Daily Microblog #4 (Stealing Compute)

This is a bulletin imported to my Blogs.

I remember stealing compute from a Korean AI art service. (https://toonsquare.co/home, they are a pretty established company in Korea.)

It was so exhilarating for me! Mostly because of how awesome tapping into the power of big compute was, normally a consumer-grade GPU would compute an image in ~12-22 seconds. While this datacenter's T4 tensor core would complete it in ~0.4-3 seconds.

https://www.nvidia.com/en-us/data-center/tesla-t4/

Although, the worst thing was that I could see how much fucking PORN people were making... On A1111 instances, you can remotely view and modify your outputs. But I was too foolish to make the mistake of showing it off on stream. I accidentally showed the power of AI to like three people...

Remember! Secure your instances; do not let me find them, for I will tune them to train all the LoRAs I want!