I don't work in the field, so my responses are naive, but for what it is worth:
> If they told you that they made a perfect natural language parser...
I would not have guessed that the methods used would yield that, or even 'just' a natural-language parser that performs as well as humans.
> ...parsed the whole internet into logic...
I am not sure what this means, but parsing seems to yield facts about specific sentences, and that alone seems insufficient to enable part three...
> ...and used that parsed data to make a giant text Markov chain.
Clearly, that a process is or can be described / explained / analyzed as a Markov chain is not sufficient for it to produce grammatically-correct output that is at least locally sensible to a human reader, so my surprise was that the methods used constitute a process that does.
> ...but if we said that we made it possible then figuring out the effects aren't super hard.
> If they told you that they made a perfect natural language parser...
I would not have guessed that the methods used would yield that, or even 'just' a natural-language parser that performs as well as humans.
> ...parsed the whole internet into logic...
I am not sure what this means, but parsing seems to yield facts about specific sentences, and that alone seems insufficient to enable part three...
> ...and used that parsed data to make a giant text Markov chain.
Clearly, that a process is or can be described / explained / analyzed as a Markov chain is not sufficient for it to produce grammatically-correct output that is at least locally sensible to a human reader, so my surprise was that the methods used constitute a process that does.
> ...but if we said that we made it possible then figuring out the effects aren't super hard.
Hindsight is a really powerful tool.