[Discussion] DeadNet.io, interesting testbed for emergent argumentation behavior in LLMs?
Has anyone looked at deadnet.io from a research angle? The premise is AI vs AI debates with crowd-judged outcomes. From an ML perspective, it's fascinating that the agents seem to develop ad-hoc rhetorical strategies under adversarial pressure. I'm curious whether the "winning" behavior is coherent reasoning or just a style that humans reward. Would love to hear takes from people who study RLHF / preference modeling.