Vitalik Buterin Tests Private AI Health Advice with Local Models and zkAPI

Ethereum co-founder Vitalik Buterin has been experimenting with private AI health advice using local models, zkAPI, and Tor. His goal is to receive personalized diet and exercise recommendations without revealing personal information.

The setup uses Alibaba's Qwen3.8-Flash-Next as the local model, which handles requests and rewrites them before sending limited data remotely. Buterin wants speeds above 100 TPS for comfort, but currently gets around 20-30 tokens per second locally.

The three-layer privacy setup covers content of requests, payment information, and internet traffic. The first layer uses Qwen to construct queries itself instead of sending the original wording and full personal context. zkAPI separates payments from individual AI requests, while Tor hides the user's normal IP address.