Markus Schreiber/AP, Tyler Le/BI
ChatGPT is looking a little unhinged right now.
Some users have been wondering what the heck is going on with OpenAI’s chatbot after it started responding to their queries with a whole lot of gibberish on Tuesday.
Sean McGuire, a senior associate at global architecture firm Gensler, shared screenshots on X of ChatGPT responding to him in nonsensical “Spanglish.”
“Sometimes, in the creative process of keeping the intertwined Spanglish vibrant, the cogs en la tecla might get a bit whimsical. Muchas gracias for your understanding, y I’ll ensure we’re being as crystal clear como l’eau from now on,” ChatGPT wrote.
It then descended into a whole lot more nonsense: “Would it glad your clicklies to grape-turn-tooth over a mind-ocean jello type?” It then followed up with references to jazz pianist Bill Evans before repeating the phrase “Happy listening!” nonstop.
Another user asked ChatGPT about the difference between mattresses in different Asian countries. It simply could not cope.
One user, who shared their interaction with ChatGPT on Reddit, said GPT-4 “just went full hallucination mode,” something they said hasn’t really happened with this severity since “the early days of GPT-3.”
OpenAI has admitted there’s an issue. Its status dashboard first noted it was “investigating reports of unexpected responses from ChatGPT” on February 20. It has since been updated to say the issue has been identified and is being monitored.
It’s an embarrassing moment for the company, which has been held as a leader in the AI revolution and received a multi-billion dollar investment from the likes of Microsoft. It's also enticed enterprises into paying it to use the more advanced version of its AI.
OpenAI did not immediately respond to a request from Business Insider about ChatGPT's hiccups.
That hasn’t stopped people from speculating about the cause of the problem.
Gary Marcus, an NYU professor and AI expert, started a poll on X asking users what they think the cause might be. Some think OpenAI got hacked, while others reckon hardware issues could be to blame.
A majority so far, however, say “corrupted weights.” Weights are a fundamental part of AI models, helping service the predictive outputs that tools like ChatGPT gives to users.
Would this be an issue if OpenAI was more transparent about how its model works and the data it’s trained on? In a Substack post, Marcus suggested the situation was a reminder that the need for technologies that are less opaque is “paramount.”
