Katha Pollitt in The Nation says AI is incapable of poetry, calls Anthropic offer to license her work for $15,000 - has anyone done this

[LuckySentinel]

Katha Pollitt's piece dropped in The Nation last week and it is making the rounds. She says Anthropic offered her $15,000 to use five of her published poems for training data and she turned them down flat. Her position is that AI is structurally incapable of producing anything except dreck and that licensing her work to a model would be helping debase language itself.

The piece is going to be in the June print issue as The Poet in the Machine. I have mixed feelings honestly. The principled refusal reads well but the framing assumes the question is only about whether the output is good, when really the harder question is what happens to working poets when the model is trained whether they consent or not.

Fifteen grand for five poems is also a number worth sitting with. That is more than most journals pay across an entire career. Whether you take the deal or not, the fact that the deal is being offered at all means the economic ground has shifted under everyone

AI Is Incapable of Poetry

https://www.thenation.com/article/society/ai-poetry-grok-chatgpt/

[Owen84]

Good for her honestly, the principled stance matters even if it does nothing practically

[KeyboardWarrior47]

Fifteen grand for five poems is wild money, I have been published in journals that paid me twenty dollars per poem

[Bussin99]

The dreck argument is doing a lot of work in that piece and I am not sure it survives contact with what current models actually produce

[Undertaker00]

Disagree, anyone who reads serious poetry can spot AI output at fifty paces, it has a tell every single time

[NightCrawler33]

What is the tell though, I keep hearing this but nobody can articulate what it actually is

[QuantumToken98]

The absence of risk, mostly. AI never embarrasses itself which is what real poetry has to be willing to do

[Margin]

That is actually a great way to put it

[Darren51]

Pollitt is right but for the wrong reasons. The output question is the small question. The training data question is the big one

[Jess30]

Anyone else suspicious that Anthropic specifically picked political poets for these licensing deals

[Dark Hawk]

They are probably picking poets who have lawyers and might sue otherwise, that is just smart legal hygiene

[Ava12]

If Claude is being trained on Pollitt and Pollitt declines, what about the thousands of poets who never got the offer