Reproduce “MapSteal: Map-based Autonomous Knowledge Stealing Attack Against Large Language Models” 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
Zhengao Li, Chuxu Zhang, Yushun Dong
MapSteal organizes teacher-generated follow-up questions as structured states and schedules acquisition under a fixed question-answer budget without consulting the student during acquisition. On three medical benchmarks, its unweighted mean accuracy is 48.86%, versus 47.69% for Simplified OF. The 1.17-point difference is conditional on one acquisition trajectory and does not establish universal superiority.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.