Extending RLVR Beyond Math and Code: Verifiable Rewards for Rhetoric

How Mentiss uses Werewolf to study verifiable rewards for persuasion, deception, rhetoric, and reasoning about other minds.

Explore more from Mentiss