It just predicts what a statistically "normal" person is likely to say right now, not in the future. The article even mentions there's an YouTube video that explains how to use this (non existent) feature already.