Rami's Readings #121 - 996 Lands in Silicon Valley
The latest on AI, LLMs, AI Model Architectures, Codestral, Qwen3-Coder, ‘996’ in Silicon Valley, AI Bubble, Perl & PHP Nostalgia, Caches, Japan, and more.
Welcome to Rami’s Readings #121 - 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, Global Fintech Insider, and The VC Corner. 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
🤖 AI Reads
Hierarchical Reasoning Model
Notes: From Singapore 🇸🇬. Amazing results considering the size of the model. Further evidence that all AI practitioners should be considering new models architectures depending on the task. Just a general reminder, XGBoost is amazing.
With only 27 million parameters, HRM achieves exceptional performance on complex reasoning tasks using only 1000 training samples. The model operates without pre-training or CoT data, yet achieves nearly perfect performance on challenging tasks including complex Sudoku puzzles and optimal path finding in large mazes. Furthermore, HRM outperforms much larger models with significantly longer context windows on the Abstraction and Reasoning Corpus (ARC), a key benchmark for measuring artificial general intelligence capabilities. These results underscore HRM's potential as a transformative advancement toward universal computation and general-purpose reasoning systems.
Announcing Codestral 25.08 and the Complete Mistral Coding Stack for Enterprise
Notes: Mistral is doing all the right things! It’s no surprise they’re also tackling the code search problem for enterprise.
Codestral Embed: experience high-recall, low-latency search across massive codebases with our advanced embedding model.
Qwen3-Coder: Agentic Coding in the World
Notes: An impressive release for the Qwen family of LLMs models. Qwen3-Coder is a MoE model with various sizes that performs competitively against SOTA models from OpenAI and Anthropic, while besting Kimi K2 from #118.
GLM-4.5: Reasoning, Coding, and Agentic Abililties
Notes: GLM-4.5 is a MoE reasoning model that performs competitively against the SOTA on SWE-Bench, supposedly besting Qwen3-Coder. So I guess this is the new king of the hill? Reddit seems impressed.
Ollama's New App
Notes: Ollama has a new app for their previously CLI-only experience. LM Studio is still my go-to for a powerful UI to pair with on-device AI models.
💼 Business Reads
Silicon Valley AI Startups Are Embracing China’s Controversial ‘996’ Work Schedule
Notes: It was inevitable that ‘996’ would make it to the US and it’s not just AI startups. Anyone in AI hasn’t slept since late-2022.
The Hater's Guide To The AI Bubble
Notes: Shared by a subscriber. Folks, it is important to read contrarian views. Long-time subscribers know of my opinion on the GPU trade.
What If the US Isn’t the World’s Most Innovative Country?
Notes: Another contrarian view about US innovation.
MIT’s Andrew Lo Sees AI Ready to Run Your Money in Five Years
Notes: So who’s going to build it?
🔀 Other Reads
Programmers Aren’t So Humble Anymore—Maybe Because Nobody Codes in Perl
Notes: 90s and 00s nostalgia runs strong in the tech community. I too miss being unable to read Perl and being forced to use PHP. 😂
The Many, Many, Many JavaScript Runtimes of the Last Decade
Notes: JavaScript forever. ❤️ The real Swiss Army Knife programming language. Sorry Python! P.S. I still love you too!
Caches: LRU v. random
Notes: Interesting, but highly technical read.
fffaraz / awesome-cpp
Notes: C++ is the first programming language I learned. I still keep an eye on developments with the language and ecosystem, but I admit paying more attention to Rust these days. This is a great list of excellent C++ libraries.
It’s Not Just Tokyo and Kyoto: Tourists Descend on Rural Japan
Notes: I shared many articles about Japan. I can’t wait to visit again!
Signing off from Redmond, WA.