@Horace said:
If a human can describe the logic in natural language then the LLM can code it. It seems you’re talking about the creative thinking behind generating abstractions but that’s not the same thing as generating code from well specified natural language. No matter how abstract the library, if a human can describe it unambiguously then an LLM can code it. If a human can code something then an LLM can too. Because if a human can code something then they can describe it.
True, but the hard part is not to translate logic in natural language to code, but it is in coming up with the logic. I'd argue that this is the actual coding.