Stake Chess: LLM-Guided RL Training for Human Tendency Exploitation
WeaveHacks 3: Self-Improving Agents Hackathon with Weights & BiasesWe introduce Stake Chess, a novel chess variant that adds hidden information through a biased staking mechanism, reshaping strategic play under uncertainty. The game is paired with an LLM-guided reinforcement learning architecture designed to model human psychological behavior. Using an 8-dimensional player profiling system powered by Qwen-3, our method accelerates convergence toward Nash-equilibrium strategies while outperforming human opponents. The approach show promise compared to traditional self-play methods and is grounded in game-theoretic analysis.
