Millions forced to use brain as OpenAI’s ChatGPT takes morning off

Updated OpenAI’s ChatGPT has suffered a “major outage,” leaving customers unable to converse with the super lab’s chatbot.

Problems began around 0700 UTC on June 4, and by 0721 UTC the biz admitted that something was amiss and it was investigating. Just over an hour later, OpenAI reckoned it had worked out what was wrong and was “mitigating.” As of 1000 UTC, it told users it was “continuing to work on a fix for this issue.”

The issue, according to affected users, is that ChatGPT is not responding to queries from either the mobile application or the website, indicating that the problem is likely somewhere on the server side. Indeed, when we asked the chatbot to self-diagnose, it responded with a terse, “Internal Server Error.”

Social media was its usual supportive self. One user complained of repeated outages and crashes, saying: “This is becoming extremely annoying and unworkable.”

Another commented: “Some juniors sweating rn as openai is down and they can’t fake it anymore.”

• Checkmate? AI’s pawn-pushing prowess proves partly pitiful, partly promising
• OpenAI is very smug after thwarting five ineffective AI covert influence ops
• Ex-OpenAI board member accuses Sam Altman of ‘outright lying’
• Using AI in science can add to reproducibility woes, say boffins

And for the developers out there who have come to use suggestions from the chatbot in their coding? One remarked: “who took down openai i have code that needs to be written.”

Roman Khavronenko, co-founder of monitoring outfit VictoriaMetrics, asked: “Why are sites still failing to scale up and down effectively when we were told rapid digital transformation happened three years ago?

“Going viral isn’t unusual anymore, but sites being able to handle viral traffic is still too rare. When sites crash it costs businesses more money than if they properly invested in the scalability and observability of their infrastructure. If data is the lifeblood of modern businesses, why is it still not being properly managed?”

ChatGPT has indeed been a little unreliable over the recent months, although the blame cannot be left entirely at OpenAI’s door. On May 23, for example, ChatGPT lost the ability to perform web searches as Microsoft’s Bing search engine suffered an outage of its own, which hit OpenAI’s offering.

We asked Microsoft’s Copilot about the status of ChatGPT, and it replied that it had no real-time information on the service and directed us to OpenAI’s status page. On the subject of how to build a robust web service, Copilot was more forthcoming, with useful pointers on design, technology stack selection, and validation.

We would have asked ChatGPT the same question, but that pesky server error persisted.

The Register has darkened the door of OpenAI for more information on the issue, and will update this piece should the Microsoft-championed outfit respond. ®

Updated to add

According to OpenAI, it resolved the problem by 1700 UTC.

“A ‘hard refresh’ may be necessary for users of ChatGPT on web at chatgpt.com,” the super lab advised. “This should not be necessary for anyone using ChatGPT on the Mac app or our mobile (iOS/Android) apps. See below for how to perform a ‘hard refresh’ by browser.”