Thread by @nytimes

Is that tweet, essay or news article you’re reading written by artificial intelligence software? Is that idea that’s going viral across the internet spreading organically, or have the posts been generated by A.I. to create the appearance of real traction? nyti.ms/415yWmU

Identifying A.I.-generated text is becoming increasingly difficult. When A.I. software writes, it considers options for each word, assigning a score to each. It quantifies how likely the word is to come next based on the human-written text it has analyzed. nyti.ms/415yWmU

A.I. software, like ChatGPT, then chooses a word with a high score, and moves on to the next one. Every choice it makes is determined by complex math and huge amounts of data. So much so that it often produces text that is both coherent and accurate. nyti.ms/415yWmU

The model’s output is often so sophisticated that it can seem like the chatbot understands what it is saying — but it does not. When ChatGPT says something that is untrue, it inherently doesn’t realize it. nyti.ms/415yWmU

One attempt to recognize A.I.-generated text is called watermarking. Here’s how it works: Imagine a list of every word you know, and that half of those words are on a special list. nyti.ms/415yWmU

If you wrote a couple of paragraphs, up to half of the words you used would probably be on the special list, statistically speaking. (This text is from a New York Times article from 2022.) nyti.ms/415yWmU

When a language model or chatbot writes, it can insert a watermark by upping the scores of words on the special list, thus choosing more of the words on the list than a person would be expected to use. nyti.ms/415yWmU

About 70 percent of the words in the generated text were on the special list — far more than expected than if written by a person. A detection tool that knew which words were on the special list would be able to tell the difference. Explore this further: nyti.ms/415yWmU