Aligning language models with human preferences

Tomasz Korbak

Language models (LMs) trained on vast quantities of text data can acquire sophisticated skills such as generating summaries, answering questions or generating code. However, they a…