Rami's Readings #80 - ✨OpenAI, Mistral, & Founder Mode
The latest on AI, LLMs, OpenAI's o1, Pixtral 12B, DeepSeek-v2.5, Fei-Fei Li, SLMs, Founder Mode, China's Startup Scene, Leicas, and more.
Welcome to Rami’s Readings #80 - 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.
🤖 AI Reads
Introducing OpenAI o1-preview
Notes: o1 is impressive. I guess Chain-of-Thought is the standard now? Inference time is longer, latency matters to users. Read the Technical Report and TheVerge’s article.
Mistral AI Released Pixtral 12B
Notes: Classic Mistral AI antics! They dropped their first multimodal model in a tweet with a Magnet link (again). 😂
DeepSeek-v2.5 Released!
Notes: Enhanced writing, instruction-following, function calling, JSON output, etc. Somehow I missed this last week. Frequent readers of newsletters know I use the DeepSeek family of models quite regularly.
Fish Speech
Notes: New Text-to-Speech model with instant voice cloning.
'AI Godmother' Fei-Fei Li Raises $230 Million to Launch AI Startup
Notes: Fei-Fei Li is building an ambitious world understanding model - a model focusing on “spatial intelligence”.
Open Source LLM Tools from Chip Huyen
Notes: Continually updated list of open source AI tools on GitHub.
The French Gen AI Ecosystem
Notes: Y’all can guess that I really like Mistral. This infographic was reposted by Yann LeCun.
What is the Role of Small Models in the LLM Era: A Survey
Notes: Great survey on the role of SLMs. Bert is underappreciated.
DataGemma: Using Real-World Data to Address AI Hallucinations
Notes: Read the paper too.
General OCR Theory: Towards OCR-2.0 via a Unified End-to-end Model
Notes: From StepFun (China).
Theory, Analysis, and Best Practices for Sigmoid Self-Attention
Notes: Huge code drop from Apple speeding inference speeds by 16% on H100 GPUs.
Establishes best practices for sigmoid attention as a drop-in softmax replacement in transformers.
OneGen: Efficient One-Pass Unified Generation and Retrieval for LLMs
Notes: From the Ant Group (China). Huh! It seems too good to be true?
💼 Business Reads
Founder Mode
Notes: ‘Founder Mode’ took the Internet by storm. I have a few strong opinions on the general idea, but this newsletter is not the right place. Also, “skip-level” meetings are extremely common in my organization. I am surprised PG thinks of them as an unusual practice. Coffee anyone? ☕
Which U.S. Stocks Generated the Highest Long-Term Returns?
Notes: The tables at the end of the paper are fun to read. These are my favorite companies sorted by cumulative compound return. 😉
Scale and Scope in Early American Business History: The “Fortune 500” of 1812
Notes: Working paper related to the previous paper. Banks and more banks.
The Chinese Startup Scene Collapsed
Notes: I have my own first-hand sources (ex-Chinese VCs) that have confirmed as much. The statement on depression is accurate. Unless you are an AI startup, you are not getting funded.
China Raises Retirement Age for the First Time Since the 1950s
Notes: In Quebec and France, protests would have erupted at this change.
Intel Has Only Tough Options After Its Long and Stinging Fall From Grace
Notes: When will Intel simply be nationalized?
🎨 Culture Reads
Today’s Parents: ‘Exhausted, Burned Out and Perpetually Behind’
Notes: Everything is a competition. As a kid, I used to spend whole summers “doing nothing” e.g. playing on my own. I suspect that level of freedom is culturally unacceptable now.
‘The Diplomat’ Season 2: Teaser Trailer, Cast, Release Date, and More
Notes: I loved Rufus Sewell in Season 1.
Demand For High-End Cameras Is Soaring
Notes: I love cameras so I’m happy to see the market is still growing! I would have loved to purchase a Leica, but given how frequently I pull out my camera, a retro-styled Fujifilm made more sense.
Signing off from Redmond.