Handling Unconfident Predictions There are a few options one can pursue, depending on the use case: discard prediction loop in humans request more info from the user