“The compute used to train large language models has grown by a factor of 10x every year for the last 10 years.”