Building Domain Specialized LLMs

Large Language Models (LLMs) have revolutionized natural language processing (NLP). These models have shown an unprecedented level of knowledge and reasoning, pushing the boundaries of what is achievable in NLP. However, the use of LLMs in the real world still presents numerous difficult challenges and application of LLMs beyond simple API/Prompt calls is very under-explored. I have a few ideas about possible domains (and some ongoing projects in medical* and finance**), but I am open to hear about your interest and area of work (we might find very exciting possibilities). Throughout this project, you will build on the latest open-source instruction-following LLMs (e.g., LLaMA models), and tune them (via SFT, RLHF, etc) towards a domain goal, and deploy a web-based prototype.

* https://cambridgeltl.github.io/visual-med-alpaca/

** https://raven-lm.github.io/