Hi @everyone we’re going to do a short (4 to 24 hour test) of a new** AI Moderator system** (only applies to V5 models)
**What’s it for?**
This new AI moderator replaces our old system relying on ‘blocked words’. The new system understands words in context and should be a lot smarter and nuanced. It should (hopefully) let through all the innocent prompts that the old system would throw out.
**How’s it work?**
First we use a fast AI to scan all queries. If it says “no” you’ll have the option to click a **[I would like to appeal]** button and it will get sent to a larger, smarter, and more powerful AI.
The large AI is quite expensive so we’re limiting to 5 appeals per 24 hours for now. We’ll adjust this number dynamically based on how much traffic we’re seeing (hopefully to the highest number we can).
We’re hoping that this two layer system gives people an almost 100% chance of getting innocent prompts through.
**What we need your help on**
Our biggest focus on this test is to make sure that our AI Moderator doesn’t have any weird edge cases that block big areas of prompt-space that shouldn’t be blocked.
This test isn’t a excuse to try to make terrible images. The goal is that this update should just be ‘invisible’ for everyone.
If you find that the system does something bad please let us know by clicking **[Notify the developers of a incorrect decision]** after a failed appeal process, or talk to us via <#1100553606761041991>
Thanks for your help! ❤️