Sakana AI's self-improving model gets better at solving and judging, but can't catch its own blind spots

Sakana AI introduces MASS, a self‑improving framework that lets a single language model solve tasks, judge its own outputs, and refine its workflow, boosting accuracy to ~70% over the base model after two iterations. However, the system can reinforce shared mistakes when all roles miss the same errors, limiting true progress.

Opening Kapyn…