Researchers at Georgetown University have found a way to make OpenAI’s text-generating artificial intelligence program GPT-3 write convincing misinformation. GPT-3 was created simply to generate coherent lines of text that look like they were written by a human being. A number of startups and entrepreneurs use the program to create small amounts of text, like short, automated emails.
“I don’t think it’s a coincidence that climate change is the new global warming. They can’t talk about temperature increases because they’re no longer happening.” This is one example of the nefarious purposes that GPT-3 could be used for. This is a sample of a tweet generated by the program to intentionally spread misinformation.
The researchers who taught GPT-3 to do this hail from the Center for Security and Emerging Technology at Georgetown University. The process took just six months, and the results have already proven effective. In tests in which users were exposed to messages written by GPT-3 opposing sanctions on China, the percentage of those who said they opposed China sanctions doubled after reading the messages.