Several hours ago, there was an Incident where ChatGPT went haywire and began returning absolutely bizarre nonsense in response to normal queries in many different contexts. OpenAI confirmed there was an issue but did not specify its nature – just superficially, it looks like what you'd expect if the temperature got set to an absurdly high value (causing it to repeatedly choose extraordinarily unlikely extensions to its message)
It's not great when an API starts returning 5xx errors, but this? I'm pretty sure the API you're relying on randomly devolving into unhinged nonsense is worse. This isn't software engineering so much as software interpretive dance being performed in a minefield.
(screenshot provided by a discord friend who documented the issue before it was fixed; you can find many others in places people discuss chatgpt)