"You can train an LLM on all the notes in classical music up to the 18th century -- but will it ever invent jazz?"
That's right. But 99% of what software devs have been paid to do is just churn out classical music notes in slightly different orders, over and over again.
There's been the very occasional breakthrough, but it could have been done by just a handful of brilliant people at key moments. The rest was just filler and AI will do it better than we have been anyway.
But most economically valuable software is not inventing jazz, it's applying well-established rules to new or neglected domains.
loading story #50567228
How do humans innovate?
They only know about things they’ve seen before, how could they ever come up with something new?
"You can train an LLM on all the hip hop up to 2000 -- but will it ever invent mumble rap?"