This is kind of the opposite of LLMs, they seem to derive meaning mostly from the variable names, not from what the code actually does.