We’ve simplified how you configure the data safety settings.
Now, you can set your preferences once and have them apply to every chat interaction.
When you launch your first chat you’ll now see this dialog:
The default settings (recommended) are as shown above.
-
Mask column names and Values replaces the values with simple identifiers prior to sending to the LLM and converts them back again when the response is received, that way no identifiable data at all is sent to the LLM. This is the most private option but has the downside that the LLM will not be able to bring inferences from its wider training.
For example, if you have a “Country” column populated with country names, the LLM may be able to include characteristics of the countries relevant to your questions to enrich its responses. But, if you mask the values the LLM will just see “Column1”, “Value1” etc and obviously won’t be able to add as much value. -
Filter out values where the count is less than x applies k-anonymisation to group small population clusters together.
-
Show details on what data is sent on each prompt lets you see the data Aible sends to the LLM
-
Show token usage shows an estimate of the tokens and cost for each query.
If you check the “Don’t show this again” option the setting is saved and you won’t be prompted again during this chat session and subsequent sessions but you can review and change the options at any time via the “Settings” button.