The Guardian publishes an op-ed written by GPT-3

The Guardian published an opinion piece, headlined “A robot wrote this entire article. Are you scared yet, human?”, presented as GPT-3 making the case that it posed no threat to humanity. The piece argued, in the model’s voice, that it had no desire for power and existed only to serve people.

A note appended to the piece disclosed the production process: editors gave GPT-3 a prompt and opening line, asked it to write roughly 500 words, and ran the query eight separate times. Rather than publishing one output unedited, staff selected what they judged the strongest passages from across the eight runs and assembled them into a single roughly 1,100-word piece, cutting and reordering lines as they said they would for a human-written op-ed.

The disclosure drew nearly as much comment as the article’s content. Critics argued the framing — “a robot wrote this entire article” — misrepresented a process that was closer to human editors curating and arranging AI-generated fragments than to a machine authoring a finished text unassisted, and that presenting the result as the model’s unmediated argument for its own harmlessness overstated both GPT-3’s coherence and its capacity to hold a position at all. The episode became an early, widely cited example of a tension that recurred through the following years of generative-AI coverage: how far editorial selection and arrangement can go before a claim that an AI “wrote” something stops being accurate.