https://ishikura-a.github.io/posts/Optim-LLM-is-CI/
Optimizing Language Models for Human Preferences is a Causal Inference Problem - Zihao Tang