Rami's Readings #65 - š NY TechWeek
Announcing upcoming talk at NY #TechWeek and the latest on AI, LLMs, FineWeb Dataset, Codestral, AI Macroeconomics, ICQ, Singaporean Intern Horror Stories, ICQ, and more.
Welcome to Ramiās Readings #65 - 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.
Upcoming Talk at NY #TechWeek
On Monday, I will be speaking at a MIT Alumni event, part of NY #TechWeek: The A to Z of AI ā The Artificial Intelligence Landscape and its impact on business and society. The event is hosted by the great Jag Gil and will feature a number of MIT-affiliated founders, operators, AI researchers, and industry leaders.
If you are in NYC and want to grab a coffee, feel free to reply to this email.
ApƩro & Intellect Is Coming Boston in June!
ApĆ©ro & Intellect is a curated series of intimate gatherings discussing Artificial Intelligence designed to stimulate both your mind and your palate. I am pleased to announce ApĆ©ro & Intellect #7 will take place in š½New York City on Tuesday, June 4th at 8am. SOLD OUT!
I am pleased to announce ApĆ©ro & Intellect #8 will take place in š Boston on Tuesday, June 6th at 5pm.
Additional events will be announced next week, with locations on the West Coast and in Canada. If you have friends in these areas, please share this newsletter with them so they don't miss out!
š¤ AI Reads
š· FineWeb: Decanting the Web for the Finest Text Data at Scale
Notes: Great article about how HuggingFace built the FineWeb dataset.
What We Learned from a Year of Building with LLMs (Part I)
Notes: This article is excellent. I strongly recommend it for AI builders especially the evaluation and monitoring section.
Mistral Released Their First Coding Model: Codestral
Notes: A 22B model that supports 80 programming languages. Weāll see if itās better than DeepSeek-Coder. The size is giving me pause about running this locally on my M1.
NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models
Notes: Nvidiaās new open source embedding model built on top of Mistral-7B, ranking first on the MTEB benchmark.
Transformers Can Do Arithmetic with the Right Embeddings
Notes: Hmmmā¦
From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step
Notes: From the Allen Institute, UW š , Harvard, and Waterloo š (Yay!)
Symbolic Chain-of-Thought
Notes: Another great CoT paper this week from NUS, UCSB and University of Auckland.
huggingface / optimum-nvidia
Notes: Speeds up HuggingFace Pipelines inference on the Nvidia platform.
š¼ Business Reads
The Simple Macroeconomics of AI
Notes: Interesting paper from MIT arguing that AIās overall impact on the economy will be less than anticipated.
Americaās Best Decade, According to Data
Notes: Fantastic data visualizations throughout.
šØ Culture Reads
ICQ, One of the Oldest Instant Messengers, Is Shutting Down
Notes: Who remembers ICQ? Or MSN Messenger?
True Singapore Intern Stories: Horror Stories from Interns Past and Present
Notes: More on Reddit.
In Shanghai, a CafĆ© Blends Life, Death ā and Coffee
Notes: Breaking taboos in China.
How a Self-Published Book Broke āAll the Rulesā and Became a Best Seller
Notes: TikTok is an incredible force for the book industry. #BookTok
Thank you to all my wonderful subscribers for your warm messages of congratulations, thoughtful ideas, and collaboration offers. I am genuinely excited to explore the fantastic projects you've proposed in June.
That is all for this week. Signing off from New York City.