Well, LLMs aren't just randomly spitting out garbled text either. Humans create and curate the training data, humans design the models, and humans train them.