Welcome to Rami’s Readings #154 - 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 recommendations from The AI Ethics Brief, Product Byte, The VC Corner and from the AI in Action Conference in Montréal. 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.
🤖 AI Reads
GLiNER2.5-Decide: An Open-Weight Model for Structured Decision Making
Notes: I’ve loved GLiNER’s classifiers and they now just launched an open-weight model for structured decision-making directly competing with Jev. Incredibly fast on Apple GPUs or old-school CPUs. This model was launched within days of Jev’s release.
Open Models Caught Up With Jev
Notes: Models are not a moat. I shared OpenJev (now known as SemIf) last weekend. There are even more models released now.
Contrastive-LM / CLM-v0.1-8B
Notes: Much much faster than Jev. Open-weight too.
jaredpalmer / kev: Jev-Like Family of Decision Models
Notes: From Jared Palmer. Built on top of Qwen3.5/3.8. Make your own Jev.
innylarouge / jevlike: Train a Small Model Like Jev
Notes: Another make your own Jev.
Introducing MiMo-V2.6 series
Notes: Xiaomi’s MiMo model is competitive with Fable and Opus on a number of benchmarks.
FreeToken: Efficient Edge-Native MoE Serving with Bandwidth-Adaptive Execution
Notes: I’ve already seen variations, but it’s still neat! Here’s a demo.
Memory Compression for High-Fanout Agent Sandboxes
Notes: Interesting paper for those building deep research-style agents.
Harness as a Language: A Minimalist Agent Framework With Maximal Expressivity
Notes: Should be easy to implement in Pi.dev if anyone has free time / tokens.
doximity / mvcc: Compile and Run CUDA on Apple M4-M5 GPUs
Notes: If this project is successful, it would improve model inference compatibility on Apple’s M chips.
💼 Business Reads
What China’s Vibe-Coding Capital Can Tell Us About the AI Boom
Notes: Just vibes. Good for getting a feel of what’s happening on the ground.
Who’s Ahead in the Global AI Talent Race?
Notes: Results are not surprising. AI is a story of two countries.
Financial Stress Spreads in Europe as Leverage Mounts, BCG Says
Notes: Any insights from my European subscribers?
Signing off from Redmond, WA.