A fake blog written by AI shot to the top of Hacker News after people thought it was real — here's how a college student made it

OpenAI
  • A college student made a fake blog post using an AI text generator and it was upvoted to the top of Hacker News by people who thought it was real.
  • University of California, Berkeley student Liam Porr created several blog posts using OpenAI's GPT-3 text generator. Several people subscribed to his account, believing he wrote the posts himself.
  • Porr told MIT Technology Review that he guided the AI to write articles about productivity and self-help because "it's not very good at being logical."
  • The episode demonstrates the strength of GPT-3, which can appear convincingly human under some circumstances.

Last month, a blog post appeared on Hacker News offering advice for more effective work: "Feeling unproductive? Maybe you should stop overthinking."

The post was upvoted nearly 200 times and garnered 71 responses from Hacker News' tech- and VC-focused commenters. What most commenters didn't realize: The post was generated entirely by artificial intelligence.

Liam Porr, a University of California, Berkeley student, created a slew of posts over the course of the past month using GPT-3, a language-generating AI tool developed by OpenAI. All Porr had to do was feed the AI a headline and it would generate an entire blog post, which Porr in turn would post on forums like Hacker News.Advertisement

Porr intentionally guided the AI to write about productivity and self-help because "it's quite good at making pretty language, and it's not very good at being logical and rational," he told MIT Technology Review.

The blog posts gained traction on Hacker News and other forums, and Porr's robot blogger even gained some subscribers, Porr said in a blog post explaining his methodology.

The episode demonstrates the strength of GPT-3, which has shown itself to appear convincingly human under some circumstances.
Advertisement

OpenAI has made GPT-3 available to a few developers via a private beta as of last month. It's one of the most powerful language-generating tools ever developed, and has been heralded for its ability to produce text reflecting a wide range of styles and topics. OpenAI plans to sell the product later this year as a subscription-based cloud AI tool.

After more than two weeks of near-daily posts, only a handful of people had voices suspicions that the posts were written by GPT-3, but those comments were frequently downvoted by other readers of Hacker News. "This is either something written by GPT-3, or the human equivalent. Zero substantive content, pure regurgitation," one commenter wrote in response to an AI-generated post. The commenter was downvoted and admonished by other readers.Advertisement

"If you disagree, be civil and give reasons rather than throw insults," someone replied.

{{}}