I thought the point would be that the number of lines of code a LLM produces to solve a given problem is... not low.