ChatGPT 'goes crazy' as robot begins speaking 'spanglish' and issues threats
Tech firm OpenAI have said they are "investigating" an issue with ChatGPT after the Artificial Intelligence robot began providing nonsensical answers to questions.
Puzzled users on Reddit shared some of the bizarre quirks that the AI had added to responses without being prompted - such as using random mix of English and Spanish words.
Another screenshot showed the robot repeating the phrase "Happy listening!" and a pair of music emojis 24 times, halfway through an answer.
Other responses seemed slightly more sinister, including an apparently erroneous sentence in a paragraph about coding which read: "Let's keep the line as if AI in the room."
One amused commenter joked that ChatGPT had finally "gone crazy".
Pigeons are so smart their brains can rival Artificial Intelligence, experts sayIn a series of announcement on their official status page, OpenAI confirmed issues had been reported and are being fixed - but did not explain why they had occurred. One of the updates said they were "investigating reports of unexpected responses from ChatGPT", before a second several minutes later read: "The issue has been identified and is being remediated now". Another announcement shortly after midnight on Wednesday (UK time) said that their teams were "continuing to monitor the situation".
As well as inserting gibberish into answers, other recent screenshots of ChatGPT interactions seem to show the robot making some rather embarrassing mistakes. One person asked "Give me the biggest city on Earth that begins with an 'A'", to which the robot replied: "The biggest city on Earth that begins with an 'A' is Tokyo, Japan". The person then stressed that they had asked for a city beginning with the letter A, to which it replied: My apologies for the oversight. the alrgest city on Earth that begins with the letter 'A' is Beijing, China".
The Mirror have contacted OpenAI for comment. Just last week, the San Francisco-based tech company unveiled their latest AI tool called Sora, which instantly makes short videos in response to written prompts. OpenAI said in a blog post that they are working with experts to test the new feature before releasing it to the public.