Rami's Readings #71 - 📢 FlashAttention-3 Released 🔥
The latest on AI, LLMs, FlashAttention-3 🔥, RankRAG, FDIC & Tech Banks, Vertical Software, Chipotle Rout, New Business Class, and more.
Welcome to Rami’s Readings #71 - 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 and a warm welcome to our community!
First off, a hearty thank you for subscribing to Rami's Readings! I am thrilled to have you on board. Every edition of this newsletter is a labor of love, tailored to provide you with enlightening reads, thought-provoking insights, and a fresh perspective on various topics.
📈 Top 3 Newsletters According to Substack:
Remember, reading isn't just about absorbing information—it's about growing, understanding, and expanding perspectives. Welcome and happy reading! 📚✨
🤖 AI Reads
FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-Precision
Notes: FlashAttention brought huge LLM inference performance improvements. Just released version 3 doubles the speed of FlashAttention-2! 🔥
We’re excited to release FlashAttention-3. It’s 1.5-2.0x faster than FlashAttention-2 with FP16, up to 740 TFLOPS, i.e., 75% utilization of H100 theoretical max FLOPS. With FP8, FlashAttention-3 reaches close to 1.2 PFLOPS, with 2.6x smaller error than baseline FP8 attention.
RankRAG: Unifying Context Ranking with Retrieval-Augmented Generation in LLMs
Notes: New fine-tuning framework from Nvidia.
Mixture of A Million Experts
Notes: Impressive solo author research paper from Google DeepMind. A promising new direction for very large LLMs.
A Survey on Mixture of Experts
Notes: Survey papers are always great to read. This paper focuses on MoE models.
💼 Business Reads
What Happens When Your Bank Isn’t Really a Bank and Your Money Disappears?
Notes: Reminds me of the Silicon Valley Bank failure. Always check for FDIC deposit insurance, and be skeptical of new banks or attractive interest rates. Here is a quote from my newsletter #11 on SVB:
Having grown up with knowledge of the currency failures and bank runs that my parents experienced, I have always maintained multiple accounts at multiple banks in two countries to benefit from insurance in the case of bank or economic collapse. It is a pain during tax season due to the FBAR filing, but I guess it is worth it in today’s risk environment. What can you do today? Open an account in a Global Systemically Important Bank and spread your cash. You don’t need more than 3 accounts.
Vertical Software in 2024
Notes: More evidence for Bill Aulet’s Disciplined Entrepreneurship.
How Much Is Sold in a SAFE Round?
Notes: For my startup founder subscribers, the data is very interesting to see.
AT&T Data Hack Prompts FCC Probe, Raises Broad Security Concerns
Notes: Cybersecurity continues to be under-invested in.
🎨 Culture & Travel Reads
Chipotle Rout on Portion Complaints Is Chance to Buy, Analysts Say
Notes: 😂 Social media caused this rout, although most analysts agree it will be temporary. In my opinion, this a a symptom of general consumer malaise with shrinkflation manifesting in retail investors selling the stock. Personally, I am a Sweetgreen fan (a holdover from my NYC-living days).
The Best New Business-Class Cabins to Know About Now
Notes: Lufthansa finally shipped their product!
Rich Americans Flex-Working on French Riviera Fuel Property Boom
Notes: Not surprising. The French Riviera has always been highly prized. 🥖
Signing off from Mr.West Cafe Bar in Seattle.