Researchers from UC San Diego got large language models (LLMs) to play the popular fantasy tabletop roleplaying game Dungeons & Dragons, in an attempt to evaluate their performance over time.