Comment by weregiraffe

Let's be accurate. LLMs are large text-corpus models. The texts are encoded as tokens, but that's just implementation detail.