A lightweight two-feature embedding based on Nash equilibrium entropy and optimal response sensitivity predicts how fine-tuning on one normal-form game affects performance in another. This method outperforms structural embeddings that merely memorize game identities.
HOW THIS AFFECTS YOU
●
researcherYou can use behavioral embeddings to better predict how fine-tuning will impact an agent's reasoning across different strategic tasks.