New CIDER dataset aids LLMs in aligning with user privacy preferences

Researchers have introduced CIDER, a new dataset designed to help large language models better align with individual privacy preferences. The dataset contains over 14,000 human annotations across various communication scenarios, detailing users' willingness to share personal information. Experiments show that models like GPT-5.4 and Claude Sonnet 4.6 can improve prediction accuracy by up to 11.41 percentage points when personalized with contextual information, though this can sometimes lead to imbalanced error rates. AI

IMPACT This dataset could lead to LLMs that better respect user privacy in real-world interactions.

RANK_REASON The cluster describes a new dataset and research paper published on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX
• CIDER
• Claude Sonnet 4.6
• DagsHub
• Gotit.pub
• GPT-5.4
• Hugging Face
• ScienceCast

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →