Rami's Readings #104 - Llama 4 ✨
The latest on AI, LLMs, Llama 4, Thoughts of LLMs, Loyalty Points, and more.
Welcome to Rami’s Readings #104 - 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
Apologies for the late night delivery, I was traveling back to Redmond this weekend after a very eventful week. This newsletter will be a tad shorter than usual. It is likely the next two newsletters will be short too.
🤖 AI Reads
Just a reminder, I do not share business news related to my employer (Microsoft AI) or my team’s products in this newsletter. See my disclosure statement in #7. However, I do encourage you to follow me on LinkedIn to stay updated on my most recent projects. 😉
Llama 4: Massive Model Release ✨
Notes: The Llama 4 family was released on Saturday. These models are massive. All three are MoE models with the two smallest activating 17B parameters. The smallest model (Scout) is 109B parameters (total - not activated), and supports a context window of 10M tokens using a technique they are calling iRoPE. In theory, you should be able to run Scout on an overpowered local machine… maybe... The models are available on HuggingFace. Since they were released yesterday, I haven’t had time to deep dive just yet.
Tracing the Thoughts of a Large Language Model
Notes: Interesting read from Anthropic.
Lightpanda: New Headless Browser
Notes: A new startup from France is building a headless browser from scratch with support for JavaScript execution.
PaperBench: Evaluating AI's Ability to Replicate AI Research
Notes: New benchmark from OpenAI for AI Agents.
🔀 Other Reads
Loyalty Points Aren’t Worth What They Used to Be
Notes: Points devaluations are extremely frustrating to consumers and they are happening more frequently. I have started spending my points as soon as enough collect. I suggest you do the same or opt-out of the game.
Stop Syncing Everything
Notes: Interesting open source transactional storage engine optimized for lazy, partial, and strongly consistent replication. Think of applications where distributed clients typically go offline. Built in Rust!
Signing off from Redmond.