OpenAI has unfortunately moved from 3 9s of uptime in October to 2 in November, and now only a single 9 for December. I've loved their 12 Days of OpenAI -- it's incredible to see such a large team maintain that product velocity, but this level of reliability is untenable for large organizations. When reliability enters a tailspin like this, it can be challenging to pull out. Wishing the OpenAI team the best in improving this; hopefully it's a serious priority for leadership.
Do you think enterprises are relying on OpenAI vs a self managed version in Azure?
I appreciate their transparency at this point. If the numbers improve too quickly, I wouldn't believe it was a change in infrastructure as much as a change in reporting policy.
What a shame, good thing that there are many other alternatives which are cheaper!
On device AI has around 6 9s of uptime. Do you believe more companies will consider this?
Technically there are still 2 9s in December so far, just not in the ones place you are looking at.
Factor in the absurdly large (but inconsistently sized) prompts that are going to make scaling extraordinarily difficult and unpredictable. This architecture might come crashing down in flames soon enough.
There’s more to the AI world than just OpenAI. I wouldn’t put all my eggs in one basket, especially considering there are other models on the market that can serve as backups when OpenAI experiences hardware failures or overloads.
Co-founder & CEO @ Parity (YC S24) | MIT CS
1dUpdate: OpenAI released a retro on the incident that I break down here https://www.linkedin.com/posts/wilsonspearman_openai-published-a-public-incident-report-activity-7274537493952143361-E-EC