It doesn't. Consider the first 10k words in my /usr/dict/words equivalent.

They start:

    A
    a
    aa
    aal

and end:

    anticonductor
    anticonfederationist
    anticonformist
    anticonscience

How do you want to describe those with a picture?

(ok, these days one could make a long, thin screen shot, but a non-trivial picture, that doesn't factor through a textual representation? Even representing the 8 strings given here via graphics sounds iffy to me...)

That epigram is referring to a fundamental asymmetry of information between words and pictures. Having a machine that can come up with many different pictures from a few words doesn’t only not say anything about how to make a picture from any 10K words, it somewhat reinforces the epigram’s point.

Those models may be more adequately described by #63: When we write programs that “learn”, it turns out that we do and they don’t.

It doesn't need to be rewritten at all. I can ask Dall-E for a picture that conveys the meaning of the Sermon on the Mount, say, or of Einstein's 1905 paper on special relativity. Whatever I get will not convey the content that the sermon or the paper convey - not even close.