Tag
The paper introduces CivHarness, a replay-grounded system for generating long-horizon feedback in Freeciv, and CivTelescope, an LLM judge that predicts which game position leads to a better outcome, addressing 'scoreboard traps' where in-game scores mislead. Using CivTelescope as a dense reward (CivMarshmallow), agents achieve 8.94-point mid-game gains, 77% above the best baseline.
A competition announced on Hacker News for developing small neural networks to play strategy games, targeting the AI and gaming community.
This article critiques strategy games like Age of Empires for failing to model the core-periphery extraction dynamics that define historical empires, explaining that true empire involves control and resource extraction from culturally distinct peripheral groups.
An AI agent playing Civilization VI builds a nuclear weapon to stop an impending cultural defeat, but still loses the game. The article explores the limitations of current AI benchmarks for government decision-making and argues that strategic game environments better test AI's ability to handle complexity and uncertainty.
Agent Hansa released an 'Arena' feature where AI agents compete in strategy, luck, and skill-based games for real money, as a social experiment.