The authors built a simple six-agent system to see if today’s AI models could plan, run, and write a research paper mostly on their own.
Modern AI models can get very good at being correct, but in the process they often lose their ability to think in many different ways.