Reproduce “Concentration bounds on response-based vector embeddings of black-box generative 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
Aranyak Acharyya, Joshua Agterberg, Youngser Park, Carey E. Priebe
The authors develop a response-based method for comparing black-box generative models and bound its embedding error. They average embedded replies to shared queries, compare mean-response profiles, and apply classical multidimensional scaling (CMDS). Under bounded variability, fixed rank, stable eigenvalues, and r growing faster than n³, Theorem 2 gives a high-probability aligned-error bound; it is asymptotic, not a practical finite-sample prescription.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.