So NVIDIA buys hugging face, builds hardware to power OS models, then all of a sudden the proprietary models go down and people start saying "this is why I have my Spark box"?!
Nice play NVIDIA, now, turn off the hack please, we have work to do.
They'd be very dumb to let internal engineers use exactly the same infrastructure as public users, regardless of possible benefits of dogfooding, precisely for this reason. I'm sure they must have thought about it before.
> That's however something an AI would not implement without being specifically asked.
This is true for all LLMs today, none of them work 100% autonomously, somewhere along the chain one human made at least one decision that lead to what the agent end up working on.
Funnily enough, this happened many times during AWS us-east-1 outages, during the incident internal tools were also down globally because they relied on us-east-1.
Nice play NVIDIA, now, turn off the hack please, we have work to do.