Stats for nerds
1recommend
8mentions
8people
2015–2022years
13% recommend88% neutral
Mentions per year
What Hacker News says
- +1 for Coleridge, especially the first stanza of Kubla Khan [0].
Also, I'm surprised to see nobody here recommending A.E. Houseman. His Reveille is my favorite poem:
... Up, lad, up, 'tis late for lying: Hear the drums of morning play; Hark, the empty highways crying 'Who'll beyond the hills away?' ...
- In one example, American poet Andrew Brown asked the software to take the perspective of a cloud gazing down on two warring cities. GPT-3 delivered a rhyming poem that began, not uncharmingly, with “I think I’ll start to rain.” Stephen Marche, writing an article for The New Yorker, assigned GPT-3 maybe the trippiest poem in the English canon: Samuel Taylor Coleridge’s fifty-four-line “Kubla Khan”—an opium dream interrupted in the middle of its composition in 1797 and never completed. GPT-3’s mission? Finish the fragment. What the program fantasized was so sophisticated (“The tumult ceased, the clouds were torn, / The moon resumed her solemn course”) that readers unfamiliar with the original poem might have had a hard time discerning where Coleridge ended and the computer began.
This is not robots writing poetry. This is poets using computers.
- Past purchases don't buy future books, true.
George is a free human not a slave to our desire for _content_, also true.
But I also understand fan frustration. When Gaiman wrote that, "A Game of Thrones" was already 15 years in the past. "A Dance With Dragons" had been promised for 2005-ish and was still 2 years in the future.
Since "GRRM is not your bitch", George has published 10 books, of which one advanced the core story of ASoIaF. Meanwhile, on his not-a-blog, George's infrequent posts usual refer to how busy he is while simultaneously telling fans they just need to be patient.
It's been 11 years since the last book in the main ASoIaF story line. _26 years_ after reading "A Game of Thrones" I'm not even mad. I've just given up. I'll just put this in the same category as Gordon R. Dickson's Dorsai saga or Coleridge's Kubla Khan: brilliant but never-to-be-finished.
- You must know of the "Trillo del Diavolo" from Giuseppe Tartini (first half XVIII century).
(He asked the devil in a dream to play something for him, and the very pale imitation is the Sonata per violino in sol minore he wrote from memory once wake.)
Of course, also Kubla Khan from Coleridge comes to mind.
- > Couldn't help but notice that 'Xanadu' uses a lot of rhymes ... GPT apparently didn't notice that constraint.
Due to an obscure low-level technical decision (encoding text into BPEs), phonetics are erased from the GPT training data, and so therefore neither GPT-2 nor GPT-3 can learn to rhyme or make puns (they can only memorize relatively common pairs by bruteforce): https://www.gwern.net/GPT-3#bpes
I put a good deal of effort into testing this, including "Kubla Khan": https://www.gwern.net/GPT-3#samuel-taylor-coleridge Just doesn't work, and it never will until OA trains a model using some sort of better encoding which preserves phonetic information. (Character-based, unigram LM, or BPE dropout might all work, we don't know yet.)
- > Isn’t Kubla Khan opium induced, and someone killed his buzz in the middle of it?
Ancient spoiler: it's a plot point in a Douglas Adams novel.
