> It is always easier to troll and derail than it is to employ such techniques for good.

Curious-- what possible good comes with the ability to generate grammatically correct text devoid of actual meaning?

Sure, you could generate an arbitrary essay, but it's less an essay about anything and more just an arrangement of words that happen to statistically relate to each other. Markov chains already do this and while the output looks technically correct, you're not going to learn or interpret anything from it.

Same goes with things like autocomplete. You could generate an entire passage just accepting its suggestions. It would pass a grammar test but doesn't mean anything.

Chatbots are an obvious application, but how is that "good," or any different than the Harlow experiment performed on humans? Fooling people into developing social relationships with an algorithm (however short or trivial) is cruel and unethical beyond belief.

A photoshopped image might be pleasant or stimulating to look at, and does have known problems in terms of normalizing a fictitious reality. But fake text? What non-malicious use can there possibly be?