Nice to have you back Eliezer.
This is by far the silliest part of the sequences for me. Within this blog post, Yudkowsky briefly went insane and decided thought experiments have to be “probable” or “realistic” in order to be engaged with. He then refuses to answer the prompt until the last four sentences, wherein he basically admits that he doesn’t have a framework for answering it.
Suppose someone, that someone indeed being a conscious agent, creates a GLUT and then swiftly dies a horrible death so you can stop focusing on the person who made the GLUT or how it got there and answer the damn question. Is the GLUT conscious?
- German Jew, 1920
I think the multi-hour computer hacking gauntlet probably trumps any considerations of account creation in terms of obstacles to new users. Just in considering things we could pare down. We also need some way to prevent computer hackers from scraping all of the exam boxes, and that means either being enormously creative or at some point requiring the creation of an account that we KYC.
I just launched a startup, Leonard Cyber. Basically a Pwn2Job platform.
If any hackers on LessWrong are out of work, here are some invite codes:
And this is why I think people don’t naturally do it this way. Lots of arguments have a “common body” of thought that it gets repetitive to include with each comment. Even when they don’t, people tend to just not think of arguments as “graphs” of justifications. They think of them like a serial back and forth of people on a podium giving speeches and engaging in “rhetorical battle”, and it’s more fun and engaging to write them that way on the internet.
This sounds like valuable progress in your thinking! Probably just writing about it has helped you see things more clear. I hope you’ll go back to my edited post. I shared some ideas about starting a journaling habit for precisely this reason!
Regarding learning a foreign language, I’m not sure what I can say. I speak Japanese, and I run a company which makes some top selling Japanese language learning products. So I know something about this topic. You’re right, learning a foreign language is a big commitment. So isn’t it obvious that the longer the required commitment, the most likely it is that people will drop out?
Yes. It’s very obvious. Which is why it’s that much more unusual that people make these commitments without more consideration than they tend to put forward. People will intuit that they should do some significant (meaning more than a couple hours) hard research into whether or not it’s worth it to get their masters, and where to get it; they tend not to realize that successfully learning a language as remote as Mandarin requires similar time investment as getting the masters degree, even though a necessary part of any plan to do so would be to figure that out, and I don’t know why. I suspect that many of these people are just going through the motions with no real objectives in mind, and that’s a running theme of my OP.
In the case of learning a foreign language, maybe over time the quitter just decided that the effort was no longer worth it. Maybe that want to prioritize other hobbies and interests. Or maybe they just don’t enjoy memorizing hanzi. Maybe the idea moving to China or the important of talking to people in Chinese has lost its luster. What’s wrong with changing your mind about these things?
As you say, nothing. The problem, at least in hindsight, was the decision to start in the first place. You can say the decision was correct given the information they had, but I’m trying to make the point that much (most?) of the time this is clearly not the case, and people engage in studying habits that will not ensure they learn the language in the next forty years.
Which is not to say that people shouldn’t use your product or service. I personally think language acquisition has some pretty significant positive externalities—languages are more useful the more people speak them, after all—but we don’t even need to go there. It’s just that the way most people set about learning them often implies they’re not even trying, to a degree that’s difficult to make sense of, at least for me.
Even if you take for granted that what you’re pointing out—leadership ability—is the main driver of success above and beyond all other factors in baseball management, and that the correct choice of manager is one selected for this trait to the negligence of all others, I don’t really understand why you’d even expect the Mets to succeed regularly in doing this. Are professional baseball teams really that good at identifying leadership ability as you’re implying?
Regarding the way Etsy was prioritizing development, it sounds like even their late stage “idea > validate > prototype” cycle is wrong. How can you “validate” before you have a prototype to get feedback on?
You need to be a little creative. I’ll give a trivial example that will probably be more enlightening if you went through the proofs C.S. bachelor students have to do of evaluating upper bounds of algorithms.
Lets say you hear about how matrix multiplication is a big problem in ML, and think you have a way to build an algorithm that should give you O(n*sqrt(n)) performance. Your idea happens to be an unusually complicated algorithm and thus a involves a complicated product; you think it will take you around 50 +-20 hours to prototype, and you don’t expect to achieve O(n*sqrt(n)) your first try. You could start your project now and build a prototype as quickly as possible, so that you can get a concrete impression of your algorithms viability. Or, you could first sit and consider for a couple hours and see if there are any ways to test or debunk the feasibility of your algorithm. If you choose the first option, you might get twenty hours in and realize your mistake only then. If you choose the second option, you might realize, just by pondering constraints for an hour, that matrix multiplication requires at least O(n^2) reads, and that your algorithm can’t in any universe work as fast as you expected it to, so no need to build a prototype without further considering where you’re going wrong.
Some ideas aren’t very easily validated. Art (I assume though I am not an artist) is an area where you pretty much just have a difficult to convey imagined finished product, and the best way to start validating is to build it and see if people like it. Lots of seed stage startups are founded to build things no one wants. The reason the team got funding anyways is often because the utility of the product they’re trying to create is difficult to “debunk” from the outset, or fails in hard-to-anticipate ways that aren’t caught by standard sanity checking.
However, outside of a few of those specific cases, most good ideas can be thought about. It might be impossible to completely rule out or in your hunch without building it first. But just by thinking about a proposed solution for the first twelve hours, you can usually establish some sort of probabilistic upper bound on the expected value and lower bound on the cost of implementation/maintenance. For a project that you think is going to take several months, even if you think there’s only a small chance you will figure out a flaw in its viability without prototyping, it’s worth it to sit down and do that sort of research.
None of this excludes trying to improve the source of your ideas, prototyping, talking to users, or fast iteration. All ideas are validated, though. The reason ideas come to your attention and stay there in the first place, among other things, is because they seem upon reflection plausible enough to be a good bet. The point of adding validation to that iteration process explicitly is to increase the thoroughness of your search for existing information that could encourage or discourage your efforts proportional to the costs associated with attempting.
But just because Etsy lacked some better ideas about how to optimize their product doesn’t mean they didn’t get important shit done. Within one year of its initial release, Etsy had gained 10,000 artists (craft makers), pitching 100,000 items, and had a market of approximately 40,000 buyers.
I agree, and that’s why I clarified with
And clearly they were unimaginably successful anyways. Part of my point in the post is to show how poorly you can do things on the “make sure your decisions are correct” axis and still win hard.
I realize how hyperbolic my criticism of Etsy sounds now that I needed this long ass post to explain their judgement errors, but I really think going into this much detail is making the technique seem much more complex than it really is. If you’re planning on spending two months improving the revenue created by feature X by 3%, do the napkin math to see if existing revenue coming from X justifies two months salary. If you’re planning on making a widget that improves the experience of furniture purchasers on your site, make sure there are enough furniture purchasers (taking into consideration the new ones you will be enabling by said widget) to justify maintenance/implementation. According to the guy I linked to, those were the kinds of sanity checks that Etsy wasn’t doing, which personally seems difficult to imagine for people that are otherwise obviously intensely competent.
See commenting guidelines
That’s quite an assertion. What’s your source? Evidence? Or is this mere conjecture (here, in this sacred space!)?
Everything Robin Hanson/Eliezer Yudkowsky/Richard Thaler has written + https://i.kym-cdn.com/photos/images/facebook/001/820/208/0d5.jpg
Jesus! Failing to master a foreign language is a “pathology”? Lighten up, man. People try things to see if they’ll enjoy them.
With regard to foreign languages, I disagree. Most people study foreign languages so they can use them, or at the very least most wouldn’t try if they knew from the outset that there’d be no lasting utility to doing so. Language learning usually involves large time commitments and tedious amounts of memorization. If you study a foreign language because you think it will be useful or cool, and then don’t end up speaking it to a degree that makes up for the time invested, then yes, that is a mistake in hindsight. That happens to be the experience of everyone I know, save a linguist friend of mine, that has studied a new language without moving to the country where they speak it. None of your statements are really analogous, because clearly you accomplished lots, and consider the journey worth the effort.
Have you ever run a fast-growing startup? There are a metric-shit-ton of things you could be working on. Multiple fires are burning. No one has the time or the resources to do all the things they could be doing. You also have to to look at all the things Etsy did instead of A/B testing, like working on their platform, building their company culture, fund-raising, etc. I’m running multiple e-commerce companies with a team of 20. I’ve never done a proper A/B, yet I’ve served tens of thousands of customers in 100+ countries. I could get to work right away on those A/B tests, but everything comes with opportunity costs. Times I spend on the A/B test project is time I could spend doing R&D for our new data model, mentoring other managers, doing financial analysis, or streamlining operations.
I think you mistook the A/B test thing as the only thing in that slideshow that was wrong with the way Etsy was prioritizing development for those first five years. A/B testing can absolutely be a wrong move depending on how many data points you could be collecting and what your programmers could be doing otherwise. I’m trying to build a startup right now and we don’t do A/B testing because we don’t have enough “interactions” on a regular basis to make it cost effective. But if you let your programmers start new, multi-month projects without asking quickly answered questions like “do we have enough customers who buy furniture for this feature to be worth maintaining or building”, then that’s clearly bad. It wasn’t just that Etsy was too busy preventing the house from burning down to worry about building an A/B test infrastructure, they were advancing new, spurious features without thinking about them.
I don’t follow this line-of-argument. Are you saying you didn’t know that you’d be weaker if you stopped working out? Surely you knew that. This doesn’t sound like an a matter of “having correct information,” but an issue of resolve. Motivation is complex. We can’t just reach into our brains and flip a switch—and that’s a good thing. It’s how nature prevents one part of the mind from shutting down all the other parts. Instead, when motivating one’s self, we have to rely on some of the same strategies and tactics we use when motivating others.
My point with the anecdote is that much of the time this is nonsense. I spent so much time twisting my brain into an M.C. Escher painting to try and push myself to take actions I knew were good for me, that I often forgot to try honestly bringing the costs and benefits of the thing to the forefront of my mind and seeing if that would be enough. It turns out, quite often enough it is, and even further sometimes I realize that I don’t actually value the thing I’m trying to convince myself to do enough to do it.
Most rationalists have lots of problems and believing that the biggest problem is consistently lack of willpower is the sort of blinding optimism I spoke about in the OP. I don’t see why taking incorrect actions wouldn’t make akrasia problems worse, and at best they’re neutral because they’re not advancing anything.
I’m a simple man. I see putanumonit post, I upvote.
Your tautologically efficient market is not what economists or anyone else in the world is talking about when they talk about a market being efficient. In the trivial sense you’re posing the problem you would describe your market as satisfying people’s needs even if people had parasitic worms in their brains that tricked them into thinking that mouse traps cured cancer. In the real world, people buy mousetraps because they *expect* those mouse traps to be more valuable than the price, for some definition of value that does not include “they bought the mouse trap”. A pillar of one such value system might be “effectiveness at catching mice”. In this scenario, people might be mistaken—they could have bounded rationality, they could have exploitable cognitive biases abused by advertisers, etc. The thing that makes financial markets resilient against this sort of problem is the nice property that rational actors are incentivized to take advantage of other peoples’ missteps. Mouse trap buyers cannot do this—you can’t buy a hundred mouse traps from an underrated dealer to correct the market, and you can’t short a mouse trap dealer’s traps because they are running deceptive advertising.
Somehow you and I are taking the exact opposite conclusions from that paragraph. He is explicitly noting that your interpretation is incorrect—that approximations are necessary because of time constraints. All he is saying is that underneath, the best possible action is determined by Bayesian arithmetic, even if it’s better on net to choose the approximation because of compute constraints. Just because general relativity is more “true” than newtonian mechanics, doesn’t mean that it’s somehow optimal to use it to track the trajectory of mortar fire.
Not even. An analogy there would imply you could take advantage of people’s lack of support of certain theories and collect against their underpricing of say, TDT. That would make it more efficient than the mouse trap market. Academics are normally just motivated wildly perpendicular to whatever consequences or value their work produces.