ChatGPT has bewildered users by answering questions in Spanglish, as the much-hyped artificial intelligence (AI) tool was struck by an embarrassing glitch.
Reports of the OpenAI-owned chatbot talking gibberish have emerged on social media, with the US tech giant admitting that users were receiving “unexpected responses”.
Among the bizarre responses were messages written in a mixture of English and Spanish, while others repeated the same word indefinitely.
One example said: “Let me encylopease me si there’s more wonderenda tu articulation’s hungry for!”
Gary Marcus, an AI expert and emeritus professor at New York University, said ChatGPT had “gone berserk”.
Writing in his newsletter, Mr Marcus said: “These systems have never been stable.
“Nobody has been able to engineer safety guarantees around them. We are still in the age of machine learning alchemy.”
On the company’s forum, one user said: “As of about three hours ago all of my conversations with GPT4 devolve very quickly into garbage.”
Several users said ChatGPT would begin as normal before the response became nonsensical.
In one post, ChatGPT said: “A day. A high. A chalk. A spark. A line. A ken. A vire. A vane. A byre. A bye. A been. A balk.”
OpenAI admitted that some users had experienced problems but said: “The issue has been identified and is being remediated.”
The problem has since been fixed.
ChatGPT has been held up as a major advance in AI development, able to deliver human-sounding answers to detailed questions.
Businesses have been turning to the bot to automate tasks, using it to write emails, summarise reports, or answer customer queries.
Tech giants have invested billions of dollars in AI companies to develop “large language models”, which are trained on vast databases of billions of articles and thousands of books.
Microsoft has pledged to invest $13bn in OpenAI alone, while Amazon and Google have each invested billions of dollars in rival Anthropic.
Elon Musk has also launched a rival venture called xAI.
However, sceptics have long pointed out that chatbots are prone to delivering strange responses, particularly as bots can make up facts – a process known as “hallucinating”.
Last week, OpenAI launched a new AI tool designed to generate photorealistic videos from text prompts alone.
However, some of its videos feature uncanny distortions, such as objects, animals or people appearing from thin air.
Sam Altman, the chief executive of OpenAI, has previously claimed AI could pose an existential threat to the future of humanity.
At the same time, he has been raising billions of dollars to pursue ever more powerful versions of the technology.
Concerns over the unknown risks posed by AI were partly behind the Government’s AI Safety Summit in November and led to an agreement to set up a safety institute to monitor the threat.
The Government has been experimenting with using ChatGPT on its own website.
chatgpt-baffles-users-by-speaking-spanglish-as-ai-goes-rogue
