LLM agents build a Hearthstone AI that outperforms top players
Build, Don't Play: Self-Improving LLM Research Agents Develop a Hearthstone Battlegrounds AI That Rivals Top Players
Artificial IntelligenceMachine LearningMultiagent Systems
Summary
Building a strong AI for complex games with hidden information and randomness is challenging. The authors created AutoTavern, a system where AI research agents collaborate and improve themselves over time by playing the game Hearthstone Battlegrounds. In about three weeks, these agents developed a full game simulator and an AI player that beat the best human players on the public leaderboard. The AI also excels at making fast and accurate decisions during live play, showing it can build better strategies than traditional language models can play. The code and benchmarks will be publicly available.
What this means in practice
- •For game developers: Use automated self-improving agents to develop and test complex game AI that adapts to real player behaviors and game changes.
- •For robotics software teams: Adopt self-improving multi-agent systems with shared memory to iteratively refine simulators and policies for real-world tasks with uncertainty and hidden data.
Authors
Changlian Ma, Limin Wang
Abstract
Can LLM agents carry a hard real-world problem from raw data to a system that holds its own against people? We study this in Hearthstone Battlegrounds, an eight-player auto-battler with stochastic shops and battles, hidden information and nearly a thousand interacting cards. We present AutoTavern, a self-improving research system: specialised LLM research agents share a typed research memory, every claim passes verification gates, and live games close the loop, so the agents improve the simulator, opponents and policy and, from the same feedback, their own research process; a human sets goals and rules on ambiguous mechanics. In 20 days AutoTavern built TavernGym, the first faithful, full-content reproduction of the live game, and delivered 23 versions of TavernAgent, trained with PPO against opponents calibrated to real games. TavernAgent plays complete games on the live ladder by itself: its rating rose from 6,000 to about 15,000 and, on the public leaderboard's measure, its last 20 records average a placement of 1.60 with 80% first places, better than the server's ten highest-rated players on every statistic (2.34, 59%). TavernBench, 960 tasks in 40 scenarios, shows that the system the agents built is the strongest player: TavernAgent answers 91% of the play tasks against at most 63% for frontier LLMs, and decides in 0.03 s where they need up to 26 s; within the second a live turn allows, it keeps 90% and no LLM more than 26%. LLM agents build far better than they play. We will open-source TavernBench and TavernGym at https://github.com/MCG-NJU/AutoTavern.