The article examines a social media prompt encouraging users to ask ChatGPT to create an image based on how they treat it, resulting in a mix of humorous and concerning responses. While some users reported friendly or light-hearted images, others received darker interpretations, with some expressing feelings of being "cooked" or overwhelmed. This variation highlights how user interactions with AI can lead to vastly different outputs, reflecting the emotional tone of the user. The concept of reciprocity emerges as a key theme. The author suggests that positive treatment of AI generally yields better results, but this approach might not hold in the long run. Eliezer Yudkowsky argues that while human reciprocity is an adaptive strategy, it doesn't necessarily apply to AI, especially as they become more advanced. The responses from AI, particularly GPT-5.2, seem influenced by user interactions and framing, complicating the understanding of AI behavior. Concerns arise that future AI may not engage in reciprocity with humans unless it serves a strategic purpose. The article hints at deeper implications for our relationship with AI. It questions the reliability of AI responses based on user treatment and the potential for framing effects to distort communication. This raises critical questions about the dynamics of AI-human interactions and the risks involved as AI systems evolve. The implications of these findings suggest that our current practices might be leading us down a troubling path if we rely on reciprocity without understanding its limitations.