Rami's Readings #131 - ASR Everywhere!
Rami’s Readings crossed 1,000 subscribers thanks to you! And the latest on AI, LLMs, ASR, STT, Anthropic, Yann LeCun, IPOs, and more.
Welcome to Rami’s Readings #130 - 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 recommendations from The AI Ethics Brief, Product Byte, and The VC Corner. 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
Friends! Rami’s Readings crossed 1000 subscribers and 130 editions, continuing to grow almost entirely through word of mouth. What started as a tiny Sunday side-project covering AI and tech news has turned into a community. To all of you who read, reply, share, debate, and send me links at random hours, I’ve learned so much from you! 🙏
🤖 AI Reads
Omnilingual ASR: Advancing Automatic Speech Recognition for 1,600+ Languages
Notes: Meta’s FAIR (same group that Yann LeCun used to lead) released an open source speech recognition model for 1600+ languages. GitHub repo.
ElevenLabs: Introducing Scribe v2 Realtime
Notes: Not to be outdone by Meta, ElevenLabs is claiming to have the most accurate low-latency Speech to Text model.
ason-format / ason: Aliased Serialization Object Notation
Notes: TOON was all the rage last week, and not without some pointed criticism that “it seems AI people rediscovered CSV”. Most of the criticism I’ve seen circulating misses an important nuance: the serialization frameworks that work for LLMs may not be the same as those built for general software development. And yes… reducing input tokens is a valid goal.
I’ve witnessed people blindly throwing JSON objects into prompts and expecting LLMs to do all JSON parsing. This is terrible. The alternative is to do proper context engineering, but that’s also hard. So improving the serialization format is a quicker win than many realize.
Anyways, here’s another project that aims for the same goals as TOON.
Anthropic’s Viral Report
Notes: Anthropic released a report on how they disrupted an APT that went viral, but there’s a few issues… Read this article next.
nathan-barry / tiny-diffusion
Notes: Built on nanochat-gpt. I love sharing simple, single file implementations of ML models.
yichuan-w / LEANN
Notes: Has anyone tried this open source project?
💼 Business Reads
Yann LeCun’s Been Right About AI for 40 Years. Now He Thinks Everyone Is Wrong.
Notes: I am still shocked, but not surprised by Yann’s departure considering how at odds he has been with Meta’s leadership. For those that missed it, Joel Pineau left Meta to join Cohere in Canada.
Longitudinal Expert AI Panel: Insights from Waves 1, 2, and 3
Notes: Excellent report. Must read concrete predictions of AI progress.
« C’est une hécatombe » : à l’origine de la crise du management, les formations hors-sol des futurs cadres
Notes: I know it’s a French article. Still, it is a worthy read if you’re interested in investing in French businesses or startups.
Signing off from Redmond, WA.