Gwern Branwen

It can't decide for you which kind of ice cream you like.

Only you can do that.

And if anything else did, it would just be worthless, basically, because it's not your particular preference.

And that's kind of the rubric for me, right?

Like, is this something that I want to do regardless of any future AI because I enjoy it?

Or is it something where I'm doing only the human part of it, maybe, and the AGI can later on do it?

Or is this writing down something that's unwritten today and thus helping kind of the future AI versions of me?

So if it doesn't fall under one of those three, I've been trying to basically not do it.

And if you look at it that way, I think many of the projects that people do right now basically have no lasting value.

They're doing things that they don't enjoy, which record nothing ephemeral kind of a value that couldn't be inferred or generated later on.

And I think they're at best kind of getting two or three years of utility out of whatever they're doing before it could have been done by an AI system.

I mean, I have ideas about how to make it possible.

Yeah.

Which might not require AGI if it kind of combined my entire corpus.

But I think many potential essay ideas are already basically mostly done in my corpus.

So you don't need to be like super intelligent to pull it out.

But, I mean, let's talk about AGI in general.

I think the anthropic timeline of 2028 seems like a good kind of personal planning starting point where even if you're wrong, you probably weren't going to do a lot of projects within the next three years anyway.

So it's not like you really lost much by instead just writing down the description.

you can always kind of go back and do it yourself later if you're wrong.