What's your opinion on canned responses?

Scenario: User enters a gibberish input that will be used in the prompt to make the API call to gpt-3.
For example, the User enters “test” or “sshdhdh” or something like that.

However, we already know GPT-3 output is going to be very unpredictable with this kind of prompt.

Should we just check for this kind of input and instead of calling the API, show some canned responses to the user?

That’s an interesting question. There’s no formal guideline on this, and it’s totally up to the business. That said, I think I’d let users type in any safe input - after all, they should be rate limited anyway, so it’s up to them how they want to use the money they spend on the application. Just my 2 cents.

3 Likes

Agreed with this. Random gibberish is often used by malicious pages for re-linking non-save contents. Thus, such prompts should trigger safe-mode, imho.

1 Like