I read the article, and I think the person you replied to is spot on. This paper, and the discussion around it, is a lot of handwringing over nothing. The resu

I read the article, and I think the person you replied to is spot on. This paper, and the discussion around it, is a lot of handwringing over nothing. The results generalize to two things: 1) if you build a system that is stable, then a minor perturbation will not destabilize it. 2) if you have a system that probabilistically models approximate human language, then it isn't surprising when it outputs text that approximates what a human likes to imagine they might do in a given situation.