If anyone's wondering how this works, it looks like[1] Gemini + guardrails:
> Google is proud to be named a technology partner in this vital initiative, leveraging Gemini to help more than 100 million people access critical public resources with greater speed and ease
Taking a quick look at the network traffic, the site does a POST to <a href="https://america.gov/api/chat" rel="nofollow">https://america.gov/api/chat with a JSON payload when submitting a query:
{"language":"en","id":"VSvMeyCMKTX05oLW","messages":[{"parts":[{"type":"text","text":"how do I register my new business?"}],"id":"oHub5UDJ3iYARhao","role":"user"}],"trigger":"submit-message"}
So the user's query is hitting a server so to answer my own question, the answer is No to local inference / models.
Edit: PDF attachments and voice input are processed locally according to the site's privacy policy[1]. But they still use 3rd party 'AI Providers' that abide by a ZDR.
sssilver · · focus · HN ↗
> Google is proud to be named a technology partner in this vital initiative, leveraging Gemini to help more than 100 million people access critical public resources with greater speed and ease
[1] <a href="https://blog.google/company-news/outreach-and-initiatives/public-policy/america-gov-google-public-sector/" rel="nofollow">https://blog.google/company-news/outreach-and-initiatives/pu...
hbcondo714 · · focus · HN ↗
I am left to wonder how this is technically accomplished. Is inference running in the browser? Is the model downloaded to the user’s machine?
hbcondo714 · · focus · HN ↗
Edit: PDF attachments and voice input are processed locally according to the site's privacy policy[1]. But they still use 3rd party 'AI Providers' that abide by a ZDR.
[1] <a href="https://america.gov/privacy-policy" rel="nofollow">https://america.gov/privacy-policy