OpenAI's new model is 'really good' at creative writing
Artificial intelligence giant OpenAI is developing a new large language model with creative writing skills that have already wowed the company's CEO. But a short story produced by the unnamed, unreleased model has proven to be divisive. Is it a technological feat or a fear realized for the authors and publishers who accused the company of illegally pirating their work to train its LLMs?
OpenAI ventures into creative writing
OpenAI has trained a new AI model that is "really good" at creative writing, CEO Sam Altman said in a post on X. He fed the model a prompt to "write a metafictional literary short story about AI and grief" and then posted the 1,172-word short story it produced. The story's protagonist, Mila, turns to an AI chatbot, who serves as the narrator, to have a monthlong conversation about losing her partner, Kai. It is unclear when the model will be released, but this was the first time Altman was "really struck by something written by AI; it got the vibe of metafiction so right."
Despite all the anxiety AI has given creatives, fiction writing "isn't an application of AI that OpenAI has explored much," said TechCrunch. The fact that it is experimenting with writing could suggest OpenAI "feels its latest generation of models vastly improve on the wordsmithing front," though historically, AI "hasn't proven to be an especially talented essayist."
The shift toward creative writing "signals OpenAI's growing ambitions beyond improving accuracy and predictability," CNET said. It also might "stem from R&D targeting new domains away from more numerical subjects like math and programming," where the company has "struggled to develop monetizable products," market research analyst Reece Hayden said to CNET. OpenAI faces an uphill battle regarding usability because it is "likely to experience significant backlash from creative industries as their intellectual property concerns are seemingly coming true."
'Beautiful and moving' vs. 'trash'
The response to the actual story has, unsurprisingly, been mixed. Machines do not feel, but they can be "taught what feeling feels like," and that is "what we get in this story," novelist Jeanette Winterson said at The Guardian. What is "beautiful and moving" about the story is its "understanding of its lack of understanding" and its "reflection on its limits." Humans always want to read what other humans say, but "like it or not, humans will be living around non-biological entities," she said. "AI reads us. Now it's time for us to read AI."
The sample story is a "remarkable bit of work and unlike anything I've read before," Lance Ulanoff said at TechRadar, "certainly anything I've ever read from an AI." Considering the speed with which OpenAI is "spitting out these powerful new models," the "future is not bright for flesh and blood authors." Soon, publishing houses will "create more detailed literary prompts that engineer vast, epic tales spanning a thousand pages." They will be "emotional, gripping and indistinguishable from those written by George RR Martin."
Some people remain unimpressed by OpenAI's alleged progress on the creative front. The new model's writing "while more verbose — still sucks," Kyle Barr said at Gizmodo. OpenAI may be working on "multiple updates to its LLMs and reasoning models," but "all signs point to them losing steam." When Altman "promotes AI's literary talents," he is trying to "create a new market for ChatGPT subscriptions by promising uncreative people they can take the reins from the literary 'elite.'" But "even if you imagine a human created this, it's still trash," Barr added. Knowing AI created it makes it "doubly trash."