I am interested in understanding inner workings of LLMs and understanding how can we improve models even more.