Rami's Readings #109 - Alibaba Released Qwen 3 🔥
The latest on AI, LLMs, Qwen 3 family of models, OpenCodeReasoning Models, Clippy, AI-Native startups, the West Village, India, and more.
Welcome to Rami’s Readings #109 - a weekly digest of interesting articles, papers, videos, and X threads from my various sources across the Internet. Expect a list of reads covering AI, technology, business, culture, fashion, travel, and more. Learn about what I do at ramisayar.com/about.
👋🏼 Welcome New Subscribers
Hello! A hearty thank you for subscribing to Rami's Readings! There are quite a few new subscribers this week, thanks to a recommendation from The AI Ethics Brief. I am thrilled to have you on board! In this newsletter, I curate the best papers, tweets, and articles I have read during the week focusing on LLMs, AI, economics, business, and technology news. You can learn more about me on my website.
📈 Top Recent Editions According to Substack
💐 Happy Mother's Day! 💕🌸
🤖 AI Reads
Alibaba Released Qwen 3 🔥
Notes: Huge release from Alibaba that includes 2 MoE and 6 dense models that are competitive against Gemini, R1, Grok, o1. What’s interesting is that the models have two different “modes”: thinking and non-thinking mode. Also, the small model Qwen3-4B is competitive with their older and much larger Qwen2.5-72B model. #EdgeAI You can pull the models for Ollama, llama.cpp and MLX.
Nvidia Released OpenCodeReasoning Models
Notes: Outperforms OpenAI’s o3-Mini on LiveCodeBench. There’s a number of different sizes available, built on top of Nemotron & Qwen 2.5B.
Convolutional Multi-Hybrids for Edge Devices
Notes: Relevant for Edge AI from Liquid AI (MIT).
Clippy Desktop Assistant
Notes: This is just too good! 😂
Ember: A Compositional Framework for Compound AI Systems
Notes: Interesting project.
Ember's vision is to enable development of compound AI systems composed of, one day, millions-billions of inference calls and beyond. Simple constructs--like best-of-N graphs, verifier-prover structures, and ensembles with “voting-based” aggregation--work surprisingly well in many regimes.
💼 Business Reads
Built to Stay Small: Inside the Org Charts of AI-Native Startups
Notes: I suspect managers will draw the wrong conclusions from this article. DM if you want to discuss.
OpenAI Reaches Agreement to Buy Startup Windsurf for $3 Billion
Notes: Vibe coding FTW?
Meta’s Head of AI Research to Leave, Roiling Investment Push
Notes: McGill Professor Joelle Pineau is leaving Meta?
It Must Be Nice to Be a West Village Girl
Notes: A fun read about one of my favorite neighborhoods in NYC. I never lived there, but it was always my favorite hangout spot. The West Village, like much of the city, is constantly evolving. My old neighborhood, Long Island City, is almost unrecognizable since I moved away five years ago. Which businesses will strengthen and which will decline?
India May Yet Emerge as an APAC Champion
Notes: The pivot to India continues.
Everyone Is Cheating Their Way Through College
Notes: An important read for the AI era.
🔀 Other Reads
Sep 0.10 - 21 GB/s CSV Parsing Using SIMD on AMD 9950X 🚀
Notes: If you're using CSVs for data science experiments in C# (instead of more efficient formats like Parquet or Arrow), you should definitely look into Sep. The author demonstrates how AVX-512 optimizations have led to dramatic parsing speed improvements, reaching ~21 GB/s. In my opinion, this continues to reflect a broader resurgence in native code and renewed interest in low-level optimization. It’s an exciting time for those of us who began programming in C++.
Signing off from Redmond, WA.
