Explicitly filter out text that could be harmful or questionable. Can be done in two ways: 1. Simple keyword-based filtering to avoid harmful text as O/p to the end-user.
2. Running it through a content filtering model side-by-side just like the one you see in OpenAI playground.

→ View original post on X — @saboo_shubham_,