So does that mean that LLM inference could go down significantly in price and/or context length would dramatically increase?