Training LLMs is literally finding sets of numbers that make them better at mimicking human language.