Rami's Readings #50 - 📢 AI Announcements

The latest on AI, LLMs, Sora, AI Startups, Ollama, Nvidia, Edge AI, Sand Trafficking, Rare-Earth Resources, Dutch Water Engineering, Air Canada, and more.

Welcome to Rami’s Readings #50 - a weekly digest of interesting articles, papers, videos, and X (formerly known as Twitter) 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.

---

---

🤖 AI Reads

OpenAI Sora Announced

Notes: 🤯 Technical article.

Gemini 1.5 Pro Announced

Notes: Adopted Mixture of Experts like Mixtral. A massive context window is less useful than most people realize. There are diminishing returns to increasing the context window, but neat regardless.

ChatGPT Is Getting ‘Memory’ to Remember Who You Are and What You Like

Notes: I am intrigued by how they'll manage memory and the ensuing personalization that will follow.

Guardrails AI raised $7.5M in seed funding!

Notes: Extremely impressive hub for guardrail components for an LLM-powered app.

Nvidia’s Chat with RTX Is a Promising AI Chatbot That Runs Locally on Your PC

Notes: Even Nvidia is “releasing apps” designed to run AI locally. I gave it a try and the UX is very basic. It comes bundled with a couple of LLMs, so the download size is insane (>30GB IIRC). It is a tech demo mostly. See the GitHub project. #EdgeAI

Ollama is Now Available Natively on Windows

Notes: Ollama 🇨🇦 remains the simplest solution I've found for non-technical individuals to run a large language model on their local machines. You no longer need a Docker container to make it run on Windows with this release. Once step closer for #EdgeAI.

Introducing Aya by Cohere for AI

Notes: New open-source LLM from the team at Cohere.

Eagle 7B: Soaring past Transformers with 1 Trillion Tokens Across 100+ Languages (RWKV-v5)

Notes: Alternative LLM architectures are emerging. The point in #49: it is still too early in the technology curve for custom silicon that will not be outdated within a year. How about FPGAs? Are they even cost-effective at the size you need to fit some of these models compared to renting a GPU cluster? Also, my memories of programming Verilog and VHDL still haunt me from college. I doubt the developer experience is much better today.

Keyframer: Empowering Animation Design using Large Language Models

Notes: Neat. Apple’s Latest Prototype AI Tool Can Animate Images Using Text Descriptions

From Nabeel Qureshi

Moore's Law For Intelligence

Generative Representational Instruction Tuning

Notes: Interesting.

F*** You, Show Me The Prompt

Notes: Much of the magic in many LLM frameworks and libraries is just good prompt engineering.

💼 Business Reads

Inside the Crime Rings Trafficking Sand

Notes: Not covered in this fascinating article is “beach-grade” sand theft from existing resorts in the middle of the night to place the sand at a different resort. Sand quality in construction material contributed to the damage in the recent Turkey and Syria earthquakes:

Such shortcuts worsened the damage from the disastrous February 2023 earthquake that shook Turkey and Syria, says Mette Bendixen, a physical geographer at McGill University who has investigated the effects of sand mining since 2017.

Wyoming Hits the Rare-Earth Mother Lode

Notes: Great news for everyone.

This River Is a Model - Dutch Water Engineering

Notes: Depoldering is a new word I learned today.

Air Canada Must Honor Refund Policy Invented by Airline’s Chatbot

Notes: 😂

Better Call GPT, Comparing Large Language Models Against Lawyers

Notes: I am interested in hearing the opinions of my lawyer subscribers.

🔀 Other Reads

Good advice even if you are smart or talented.

How to do things if you're not that smart and don't have any talent

That is all for this week. Signing off from Redmond, WA.