Users of the intelligent chatbot ChatGPT of the American company OpenAI drew attention to a strange malfunction in its neural network that occurred on February 20-21. ChatGPT responded to requests with reports and meaningless expressions and incomplete sentences for 16 hours portal Tech Xplore, with reference to data from the Agence France-Presse (AFP) news agency.
According to those who encountered the problem, the chatbot produced non-existent words, randomly switched to different languages, constantly corrected itself, stuck on the same sentences, and generally behaved like it was “possessed by ghosts”.
In some cases, the neural network’s responses resembled the works of avant-garde poets.
“The chick of the product rejected by the waves of the photo set is deep, as if a gloomy nanny in the color of the manager”, this is what a copy of ChatGPT looked like during a failure in translation into Russian.
OpenAI did not comment on the problems, only stating on its website that the company was investigating complaints about ChatGPT’s unexpected behavior. After more than half a day, the AI returned to normal. The company also refused to answer journalists’ questions about the reasons for the incident.
Previous chatbot creators warned about the phenomenon of “hallucinations” of artificial intelligence.