We hosted Caleb Ziems, Dora Zhao, and Diyi Yang for a live discussion of their new paper, Reflections and New Directions for Human-Centered Large Language Models.
Growing out of Stanford’s CS329X Human-Centered LLM class, the paper distills 60+ pages of insights — spanning design, data sourcing, training, evaluation, and deployment — into a roadmap for building LLMs that humans can meaningfully live and work with.