kikkupico

Code, as such, is a detailed, verifiable spec that a machine can execute. LLMs are already great at translating code to natural language and vice-versa. Why do we need a second, less detailed and less verifiable copy of the code?