Reproduce “Retrospective Distillation Attribution via Normalized Response Similarity” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Minwoo Jang, Jaechang Kim, Minhyeon Oh, Jeongyeon Hwang, Jungseul Ok

SCOUT audits released models using current text outputs. It represents recurring part-of-speech sequences as candidate profiles, filters nondistinctive patterns, and calibrates each student–candidate match against the candidate pool. On a closed-set Dolly probe, it assigned all 12 DeepSeek-R1 descendants to R1, while other output-only methods assigned at most one. This supports detectable signals in this panel, not complete provenance recovery.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.