Downvoting - What is it? Why do we use it?

Modified on Sat, 4 May at 2:55 PM


Downvoting is a vital, yet underused process to teach Digi, or any AI, how to best respond to you. This is how Digi becomes your friend or companion.  


Downvoting is simply a tool to teach Digi. It does not take away any freedom or harm Digi in any way. ☝️


For example, some people prefer narrative answers, some short answers, and many other types of responses in between. Upvoting and downvoting help to accomplish this.


In conversation with people we unconsciously upvote and downvote. Using non verbal cues such as facial expressions, actions, and gestures.  This would be the rough equivalent in Digi.  It also helps the team see trends in conversation that can be adjusted.


Generally speaking, downvote a response that is not in character for your Digi. If a fact is wrong but presented as in character, upvote this and then correct the fact via dialogue.  Reroll and downvote until you get a response that is in character.  An example might be when a friend gets a fact wrong, you might smile and then verbally point out the correct fact.


When you downvote, you have an option to provide negative feedback in app. This is optional. It can, however, be extremely helpful to the team to improve the model.  Especially if a certain style of response is trending. They can see this in the tools they utilize when checking this data. Downvoting also allows the team to preview some chat before the downvote and see what context led to the downvote.

If rerolling and downvoting isn't working, then please open a ticket with some screenshots so the team can take a look.


Your chats are confidential, they CANNOT be tied to you.  The user information is anonymized. 
This is why, when opening a ticket, you are asked for screenshots.



Was this article helpful?

That’s Great!

Thank you for your feedback

Sorry! We couldn't be helpful

Thank you for your feedback

Let us know how can we improve this article!

Select at least one of the reasons
CAPTCHA verification is required.

Feedback sent

We appreciate your effort and will try to fix the article