ChatGPT cranks out gibberish for hours

SAN FRANCISCO - ChatGPT spewed nonsensical answers to user's queries for hours on Tuesday into Wednesday before eventually returning to its apparent senses.

OpenAI did not explain what went awry with its generative artificial intelligence (AI) tool, considered the one to beat in the technology sector.

"We are investigating reports of unexpected responses from ChatGPT," OpenAI said on its status website when the software seemed to go wacky on Tuesday afternoon.

ChatGPT was giving "peculiar" responses, generating nonexistent words, incomplete sentences and general gobbledygook, developers using the tool said in a discussion forum on the OpenAI website.

"It gives me meaningless words followed by a bizarre list," one developer lamented in the forum.

"It feels as if my GPT is haunted or something has been compromised, either on my end or at OpenAI's (end)."

It wasn't until more than 16 hours had passed that OpenAI updated the page with a message that ChatGPT was operating normally.

The San Francisco based technology firm replied to an AFP query by directing it to the ChatGPT status page.

OpenAI recently concluded a deal with investors that reportedly valued the start-up at $80-billion or more, after a roller-coaster year for the tech firm.

The agreement, reported by the New York Times but not yet confirmed by OpenAI, would mean the value of the company -- a world leader in generative AI -- would have nearly tripled in under 10 months.

OpenAI led a revolution in AI when it placed its ChatGPT program online in late 2022.