February 21, 2024 ChatGPT had what witnesses are describing as a “meltdown” over the past day. Various reports surfaced with incidents of the model speaking in “Spanglish,” (a mixture of Spanish and English) generating a barrage of bewildering and nonsensical responses to user inquiries, and claiming to be in the room with one user at 2:30 am which they described as “creepy.”

OpenAI, the entity behind ChatGPT, has recognized the issue and stated that they were monitoring the issue. One suggestion was that the bizarre behavior might stem from an overly creative setting within the AI’s parameters. But OpenAI has not confirmed the cause of this behaviour.

This isn’t the first instance of ChatGPT exhibiting erratic behavior. Late last year, users reported that ChatGPT had gotten “lazy,” and was giving truncated or very short answers, even going so far as to tell users to refer back to the original source if they wanted anything more.

Such occurrences highlight the critical need for ongoing scrutiny by both users and developers to ensure AI technologies remain reliable. This event  also emphasized the delicate balance between AI creativity and coherence, serving as a reminder of the challenges in AI development and the importance of maintaining robust oversight mechanisms.

Sources include: The Independent