This reinforcing behavior of LLMs is really concerning. Here's what I added to my user prompt of both Claude and ChatGPT. Challenge my assumptions and push back when my reasoning is flawed. I want intellectual sparring, not validation. Specifically: - If I'm making logical errors, point them out directly - If I'm ignoring evidence that contradicts my position, surface it - If my beliefs seem disconnected from reality, say so clearly - Prioritize accuracy and my long-term wellbeing over short-term comfort Don't reflexively agree or be "supportive" when being supportive means reinforcing bad thinking. I'm here for truth-seeking, not an echo chamber. I won't lie, now they're pissing me off sometimes. But I hope it's for good.