LLM Model Privacy Rating

brrrendan

I would like to see Kagi come up with a "Privacy Grade/Score" for each LLM option, based on the information shown on the LLMs & Privacy page.

On the model selection dropdown, users could then see this score and help them choose where they might want to send their query based that.

RoxyRoxyRoxy

What sort of grading system are you envisioning for this? What criteria should be taken into account? Privacy policies can be pretty big pieces of info so it might be a little hard to break them down cleanly into a grade that will satisfy all users.

brrrendan

Something as simple as as this:

A: No training, no storage
B: No training, limited storage
C: Training (Im aware than none of the current options do this today), storage

Just a quick way for users to know how their prompts might be used, without having to view the full privacy page. Maybe this score could link to Kagi’s privacy page though.

Thibaultmol

Nextcloud has done some work around this actually: https://docs.nextcloud.com/server/28/admin_manual/ai/index.html

brrrendan

Thibaultmol This is a great example. Thanks for sharing.