Perhaps due to sizes? AI/NN models before LLM were magnitudes smaller, as evident in effectively all LLMs carrying "Large" in its name regardless of relative size differences.