In humans, the truth is sometimes lost due to prejudices, limited information and emotional reactions, writes American author Steven Rosenbaum in his book published earlier this month, which was initially positively received The Future of Truthabout the truth in times of artificial intelligence. “Machines introduce a different kind of distortion: a Truth that is so mathematically complex, so rapidly generated and validated, that it becomes incomprehensible to human perception.”
That’s exactly what happened to him: his book contains several misattributed or even fabricated (AI-hallucinated) quotes, revealed The New York Times exactly one week after the publication date.
For example, American tech journalist Kara Swisher is quoted as saying: “The most advanced AI language model is like a mirror. It reflects our own morals back to us, polished and articulate, but ultimately empty beneath the surface.” She herself said yes The New York Times that she had never said that. “I also sound like I have a stick up my ass, according to ChatGPT,” she said in a message to the newspaper.
You can clearly recognize the erratic virtual hand of the citation errors large language models. Emotion professor Lisa Feldman Barrett is credited with a made-up quote about emotions that she believes is incorrect. A quote from American data journalism professor Meredith Broussard is attributed to a book of hers as it came from a radio interview.
The fact checker and two editors from the publishing house didn’t notice the misquotes. It is unclear how many there are. The New York Times speaks of “more than half a dozen”, the author himself speaks of “a handful” in a response to that newspaper. There could be more as the investigation continues.
It was an accident, Rosenbaum said. He had used AI programs Claude and ChatGPT in his research and during writing and editing; that was also stated in his book, but he had not intended to fabricate points of view. He takes full responsibility for the mistakes. New editions of the book will be corrected.
Fixed. Meanwhile, publishers and media are struggling with writers using AI. This month became one short story that a competition in literary magazine Granta won, suspected by many to be written by AI. Last year AI came up with a list hallucinated summer books for the Chicago Sun Times. In science, the number of references to non-existent articles is increasing sharply.
By the way, Rosenbaum has no plans to stop using ChatGPT and Claude, he told news site ArsTechnica. And against both Ars Technica as the The New York Times he dared to say that his mistakes are precisely the warning about the risks of AI that prompted him to write the book. Although he would have preferred to receive attention for it in a different way.
