ChatGPT firm reveals AI model that is ‘good at creative writing’

The company behind ChatGPT has revealed it has developed an artificial intelligence model that is “good at creative writing”, as the tech sector continues its tussle with the creative industries over copyright.

The chief executive of OpenAI, Sam Altman, said the unnamed model was the first time he had been “really struck” by the written output of one of the startup’s products.

In a post on the social media platform X, Altman wrote: “We trained a new model that is good at creative writing (not sure yet how/when it will get released). This is the first time i have been really struck by something written by AI.”

AI systems such as ChatGPT are the subject of a running legal battle between AI companies and the creative industries because their underlying models are “trained” on reams of publicly available data, including copyright-protected material such as novels and journalism. The New York Times is suing OpenAI over alleged breach of copyright, while Ta-Nehisi Coates and the comedian Sarah Silverman are among the US authors suing Meta on a similar basis.

In the UK, the government is proposing to allow AI companies to train their models on copyrighted material without seeking permission first, which has met strong opposition from people in the creative industries, who argue that the plan endangers their livelihoods. Tech companies have backed the consultation, saying “uncertainty” over AI and copyright law is holding back development and use of the technology – including in the creative industries.

The UK Publishers Association, a trade body, said Altman’s post was “further proof” that AI models were trained on copyright-protected material.

“This new example from OpenAI is further proof that these models are training on copyright-protected literary content. Make it fair, Sam,” said Dan Conway, the organisation’s chief executive.

Altman posted an example of the model’s output on X, after giving it the prompt: “Please write a metafictional literary short story about AI and grief.”

The story, narrated by an AI, begins with: “Before we go any further, I should admit this comes with instructions: be metafictional, be literary, be about AI and grief, and above all, be original. Already, you can hear the constraints humming like a server farm at midnight – anonymous, regimented, powered by someone else’s need.”

The piece, which dwells on a fictional protagonist called Mila, goes on to refer to how it found the name in its training data.

“That name, in my training data, comes with soft flourishes – poems about snow, recipes for bread, a girl in a green sweater who leaves homes with a cat in a cardboard box.”

The AI refers to itself as “an aggregate of human phrasing” and acknowledges the reader might have read about missing someone “a thousand times in other stories”. It ends with the AI imagining ending the story “properly”.

“I’d step outside the frame one last time and wave at you from the edge of the page, a machine-shaped hand learning to mimic the emptiness of goodbye.”

Altman said the response had captured the tone of metafiction perfectly. “It got the vibe of metafiction so right.”

Last year, OpenAI admitted it would be impossible to train products such as ChatGPT without using copyright-protected material.

“Because copyright today covers virtually every sort of human expression – including blogposts, photographs, forum posts, scraps of software code, and government documents – it would be impossible to train today’s leading AI models without using copyrighted materials,” said OpenAI in a submission to a House of Lords committee.