Hacker News new | ask | show | jobs
We built a customer support agent that learns (vectorize.io)
2 points by cbartholomew 16 days ago
2 comments

Interesting distinguish learning from prior interactions from authorization to take a current customer-facing action, curious how learned preferences are prevented from expanding action authority? whether send/refund/account-change actions have separate execution controls.
How do you make sure that customer PHI is not bleeding into the genrated response?