Bot Moderation

Sometimes a message is sent to the bot that's not appropriate. The reasons for this are large and complex, much outside of our developer's controls. When you submit a request to the bot, it will leverage several moderation APIs to ensure that your request is appriorate before sending it to the OpenAI API. Based on the command, there are different reasons your request will be rejected.

Rejected requests do not consume snaks.

/prompt

The /prompt command runs first against our internal moderation API. If it passes, it will then run against the OpenAI API's moderation API. If it passes, it will then send the request to the OpenAI API.

/img

The /img command runs first against our internal moderation API. It then runs against a simple word filter based on banned words from the OpenAI API. It then runs it through a OpenAI moderation check. Finally it runs it through an AWS moderation check. If it passes all of these checks, it will send the request to the OpenAI API.

The bot will validate that the image is safe before posting it. If the image is deemed unsafe (the generated image), the bot will reattempt to generate the image for free.

There are a variety of words that will cause the bot to reject the requests. These words aren't available to the public. As the bot finds them, it attempts to expand its list of words to avoid failed queries.

If it's a word based block, the bot will tell you. If the bot says that it's bad vibes, then the moderation check failed.