Easy, I can also make a nanoGPT run for 7 hours when inferring on a 68k, and make it produce as much value as I usually do.