Question on AI GuardRails

[Amit Mukherjee]

Hello CBioPortalers,
My name is Amit and I work with Fredrick National Lab for CCDI initiative. I had a bit of play around with the chat feature and noticed that it can be prompted to use abusive language and also allow persistence of PII in Chat Sessions. Wanted to see if you guys are building any guard rails around those. 
The excessive compute guardrail seems to work well so kudos around that! Much appreciated!!

Regards

Amit

[Ruslan Forostianov]

Hi Amit,

Thank you for reaching out and for your willingness to help improve the chat system. Could you please clarify what types of responses you would expect in your specific case when the user uses abusive language or provides personal information? Additionally, have you tested similar scenarios with the standard version of ChatGPT or with Claude?

Regards,

Ruslan

[Amit Mukherjee]

Hey Ruslan,

Apologies, did not get ur response earlier.I would expect something like, it is an abusive language pattern and I shall not do that. I have seen it work successfully with other bots e.g. Quickbooks, and even with AI Bedrock Guardrails in a local bot we developed internally.

Regards

Amit