To give my own take despite it not being much different from Rohin’s: The point of an inside view is to generalize, and the flaw of just copying people you respect is that it fails to generalize.
So for all the parts that don’t need to generalize—that don’t need to be producing thoughts that nobody has ever thought before—deferring to people you respect works fine. For this part I’m totally on board with you—I too think the inside view is overrated.
But I think it’s overratedness is circumscribed. It’s overrated when you’re duplicating other peoples’ cognitive work. But it’s definitely not overrated when you’re doing your own thinking!
The advice that I’d like new people to try (and give me feedback on) is to not worry about being able to do first-principles reasoning about AI safety in its entirety. Pick people you respect and try to go where they’re pointing. But once you’re there, try to learn what’s going on for yourself—generate an inside view centered on solving a particular piece of the problem, with tendrils extending along other directions of your interest.
Such an inside view would look like a combination of technical understanding of the details (e.g. if you want to interpret model-based RL agents, you should understand them and understand the approaches to interpreting them and what algorithms would be used), along with intuitions about the course of history in this sub-field.
To give my own take despite it not being much different from Rohin’s: The point of an inside view is to generalize, and the flaw of just copying people you respect is that it fails to generalize.
So for all the parts that don’t need to generalize—that don’t need to be producing thoughts that nobody has ever thought before—deferring to people you respect works fine. For this part I’m totally on board with you—I too think the inside view is overrated.
But I think it’s overratedness is circumscribed. It’s overrated when you’re duplicating other peoples’ cognitive work. But it’s definitely not overrated when you’re doing your own thinking!
The advice that I’d like new people to try (and give me feedback on) is to not worry about being able to do first-principles reasoning about AI safety in its entirety. Pick people you respect and try to go where they’re pointing. But once you’re there, try to learn what’s going on for yourself—generate an inside view centered on solving a particular piece of the problem, with tendrils extending along other directions of your interest.
Such an inside view would look like a combination of technical understanding of the details (e.g. if you want to interpret model-based RL agents, you should understand them and understand the approaches to interpreting them and what algorithms would be used), along with intuitions about the course of history in this sub-field.