Research map / Large language models / Language, safety and interpretability
Operationalizing Data Minimization for Privacy-Preserving LLM Prompting
Jijie Zhou, Niloofar Mireshghallah, Tianshi Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| mask spans (name) via minimization | data minimization · WildChat type-wise recovery | Hit@1 %: 90.3 -> 0.0 | Improved |
| mask geolocation spans | data minimization · WildChat type-wise recovery | Hit@1 %: 89.8 -> 2.2 | Improved |
| mask occupation spans | data minimization · WildChat type-wise recovery | Hit@1 %: 85.4 -> 8.0 | Improved |
| mask affiliation spans | data minimization · WildChat type-wise recovery | Hit@1 %: 83.0 -> 1.9 | Improved |
| use abstraction instead of redaction | minimized prompts · span-wise recovery | p_corr: 5.6--14.9% vs 2.7--7.7% | Hurt |
| use larger frontier response model (gpt-5 vs qwen2.5-0.5b) on open-ended tasks | data minimization oracle · open-ended | optimal REDACT %: 19.3% -> 85.7% | Improved |
| use larger model on closed-ended tasks (gpt-4.1 vs qwen2.5-0.5b) | data minimization oracle · closed-ended | optimal REDACT %: 32.1% -> 98.0% | Improved |
| replace zero-shot o3 judge with finetuned Qwen2.5-7B comparator | privacy comparator · human-label alignment | latency (s): 6.37 -> 0.31 | Improved |
| finetune Qwen2.5-7B comparator vs o1 zero-shot | privacy comparator · consensus>=0.8 | accuracy %: 90% -> 89% | No clear effect |
| predict actions directly with LLM instead of comparator-guided search | LLM predictors · prediction vs oracle | oversharing | Hurt |
| prefer ABSTRACT over REDACT in direct prediction | LLM predictors · everyday prompts | privacy protection | Hurt |
Closest papers by meaning
- Can LLMs Keep a Secret? Testing Privacy Implications of Language Models via Contextual Integrity Theory (ICLR 2024 · similarity 0.91)
- Beyond Memorization: Violating Privacy via Inference with Large Language Models (ICLR 2024 · similarity 0.91)
- Open LLMs are Necessary for Current Private Adaptations and Outperform their Closed Alternatives (NeurIPS 2024 · similarity 0.90)
- PrivacyLens: Evaluating Privacy Norm Awareness of Language Models in Action (NeurIPS 2024 · similarity 0.89)
- PrivAuditor: Benchmarking Data Protection Vulnerabilities in LLM Adaptation Techniques (NeurIPS 2024 · similarity 0.88)
- Underestimated Privacy Risks for Minority Populations in Large Language Model Unlearning (ICML 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
