AI Safety Thursday: Monitoring LLMs for deceptive behaviour using probes — LessWrong