Rami's Readings #114 - Mistral Released Magistral 🔥
The latest on AI, LLMs, Magistral, RL Training, Apple Foundation Models, Gaokao (高考), UK taxation, Apple's Containerization Efforts, and more.
Welcome to Rami’s Readings #114 - 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
🤖 AI Reads
Mistral Released Magistral!
Notes: It is no secret that I am a fan of Mistral 🇫🇷 to my newsletter subscribers. Mistral released their first reasoning model: Magistral. There is one flavor that is open-source (24B parameter) released under an Apache 2 license (already available on Ollama). Mistral continues to impress me with their wide selection of specialized models and workflows. Accompanying this release, their paper discussed their distributed RL training system. Highly worth a read.
LlamaRL: A Distributed Asynchronous Reinforcement Learning Framework for Efficient Large-scale LLM Training
Notes: Speaking of RL training systems, Meta GenAI published a paper detailing their distributed asynchronous RL framework.
Meta Set to Throw Billions at Startup That Leads AI Data Market
Notes: Scale AI is the hidden hero behind much of the LLM AI revolution. I am surprised that Meta is investing in this way - not the path I expected for Scale.
Text-to-LoRA: Instant Transformer Adaption
Notes: A great paper from Sakana AI. 🇯🇵
apple / foundation models
Notes: Apple opening up their LLMs to app developers on iOS and macOS provides further evidence that Edge AI is happening. See predictions in #94.
The Foundation Models framework provides access to Apple’s on-device large language model that powers Apple Intelligence to help you perform intelligent tasks specific to your use case. The text-based on-device model identifies patterns that allow for generating new text that’s appropriate for the request you make, and it can make decisions to call code you write to perform specialized tasks.
Self-Adapting Language Models
Notes: From MIT, I like this paper.
💼 Business Reads
Alibaba, Tencent Freeze AI Tools During High-Stakes China Exam
Notes: The gaokao is incredibly important in China, far more so than the SAT or ACT are in the US. Honestly, Alibaba & Tencent made the right call by freezing their AI. The only students harmed are those planning to cheat, it reduces liability for the AI providers, and it reduces doubts about high scoring students. Obviously, this freeze will not eliminate all occurrences of cheating, but it does shut down one(read about efforts to reduce cheating in 2024).
Britain Counts the Mounting Cost of Taxing Wealthy ‘Non-Doms’
Notes: Not surprising (see #81). Milan, the US, and the UAE are favored destinations.
🔀 Other Reads
apple / containerization
Notes: I am perplexed by this project. I don’t understand Apple’s objective. I don’t know of any organization running Linux containers on macOS in production. I guess the local container development experience on macOS will improve? Not a bad outcome. Docker is silly slow on macOS compared to Linux or Windows (with WSL 2).
Signing off from ☀️ sunny Redmond, WA.