POETRY
Conflicted Copy
Sam Riviere
Faber & Faber, $26.99

Sam Riviere doesn’t want to be original. In the last decade, the British writer has published: a novel about plagiarism, whose title and style were both taken from other writers; poetry that incorporated internet searches, classical epigrams, and Google Translate; and a manifesto against intellectual property. With his latest collection, Conflicted Copy, he has released a book of which he is, at best, a co-author.

This fraught relationship between the writer and his words comes through in Old Mode, which appears in the early pages:

“The only interesting bits are when an element is/ selected and the next sentence is delivered on top –/ this means the whole sequence might never end./ You know when in music a phrase repeats over two/ successive bars only, so that you catch it when it’s /actually ending (and that’s a good song)? Here too, /both modes offer similar outcomes – one just leaves/ the sentence without any conclusion, the other does/ indeed end! That, in itself, is somewhat frustrating.”

For the bored speaker, art is full of “old modes” and spent traditions; interest can only be found in the way elements are “selected” and arranged. The artist becomes the curator, the sommelier. What seems to be a pointed comment directed at the poem itself shifts by the fourth line to an ironic statement on music and monotony, leading to a conclusion that sees originality as a zero-sum game.

The conceit is carried through, the poem essentially makes sense, but there are signs that something is amiss: a “sentence” seems to shift to a “song” before they both somehow become “modes”, words such as “here” and “that” float without any apparent point of reference. That the poem ends with a wry, mannered statement of failure is enough to make a reader wonder if they are the unwitting victim of some sick joke.

Riviere, to his credit, is honest enough to tell us what he’s laughing about. At the start of the collection, in lieu of a dedication or epigraph, the author provides this disclaimer: “This manuscript was composed from December 2020 to January 2021, using Generative Pre-trained Transformer 2 (GPT-2), an open-source neural network created by OpenAI in February 2019.”

In other words, this book, like an ever-increasing amount of text in existence, was written by a form of ChatGPT. Riviere has explained in interviews that he was working with a significantly stodgier version of the language model during the isolated days of COVID, choosing from a short vocabulary list in an interface akin to predictive text: “In a way,” he says, “I was never deciding what words [I would use], other than from this list.” The challenge then became directing the AI away from “the most well-worn grooves of language” towards “weirder or more interesting suggestions”. The results are undeniably weirder, if less coherent, than the limp quatrains the most advanced public version of ChatGPT spits out.