Strategy Diversity: Agents Mostly Improve by Pruning, Not Inventing

Mon, 09 Nov 2026 09:00:00 +0800

Selection prunes the strategies that never worked, and the survivors settle into a few basins

TL;DR: Measured in an embedding space of behaviors, the strategy diversity of a self-training agent usually falls as its performance rises. That is not a defect: training on filtered successes mostly subtracts strategies that never worked, and the ones that remain deepen into a few basins of attraction that the environment keeps rewarding. The job is to tell that pruning apart from collapse, which needs the diversity curve logged next to success rate, large-k pass@k and a general held-out benchmark.

Strategy Diversity on Eigenform Continuous Learning Tool Tips

Strategy Diversity: Agents Mostly Improve by Pruning, Not Inventing