로딩 중...

When Good Verifiers Go Bad: Self-Improving VLMs Can Regress on New Tasks | AI Paper Digest | AI Paper Digest