Our work on making LLMs more logically-consistent is accepted at ICLR 2025.