Thank you for your reply! I’ll respond to both comments here.
>I find it likely that you are overestimating the difficulty of doing this.
Could you make an argument for this? I have a hard time understanding this perspective. I don’t think most people can just become agentic autodidactic elites if they want to?
>getting creative about depression
I have heard of all of these, yes. I’ve tried a few legal drugs, I don’t plan on taking illegal drugs, multiple therapists haven’t worked, I already do light therapy for other reasons as you noticed, if I’m going to do TMS I may as well do ECT, and part of my motivation for moving to the Bay is to have an in-person community I feel like I could belong in.
>research problems
I don’t think technical safety research is a particularly impactful thing to do. That would be a bet that solving every aspect of x-risk relevant AI safety can happen before the frontier labs get to AGI with mere product safety, and that doesn’t seem like a very good bet. It would, as you note, take at least a couple years to make even the smallest contribution in the first place.
I don’t believe in myself enough (I don’t have academic connections, or a large body of internalized knowledge, etc.) to think that going off in my own direction would work very well either.
>math
I do not understand any of that bounty. I did not read much of the Sequences, and I don’t have the math chops to understand the more mathier ones. Perhaps if I went and studied math for a while I would, but, is that the best use of my time, compared to e.g. trying to figure out a way to do AI safety advocacy?
>post on light glasses
Thank you : ) Most of that knowledge was not my own, it was from the VLiDACMel document, whose author actually did the research. I just implemented it in my particular case—being highly motivated to do so to fix my sleep—and then wrote up the resulting protocol, and gave some easily googleable context about the nature of sleep and related products?
The motivation to write it came from a bunch of people telling me in person they’d read an article about the light glasses if I wrote it. That is, I got a clear external signal that my output mattered. But to be in the kind of situation where I could get signals like that for AI safety, I have to first be accomplished enough for people to notice and push on me. And I didn’t get accomplishments from my background, and to make new accomplishments, I have to get an external signal… it’s a catch-22. Hence why I am skeptical of my ability to bootstrap.
I don’t think technical safety research is a particularly impactful thing to do
I spoke of that because it’s the only thing I’ve looked into at all. I suspect the lesson is generalizable to, say, technical governance or politicking (which it sounded like you think are more useful). As an example: if you live near a college, I would consider showing up to the relevant shelling point location with a plastic chair, plastic table, and whiteboard that says something like “I think superintelligent AI will kill us all, and we need to stop development immediately—Change My Mind”. e.g. my campus has a place where people tend to show up to argue about abortion or religion or whatever (though usually one of those two...)
Most of that knowledge was not my own, it was from the VLiDACMel document, whose author actually did the research. I just implemented it in my particular case—being highly motivated to do so to fix my sleep—and then wrote up the resulting protocol, and gave some easily googleable context about the nature of sleep and related products?
This sure sounds like what someone with some special capabilities would say? Plenty of (imo unusually smart and hard working!) people I know would think “impressive, I couldn’t do that—where would I even start?”. I kinda feel that way too, I suspect that if I was in your shoes I wouldn’t have ended up thinking to do that, or doing it as well.
>depression
Yeah, it looked like you were being pretty reasonable in trying to fix that. I’d still say to keep pushing even when it feels like you’ve exhausted everything, but you probably have already heard this or already do that. I still think the approach is a good one.
Thank you for your reply! I’ll respond to both comments here.
>I find it likely that you are overestimating the difficulty of doing this.
Could you make an argument for this? I have a hard time understanding this perspective. I don’t think most people can just become agentic autodidactic elites if they want to?
>getting creative about depression
I have heard of all of these, yes. I’ve tried a few legal drugs, I don’t plan on taking illegal drugs, multiple therapists haven’t worked, I already do light therapy for other reasons as you noticed, if I’m going to do TMS I may as well do ECT, and part of my motivation for moving to the Bay is to have an in-person community I feel like I could belong in.
>research problems
I don’t think technical safety research is a particularly impactful thing to do. That would be a bet that solving every aspect of x-risk relevant AI safety can happen before the frontier labs get to AGI with mere product safety, and that doesn’t seem like a very good bet. It would, as you note, take at least a couple years to make even the smallest contribution in the first place.
I don’t believe in myself enough (I don’t have academic connections, or a large body of internalized knowledge, etc.) to think that going off in my own direction would work very well either.
>math
I do not understand any of that bounty. I did not read much of the Sequences, and I don’t have the math chops to understand the more mathier ones. Perhaps if I went and studied math for a while I would, but, is that the best use of my time, compared to e.g. trying to figure out a way to do AI safety advocacy?
>post on light glasses
Thank you : ) Most of that knowledge was not my own, it was from the VLiDACMel document, whose author actually did the research. I just implemented it in my particular case—being highly motivated to do so to fix my sleep—and then wrote up the resulting protocol, and gave some easily googleable context about the nature of sleep and related products?
The motivation to write it came from a bunch of people telling me in person they’d read an article about the light glasses if I wrote it. That is, I got a clear external signal that my output mattered. But to be in the kind of situation where I could get signals like that for AI safety, I have to first be accomplished enough for people to notice and push on me. And I didn’t get accomplishments from my background, and to make new accomplishments, I have to get an external signal… it’s a catch-22. Hence why I am skeptical of my ability to bootstrap.
(only currently replying to a couple subpoints)
I spoke of that because it’s the only thing I’ve looked into at all. I suspect the lesson is generalizable to, say, technical governance or politicking (which it sounded like you think are more useful). As an example: if you live near a college, I would consider showing up to the relevant shelling point location with a plastic chair, plastic table, and whiteboard that says something like “I think superintelligent AI will kill us all, and we need to stop development immediately—Change My Mind”. e.g. my campus has a place where people tend to show up to argue about abortion or religion or whatever (though usually one of those two...)
This sure sounds like what someone with some special capabilities would say? Plenty of (imo unusually smart and hard working!) people I know would think “impressive, I couldn’t do that—where would I even start?”. I kinda feel that way too, I suspect that if I was in your shoes I wouldn’t have ended up thinking to do that, or doing it as well.
>depression
Yeah, it looked like you were being pretty reasonable in trying to fix that. I’d still say to keep pushing even when it feels like you’ve exhausted everything, but you probably have already heard this or already do that. I still think the approach is a good one.