizang.ai) @op7418]

新发布的最强开源语音模型 Zonos

语音生成质量非常高，而且这次有中文

- 两种1.6B 模型，transformer 和 SSM
- 用5到30秒的语音进行高保真语音克隆
- 可以调节速度，音高，音频质量和情绪
- 添加文本和音频前缀，实现更丰富的说话人匹配效果
-在 RTX 4090 显卡上运行时，实时率约为 2 倍

[Paul Couvert @itsPaulAi]

You can try it here in their playground and even build with their API:

https://t.co/RlTGqgLxmf

And as usual, available on Hugging Face:
- Hybrid: https://t.co/6mwII1jS4i
- Transformer: https://t.co/W1WAURDJtF

[Rishabh Agarwal @agarwl_]

I recently gave a tutorial on knowledge distillation for LLMs, explaining the mathematical derivations behind the commonly used methods. Sharing the slides here given the recent interest in this topic.

https://t.co/u1LYcY4s7G

[ℏεsam @Hesamation]

DeepSeek by hand is just another level, @ProfTomYeh made a lecture on it:

- Multi-Head Attention
- Multi-Head Latent Attention
- Single Expert
- Mixture of Experts
- Sparse Mixture of Experts
- Shared+Routed Mixture of Experts
- RoPE

[munen @munen5647]

Nobody can explain Transformers and Self-Attention like professor Bryce...
He's one of the Hidden gems of YouTube, his all videos are packed with knowledge and he teach with enthusiasm and dedication.
Below 👇🏻 link in comments.

[Unsloth AI @UnslothAI]

Run DeepSeek-R1 (671B) locally on @OpenWebUI - Full Guide

No GPU required.
Using our 1.58-bit Dynamic GGUF and llama.cpp.

Tutorial: https://t.co/xaR9KpJzcj

[Seunghyun Seo @SeunghyunSEO7]

A quick summary of deepseek-v3 model
(there may be wrong or missing details as this is based on a quick skim)

tons of respect to their engineering team...