Rami's Readings #56 - ✨ AI OSS ✨, DBRX, and the XZ Backdoor Insertion
The latest on AI, LLMs, AI Papers, DBRX, ✨ Open-Source AI Projects ✨, XZ Backdoor Insertion 😧, Global AI Talent, AI Jobs, and more.
Welcome to Rami’s Readings #56 - 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.
Apéro & Intellect Is Coming to Toronto!
Apéro & Intellect is a curated series of intimate gatherings discussing Artificial Intelligence designed to stimulate both your mind and your palate. After receiving great feedback from the three pilots last year, I am expanding the event series this year to gather a slightly larger group of people. Many of you desired more time among each other, the event will now include fantastic food (FREE) to match the beverages.
I am pleased to announce Apéro and Intellect #6 will take place in 🏙 Toronto, Canada 🍁 on Tuesday, April 9th at noon.
🤖 AI Reads
Introducing DBRX: A New State-of-the-Art Open LLM
Notes: A new state-of-the-art model from the Databricks team. It is decoder-only with 132B total parameters but with 36B active at any time. It is also a Mixture-of-Experts model but with many smaller expert models compared to Mistral and Grok. This is fantastic!
MyScaleDB, SQL Vector Database, Goes Open-Source
Notes: A win for open-source AI.
Neuml/txtai: All-in-one open-source embeddings database for semantic search, LLM orchestration and language model workflows
Notes: An interesting open-source project.
Dzhng/DeepSeek: LLM Powered Retrieval Engine
Notes: Another interesting open-source project.
danswer-ai/danswer: Gen-AI Chat for Teams - Think ChatGPT if it had access to your team's unique knowledge.
Notes: Hmmm…
Lightning-AI/litgpt: Hackable Implementation of State-of-the-Art Open-Source LLMs Based on NanoGPT
Notes: Another interesting open-source project. You can kinda infer where my explorations are going these days…
Supports flash attention, 4-bit and 8-bit quantization, LoRA and LLaMA-Adapter fine-tuning, pre-training. Apache 2.0-licensed.
Building AI Models Is Faster and Cheaper than You Probably Think
Notes: From YC.
Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews
Notes: 😂 at this Tweet. I guess I can no longer use the word ‘innovative’.
Lost in the Middle: How Language Models Use Long Contexts
Notes: Published November 2023.
Universal and Transferable Adversarial Attacks on Aligned Language Models
Notes: Published December 2023.
RAT: Retrieval Augmented Thoughts Elicit Context-Aware Reasoning in Long-Horizon Generation
Notes: Neat!
Huge PDF OCR Datasets on HuggingFace
🛡️CyberSecurity Reads
In the past two days, the cybersecurity community has experienced upheaval following the discovery (by a Microsoft engineer ❤️) of a backdoor inserted into the widely-used open-source library XZ Utils. 😧 The compromised code was incorporated into a few Linux distributions before discovery. It is unclear how many servers are comprised. Given how this backdoor compromises the SSH daemon on your system, the severity of the hack is huge! This hack is reminiscent of the SolarWinds hack in 2021 and further highlights the importance of software supply chain monitoring.
Original Email from Andres Freund (Microsoft)
Everything I Know About the Xz Backdoor
Great Visualization on Twitter
Red Hat CVE
💼 Business Reads
The Global AI Talent Tracker 2.0
Notes: I keep saying Chinese AI talent is under appreciated and there is incredible research being conducted in China. Happy to see data emphasizing the size of the Chinese AI talent pool. MIT ❤️ continues to be in the top 5 of Top-Tier AI Research Institutions. The data comes from NeurIPS.
Because China produces a sizable portion of the world’s top AI researchers—rising from 29% in 2019 to 47% in 2022—it is no surprise that more Chinese talent are working in domestic industry.
Will AI Replace Jobs or Can They Be Used to Enable Employees?
Notes: Good roundup of AI startups with large raises.
That is all for this week. Signing off from Redmond.