"If Anthropic had offered me $15,000 to use five pieces of my published work to train its large language models, I would have said no way. Why should I help Dario Amodei, the company's CEO, to debase language, imagination, individuality, art? Aren't people stupid enough already? Perhaps anticipating this response from writers, a proud and prickly lot, Anthropic simply took some 500,000 books and articles and used them without the authors' knowledge or permission. That's plagiarism."
"Three authors filed a lawsuit, and a court ruling requires Anthropic to pay out $1.5 billion-lunch money for the company, which is valued at $380 billion. I filled out the online forms to collect my $15,000 share-well, actually my brilliant young assistant, who describes herself as a "digital native," did that. The forms were pretty complicated, probably just to remind us old-fashioned scribblers who really holds the upper hand here. After all, the damage is done, and no amount of settlement money can undo it."
"About a year ago, I asked ChatGPT to write a poem "in the style of Katha Pollitt." The result was fairly ridiculous: more like a greeting-card jingle than a poem by anyone over the age of 10. Whew! I tested ChatGPT again just now. Apparently it has been taking poetry workshops. Singsong rhyme and meter are out; free verse and wistfulness are in. Here's the beginning:"
"The meeting runs long, as meetings do- a table of voices, mostly baritone, interrupted by a careful soprano that learns to fold itself between commas. Outside, the city conducts its own debate: sirens insisting, buses sighing dissent, a woman on the corner counting tips like a rosary of small survivals. This has all the tics of contemporary mediocre poetry: the knowing nudge ("as meetings do"), the look out the"
A refusal to help train language models is framed as a defense of language, imagination, individuality, and art. Anthropic is described as using hundreds of thousands of books and articles without authors’ knowledge or permission, leading to plagiarism claims and a lawsuit. A court ruling requires a $1.5 billion payout, and the company’s value is noted. The writer describes receiving a $15,000 share through online forms, with an assistant handling the process. Experiments with ChatGPT are used to show poor poetic results, including earlier greeting-card-like verse and later free-verse output with contemporary poetic mannerisms. The damage is presented as irreversible despite settlement money.
#large-language-models #copyright-and-plagiarism #creative-writing #poetry-generation #ethics-of-ai-training
Read at The Nation
Unable to calculate read time
Collection
[
|
...
]