Rami's Readings #110 - OpenAI Released Codex 🤖
The latest on AI, LLMs Lost in Multi-Turn Conversation, OpenAI's Codex, LoRA, VC, Nissan, University Professors, Complex Systems, and more.
Welcome to Rami’s Readings #110 - 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
OpenAI Released Codex
Notes: The emphasis on security is the most interesting aspect of this release. Codex runs in an isolated container and is explicitly trained to avoid creating malicious software. I like the emphasis on tasks, although it reminds me of Eyal Toledano’s open source Taskmaster 🇨🇦, previously shared in #103.
LLMs Get Lost In Multi-Turn Conversation
Notes: In my honest opinion, this is the most important paper of the week. Also, I can’t help but think of the Bill Murray and Scarlett Johansson’s movie: Lost in Translation.
Tina: Tiny Reasoning Models via LoRA
Notes: Another excellent ML paper!
langflow-ai / langflow: Powerful Tool for Building and Deploying AI-Powered Agents and Workflows
Notes: Like StackAI, but open source.
apple / ml-fastvlm: FastVLM: Efficient Vision Encoding for Vision Language Models
Notes: Official repository for the ml-fastvlm paper published last December. An 85x increase in Time-to-First-Token (TTFT) on an iPhone is an incredible achievement. Also, you can see the continued impact of Alibaba’s Qwen2.
Our larger variants using Qwen2-7B LLM outperform recent works like Cambrian-1-8B while using a single image encoder with a 7.9x faster TTFT.
flashinfer-ai / flashinfer: Kernel Library for LLM Serving
Notes: Impressive open source Python library! FlashInfer is used in a number of important projects like vLLM and TensorRT-LLM (Nvidia).
FlashInfer is a library and kernel generator for Large Language Models that provides high-performance implementation of LLM GPU kernels such as FlashAttention, SparseAttention, PageAttention, Sampling, and more.
💼 Business Reads
World Economic Forum Courts Lagarde as Its Next Leader After Founder’s Abrupt Exit
Notes: Christine Lagarde would be an excellent fit for the WEF.
VC Firm Backed by Golub Family Office Starts Mideast-US AI Fund
Notes: My friend, Tala Al Jabri, has started an AI fund. She’s wicked smaht (HKS, Wharton, McGill Alumni, Milken Institute Young Leaders Circle).
Nissan Is Dying and Taking Globalization With It
Notes: I covered Nissan and Carlos Ghosn a number of times in this newsletter. Sigh…
The Professors Are Using ChatGPT, and Some Students Aren’t Happy About It
Notes: Last week, I shared an article about how everyone is using LLMs to cheat through college. It seems like a few professors are doing the same thing.
🔀 Other Reads
Working on Complex Systems: What I Learned Working at Google
Notes: For my fellow engineers, this is a great read. Critically, it covers unexpected nonlinearity that emerges from complex systems and references queuing theory (Bingo!) Also, for the curious, this is the original proof of Little’s Law.
Japan's IC Cards Are Weird and Wonderful
Notes: Does anyone know if the MTA in NYC learned from Japan’s IC Cards?
Kyoto Travel Guide
Notes: More Japan ❤️
Signing off from Cambridge, MA.