OpenAI’s popular language model, ChatGPT, has been at the center of an unusual incident involving erratic and nonsensical responses.
Users have reported an array of incoherent messages, ranging from “Spanglish” utterances to eerie affirmations of being present in the same room.
The underlying cause of these anomalies has yet to be disclosed. Nevertheless, OpenAI has acknowledged the issue, stating the team is actively investigating the matter. Updates on OpenAI’s official status page indicated progress in identifying the problem, assuring users about ongoing monitoring.
Examples of the erratic behavior surfaced on various online platforms, including Reddit, where users shared their perplexing interactions with ChatGPT. A discussion on jazz albums abruptly turned into the AI shouting “Happy listening!” over and over, one person wrote. Another user’s simple inquiry about a computer’s definition led to a verbose yet unintelligible response.
Further complications arose as ChatGPT seemingly struggled with language consistency, with instances of mixing Spanish and English, deploying Latin terms, or fabricating words entirely. In one response, it wrote: “Let me encycloplease me si there’s more wonderenda tu articulation’s hungry for!”
The strangeness of these outputs extended into the practical suggestions provided by ChatGPT. For example, when prompted on how to make sun-dried tomatoes, the advice included the incomprehensible step “utilise as beloved,” leaving users bemused, The Independent reported.
Some users described their interactions with the AI as unnerving, especially those receiving odd messages late at night. One particular encounter involved ChatGPT responding to a coding assistance request with a tangential and bewildering message, including references to keeping an AI “in the room.”
On its official status page, OpenAI noted the issues with ChatGPT, but offered no reason for its odd behavior.
“We are investigating reports of unexpected responses from ChatGPT,” an update said, followed by another that announced the “issue has been identified.” “We’re continuing to monitor the situation,” the latest update read.
• Staff can be reached at 202-636-3000.
Please read our comment policy before commenting.