Resolves according to https://status.openai.com
For this question ‘incident’ is more widely defined to include both incidents OpenAI reports and also partial or complete downtime that are not incidents per OpenAI.
Related questions
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ3,867 | |
| 2 | Ṁ3,658 | |
| 3 | Ṁ2,431 | |
| 4 | Ṁ642 | |
| 5 | Ṁ539 |
@Bayesian Calvin trade to buy 1000 up to 94% 10:32:48 PST. OpenAI page reports start of investigating 10:32 PST Calvin didn't trade it up to 98% until 10:43. Bot for first trade then a human check before second trade I would think? or is that a bot reaction to page saying fix implemented at 10:42?
@ChristopherRandles that's smart. but also it could be that the bot doesn't fully trust the announcement before it's been on the page for a certain time. @Calvin6b82 care to enlighten us? feel free not to tho LOL, don't wanna cost you some alpha
@Bayesian It has screwed me over before so you can totally beat my bot. There has been multiple times where there was a "phantom outage" - a outage is posted but then quickly deleted. Probably because they had some metrics based trigger misfire. You can see which ones are bot trades, they're clearly labeled.
Report an incident 21st?
They have said after monitoring it is working normally. So has an incident been ***reported*** on 21st? People seem to be betting/assuming this resolves yes but there doesn't seem to be anything reported other than monitoring and concluding that it is operating normally. That doesn't seem like reporting an incident to me.
@ChristopherRandles - I'd classify this as an incident. According to the postmortem [0],
On February 20, 2024, an optimization to the user experience introduced a bug with how the model processes language.
LLMs generate responses by randomly sampling words based in part on probabilities. Their “language” consists of numbers that map to tokens.
In this case, the bug was in the step where the model chooses these numbers. Akin to being lost in translation, the model chose slightly wrong numbers, which produced word sequences that made no sense. More technically, inference kernels produced incorrect results when used in certain GPU configurations.
Upon identifying the cause of this incident, we rolled out a fix and confirmed that the incident was resolved.
@LukasDay Sure that is an incident but it happened on 20th. What happened on 21st some monitoring and conclusion that it was operating normally.
Edit according to latest resolved time which has changed not even that was done on 21st.
Feb 21, 2024
Unexpected responses from ChatGPT
Postmortem - Read details
Feb 21, 17:03 PST
Resolved - ChatGPT is operating normally.
Feb 21, 08:14 PST
Monitoring - We're continuing to monitor the situation.
Feb 20, 16:59 PST
Identified - The issue has been identified and is being remediated now.
Feb 20, 15:47 PST
Investigating - We are investigating reports of unexpected responses from ChatGPT.
Feb 20, 15:40 PST
Right now it looks like this
Feb 20, 2024
Unexpected responses from ChatGPT
Postmortem - Read details
Feb 21, 17:03 PST
Resolved - ChatGPT is operating normally.
Feb 20, 23:14 PST
Monitoring - We're continuing to monitor the situation.
Feb 20, 16:59 PST
Identified - The issue has been identified and is being remediated now.
Feb 20, 15:47 PST
Investigating - We are investigating reports of unexpected responses from ChatGPT.
Feb 20, 15:40 PST
@Calvin6b82 ooh changed the resolved time to exactly 9 hours earlier.
I feel like someone has a friend who works on-call at OpenAI and asked them to delay resolving some incidents and delete incidents after they're resolved just to mess with us.
16th: Re API Partial outage1 hrs10 mins
No incidents or maintenance related to this downtime.
Later
Feb 16, 2024
No incidents reported today.
Seems like partial downtime but not an incident.
Which matters? why yes if no incidents reported?
I didn't bet on this day and I don't mind how it is resolved if you want to add to description that for this question incident is more widely defined to include both incidents openAI reports and also partial or complete downtime that are not incidents per openAI then I think that would be fine as a clarification.