Back to papers
June 12, 2026cs.CRcs.AI

When Good Verifiers Go Bad: Self-Improving VLMs Can Regress on New Tasks

Categories

cs.CR, cs.AI