This article is reproduced from: College students successfully deceived everyone by generating fake articles with GPT-3
GPT-3 was widely discussed when it came out. Not long ago, Liam Porr, a student from the University of California, Berkeley, conducted an experiment to prove the possible impact of GPT-3 on the way people write and read.
He opened a blog under the pseudonym of Adolos , where he published articles generated by the GPT-3 model. The first article was topped on the Hacker News headline, and was also reposted by major platforms such as medium. The blog quickly gained 60 followers and approximately 26,000 visits. But noticed that the article was generated by artificial intelligence, and only a handful of people questioned it.
This article is entitled " Feeling Low Productivity? Maybe you should stop overthinking " the article attracted a lot of netizens empathy and echoed. One of the commenters pointed out that "this may have been written by GPT-3, with zero substance content", but was rebutted by others, was asked to provide reasons, and "keep civilized and do not insult others".
In fact, Porr himself only specified the title and introduction, and all other content was completed by GPT-3. He found that GPT-3 is not good at logic and reason, but can create beautiful sentences. Such articles cannot withstand careful scrutiny, but it does not prevent a large number of people from still reading them, and it is often enough to throw a popular topic.
GPT-3 is considered to be the largest AI language model to date, consisting of 175 billion parameters. This is an unprecedented and huge language model. It can do almost any work that can be expressed in words, and can guide it to answer questions, write articles, write poetry, and even write code.
The developer behind it, OpenAI, has only opened a small number of GPT-3 API test qualifications, and Porr found a qualified doctoral student to cooperate with the Berkeley AI community. From generating this idea to using GPT-3 to generate articles with viral potential, Porr only took a few hours. In the week before that, he did not understand what GPT-3 was.
The experiment lasted for two weeks, and finally ended in the last article published on the Adolos blog " If I had no morals, what would I do with GPT-3 ", the article pointed out that "the method of abusing this technology is more subtle than creating Nazi propaganda. Much". Of course, whether this article was generated by GPT-3 or written by humans is also a mystery.
Porr himself finally explained the entire experimental process in his blog . In addition to alerting the current network content quality and people's ability to distinguish, he also proposed the possible impact of technologies such as GPT-3 on media organizations. He believes that even if labor costs can be drastically reduced, the traditional and established media will still resist them due to their professionalism, and the new media may gain space from it.
About GPT-3, you can also read: GPT-3, the most powerful algorithm model in history, comes out