Google Research has introduced a new system called Provably Private Insights (PPI) designed to analyze how users interact with on-device generative AI features. This system combines large language models (LLMs), differential privacy (DP), and trusted execution environments (TEEs) to ensure that user data remains anonymous and individual data is not inspectable. The PPI system, built on confidential federated analytics (CFA), allows developers to gain insights into AI usage patterns without compromising user privacy, and Google has open-sourced the technology for external verification. AI
IMPACT Enhances privacy for on-device AI interactions, enabling developers to gain insights without compromising user data.
RANK_REASON The item details a new system for analyzing AI usage with privacy guarantees, including open-sourcing the technology. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Google AI / Research →
- confidential federated analytics
- differential privacy
- Gboard
- Gemma
- Generative AI
- Google Parfait
- Google Research
- large language models
- Provably Private Insights
- Recorder
- Rekor
- trusted execution environments
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
