The predictive text Harry Potter chapter is funny but it is also sadly fake, or at least heavily human-edited.
@yeshomo how do you figure? Seems plausible enough to me, if they taught a predictive text off of all of Harry Potter
@CFHistory too much continuity between sentences. Text prediction algorithms aren't smart enough yet to know that someone who's torn their eyes out is unable to see, for instance. They understand that certain words get repeated in sentences near each other but I've never seen one that could do that for a word and its related words, like "eyes" and "see", let alone grasp the meaning of "tearing them out" enough to phrase "seeing" as a negative.
@yeshomo who knows? Considering it was only trained using a series of novels, which are more gramatically correct than the average text chain by nature, I couls see seven books of text teaching it certain patterns. I agree at least the ends of sentences could be curated but I'd believe its a real thing. They included links to the predictive text filed they produced after all
@CFHistory Honestly I would believe they generated a large number of sentences and then curated and arranged them to form this, but that's what I meant by human editing.
@yeshomo eh maybe. Still funny though
@CFHistory Part of my gut reaction was "these feel too much like human jokes" but it's hard in the era of meme culture and absurdist humor to differentiate between arbitrary word combinations created by a machine and created by a human.