PAPILLON: Privacy Preservation from Internet-based and Local Language Model Ensembles

TL;DR

This paper introduces PAPILLON, a multi-stage LLM pipeline that balances user privacy with response quality by combining API-based and local models, and presents a new benchmark called PUPA for evaluating privacy-preserving LLM interactions.

Contribution

It proposes Privacy-Conscious Delegation, a novel approach for chaining API and local models, and introduces PUPA, a benchmark for privacy-preserving LLM evaluation.

Findings

Maintains high response quality for 85.5% of queries.

Restricts privacy leakage to 7.5%.

Provides a new benchmark and pipeline for privacy-preserving LLMs.

Abstract

Users can divulge sensitive information to proprietary LLM providers, raising significant privacy concerns. While open-source models, hosted locally on the user's machine, alleviate some concerns, models that users can host locally are often less capable than proprietary frontier models. Toward preserving user privacy while retaining the best quality, we propose Privacy-Conscious Delegation, a novel task for chaining API-based and local models. We utilize recent public collections of user-LLM interactions to construct a natural benchmark called PUPA, which contains personally identifiable information (PII). To study potential approaches, we devise PAPILLON, a multi-stage LLM pipeline that uses prompt optimization to address a simpler version of our task. Our best pipeline maintains high response quality for 85.5% of user queries while restricting privacy leakage to only 7.5%. We still leave a large margin to the generation quality of proprietary LLMs for future work. Our data and code is available at https://github.com/siyan-sylvia-li/PAPILLON.