“Gavin Baker estimates that only 10 basis points (0.1%) of the population are currently using large language models, while there is already a shortage of compute.”