If you don't write in the moment of inspiration it either won't get done or will turn into very boring report by committee style prose. IME.
I don't see Szabo linked:
http://unenumerated.blogspot.com/2010/09/malthusian-mystery.html
http://unenumerated.blogspot.com/2010/12/some-conjectures-and-facts-regarding.html
http://unenumerated.blogspot.com/2010/10/malthus-and-capital.html
http://unenumerated.blogspot.com/2013/11/european-asian-divergence-predates.html
http://unenumerated.blogspot.com/2013/07/a-very-underrated-invention.html
Posts like this are deceptively hard to write, so I really appreciate how well done this is.
Providing reasons feels fractal, or ship of theseus like to me. The metaphor that comes to mind is something like
Imagine two martial artists sparring, you are listening to a commentator describe the match over a radio. Two commentators would describe the match differently. In principle, a fight between two novices and a fight between two masters might sound very similar if the commentary captures a low enough resolution of events. When trying to communicate, we're something like the commentator looking directly at the mashing together of felt senses and using various mental moves to carve up the high dimensional space differently. Groups of people will fall into commentator norms to improve bandwidth, but these choices carry (usually unacknowledged) trade offs. Reification at one particular abstraction level forces a lot of structure on things that is a result of the choice of level as much as a result of the territory.
This is one of the reasons for Chapman's 'if a problem seems hard, the representation is probably wrong.' Different initial basis choices tend to push the complexity around to different parts of the model. And this process isn't even always perverse. Often the whole point is that you really can shove the uncertainty somewhere where it doesn't matter for your current purposes.
https://plato.stanford.edu/entries/structural-realism/
see also
I don't get the liar's paradox, why is 'P = ¬P' an interesting statement?
The AI Impacts piece reads like something that has a bottom line written first rather than trying to deconfuse the issue. It looks like it is aping some deconfusion patterns but always in a single direction with a single exception (awesome alphazero, which is also the most concrete. This argument should be fleshed out in more detail since it has the most factual material available).
Text is a bit laborious. Happy to chat about various simulation/multiverse parameters some time if it seems alive.
Someone had to experience the worst possible timeline that is still worth simulating. Thankfully, you volunteered. The rest of the Raemon-verse considers you a hero.
The actions of others being more optimized than one thinks is a generally good frame. There is an obvious objection that no, really, a lot of actions are un optimized serves as a curiosity stopper on considering which non conscious or non agentic process might have optimized what you're seeing.
Meaning making is a skill and you have to invest in it before you can come up with stuff that is as satisfying to you as the most popular mass market competitors. The mass market competitors are unsatisfying if you aren't their target market.
Trying to do things in the most expensive/competitive places to live is often needlessly punishing. Even if you have slack, you'll be trying to coordinate with people who don't. Plus, mimesis.
edited
I'd guess it's a classic bias-variance tradeoff. Rolling novel causal models is high variance while outside view considerations can be biased in ways you are blind to but can be good enough for coarse analysis when you just need to get the sign right.
Might be worthwhile to note that this strongly tilts towards the inside view and a suggestion for a strong counterpoint (statistical analysis of major trends that potentially gave rise to various viewpoints here).
Read as many such critiques as possible, take notes, and do iterated compression/summarization of the notes. This way you'll build your own toolbox of heuristics for evaluation that you deeply understand rather than aping the experts without really understanding.
Another reason not to integrate is that integration is actually just bad in some circumstances. You don't want all your heuristics to propagate to all possible domains all at once since they wouldn't be applicable and too many options would likely make your decision making capabilities worse. Some kinds of drug experiences demonstrate this.
I have to trade off the cost of following high complexity decision theory against the risk of being dominated*the badness of being dominated.
Great to see these points being made to a broader audience. My take from a similar investigation into science funding is that there is a common pattern to these really high impact researchers that have trouble getting funding: they're often doing methods innovation rather than object level progress on some area.* It's really hard to get grantors to understand the potential value of methods research even though it underlies scientific advancement. Big shots like the aforementioned Nobel winner, Douglas Englebart, and many others push for direct methods research only to have it seemingly fall on deaf ears even given their past accomplishments. I think part of the reason is that the benefits to major methods breakthroughs are basically unbelievable from the perspective of normal scientific work, and that people's ability to think coherently about hits based research isn't great. If we want breakthroughs the world desperately needs a billionaire who understands the value of methods work. I was really hopeful for Moskovitz to be this person given his blog posts around Asana and solving the meta problem, but have been disappointed by OpenPhil seeming to move in the direction of other foundations in terms of the range of grants they give out. What I mean by that is that glancing through their grants list, you could transplant most of them to the grants list from other foundations and no one would bat an eyelid. Thankfully there are a few exceptions, and people in methods have to take any concessions they get. The Templeton Foundation is another grantor in this space that at least has tried a little bit.
*Yes, there are arguments to be made about whether methods work is better thought of as something that could be pursued as it's own thing vs something that must generally arise out of object level work. And I'd be thrilled if that argument *was actually happening*.
(QRI is working on the consciousness meter btw ;)
Regulatory capture, in practice, means that if you circumvent the existing players they can have you arrested. Many many people are trying to figure out how to supply insulin to diabetics in the US, but no dice so far.
One of the reasons feedback feels unpleasant is when it fails to engage with what actually interests you about the area. When you receive such feedback, there will then be the feeling of needing to respond for the sake of bystanders who might otherwise assume that there aren't good responses to the feedback.
Keep in mind doctors are optimizing for patients of average ability wrt not acting insanely on their instructions. I found a lot more sympathy for people in positions of authority when I gained experience with the breath taking number of ways people can alter what seem to be very simple instructions.
If it were in person the nurse may even have smiled at him.
Ah, mimicking of the post-rigor state, and that being sufficient to get points in interactions with the pre-rigorous is what is babbly about babblers.
I think the hard reification of villagers and werewolves winds up stopping curiosity at the wrong places in the abstraction stack. Seeing agents as following mixed strategies determined by local incentives which tend to be set by super-cooperators and super-defectors seems better to me. It's also a much more tractable problem and matches what I see on the ground in orgs.
That sounds equivalent to kelly criterion, that most of your bankroll is in a low variance strategy and some proportion of your bankroll is spread across strategies with varying amounts of higher variance. Is there any existing work on kelly optimization over distributions rather than points?
edit: full kelly allows you to get up to 6 outcomes before you're in 5th degree polynomial land which is no fun. So I guess you need to choose your points well. http://www.elem.com/~btilly/kelly-criterion/
It seems like at the end of a fairly complicated construction process that if you wind up with a model that outperforms, your prior should be that you managed to sneak in overfitting without realizing it rather than that you actually have an edge right? Even if, say, you wound up with something that seemed safe because it had low variance in the short run, you'd suspect that you had managed to push the variance out into the tails. How would you determine how much testing would be needed before you were confident placing bets of appreciable size? I'm guessing there's stuff related to structuring your stop losses here I don't know about.
agree, in this situation he should state that he feels incentivized to state 70% and that that's a problem.
I don't like reifying this as dishonesty when the outside view on taking ideas seriously says that it's pretty reasonable to update slowly as you gather more kinds of evidence than just logical argument.
This suggest to me that it's a good idea to power boost people who are in the upper echelons of competence in any given domain, but to be careful to not power boost them enough that they exit the domain they are currently in and try to play in a new larger one where they are of more average competence. Sort of an anti peter principle. At least if the domain is important. For unimportant domains you probably do want to skim the competent people out and get them playing in a more important domain.
unpaid internet arguing, without the reward of seeing a change positively impact someone's life. The selection effect means you wind up interacting mostly with those who want to argue rather than collaborate.
noticing what candy crush is doing.
I may have a better answer for the concrete thing that it allows you to do: it's fully generalizing the move of un-goodharting. Buddhism seems to be about doing this for happiness/inverse-suffering, though in principle you could pick a different navigational target (maybe).
Concretely, this should show up as being able to decondition induced reward loops and thus not be caught up in any negative compulsive behaviors.
I see a 2x2 in the pattern of questions and responses.
Simple question, simple answer. Only arises to the level of intention if an idiot or a very motivated argumentative person wants to use it for something that isn't really about the original question.
Complicated question, simple answer. Everyone loves these, they make dumb people feel like they're smarter than they are.
Complicated question, complicated answer. Self limiting in the effort of the people willing to engage with it.
Simple question, complicated answer. Here is where all the problems are. Even though a satisfactory answer exists the question recurs perennially because the people who ask it haven't read any of the other long responses. People's misperceptions about it go in many directions meaning that the path to gaining understanding is idiosyncratic and a person capable of understanding the answer has to hand hold arguers through the inferences necessary. Even if such a person decides to do this, they will eventually get fed up and leave. This will be taken by people as evidence that the question does not have a good answer.
An example of a similar decomposition by Shinzen Young:
"When I hear the word mindfulness without further qualification, I don’t think of one thing. I think of eight things. More precisely, I see a sort of abstract octahedron—one body with eight facets. The eight facets are:
1. Mindfulness – The Word
2. Mindfulness – The Awareness
3. Mindfulness – The Practices
4. Mindfulness – The Path
5. Mindfulness – The Translation
6. Mindfulness – The Fad
7. Mindfulness – The Shadow
8. Mindfulness – The Possible Revolution "
Related: after extensive testing, I bought a thousand dollar laptop even though I had a perfectly good one. Why? The increase in typing speed from having a mechanical keyboard was so large that the time savings more than covers the cost. This was mildly surprising to me as I had never purchased anything for more than maybe $500 except my car.
Escaping local minima by reasoning even when local evidence should keep you in it.
Many life skills don't show benefit until they become internalized enough to be deployed organically in response to circumstance. Probabilistic reasoning, factor analysis, noticing selection effects, noticing type errors, etc. are 'rationalist' examples of this, but it applies to many if not most skills.
examine.com
The relationships between maps is a neglected source of insight in my experience. Indeterminacy of translation points to why. If you don't examine them, you have tacit links between maps (eg the heuristic that determines when to switch between them). These tacit links aren't necessarily built skillfully by default.
Levels of Analysis is Marr's take on this problem.
To bridge from Adams' systems not goals: a good system regularly outputs updated plans to achieve intermediate goals that preserve or expand option value given the observed and hypothesized variance in your goals. This often looks like plans to test key assumptions in your big goals/directions/navigational tools, or deliberate practice of a skill that is useful for multiple goals.
Levels of Analysis points to relationships between maps as neglected. Indeterminacy of translation.
I think my sense of miscommunication with you is that you don't seem to have a sense of the law of equal and opposite advice + meta-contrarianism. Different things seem useful at different stages, and principle of charity means at least trying to see why what people are saying might be useful from their perspective.
Guess: human values reflect beliefs about the modularity of reality. A necessary component of the counterfactual simulator.
The counterfactual simulator, in turn, seems to be about convex optimization of tradeoff space.
Yeah, pointing at the same stuff. That clarification helped.
I'm saying that this post itself is falling prey to the thing it advises against. Better to point at a cluster that helps navigate, like Hanson's babblers than to talk about the information theoretic content of aggregate clusters.
Most valuable IMO is the idea that relational practices expose shadow sides for processing that individual practice doesn't.
I have problems with much of his stuff due to having the 'look how much more inclusive my metaphysics is' problem where the framework gives you more degrees of freedom than the phenomenon being explained, allowing you to cold read yourself. This is covered in technical explanation of technical explanations. You want your framework to have fewer degrees of freedom than the system it describes (compression), that's where your predictive constraints come from.
Thanks for the clear suggestions/feedback.
The tacit claim is that LW should be about confirmatory research and that exploratory research doesn't belong here. But confirmatory, cited research has never been the majority of content going back to LW 1.0.