CIDER Dataset: Personalized Privacy Preference Alignment
tianshi_li · x · 2026-08-26
Introduces CIDER, a dataset designed to align Large Language Models (LLMs) with personalized human privacy preferences. It contains 14,850 annotations from 169 users across 60 interpersonal communication scenarios.
Key Findings:
- With just 6 historical examples, in-context personalization improves prediction accuracy by up to 11.41 percentage points.
- Larger models like GPT-5.4 and Claude Sonnet 4.6 are better at leveraging semantic context to understand user-specific preferences.
- Smaller models tend to rely on structured heuristics based on disclosure granularity and identifiability.
More from Safety
- Presenting AI output as your own writing is plagiarism — TuhinChakr · 2026-08-26
- OpenAI bans Russia-linked accounts promoting a think tank built on copied papers — ryanmerket · 2026-08-26
- X sends C&D to Nitter: A case study in platform risk — HaktanSuren · 2026-08-26
- Paper Hypothesizes Specific Reward Hack Could Break AI Evaluations — nabla_theta · 2026-08-26
- All LLMs converge on a universal geometry of meaning, study shows — embeddings can be translated and inverted — petrusenko_max · 2026-08-26
- MolTrust MCP: Blockchain-based identity verification for AI agents — modelcontextprotocol · 2026-08-26