OpenAI can analyze text and score content among various categories. This can be useful for content editors who may have sensitive audiences as well as hook into things like Comment submissions - like flagging for review.
https://platform.openai.com/docs/guides/moderation/overview
Start by adding it to nodes.
Add a new section on node edit forms for analyzing text.
| Comment | File | Size | Author |
|---|---|---|---|
| #3 | Screenshot from 2023-03-10 11-51-34.png | 24.23 KB | kevinquillen |
| #3 | Screenshot from 2023-03-10 11-51-26.png | 35.9 KB | kevinquillen |
Show commandsStart within a Git clone of the project using the version control instructions.
Or, if you do not have SSH keys set up on git.drupalcode.org:
changes
Comments
Comment #3
kevinquillen commentedAdding this was not difficult to start out with.
Although... the scores in their documentation were well over 1.0 for no reason (despite not being flagged as violated) for technical content. In fact, when I try to add text that should violate categories, it does not at all.
Comment #4
kevinquillen commentedNext steps could be:
Comment #5
kevinquillen commentedIs it possible the OpenAI docs are outdated and the score range is no longer 0-1? You can see for example in this blog post, this persons scores came back well over 1 for categories too:
https://cobusgreyling.medium.com/adding-a-moderation-layer-to-your-opena...
Comment #6
ressaThat's a nice feature, thanks for working on it @kevinquillen. I have no opinion on the numbers ... but the ranges do look quite different between the documentation and the medium.com article.
Going the other way, checking user input for bad words and filtering out before submitting to ChatGPT can be done with https://github.com/ConsoleTVs/Profanity.
Comment #7
kevinquillen commentedIt is more than a word filter though, the scores are designed or should indicate a direction the way the content leans, which some types of sites may find useful. Particularly with user submitted content. Using this endpoint is free of charge too. Its fine for content to have curse words, lets say, if that is the context of the content.
Comment #8
bogdog400 commentedThis could also be useful as a prefilter before calling the main OpenAI API. If a question/prompt has unacceptable things in it, the call to the main API could be pre-empted.
Comment #9
kevinquillen commentedThat is true. On the upside these API calls appear to be free.
We're also considering it for a project that has user submitted content with Content Moderation module enabled. Until an individuals submission passes the score thresholds, the content will live as Draft and not come before the site managers who review content until that is satisfied.
At the same time, we could probably combine that with a "change the tone" style action after that to get content "in shape" for this particular org for display.. the possibilities are pretty cool.
Comment #10
bogdog400 commentedYes. The free price is a nice way to roll out the tech.
Comment #13
kevinquillen commentedThis seems to work alright, especially in combination with other sidebar tools:
https://www.youtube.com/watch?v=pN2GeUnxpl0