Reproduce “AdaptPrint: Response-Adaptive Fingerprinting of Black-Box LLM Services” 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
Yilin Li, Yifei Zhang, Guozhu Meng
AdaptPrint identifies the language model behind a black-box service by using the service’s answers to generate related Direct, Continuation, and Follow-up probes, then comparing responses with candidate models. Across 18 system prompts and 27 candidates, it reaches 80.6% Top-1 accuracy. In a matched 12-target comparison, it outperforms ZeroPrint, but depends on probe quality and model stability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.