Privacy summary for chat models

dune

On the same way the chat assistant presents a summary of the speed and quality of the different models, perhaps something similar could be done from a privacy perspective.

I get that Kagi doesn't log any PII, but probably users might wonder how their information will be treated depending their model choice. Perhaps a summary on the help center, and a link to the privacy terms of each LLM API provider website can be a good start?

Extra points for creating something similar to what Apple did on the App Store with the privacy cards. Examples of dimensions to highlight:

Provider data retention period
Do the provider terms states use of inputs/outputs in their model training?
Etc

Vlad

dune Great idea. What would be the UI for this? (that would fit into the existing chart)

CrunchyFritos

+1 love this idea.

Value7609

FWIW none of the providers will use your prompts for training. But data retention and company name/website of the provider could be interesting.

CrunchyFritos

Value7609 Right, but specifying that explicitly is still beneficial, unless users are expected to manually check the terms of each model they use. Also, Kagi continues to add/remove LLMs as the landscape changes, so this could change and I haven't seen the Kagi team explicitly specify that they will only use LLM APIs that don't use prompts for training.

CrunchyFritos

Vlad Maybe a column called "Privacy" with 0, 1, or 2 dots.
0: User data is used to train models
1: User data is not used to train models
2: User data is not used to train models AND User data is deleted within 30 days

These explanations would then be provided in the Assistant documentation.

Alternatively, an MVP could just be a column that links to each model API's specific terms of service, or it could just be a binary column indicating with a check mark that user data is not used for model training, though I suspect data retention is also a common concern for users.

Zambyte

@Vlad I just noticed this extra column on the AI chat tab. Is there documentation anywhere on exactly what the different dots represent? Considering there are 4 dots it doesn't seem like what @CrunchyFritos posted is exactly what is used.

frin

Also, the table under Custom assistant Setup is not showing the privacy tab.