Tag
This paper evaluates how LLMs perform in the game Taboo under lexical constraints, showing trade-offs between compliance and communicative effectiveness, and finding that models are weaker guessers than humans.