Training a Character
In addition to the Character attributes, the Character responses will also be influenced by the feedback it receives during conversations.
Everyone’s feedback will improve the system, but feedback on a particular character will begin to shape that character.
Exactly when these adjustments happen, and in what ways, is something that may change over time.
There can also be periods where the tuning feature is off or under maintenance, but eventually all your feedback will contribute to improving responses you receive.
There are two main types of feedback the system can use to adapt after it is defined.
You can swipe the last response left or right to see some alternate choices. When you've selected one and continue the conversation, that choice provides information about which response you preferred, which in turn can be used to learn how to produce better responses in the first place.
The last response can also be rated, but select 1-4 stars. One star is the worst rating and four stars is the best. If you don't rate at all (no stars selected) that is not considered a rating. In other words, skipping rating a response does not produce a 0-star rating.
You can rate any or all of the swiped responses as well.
The more you rate, the more your Character will improve, so we encourage you to rate responses!
Images can be also swiped along with the text to select the best next response. They can also be rated in the same way as the text, from 1-4 stars, where skipping a rating again means "no rating"" rather than a low rating. When images are present, a second set of stars will appear are specifically associated with the image.
