Exploring dissociations between self-reported traits and actual behaviors in large language models — a study on LLM personality, alignment, and consistency.