Change the world how?

I was one of the very first users and developers of Bitcoin, April 2009 - Dec 2015. So for better or worse I did take part in something that changed the world. And here's the thing about that: you don't want lots of people changing the world. The world is as it is for reasons and many of the people who attempted to change the world ended up downgrading it, if not for themselves then for at least some other people. Textbook example: the long history of utopian socialist revolutions that always ended up bathed in blood.

There are relatively few world changes nowadays that everyone would agree on. That's why the utopian pitch for AI is always the same tiny handful of ideas, like curing cancer. Beyond basics like curing the last diseases, there aren't a whole lot of changes you can make that are obviously positive because most problems faced by modern societies are purely socio-political problems. They aren't problems that yield to application of intelligence. They're all about clashes of values and people struggling to impose their will on each other.

Of the few where it seems intelligence might make a difference, there are mundane reasons why it doesn't help much. Medicine is bottlenecked on clinical trials. They have far more ideas than they can check already, so generating even more won't help. Physics is bottlenecked on particle accelerator power. Ain't no LLM going to fix that. What might help is figuring out how those ideas would work out ahead of time, but that's hard because we can't simulate the body or the universe and LLMs don't have any magical solution to that either.

So the biggest impact of AI will continue to be in software, a domain where we actually are intelligence bottlenecked. LLMs can systematically drain the swampy backlogs of a gazillion badly maintained projects and that could probably absorb all the datacenter capacity being built today, if it were cheap enough. They can let us try out new platform architectures and then rewrite all the software that sits on top. Etc etc. And yeah right now some of that is limited by our imaginations and some of it is limited by cost, speed and other things that will hopefully continue to improve post-bubble.