OpenAI’s popular ChatGPT artificial intelligence (AI) system experienced a public meltdown between February 20 and 21, bewildering and confounding users with gibberish and other oddities, including unprompted pseudo-Shakespeare.
As of 8:14 a.m. Pacific Standard Time on February 21, the problem appears to have been resolved. The most recent update on OpenAI’s Status page states that “ChatGPT is operating normally.” This implies that the issue was resolved 18 hours after OpenAI initially reported it.
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
Something in my data file sent ChatGPT into a spiral that ended with it… speaking like Shakespeare? To its credit, it worked…. #surcease pic.twitter.com/4wHxVzF0Pw
— NCAAB Market Picks (@cbbmarketpicks) February 21, 2024
It is unknown at this point what caused the issue, and OpenAI has yet to react to our request for comment.
Based on a preliminary analysis of the reported outputs, it appears that ChatGPT encountered some type of tokenization confusion. Because of the black box nature of huge language models constructed with GPT technology, OpenAI scientists may be unable to determine exactly what went wrong. If this is the case, the team will most likely focus on preventative measures, such as adding more safeguards against long strings of apparent gibberish.
Social media reaction appears to indicate that any damage produced by the chatbot was primarily limited to wasting the time of users who expected sensible solutions to queries.
