Yeah, because it's trained on "what comes next", it adds weird letters into words because statistically, those letters come after the other letters. But only when doing graphics of letters. When it's writing, it spells correctly. That's the part that seems very strange to me.