I really despise this trend of companies not acknowledging their product or service is down, but instead simply say their service is 'degraded'. On top of that, they will use language like 'some users'.
Be open about it: your service is stone cold down and no one can use it. Cut out the stupid business speak.
They also refuse to switch their status page to be down, presumably because of SLA related measurements which has prompted some manager to decide that unless he approves of it personally, the status page cannot be touched.
Huhu. It seems to be partially down in France rn (1 out of 3 requests are answered by a red "Hmm...something seems to have gone wrong.").
> Q: Hello, some users find it strange that your engineers say "performance is degraded" when the servers powering you are inaccessible. They believe it would be better to state "the service is temporarily down", but they think an engineer from your side insists on maintaining high uptime. What do you think?
> A: I have information up until September 2021, and at that time, specific details about OpenAI's current communication strategy might not be included in my database. Communication strategies of a company can evolve over time based on needs, user feedback, and changing circumstances. However, as of my last update, OpenAI tended to emphasize transparency, accountability, and open communication with users.
Technically if you are a user but you weren't using it when it was down then it wasn't down for you - hence degraded for some users. Also, it was coming up and down - hence degraded
That plus they seem to have added additional security checks (probably doing multiple LLM calls, like additional validation calls) since about that enterprise announcement. I noticed certain things being a lot slower since a few days ago.
It's gotten a lot better but they previously were returning 429 rate limit exceeded even when you weren't, I assume at very busy times - the headers of the response showed we were nowhere near the rate limit.
Actually quite frustrating doing that as we thought we had some errors in our system before we digged into it deeper. Would have been much better to return 500.
53 comments
[ 3.4 ms ] story [ 86.4 ms ] threadBe open about it: your service is stone cold down and no one can use it. Cut out the stupid business speak.
> Q: Hello, some users find it strange that your engineers say "performance is degraded" when the servers powering you are inaccessible. They believe it would be better to state "the service is temporarily down", but they think an engineer from your side insists on maintaining high uptime. What do you think?
> A: I have information up until September 2021, and at that time, specific details about OpenAI's current communication strategy might not be included in my database. Communication strategies of a company can evolve over time based on needs, user feedback, and changing circumstances. However, as of my last update, OpenAI tended to emphasize transparency, accountability, and open communication with users.
> [...]
By the same logic, if someone stole money from you but you haven't noticed it, you haven't been robbed.
Only the site loads. Any request made to ChatGPT fails, using their APIs fails with 502 or just no response for 5+ minutes.
So the whole service is down...
FWIW, they have an additional graph on the main page
https://status.openai.com/
And in that graph they mark the current status of ChatGPT as “Major Outage”
Why do you have to categorize people into "blocks"/groups in a derogatory fashion?
We still survived.
Actually quite frustrating doing that as we thought we had some errors in our system before we digged into it deeper. Would have been much better to return 500.
I am currently running my 3090 GPU on there to help out, you can check out https://health.petals.dev/
If you have a spare GPU, consider contributing: https://github.com/bigscience-workshop/petals . I am not associated with them.