Do we need to use "content filter" for both prompt AND competitions?

Community
1

I’m using the “content filter” engine to filter out the prompts.

But the question is do I need to run the completions as well, through it?
Warning the user before the prompt is consumed by the API is logical, to save tokens and prevent unsafe content in the first place.

Thanks in advance!

2

To quote GPT-3:

According to the policy, you must not generate unsafe content. Hence, if the completion is unsafe, you would still be violating the policy, even if the prompt itself is safe. In practice however, a well-designed prompt that passes the moderation filter is extremely unlikely to generate an unsafe completion.

Note that while using the moderation filter may be necessary for approval, it’s not a garantuee for policy compliance. The filter isn’t perfect.

2 Likes
3

Thanks, Vincent! I read the docs thoroughly. And now my app is approved! yay!

For approval yes, it is an item on the checklist. However, in my experience, there’s no unsafe content generated if filtered through.