Large Language Models and Strategic Deception - arxiv.org