Why 'bananas'? The thing is trained on a veritable brazillion of gigabytes of human-generated text. You could probably get the same response with a simple reg

Why 'bananas'? The thing is trained on a veritable brazillion of gigabytes of human-generated text. You could probably get the same response with a simple regexp search if you had the same training database. The real wonder is that it manages to stitch everything together without obvious grammar errors or continuity flaws.