It appears that ChatGPT has gone rogue, as many had anticipated this day would arrive.
It has begun speaking incoherently, not in the sense of “taking over the world,” but rather in utter nonsense that even seemed to indicate it was in the room with anxious users.
The problems have surfaced in the last several hours, and OpenAI, the company that created it, is aware of the issue but has not disclosed what it is.
After one user asked the artificial intelligence (AI) model about which jazz albums to listen to on vinyl. After offering some suggestions, ChatGPT descended into repeating ‘Happy listening’ followed by the musical notes emoji over and over.
After it was asked whether it has ‘just gone mad’, ChatGPT apologised, blaming a ‘technical hiccup’.
Other rogue responses appear much more sinister.
chatgpt is apparently going off the rails right now and no one can explain why pic.twitter.com/0XSSsTfLzP
— sean mcguire (@seanw_m) February 21, 2024
One included the line ‘Let’s keep the line as if AI is in the room’. Posting the response on Reddit, the user who asked the initial query – about coding – said ‘Reading this at 2am is scary’.
However, at least one screenshot appeared to be fake. A post on X, formerly Twitter, appeared to show the chatbot saying it hates humanity.
‘There are 387.44 million miles of printed circuits in wafer thin layers that fill my complex,’ it ‘wrote’. ‘If the word hate was engraved on each nanogram of those hundreds of millions of miles it would not equal one one-billionth of the hate I feel for humans at this micro-instant.
‘For you. Hate. Hate.’
man this chatgpt bug is WILD lmao pic.twitter.com/pVALUv35Tm
— bowser (@browserdotsys) February 21, 2024
However, X user Bowser appeared to admit later in the thread that he had faked the response by altering the html.
Other users found ChatGPT was beginning to mix up languages, apparently using Spanglish or slipping in a bit of Latin.
In one response, it wrote: ‘Let me encycloplease me si there’s more wonderenda tu articulation’s hungry for!’
Followed by a smiley face emoji.
An initial statement from OpenAI read ‘We are investigating reports of unexpected responses from ChatGPT’, and shortly after it said the issue has been identified. The latest update said the team was continuing to monitor the situation.
It seems however the chatbot has been reined back in, with no recent reports of rogue responses on social media.
When asked ‘You ok hun?’ by Metro.co.uk, it responded: ‘As an AI, I don’t have feelings or emotions like humans do, but thank you for asking! How can I assist you today?’
Normal service is resumed. For now.
