Artificial Intelligence #231

Hey, in this issue: eliminating matrix multiplication in LLMs; lessons learned from scaling to multi-terabyte datasets; how Meta trains large language models at scale a survey of LLMs for financial applications; uncensor any LLM with abliteration; and more.

The sponsors of this issue are Twilio and NeuralMagic.

• [Ars Technica] Researchers upend AI status quo by eliminating matrix multiplication in LLMs
• llama.ttf: a tinkerer integrated an LLM into a font
• [Meta] Sharing new research, models, and datasets
• [Sponsored] What can good data do for you? - Twilio Segment Customer Data Platform

Segment helps 25,000+ companies turn customer data into tailored experiences. With customer profiles that update real-time, and best in class privacy features - Segment's Customer Data Platform allows you to make good data available to every team.

• Lessons learned from scaling to multi-terabyte datasets
• [ArXiv] A survey of LLMs for financial applications: progress, prospects and challenges
• [Meta] How Meta trains large language models at scale
• [Sponsored] Webinar - Learn why vLLM is the leading inference server and how Neural Magic helps you deploy vLLM-based model services.
• [Hugging Face] Uncensor any LLM with abliteration
• [Model] Meta released the LLM that predicts multiple tokens at the same time

More than 800,000 subscribers are reading this newsletter. If you are building an AI or a data product or service, you can become a sponsor of one of the future newsletter issues and get your business featured in the newsletter. Feel free to reach out to [email protected] for more details on sponsorships.

Enjoy the newsletter? Please help us make it bigger and better by sharing it with colleagues and friends.