I was just watching this Andrew Huberman video titled "Train to Gain Energy & Avoid Brain Fog". The interviewee was talking about track athletes and stuff their coaches would have them do.
It made me think back to Anders Ericsson's book Peak: Secrets from the New Science of Expertise. The book is popular for discussing the importance of deliberate practice, but another big takeaway from the book is the importance of receiving coaching. I think that takeaway gets overlooked. Top performers in fields like chess, music and athletics almost universally receive coaching.
And at the highest levels the performers will have a team of coaches. LeBron James is famous for spending roughly $1.5 million a year on his body.
And he’s like, “Well, he’s replicated the gym that whatever team — whether it was Miami or Cleveland — he’s replicated all the equipment they have in the team’s gym in his house. He has two trainers. Everywhere he goes, he has a trainer with him.” I’m paraphrasing what he told me, so I might not be getting all these facts right. He’s got chefs. He has all the science of how to sleep. All these different things. Masseuses. Everything he does in his life is constructed to have him play basketball and to stay on the court and to be as healthy as possible and to absorb punishment when he goes into the basket and he gets crushed by people.
This makes me think about AI safety. I feel like the top alignment researchers -- and ideally a majority of competent alignment researchers -- should have such coaching and resources available to them.
I'm not exactly sure what form this would take. Academic/technical coaches? Writing coach? Performance psychologists? A sleep specialist? Nutritionist? Meditation coach?
All of this costs money of course. I'm not arguing that this is the most efficient place to allocate our limited resources. I don't have enough of an understanding of what the other options are to make such an argument.
But I will say that providing such resources to alignment researchers seems like it should pretty meaningfully improve their productivity. And if so, we are in fact funding constrained. I recall (earlier?) conversations about funding not being a constraint, rather the real constraint is that there aren't good places to spend such money.
Also relevant is that this is perhaps an easier sell to prospective donors then something more wacky. Like, it seems like a safe bet to have a solid impact, and there's a precedent for providing expert performers with such coaching, so maybe that sort of thing is appealing to prospective donors.
Finally, I recall hearing at some point that in a field like physics, the very top researchers -- people like Einstein -- have a very disproportionate impact. If so, I'd think that it's at least pretty plausible that something similar is true in the field of AI alignment. And if it is, then it'd probably make sense to spend time 1) figuring out who the Einsteins are and then 2) investing in them and doing what we can to maximize their impact.
Wow, I just watched this video where Feynman makes an incredible analogy between the rules of chess and the rules of our physical world.
You watch the pieces move and try to figure out the underlying rules. Maybe you come up with a rule about bishops needing to stay on the same color, and that rule lasts a while. But then you realize that there is a deeper rule that explains the rule you've held to be true: bishops can only move diagonally.
I'm butchering the analogy though and am going to stop talking now. Just go watch the video. It's poetic.
One thing to keep in mind is that, from what I understand, ovens are very imprecise so you gotta exercise some judgement when using them. For example, even if you set your oven to 400°F, it might only reach 325°F. Especially if you open the oven to check on the food (that lets out a lot of heat).
I've also heard that when baking on sheet pans, you can get very different results based on how well seasoned your sheet pan is. That shouldn't affect this dish though since the intent is for the top to be the crispy part and that happens via convection rather than conduction. But maybe how high or low you place the baking dish in your oven will affect the crispiness.
As another variation, I wonder how it'd come out if you used a sheet pan instead of a baking dish. I'd think that you'd get more crispy bits because of the increase in surface area of potato that is exposed to heat. Personally I'm a big fan of those crispy bits!
You'd probably need to use multiple sheet pans, but that doesn't seem like much of an inconvenience. You can also vary the crispiness by varying the amount of exposed surface area. Like, even if you use a sheet pan you can still kinda stack the potatoes on top of one another in order to reduce the exposed surface area.
I have not seen that post. Thank you for pointing me to it! I'm not sure when I'll get to it but I added it to my todo list to read and potentially discuss further here.
Scott's take on the relative futility of resolving high-level generators of disagreement (which seems to be beyond Level 7? Not sure) within reasonable timeframes is kind of depressing.
Very interesting! This is actually the topic that I really wanted to get to. I haven't been able to figure out a good way to get a conversation or blog post started on that topic though, and my attempts to do so lead me to writing this (tangential) post.
I could see that happening, but in general, no, I wouldn't expect podcast hosts to already be aware of a substantial subset of arguments from the other side.
My impression is that podcasters do some prep but in general aren't spending many days let alone multiple weeks or months of prep. When you interview a wide variety of people and discuss a wide variety of topics, as many podcasters including the ones I mentioned do, I think that means that hosts will generally not be aware of a substantial subset of arguments from the other side.
For the sake of argument, I'll accept your points about memes, genes, and technology being domains where growth is usually exponential. But even if those points are true, I think we still need an argument that growth is almost always exponential across all/most domains.
The central claim that "almost all growth is exponential growth" is an interesting one. However, I am not really seeing that this post makes an argument for it. It feels more like it is just stating it as a claim.
I would expect an argument to be something like "here is some deep principle that says that growth is almost always in proportion to the thing's current size". And then to give a bunch of examples of this being the case in various domains. (I found the examples in the opening paragraph to be odd. Bike 200 miles a week or never? Huh?) I also think it'd be helpful to point out counterexamples and spend some time commenting on them.
[This contains spoilers for the show The Sopranos.]
In the realm of epistemics, it is a sin to double-count evidence. From One Argument Against An Army:
I talked about a style of reasoning in which not a single contrary argument is allowed, with the result that every non-supporting observation has to be argued away. Here I suggest that when people encounter a contrary argument, they prevent themselves from downshifting their confidence by rehearsing already-known support.
Suppose the country of Freedonia is debating whether its neighbor, Sylvania, is responsible for a recent rash of meteor strikes on its cities. There are several pieces of evidence suggesting this: the meteors struck cities close to the Sylvanian border; there was unusual activity in the Sylvanian stock markets before the strikes; and the Sylvanian ambassador Trentino was heard muttering about “heavenly vengeance.”
Someone comes to you and says: “I don’t think Sylvania is responsible for the meteor strikes. They have trade with us of billions of dinars annually.” “Well,” you reply, “the meteors struck cities close to Sylvania, there was suspicious activity in their stock market, and their ambassador spoke of heavenly vengeance afterward.” Since these three arguments outweigh the first, you keep your belief that Sylvania is responsible—you believe rather than disbelieve, qualitatively. Clearly, the balance of evidence weighs against Sylvania.
Then another comes to you and says: “I don’t think Sylvania is responsible for the meteor strikes. Directing an asteroid strike is really hard. Sylvania doesn’t even have a space program.” You reply, “But the meteors struck cities close to Sylvania, and their investors knew it, and the ambassador came right out and admitted it!” Again, these three arguments outweigh the first (by three arguments against one argument), so you keep your belief that Sylvania is responsible.
Indeed, your convictions are strengthened. On two separate occasions now, you have evaluated the balance of evidence, and both times the balance was tilted against Sylvania by a ratio of 3 to 1.
You encounter further arguments by the pro-Sylvania traitors—again, and again, and a hundred times again—but each time the new argument is handily defeated by 3 to 1. And on every occasion, you feel yourself becoming more confident that Sylvania was indeed responsible, shifting your prior according to the felt balance of evidence.
The problem, of course, is that by rehearsing arguments you already knew, you are double-counting the evidence. This would be a grave sin even if you double-counted all the evidence. (Imagine a scientist who does an experiment with 50 subjects and fails to obtain statistically significant results, so the scientist counts all the data twice.)
I had the thought that something similar probably applies to morality as well. I'm thinking of Tony Soprano.
People say that Soprano is an asshole. Some say he is a sociopath. I'm not sure where I stand. But I finished watching The Sopranos recently and one thought that I frequently had when he'd do something harmful is that his hand was kinda forced.
For example, there was a character in the show named Adriana. Adriana became an informant to the FBI at some point. When Tony learned this, he had her killed.
Having someone killed is, in some sense, bad. But did Tony have a choice? If he didn't she very well could have gotten Tony and the rest of the mob members sent to jail, or perhaps sentenced to the death penalty. When that is the calculus, we usually don't expect the person in Tony's shoes to prioritize the person in Adriana's shoes.
It makes me think back to when I played poker. Sometimes you end up in a bad spot. It looks like you just don't have any good options. Folding seems too nitty. Calling is gross. Raising feels dubious. No move you make will end well.
But alas, you do in fact have to make a decision. The goal is not necessarily to find a move that will be good in an absolute sense. It's to make the best move relative to the other moves you can make. To criticize someone who chooses the best move in a relative sense because it is a bad move in an absolute sense is unfair. You have to look at it from the point-of-decision.
Of course, you also want to look back at how you got yourself in the bad spot in the first place. Like if you made a bad decision on the flop that put you in a bad spot on the turn, you want to call out the play you made on the flop as bad and learn from it. But you don't want to "double count" the move you made on the flop once you've moved on to analyzing the next street.
Using this analogy, I think Tony Soprano made some incredibly bad preflop moves that set himself up for a shit show. And then he didn't do himself any favors on the flop. But once he was on later streets like the turn and river, I'm not sure how bad his decisions actually were. And more generally, I think it probably makes sense to avoid "double counting" the mistakes people made on earlier streets when they are faced with decisions on later streets.
In basketball there isn't any certification. Coaches/trainers usually are former players themselves who have had some amount of success, so that points towards them being competent to some extent. There's also the fact that if you don't feel like you're making progress with a coach you can fire them and hire a new one. But I think there is also a reasonably sized risk of the coach lacking competence and certain players sticking with them anyway, for a variety of reasons.
I'm sure that similar things are true in other fields, including athletics but also in fields like chess where there isn't a degree you could get. In fields with certifications and degrees it probably happens less often, but I know I've dealt with my fair share of incompetent MDs and PhDs.
So ultimately, I agree with the sentiment that finding competent coaches might involve some friction, but despite that, it still feels to me like a very tractable problem. Relatedly, I'm seeing now that there has been some activity on the topic of coaching in the EA community.
I don't expect that the needs of alignment researchers are too unique when compared to the needs of other intellectuals. I mention alignment researchers because I think they're a prototypical example of people having large, positive impacts on the world, as opposed to intellectuals who study string theory or something.