Search papers, labs, and topics across Lattice.
The paper introduces MEMO, a self-play framework that optimizes LLM inference-time context for multi-turn, multi-agent games by coupling a persistent memory bank of structured insights with tournament-style prompt evolution. MEMO uses TrueSkill-based uncertainty-aware selection and prioritized replay to revisit rare and decisive states, improving both performance and robustness. Experiments across five text-based games show that MEMO significantly increases win rates for GPT-4o-mini and Qwen-2.5-7B-Instruct, while also reducing run-to-run variance.
LLM win rates in multi-agent games can nearly double (from 25% to 50%) simply by optimizing the context provided during inference.
Multi-turn, multi-agent LLM game evaluations often exhibit substantial run-to-run variance. In long-horizon interactions, small early deviations compound across turns and are amplified by multi-agent coupling. This biases win rate estimates and makes rankings unreliable across repeated tournaments. Prompt choice worsens this further by producing different effective policies. We address both instability and underperformance with MEMO (Memory-augmented MOdel context optimization), a self-play framework that optimizes inference-time context by coupling retention and exploration. Retention maintains a persistent memory bank that stores structured insights from self-play trajectories and injects them as priors during later play. Exploration runs tournament-style prompt evolution with uncertainty-aware selection via TrueSkill, and uses prioritized replay to revisit rare and decisive states. Across five text-based games, MEMO raises mean win rate from 25.1% to 49.5% for GPT-4o-mini and from 20.9% to 44.3% for Qwen-2.5-7B-Instruct, using $2,000$ self-play games per task. Run-to-run variance also drops, giving more stable rankings across prompt variations. These results suggest that multi-agent LLM game performance and robustness have substantial room for improvement through context optimization. MEMO achieves the largest gains in negotiation and imperfect-information games, while RL remains more effective in perfect-information settings.
