This repository contains the research code for a project investigating how altering different aspects of an LLM prompt changes the LLM's preference for power-seeking plans. This project was carried out as part of the SERI MATS Summer 2023 Training Phase. Read the full project writeup here.