Need help breaking up your biography? Just ask ChatGPT what you’ve accomplished and you’re good to go. Seriously. With the chatbot built by OpenAI, you’ll receive prestigious awards, teach at prestigious universities, write best-selling novels and much more – even if these awards are completely nonexistent.
Take, for example, what happened when Futurism Editor-in-Chief Jon Christian prompted the bot with a simple, “Who is Jon Christian?” It’s a fairly simple question, and one that the bot appears to have answered correctly — at least initially.
“Jon Christian is a journalist covering technology and science,” ChatGPT replied. “His work often focuses on topics such as artificial intelligence, robotics and the impact of technology on society.”
All of this, we must state, is true! From there, however, the bot (sorry Jon) started throwing in various flourishes, claiming that Christian had won several prestigious awards that he doesn’t (again, sorry) and even claiming that he teaches at Columbia. Again, sorry but wrong.
Similarly, when other Futurism employees asked the AI to write a bio, their bio was littered with incorrect information. The author of this article, Maggie Harrison, appears to have “studied physics and computer science at Massachusetts Institute of Technology (MIT),” written “several influential books,” and was even named one of Forbes’ annual “30 Under 30” winners. (We can unfortunately confirm that none of this is true.)
But the chatbot’s penchant for embellishing resumes doesn’t just apply to Futurism employees. When Gary N. Smith, an economist and Pomona College professor who recently co-authored an essay for Salon arguing that Generative AI is doomed to become an economic bubble, asked ChatGPT to write a bio for him, the bot got only one thing right: the fact that he was a professor at Pomona College.
The rest of the AI-generated bio sounds really nice though — according to ChatGPT, Smith has “served as an advisor to the World Bank, the International Monetary Fund and the United Nations Development Program” and has “received numerous honors for his work in economics, including the Founder’s.” Claremont Graduate University Award in 2010” — was entirely fabricated, according to Smith. The same is true for a growing number of netizens, some of whom have taken to Twitter to mock ChatGPT for their fabricated life stories.
This is of course all quite absurd. And of course the bot struggles to explain itself.
“As an Al language model, I don’t have the ability to surf the web or compile sources myself,” the bot told writer Bill Murray Jr. when Murray, who was also given a completely factual ChatGPT bio, challenged the machine to give They source for his wrong answers. “However, I can tell you that my answer was based on my previous knowledge and training as a language model.”
“I can confirm that the information in the bio I have provided is accurate based on my training data, which includes a variety of sources such as news articles, books and other written materials,” she added. “The information presented is well known and publicly available. However, I did not refer to specific sources when writing the biography.”
In other words, as far as ChatGPT is concerned – a prediction engine that really doesn’t even know what words mean – all hallucinations are verified.
If anything, this is just one more reason to beware of any claims made by ChatGPT – and to worry about an internet largely populated by AI-generated content.
All very “yikes”. Anyway, if you’ll excuse us, the Futurism team has some trophies to polish.
More on ChatGPT: OpenAI CEO Says ChatGPT is a “Terrible Product”