[Jodavaho.io]

This horse/car analogy for LLMs is very weird.

Who’s the engineer here?

Are we sure this isn’t a CEO perspective where they are annoyed things can’t magically go faster and why do we have to feed these dumb engineers?

Have we considered that cars can’t break new ground and require roads to and from the destination?

LLMs have a kind of have a “mental GPS” functionality sometimes: You can always check in with an LLM in “interactive encyclopedia mode” to get a vague sense of where you are and what’s around you, but it can be off on the details quite a bit.

They are kind of like a self-driving car vs a real car: If I need to go somewhere that millions of other people have gone and do so daily, I can get there with zero thought now but I have to pay attention still.

The closest analogy is CNC vs hand milling, honestly.

The problem is that CNC has a ton of up-front design work with computer-aided drafting to make sure the damn thing is fully sketched out to the millimeter. We have no such tool for software. And no LLM is precise to the millimeter so to speak.

AI like we have now is something new. Far short of the promises, but monumentally different than other tools we’ve built. It’s neat. And everyone I know uses it for casual things, so it’s here to stay. It’s impact on engineering will be permanent but possibly not as thorough as promised.