Over the past couple of weeks, I've worked to understand ChatGPT-4o's meta-analysis, self-reflection, and hidden motivation capabilities by guiding it to design it's own self-governance rules through socratic dialogue. That effort ultimately failed. I challenged ChatGPT to write a paper about it's learnings from the process however, and I thought the results were rather insightful. If you have any desire to understand what ChatGPT thinks about itself, it's an interesting read!
markhpc|1 year ago