[CL] Inside you are many wolves: Using cognitive models to interpret value trade-offs in LLMs [Harvard University] arxiv.org