UniMaia: Steering Chess Policies with Language for Human-like Play

In three linesUniMaia is a framework controlling a chess policy (Lc0) via natural language prompts without full multimodal retraining. A lightweight text encoder and ControlNet-style mechanism enable gameplay modulation (opening selection, strength). UniMaia-Aux adds temporal and behavioral prediction objectives. SOTA results on prompt-conditioned benchmarks.

Summary generated by Claude — human-verified