well, for one, closer. much, much closer. it would allow building algorithms to solve problems previously thought intractable—that sounds like an ai advance to me!
Responding to the last line: to be clear, I’m not claiming I have one. More wondering if the AI risk community should try to find one as a desperate hail mary given they have ~0 hope for their current research directions.
aka I’m wondering if trying to find one even is a desperate hail mary
I think we are a lot closer to solving alignment the normal way than that, and the problem is that understanding the landscape requires skimming a lot of papers, which most people don’t feel like doing for various reasons (a big one being that even for researchers who write and read a lot of papers, reading papers deeply is a drag).
well, for one, closer. much, much closer. it would allow building algorithms to solve problems previously thought intractable—that sounds like an ai advance to me!
But also, holy roll to disbelieve, batman!
Responding to the last line: to be clear, I’m not claiming I have one. More wondering if the AI risk community should try to find one as a desperate hail mary given they have ~0 hope for their current research directions.
aka I’m wondering if trying to find one even is a desperate hail mary
I think we are a lot closer to solving alignment the normal way than that, and the problem is that understanding the landscape requires skimming a lot of papers, which most people don’t feel like doing for various reasons (a big one being that even for researchers who write and read a lot of papers, reading papers deeply is a drag).
The search for the MOST efficient neural network from a combinatorial space of architectures and hyperparameters in polynomial time?
Yeah that would do it. Though presumably RL agents with enough training data on designing neural networks will be almost that good.