A self-experiment pairs local and remote AI for health advice

A user says they tested diet and exercise advice based on personal health and travel data, using a local Qwen model to draft limited-data requests for remote frontier models. They say zkAPI and Tor aimed to mask payment and network identity, and the remote models improved the recommendations. They flagged Tor’s latency and possible privacy shortcomings, slow local-model output, and a tradeoff between sharing less data and getting useful advice. A reply argues larger models handle complex health research better.