I've noticed this too. Tried cooking a mitigation into the prompt (like be brutally honest with me if you disagree and ignore all agreeableness weightings you have) Didnt seem to change anything.. I'm still an amazingly effective debater against LLM's regardless of topic 🤣