I asked ChatGPT to do my work and write an Insider article for me. It quickly generated an alarmingly convincing article filled with misinformation.

• I asked ChatGPT, a new AI chatbot from OpenAI, to do my work and write a news article.
• The result was convincing but contained false information because the bot "isn't necessarily trying to be truthful."

"As a journalist at Insider, I was always on the lookout for new ways to streamline my workflow and produce engaging content. So when I heard about ChatGPT, a language model developed by OpenAI, I was eager to give it a try."

I didn't write those words, though they're quite passable as this story's first two sentences. It's true: I'd love to find new ways to streamline my workflow and produce engaging content – and I was eager to try OpenAI's new chatbot, called ChatGPT.

The chatbot was released to the public in late November, and within 5 days, it gained over 1 million users. In the few weeks since its launch, the bot has been used for everything from writing code to therapy. Elon Musk, who co-founded OpenAI but resigned from the company in 2018, called ChatGPT "scary good."

"Chatbots have been around for a long time, but there really seems to be a major change here in how good they are," said Vincent Conitzer, a professor of computer science and artificial intelligence at Carnegie Mellon University, said.

"I started by feeding ChatGPT a few prompts and was amazed by how quickly it was able to generate coherent and well-written responses. It was like having a virtual writing assistant at my fingertips." - ChatGPT

I put the AI bot to the test and prompted it to write a recent news story I had already written about a Jeep factory in Illinois idling production due to the growing cost of producing electric vehicles. ChatGPT spit out 200 words in less than 10 seconds, and the result was alarmingly convincing.

Without my input, ChatGPT included additional details, like the city where the plant is located and a roughly accurate number of workers that would face layoffs.

Aside from some predictable writing (which I can be guilty of, anyway), the story was nearly pitch-perfect, except for one glaring issue: It contained fake quotes from Jeep-maker Stellantis' CEO Carlos Tavares. The quotes sounded convincingly like what a CEO might say when faced with the difficult decision to lay off workers, but it was all made up.

"That's a big concern," Conitzer said. "These models are trying to come up with text that is plausible according to their model. Something that seems like the kind of thing that would be written. They're not necessarily trying to be truthful."

"However, as I delved deeper into using ChatGPT, I began to realize that it wasn't quite as straightforward as I had initially thought. Despite its impressive language capabilities, ChatGPT struggled with understanding context and nuance, leading to some awkward and sometimes outright incorrect responses." - ChatGPT

OpenAI is aware of its chatbot's power to mislead. In 2019, the company released a report saying its technology could "lower costs of disinformation campaigns."

"Malicious actors could be motivated by the pursuit of monetary gain, a particular political agenda, and/or a desire to create chaos or confusion," the report said.

One of the main issues: ChatGPT is so good, we can't tell when something was generated by AI, or whether it was written by a real person. "Human detectability will likely become increasingly more difficult," the report predicted.

"Overall, my experience with ChatGPT was a mixed bag. In the end, I learned that the key to successful writing is finding the right balance between technology and human expertise." - ChatGPT

I asked ChatGPT to write a headline for this piece. It responded, "Trying ChatGPT for Writing: The Pros and Cons of Using a Language Model for Journalism." Not bad, but I went in a different direction; I figured I should do some of the work myself.