News for hackers. Stuff that matters.
Re:Friday Fun: Challenging ChatGPT-4o to write about its own... by markhpc on Friday March 07, 2025 @04:05PM (#43294651)
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!