hodlbod on Nostr: I realized today that one difference between AI-generated code and human-generated ...

I realized today that one difference between AI-generated code and human-generated code is the story implicit in it.

Code is linguistic, not merely mathematical. Its meaning is more than its function — code has a purpose and a meaning outside of itself. An LLM might write "correct" code based on some validation function, but there is no "story".

Consider:

```
function doY() {}
function doX() {}
function doZ() {
doX()
doY()
}
```

The inconsistency in how the function definitions and invocations are ordered will cause no trouble to an LLM. It transparently sees the semantics, and isn't bothered by concerns unrelated to function. But this is confusing to a human, because it interrupts the structure of the narrative.

Structuring code is a discipline by and for humans, based on human aesthetics and mind. Machines need an index; humans need concepts, meaning, and purpose. It just so happens that our conceptual boxes work well as indexes, but the converse is not true — it would be very hard to navigate a codebase indexed by functionality rather than concept.

Relationships exist in the spaces between things, and are therefore mostly implicit. Likewise, the purpose of a system is embodied by the system, but is essentially external to it. LLMs cannot code in the same way humans do, because they cannot tell a story. Telling a story means caring about what is told, for its own sake as well as its purpose.