TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Too much efficiency makes everything worse (2022)

917 点作者 feyman_r8 个月前

87 条评论

refibrillator8 个月前
I recognize the author Jascha as an incredibly brilliant ML researcher, formerly at Google Brain and now at Anthropic.<p>Among his notable accomplishments, he and coauthors mathematically characterized the propagation of signals through deep neural networks via techniques from physics and statistics (mean field and free probability theory). Leading to arguably some of the most profound yet under-appreciated theoretical and experimental results in ML in the past decade. For example see “dynamical isometry” [1] and the evolution of those ideas which were instrumental in achieving convergence in very deep transformer models [2].<p>After reading this post and the examples given, in my eyes there is no question that this guy has an extraordinary intuition for optimization, spanning beyond the boundaries of ML and across the fabric of modern society.<p>We ought to recognize his technical background and raise this discussion above quibbles about semantics and definitions.<p>Let’s address the heart of his message, the very human and empathetic call to action that stands in the shadow of rapid technological progress:<p><i>&gt; If you are a scientist looking for research ideas which are pro-social, and have the potential to create a whole new field, you should consider building formal (mathematical) bridges between results on overfitting in machine learning, and problems in economics, political science, management science, operations research, and elsewhere.</i><p>[1] Dynamical Isometry and a Mean Field Theory of CNNs: How to Train 10,000-Layer Vanilla Convolutional Neural Networks<p><a href="http:&#x2F;&#x2F;proceedings.mlr.press&#x2F;v80&#x2F;xiao18a&#x2F;xiao18a.pdf" rel="nofollow">http:&#x2F;&#x2F;proceedings.mlr.press&#x2F;v80&#x2F;xiao18a&#x2F;xiao18a.pdf</a><p>[2] ReZero is All You Need: Fast Convergence at Large Depth<p><a href="https:&#x2F;&#x2F;arxiv.org&#x2F;pdf&#x2F;2003.04887" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;pdf&#x2F;2003.04887</a>
评论 #41685494 未加载
评论 #41685144 未加载
评论 #41685600 未加载
评论 #41685275 未加载
评论 #41741681 未加载
评论 #41687384 未加载
评论 #41700970 未加载
评论 #41685060 未加载
t_mann8 个月前
The argument rides on the well-known Goodhart&#x27;s law (<i>when a measure becomes a target, it ceases to be a good measure</i>). However, it only puts it down to measurement problems, as in, we can&#x27;t measure the things we really care about, so we optimize some proxies.<p>That, in my view, is a far too reductionist view of the problem. The problem isn&#x27;t just about measurement, it&#x27;s about human behavior. Unlike particles, humans will actively seek to exploit any control system you&#x27;ve set up. This problem goes much deeper than just not being able to measure &quot;peace, love, puppies&quot; well. There&#x27;s a similar adage called Campbell&#x27;s law [0] that I think captures this better than the classic formulation of Goodhart&#x27;s law:<p><i>The more any quantitative social indicator is used for social decision-making, the more subject it will be to corruption pressures and the more apt it will be to distort and corrupt the social processes it is intended to monitor.</i><p>The mitigants proposed (regularization, early stopping) address this indirectly at best and at worst may introduce new quirks that can be exploited through undesired behavior.<p>[0] <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Campbell%27s_law" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Campbell%27s_law</a>
评论 #41686989 未加载
评论 #41688796 未加载
评论 #41689302 未加载
评论 #41685592 未加载
whizzter8 个月前
This has become a societal problem in Sweden during the past 20 or so years.<p>1: Healthcare efficiency is measured by &quot;completed tasks&quot; by primary care doctors, the apparatus is optimized for them handling simple cases and they thus often do some superficial checking and either send one home with some statistically correct medicine (aspirin&#x2F;antibiotics) or punt away cases to a specialized doctor if it appears to be something more complicated.<p>The problem is that since there&#x27;s now fewer of them (efficient) they&#x27;ve more or less assembly line workers and have totally lost the personal &quot;touch&quot; with patients that would give them an indication on when something is wrong. Thus cancers,etc are very often diagnosed too late so even if specialized cancer care is better, it&#x27;s often too late to do anything anyhow.<p>2: The railway system was privatized, considering the amount of cargo shipped it&#x27;s probably been a huge success but the system is plagued by delays due to little gaps in the system to allow late trains to speed up or to even do more than basic maintenance (leading to bigger issues).
评论 #41705909 未加载
评论 #41689285 未加载
remram8 个月前
Those are great points! Another related law is from queuing theory: waiting time goes to infinity when utilization approaches 100%. You need your processes&#x2F;machines&#x2F;engineers to have some slack otherwise some tasks will wait forever.
评论 #41684905 未加载
评论 #41684807 未加载
评论 #41684717 未加载
评论 #41685126 未加载
评论 #41684877 未加载
评论 #41685017 未加载
评论 #41686284 未加载
评论 #41689619 未加载
netcan8 个月前
Another example of this approximate law is in exercise physiology.<p>To a normal person, there are a lot of good proxy indicators of fitness. You could train sprinting. You could hop up and down. Squat. Clean and jerk.. etc.<p>Running faster,hopping higher, squatting heavier... all indicators of increasing fitness... and success of your fitness training.<p>Two points:<p>1 - The more general your training methodology, the more meaningful the indicators. Ie, if your fitness measure is &quot;can I push a car uphill,&quot; and your training method is sprinting and swimming... pushing a heavier car is a really strong indicator of success. If your training method is &quot;practice pushing a car,&quot; then an equivalent improvement does not indicate equivalent improvement in fitness.<p>2- As an athlete (say clean and jerk) becomes more specialized... improvements in performance become <i>less</i> indicative of general fitness. Going from zero to &quot;recreational weighlifter&quot; involves getting generally stronger and musclier. Going from college to olympic level... that typically involves highly specialized fitness attributes that don&#x27;t cross into other endeavors.<p>Another metaphor might be &quot;base vs peak&quot; fitness, from sports. Accidentaly training for (unsustainable) peak performance is another over-optimization pitfall. It can happen when someone blindly follows &quot;line go up.&quot; Illusary optimizations are actually just trapping you in a local maxima.<p>I think there are a lot of analogies here to biology, but also ML optimization and social phenomenon.
评论 #41686986 未加载
评论 #41689210 未加载
bilsbie8 个月前
This is why I don’t like focusing on GDP. I think a quarterly poll on life satisfaction and optimism would be a better measure.<p>If you’re curious about GDP. I my car breaks and I get it fixed, that adds to GDP.<p>If a parent stays home to raise kids, that lowers GDP. If I clean my own house that lowers GDP. Etc.<p>Unemployment is another crude metric. Are these jobs people want or do they feel forced to work bad jobs.
评论 #41687695 未加载
评论 #41687688 未加载
评论 #41688282 未加载
评论 #41687368 未加载
LarsDu888 个月前
I was trying to remember where I remember where I heard of this author&#x27;s name before.<p>Invented the first generative diffusion model in 2015. <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;1503.03585" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;1503.03585</a>
评论 #41685634 未加载
redsparrow8 个月前
This makes me think of going to chain restaurants. Everything has been focus-grouped and optimized and feels exactly like an overfit proxy for an enjoyable meal. I feel like I&#x27;m in a bald-faced machine that is optimized to generate profit from my visit. The fact that it&#x27;s a restaurant feels almost incidental.<p>&quot;HI! My name is Tracy! I&#x27;m going to be your server this evening!&quot; as she flawlessly writes her name upside down in crayon on the paper tablecloth. Woah. I think this place needs to re-calibrate their flair.
usaphp8 个月前
I think it also applies to when managers try to overoptimize work process, in the end creative people lose interest and work becomes unbearable...little chaos is necessary in a work place&#x2F;life imo...
评论 #41684997 未加载
thomassmith658 个月前
I noticed an example of this rule at my local hardware superstore.<p>Around a decade ago, the store installed anti-theft cages.<p>At first they only kept high-dollar items in the cages. It was a bit of an inconvenience, but not so bad. If a customer is dropping $200+ on some fancy power tool, he or she likely doesn&#x27;t mind waiting five minutes.<p>But a few years later, there was a change - almost certainly a &#x27;data-driven&#x27; change: suddenly there was no discernible logic to which items they caged and which they left uncaged. Now a $500 diagnostics tool is as likely to sit open on a shelf, as a $5 light bulb to be kept under lock and key.<p>Presumably the change is a result of sorting a database by &#x27;shrinkage&#x27; - they lock up the items that <i>cumulatively</i> lose the hardware store the most money, due to theft.<p>But the result is (a) the store atmosphere reads as &quot;so profit-driven they don&#x27;t trust the customers not to steal a box of toothpicks&quot; and (b) it&#x27;s often not worth it for customers to shop there due to the waiting around for an attendant to unlock the cage.<p>I doubt the optimization helped their bottom-line, even if it has prevented the theft of some $3 bars of soap.
评论 #41688998 未加载
评论 #41689046 未加载
评论 #41688991 未加载
jrochkind18 个月前
Calling this the &quot;strong version of Goodhart&#x27;s law&quot; was immediately brain-expanding for me.<p>I have been thinking of goodhart&#x27;s law a lot, but realized I had been leaning toward focusing on human <i>reaction</i> to the metric as the cause of it; but this reminded me it&#x27;s actually fundamentally about the fact that <i>any</i> metric itself is inherently not an exact representation of the quality you wish to represent.<p>And that this may, as OP argues, make goodhart&#x27;s law fundamental to pretty much any metric used as a target. Independently of how well-intentioned any actors. It&#x27;s not a result of like human laziness or greed or competing interests, it&#x27;s an epistemic (?) result of the neccesary imperfection of metric validity.<p>This makes some of OP&#x27;s more contentious &quot;Inject noise into the system&quot; and &quot;early stopping&quot; ideas more interesting even for social targets.<p>&quot;The more our social systems break due to the strong version of Goodhart&#x27;s law, the less we will be able to take the concerted rational action required to fix them.&quot;<p>Well, that&#x27;s terrifying.
raister8 个月前
This reminds me of Eli Goldratt&#x27;s quote: &quot;Tell me how you measure me, I will tell you how I behave.&quot;
评论 #41685672 未加载
评论 #41684630 未加载
inglor_cz8 个月前
When it comes to his <i>Mitigation: Inject noise into the system.</i> proposal: I would be happy to experiment with some sortition in our political systems. Citizens&#x27; assemblies et cetera.<p>Randomly chosen deliberative bodies could keep some of the stupid polarization in check, especially if your chances to be chosen twice into the same body are infinetesimal.<p><a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Sortition" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Sortition</a><p>We tend to consider &quot;democracy&quot; as fundamentally equivalent to &quot;free and fair elections&quot;, but sortition would be another democratic mechanism that could complement our voting systems. Arguably <i>more</i> democratic, as you need money and a support structure to have a chance to win an election.
orcim8 个月前
It&#x27;s an effect that exists, but the examples aren&#x27;t accurate.<p>An overemphasis on grades isn&#x27;t from wanting to educate the population; obesity isn&#x27;t from prioritizing nutrient-rich food; and increased inequality isn&#x27;t from wanting to distribute resources based on the needs of society.<p>Living a well-lived life through culture, cooking, or exercise doesn&#x27;t make you more susceptible to sensationalism, addiction, or gambling. It&#x27;s a lack of stimulus that makes you reach for those things.<p>You can argue that academia enables rankings, industrial food production enables producing empty calories, and economic development enables greater inequality. But that isn&#x27;t causation.<p>It also isn&#x27;t a side effect when significant resources specifically go into promoting education as a private matter best used to educate the elite, that businesses aren&#x27;t responsible for the externalities they cause, and that resources should be privately controlled.<p>In many ways, it is far easier to have more public education, heavily tax substances like sugar, and redistribute wealth than it is to do anything else. That just isn&#x27;t the goal. It used to be hard to get a good education, good food, and a good standard of living. And it still is. For the same reasons.
dooglius8 个月前
Overfitting may be a special case of Goodhart&#x27;s Law, but I don&#x27;t think Goodhart&#x27;s Law in general is the same as overfitting, so I don&#x27;t think the conclusion is well-supported supported in general; there may be plenty of instances of proxy measures that do not have issues.<p>I&#x27;ll also quibble with the example of obesity: the proxy isn&#x27;t nutrient-rich good, but rather the evaluation function of human taste buds (e.g. sugar detection). The problem is the abundance of food that is very nutrient-poor but stimulating to taste buds. If the food that&#x27;s widely available were nutrient-rich, it&#x27;s questionable whether we would have an obesity epidemic.
评论 #41684636 未加载
评论 #41685845 未加载
评论 #41684989 未加载
rowanG0778 个月前
I don&#x27;t think it&#x27;s unintuitive at all. 100% optimized means 100% without slack. No slack means any hitch at all will destroy you.
评论 #41684593 未加载
评论 #41685384 未加载
评论 #41685069 未加载
abernard18 个月前
The author identifies problems with a system measuring targets, but then all the proposals are about increasing the power and control of the system.<p>Perhaps the answer—as hippy sounding as it is—is to reduce the control of the system outright. Instead of adding more measures, more controls, which are susceptible to the prejudices of control, we let the system fall where it may.<p>This, to me, is a classic post of an academic understanding the failures of a system (and people like themselves in control of said system) but then not allowing the mitigation mechanisms of alternate systems to take its place.<p>This is one of the reasons I come to HN: to view the prime instigators of big-M Modern failure and their inability to recognize their contributions to that problem.
评论 #41688860 未加载
评论 #41691257 未加载
projektfu8 个月前
And that&#x27;s leaving out Jevon&#x27;s paradox, where increasing efficiency in the use of some scarce resource sometimes&#x2F;often increases its consumption, by making the unit price of the dependent thing affordable and increasing its demand. For example, gasoline has limited demand if it requires ten liters to go one km, but very high demand at 1 L&#x2F;10km, even at the same price per liter.
评论 #41685035 未加载
whatever18 个月前
When we optimize we typically have a specific scenario in our head. With the proper tools one can probably make the mathematically optimal decisions to deal with this exact scenario.<p>However: 1) This exact scenario will likely never materialize 2) You have not good quantification of the scenario anyway due to noise&#x2F;biases in measurements.<p>So now you optimized for something very specific, and the nature throws you something slightly different and you are completely screwed because your optimized solve is not flexible at all.<p>That is why a more “suboptimal” approach is typically better and why our stupid brains outperform super fancy computers and algorithms in planning.
smokel8 个月前
I was listening to an episode of the &quot;inControl&quot; podcast [1], in which Ben Recht suggested that overfitting is not always well understood.<p>Perhaps it is interesting to read his blogpost &quot;Machine Learning has a validity problem&quot; alongside this article.<p>[1] <a href="https:&#x2F;&#x2F;www.incontrolpodcast.com&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.incontrolpodcast.com&#x2F;</a><p>[2] <a href="https:&#x2F;&#x2F;archives.argmin.net&#x2F;2022&#x2F;03&#x2F;15&#x2F;external-validity&#x2F;" rel="nofollow">https:&#x2F;&#x2F;archives.argmin.net&#x2F;2022&#x2F;03&#x2F;15&#x2F;external-validity&#x2F;</a>
tpoacher8 个月前
There was no need to invent a new law named &quot;strong version&quot;, it already exists: Campbell&#x27;s law.<p>The subtle difference between the two being exactly what the author describes: Goodhart&#x27;s law states that metrics eventually don&#x27;t work, Campbell&#x27;s law states that, worse still, eventually they tend to backfire.
cb3218 个月前
I do like it when researchers try to connect the deep ideas within their work to broader more general systems, but caution is warranted to the optimism. This article is the kind of formal analogy that inspired&#x2F;drove much of the marketing appeal of the Santa Fe Institute back in the 1980s. It&#x27;s honestly always pretty fun, but the devil is usual in the details here (as is usual in making anything &quot;work&quot;, such as self-organized criticality [1] which if you enjoyed this article you will also probably enjoy!).<p>As just one example to make this point more concrete (LOL), the article mentions uncritically that &quot;more complex ecosystems are more stable&quot;, but over half a century ago in 1973 Robert May wrote a book called &quot;Stability and Complexity in Model Ecosystems&quot; [2] explaining (very accessibly!) how this is untrue for the easiest ideas of &quot;complex&quot; and &quot;stable&quot;. In more human terms, some ideas of &quot;complex&quot; &amp; &quot;stable&quot; can lead you astray, as has been appearing in the relatively nice HN commentary on this article here.<p>Perhaps less shallowly, things go off the rails fast once you have both multiple metrics (meaning no &quot;objective Objective&quot;) and competing &amp; intelligent agents (meaning the system itself has a kind of intrinsic complexity, often swept under the rug by a simplistic thinking that &quot;people are all the same&quot;). I think this whole topic folds itself into &quot;Humanity Complete&quot; (after NP-complete.. a kind of infectious cluster of Wicked Problems [3]) like trust&#x2F;delegation do [4].<p>[1] <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Self-organized_criticality" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Self-organized_criticality</a><p>[2] <a href="https:&#x2F;&#x2F;press.princeton.edu&#x2F;books&#x2F;paperback&#x2F;9780691088617&#x2F;stability-and-complexity-in-model-ecosystems" rel="nofollow">https:&#x2F;&#x2F;press.princeton.edu&#x2F;books&#x2F;paperback&#x2F;9780691088617&#x2F;st...</a><p>[3] <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Wicked_problem" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Wicked_problem</a><p>[4] <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Demarcation_problem" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Demarcation_problem</a>
kzz1028 个月前
I think of efficiency as one example where naive economic thinking has poisoned common sense. Economists view inefficiency as a problem. Because a healthy economy is efficient, therefore inefficiency is unhealthy. Any inefficient market is a &quot;market failure&quot;. Efficiency is also the primary way a manager can add value. But the problem is, efficiency assumes existence of metrics, and indeed is counter productive if your metrics are wrong.
评论 #41688102 未加载
gond8 个月前
“Though this pheonomenon is often discussed, it doesn&#x27;t seem to be named. Let&#x27;s call it the strong version of Goodhart&#x27;s law“<p>I wonder why the author called it that way when this seems to me clearly derived from Ross Ashby‘s law of Requisite Variet[1], predating Goodhard by 20 years. As I see it, it is not even necessary to put more meaning it Goodhard as there actually is. Requisite Variety is sufficient. Going by his resume, I strongly assume the author knows this.<p>Russel Ackoff, building on countless others, put into two sentences for which others needed two volumes:<p>“The behaviour of a system is never equal to the behaviour of its parts. - It is the product of their interactions.“<p>[1] <a href="https:&#x2F;&#x2F;en.m.wikipedia.org&#x2F;wiki&#x2F;Variety_(cybernetics)" rel="nofollow">https:&#x2F;&#x2F;en.m.wikipedia.org&#x2F;wiki&#x2F;Variety_(cybernetics)</a>
评论 #41686274 未加载
otherme1238 个月前
&gt; Goal: Distribution of labor and resources based upon the needs of society<p>&gt; Proxy: Capitalism<p>&gt; Strong version of Goodhart&#x27;s law leads to: Massive wealth disparities (with incomes ranging from hundreds of dollars per year to hundreds of dollars per second), <i>with more than a billion people living in poverty</i><p>Please, show me a point in all human history when we have less than 90% global population living in poverty, pre-capitalism. Yes, there are 1 billion people (out of 8 billion) living in poverty today. But they were 2 billion (of 4.5 billion total) living in poverty as recently as 1980 (<a href="https:&#x2F;&#x2F;www.weforum.org&#x2F;agenda&#x2F;2016&#x2F;01&#x2F;poverty-the-past-present-and-future&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.weforum.org&#x2F;agenda&#x2F;2016&#x2F;01&#x2F;poverty-the-past-pres...</a>).<p>Poverty is steadily going down (<a href="https:&#x2F;&#x2F;www.weforum.org&#x2F;agenda&#x2F;2016&#x2F;01&#x2F;poverty-the-past-present-and-future&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.weforum.org&#x2F;agenda&#x2F;2016&#x2F;01&#x2F;poverty-the-past-pres...</a>) since we have data. The first countries to get rid of recurrent famines were the same that first adopted capitalism. The same countries where their population started having higher expectations than to live another day.<p>Paraphrasing Churchill about democracy, &quot;[capitalism] is the worst economic system except for all other systems that has been tried from time to time&quot;.
评论 #41688215 未加载
zmmmmm8 个月前
Maybe I&#x27;m misunderstanding this but this doesn&#x27;t seem like an accurate explanation of overfitting:<p>&gt; In machine learning (ML), overfitting is a pervasive phenomenon. We want to train an ML model to achieve some goal. We can&#x27;t directly fit the model to the goal, so we instead train the model using some proxy which is similar to the goal<p>One of the pernicious aspects of overfitting is it occurs even if you <i>can</i> perfectly represent your goal via a training metric. In fact it&#x27;s even worse simetimes as an incorrect training metric can indirectly help regularise the outcome.
评论 #41685962 未加载
评论 #41692127 未加载
lynguist8 个月前
I would claim in a completely informal way that the optimal degree of utilization is ln(2)=0.693, around 70%.<p>This stems from the optimal load of self-balancing trees.<p>A little bit of slack is always useful to deal with the unforeseen.<p>And even a lot of slack is useful (though not always as it is costly) as it enables to do things that a dedicated resource cannot do.<p>On the other hand, no slack at all (so running at above 70%) makes a system inflexible and unresilient.<p>I would argue for this in any circumstance, be it military, be it public transit, be it funding, be it allocation of resources for a particular task.
评论 #41692424 未加载
评论 #41686226 未加载
hedora8 个月前
I don’t think the author understands what efficiency measures.<p>All of the examples involve a bad proxy metric, or the flawed assumption that spending less improves the ratio of price to performance.
评论 #41684634 未加载
评论 #41684582 未加载
评论 #41685206 未加载
评论 #41684599 未加载
评论 #41685059 未加载
评论 #41685355 未加载
2grue8 个月前
This is super interesting. The author states &quot;the strong version of Goodhart&#x27;s law&quot; as a fact, but does not provide a theorem which shows that it is true. This recent paper does the job.[0] The authors write about Goodhart&#x27;s law in the context of AI alignment but they are clear that their theorem is much more broadly applicable:<p>&gt; we provide necessary and sufficient conditions under which indefinitely optimizing for any incomplete proxy objective leads to arbitrarily low overall utility<p>&gt; Our main result identifies conditions such that any misalignment is costly: starting from any initial state, optimizing any fixed incomplete proxy eventually leads the principal to be arbitrarily worse off.<p>[0]: <a href="https:&#x2F;&#x2F;proceedings.neurips.cc&#x2F;paper&#x2F;2020&#x2F;hash&#x2F;b607ba543ad05417b8507ee86c54fcb7-Abstract.html" rel="nofollow">https:&#x2F;&#x2F;proceedings.neurips.cc&#x2F;paper&#x2F;2020&#x2F;hash&#x2F;b607ba543ad05...</a>
RandomLensman8 个月前
Efficiency is usually easier than effectiveness so it is optimised for much more and that spills over in the results and outcomes, of course.
teleforce8 个月前
There is a big difference between efficiency and effectiveness, and all system should focus on the latter rather than the former whether it&#x27;s AI based or not.<p>There&#x27;s a reason why the best-seller of self-help book for several decades now is the book by Stephen Covey entitled &quot;The 7 Habits of Highly Effective People&quot; not &quot;Efficient People&quot;.
godelski8 个月前
I find this article a bit odd, considering what the author is an expert in: generative imagery. It&#x27;s the exact problem he discusses, the lack of a target that is measurable. Defining art is well known to be ineffable, yet it is often agreed upon. For thousands of years we&#x27;ve been trying to define what good art means.<p>But you do not get good art by early stopping, you do not get it by injecting noise, you do not get it by regularization. All these do help and are essential to our modeling processes, but we are still quite far. We have better proxies than FID but they all have major problems and none even come close (even when combined).<p>We&#x27;ve gotten very good at AI art but we&#x27;ve still got a long way to go. Everyone can take a photo, but not everyone is a photographer and it takes great skill and expertise to take such masterpieces. Yet there are masters of the craft. Sure, AI might be better than you at art but that doesn&#x27;t mean it&#x27;s close to a master. As unintuitive as this sounds. This is because skill isn&#x27;t linear. The details start to dominate as you become an expert. A few things might be necessary to be good, but a million things need be considered in mastery. Because mastery is the art of subtly. But this article, it sounds like everything is a nail. We don&#x27;t have the methods yet and my fear is that we don&#x27;t want to look (there are of course many pursuing this course, but it is very unpopular and not well received. Scale is all you need is quite exciting, but lacking sufficient complexity, which even Sutton admits to be necessary). It&#x27;s my fear that we get too caught up in excitement that we become blind to our limitations. Because it&#x27;s knowing those limitations that is what gives us direction to improve upon. When every critique is seen as spoiling the fun of the party, we&#x27;ll never be able to have anything better. I&#x27;m not trying to stop the party, in fact, I&#x27;m worried it&#x27;ll stop.
评论 #41685345 未加载
评论 #41685179 未加载
naitgacem8 个月前
Upon reading the title at first glance, I thought this was going to be how &quot;effecient&quot; computers nowadays. Such as MacBooks and such, who started this efficient computers thing in the recent times. And they are, but as a result computers are all worse off for it. I mean soldered RAMs and everything is a system on a chip.
评论 #41685942 未加载
评论 #41686296 未加载
slashdave8 个月前
I am surprised that the author left out another mitigation. To build solutions (models) that are constructed to be more transferable (amenable to out of domain data). For example, in machine learning, using physics informed models. Perhaps this is simply a sign that the author is a proponent of generic, deep-learning.
评论 #41688186 未加载
aswanson8 个月前
You can also be too efficient in your career&#x2F;life. You can take the &quot;Inject noise into the system&quot; as injecting positive randomness into your associations with people and ideas. If something seems slightly interesting but off your beaten track, learn more about it.
js88 个月前
I have a pet theory that the state-planned economies failed not because they were inefficient (as neoclassical economics claims), but rather because they attempted to be too efficient. They tried to exactly calculate which producer needs what inputs, what they should produce and when, and a little deviation from the plan caused big cascading failures.<p>Free market is actually less efficient than direct control, but it is correspondingly more robust. This is evidenced in the big companies, which also sometimes try to control things in the name of &quot;efficiency&quot; and end up being quite inefficient. And also small companies, which are often competing and duplicating efforts.<p>The optimum (I hesitate it call that because it&#x27;s not well-defined, it is in some sense a society&#x27;s choice) seems to be somewhere in the middle - you need decent amount of central direction (almost all private companies have that) and redundancy (provided by investment funds on the free market).<p>(As aside, despite me being democratic socialist, I don&#x27;t believe the democracy matters that much for economic development, but is desirable from a moral perspective. You can have a lot of economic development under authoritarian rule, there are examples on both sides, as most private companies are also actually small authoritarian fiefdoms.)
评论 #41687011 未加载
评论 #41686557 未加载
评论 #41686534 未加载
RadiozRadioz8 个月前
This is more a meta comment about the blog itself (as is customary for HN): I like the blog, there has been a lot of work put into it, so it makes me sad that it&#x27;s hosted on GitHub pages using a subdomain of GitHub.io. When the day comes that GitHub inevitably kills&#x2F;ruins Pages, because it _will_ happen, there is no question, the links to this blog will be stuck forever pointing to this dead subdomain that the author has no control over. We just have to hope that the replacement blog is findable via search engines, and hope that comments are enabled wherever the pages link is referenced so that new people can find the blog. An unfortunate mess that is definitely going to happen, entirely Microsoft&#x27;s fault.
评论 #41688732 未加载
评论 #41685561 未加载
tikkun8 个月前
Makes me think of: some of Taleb&#x27;s ideas about just-in-time manufacturing (no slack eg covid supply shocks)<p><a href="https:&#x2F;&#x2F;www.lesswrong.com&#x2F;posts&#x2F;yLLkWMDbC9ZNKbjDG&#x2F;slack" rel="nofollow">https:&#x2F;&#x2F;www.lesswrong.com&#x2F;posts&#x2F;yLLkWMDbC9ZNKbjDG&#x2F;slack</a><p>Also, can&#x27;t recall it but a long time ago I read a piece about how scheduling a system to 60% of its max capacity is generally about right, to allow for expected but unexpected variations (also makes me think of the concept of stochastic process control and how we can figure out the level of expected unexpected variations, which could give us an even better sense of what %-of-capacity to run a system at)
leoc8 个月前
Related: &quot;Dodo-Lean&quot; by Darrell Mann <a href="https:&#x2F;&#x2F;www.darrellmann.com&#x2F;dodo-lean&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.darrellmann.com&#x2F;dodo-lean&#x2F;</a> , about systems which have been optimised into fragility.
ocean_moist8 个月前
Metrics are ambiguous because they are abstractions of success and miss context. If you want a pretty little number, it doesn’t come without cost&#x2F;missing information.<p>I don’t know if this phenomenon is aptly characterized as “too much efficiency”.
chrisweekly8 个月前
Efficiency tends to come at the cost of adaptability. Don&#x27;t put it on rails if it needs a steering wheel. So many enterprises suffer from extreme rigidity - often caused by optimizations that lead to local maxima.
michaelcampbell8 个月前
This hits home. Management has started measuring things at my workplace. They won&#x27;t admit it, but I said from the start it was because they are easy to measure, not because they are useful.
bdjsiqoocwk8 个月前
In practice this is just Goodhart&#x27;s law itself. It&#x27;s not distinct. In Goodhart&#x27;s law<p>&gt; when a measure becomes a target, it ceases to be a good measure<p>If you ask someone &quot;could you give me an example&quot; you will see that in the example the measure that becomes a target is already a proxy. Even the example that the author presents, the good that cares a lot about testing its students... How does the school test its students? With exams. But that&#x27;s already a proxy for testing students knowledge...<p>But overall excellent article.
numbol8 个月前
There is a book on this topic, &quot;Why Greatness Cannot Be Planned&quot; <a href="https:&#x2F;&#x2F;link.springer.com&#x2F;book&#x2F;10.1007&#x2F;978-3-319-15524-1" rel="nofollow">https:&#x2F;&#x2F;link.springer.com&#x2F;book&#x2F;10.1007&#x2F;978-3-319-15524-1</a> There are many youtube videos where Ken explain those ideas, this one for example <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=y2I4E_UINRo" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=y2I4E_UINRo</a>
o-o-8 个月前
&gt; FTA: This same counterintuitive relationship between efficiency and outcome occurs in machine learning.<p>The &quot;examples abound, in politics, economics, health, science, and many other fields&quot; isn&#x27;t a relationship between efficiency and outcome, but rather measuring and efficiency, or measuring and outcome. I think a better analogy is Heissenberg&#x27;s uncertainty principle – the more you measure the more you (negatively) affect the environment you&#x27;re measuring.
ansc8 个月前
Kinda surprised to not see anyone mention Jacques Ellul and his The Technological Society which highly revolve around this. Technological here does not refer to technology.
评论 #41691184 未加载
ezekiel688 个月前
This certainly tugs at all the right levers of the intuition. Not sure how to &quot;buck the trend&quot; in any established organization&#x2F;regime to adjust expectations according to the theory. Looks like this might need to be demonstrated in the wild at a new concern or as a turnaround job, where the leaders could have a strong influence on the culture (similar to how Jim Keller steered AMD and is now steering TensTorrent).
fwungy8 个月前
Efficiency means minimizing the use of the costliest components. It&#x27;s installing fault points into a system on purpose.<p>Robust systems minimize fault points. Efficient systems come at the cost of robustness, and vice versa given a fixed definition of what is being conserved, i.e. costs or energy.<p>For example, a four cylinder engine that gets 15mpg will have a longer life than one that gets 30mpg, given the same cost.
mch828 个月前
Really interesting to learn about the ML perspective of the cost of localized efficiency. Local efficiency can also make things worse from a queueing theory perspective. Optimizing a process step that feeds a system bottleneck can cause queues to pile up, decreasing system-level productivity. Automating production of waste forces downstream processes to deal with added waste.
eru8 个月前
Just add some measure of robustness to your optimization criterion. That includes having some slack for unforeseen circumstances.
评论 #41685594 未加载
评论 #41685862 未加载
Angostura8 个月前
Really interesting article. Got me pondering the extent to which the peacock’s tail is an example of overfitting and Goodhart&#x27;s<p>The female peacock is using the make peacock’s tail as a proxy for fitness - with beautiful consequences, but the males with the largest, showiest tails are clearly <i>less</i> fit, and more prone to predation.
评论 #41685720 未加载
Angostura8 个月前
Really interesting article. Got me pondering the extent to which the peacock’s tail is an example of overfitting and Goodhart&#x27;s<p>The female peacock is using the make peacock’s tail as a proxy for fitness - with beautiful consequences, but the males with the largest, showiest tails are clearly <i>less</i> fit
评论 #41685970 未加载
chriscappuccio8 个月前
While this intuitively seems likes good idea, his real life examples are severely lacking. This gets interesting when we see where the rubber hits the road, the causes and effects of what is being optimized for vs what is happening and we look deeply into improving that scenario.
baq8 个月前
See also antifragility: <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Antifragility" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Antifragility</a><p>In short, efficiency is fragile. If you want your thing to be be stronger after a shock (instead of falling apart), you must design it to be antifragile.<p>Note: it&#x27;s hard to build antifragile physical things or software, but processes and organizations are easier. ML models can be antifragile if they&#x27;re constantly updating.
评论 #41686540 未加载
idunnoman12228 个月前
The statement overfits its own idea. Testing students is not an example of an efficiency
leeoniya8 个月前
&quot;Efficiency trades off against resiliency&quot;<p><a href="https:&#x2F;&#x2F;blog.nelhage.com&#x2F;post&#x2F;efficiency-vs-resiliency&#x2F;" rel="nofollow">https:&#x2F;&#x2F;blog.nelhage.com&#x2F;post&#x2F;efficiency-vs-resiliency&#x2F;</a>
amai8 个月前
Related is the book from deMarco (2002): Slack<p><a href="https:&#x2F;&#x2F;herbertlui.net&#x2F;slack-tom-demarco-summary&#x2F;" rel="nofollow">https:&#x2F;&#x2F;herbertlui.net&#x2F;slack-tom-demarco-summary&#x2F;</a><p>“People under time pressure don’t think faster!”
failrate8 个月前
&quot;If you do not build the slack into the system, the system will take the slack out of you.&quot;
shahules8 个月前
Can&#x27;t agree with you more my friend. Another point on a philosophical level is efficiency or optimization in life, which always focuses on tangible aspects and ignores the greater intangible aspects of life.
fazkan8 个月前
Kind of reminds of the cellular automaton, about how reducing the number of components can lead to complex worlds and rules. Maybe tangentially related. Conway game of life is another example.
mooktakim8 个月前
Teaching is a terrible example. Teaching is actually more efficient when it is decentralised as the teachers can adapt to local environment and changes. With centralisation you have bad feedback loop.
MorningBell8 个月前
Premature optimization is the root of all evil. - Donald Knuth<p>Head and hands need a mediator. The mediator between head and hands must be the heart! - movie &quot;Metropolis&quot;
alexashka8 个月前
Does being a super efficient AI researcher make everything worse?
Animats8 个月前
Important subject, so-so blog post. This idea deserves further development.<p>The author seems to be discussing optimizing for the wrong metric. That&#x27;s not a problem of too much efficiency.<p>Excessive efficiency problems are different. They come from optimizing real output at the expense of robustness. Just-in-time systems have that flaw. Price&#x2F;performance is great until there&#x27;s some disruption, then it&#x27;s terrible for a while.<p>Overfitting is another real problem, but again, a different one. Overfitting is when you try to model something with too complex a model and and up just encoding the original data in the model, which then has no predictive power.<p>Optimizing for the wrong metric, <i>and what do about it</i>, is an important issue. This note calls out that problem but then goes off in another direction.
评论 #41684654 未加载
评论 #41685580 未加载
dawnofdusk8 个月前
The author is a very sharp individual but is there a reason he insists on labelling overfitting as a phenomenon from machine learning instead of from classical statistics?
评论 #41684607 未加载
评论 #41684605 未加载
评论 #41684698 未加载
natmaka8 个月前
<a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Diminishing_returns" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Diminishing_returns</a>
sillyLLM8 个月前
It seems that this is related to the Tim Harford book Messy: The Power of Disorder to Transform Our Lives, but that book is not about deep learning.
Trasmatta8 个月前
From a social &#x2F; emotional &#x2F; spiritual&#x2F; humanistic perspective, this is what I see in the &quot;productivity&quot; and &quot;wellness&quot; spaces.<p>&quot;Ahh, if only I hyperoptimize all aspects of my existence, then I will achieve inner peace. I just need to be more efficient with my time and goals. Just one more meditation. One more gratitude exercise. If only I could be consistent with my habits, then I would be happy.&quot;<p>I&#x27;ve come to see these things as a hindrance to true emotional processing, which is what I think many of us actually need. Or at least it&#x27;s what I need - maybe I&#x27;m just projecting onto everyone else.
评论 #41684787 未加载
boredhedgehog8 个月前
If a citizen recognizes or intuits this to be a deep-seated problem of the political process, and if the only concrete influence this citizen can exert on the political process is choosing one of several proposed representatives, it seems rational to choose the most irrational, volatile, chaotic and unpredictable candidate.<p>The ideal choice would be a random number generator, but lacking that, he would want to inject the greatest dose of entropy available into the system.
kazinator8 个月前
Goal: efficient applications<p>Proxy: minimizing execution time of hot loops<p>Strong version Goodhart&#x27;s: applications get incredibly bloated and unresponsive
shae8 个月前
Gerrymandering is over fitting. Mitigation: randomize the actual shape of a district when the votes are counted.
AnimeLife8 个月前
Very interesting article. I don&#x27;t get though for hotspot partitions they didn&#x27;t use a cache like Redis.
hcfman8 个月前
Well the use of the phrase “too much” already implies less than optimal. A self fulfilling prophesy by definition ?
submeta8 个月前
Remind‘s me of a quote from Donald Knuth: „Premature optimisation is the source of all evil.“
efitz8 个月前
I am skeptical of the analogy to overfitting, although I understand where the author is coming from and agree with the sentiment.<p>The basic problem is stupid simple. Optimizing a process for one specific output necessarily un-optimizes for everything else.<p>Right now much of commerce and labor in the United States is over-optimized for humans because tech businesses are optimizing for specific outcomes (productivity, revenue, etc) in a way that ignores the negative impacts on the humans involved.<p>The optimizations always turn into human goals, eg my manager needs to optimize for productivity if they want a bonus (or not get optimized out themselves), which means they need to measure or estimate or judge or guess each of their employees’ productivity, and stupid MBA shit like Jack Welch’s “fire the lowest 10% every year”) results in horrible human outcomes.<p>Sure there are people who need to be fired, but making it an optimization exercise enshittified it.<p>Same for customer service. Amazon wants to optimize revenue. Customer service and returns are expensive. Return too many things? You’re fired as a customer.<p>Call your mobile providers customer service too often? Fired.<p>Plus let’s not staff customer service with people empowered to do, well, service. Let’s let IVRs and hold times keep the volumes low.<p>All anecdotes but you’ve experienced something similar often enough to know it is the rule, not the exception, and it’s all due to over-optimization.
knodi8 个月前
Also it leads to a rigid system that is inflexible to deal with unknowns.
zug_zug8 个月前
I like the connection between Goodheart&#x27;s law and overfitting. However these examples are a reach:<p>--- Goal: Healthy population Proxy: Access to nutrient-rich food Strong version of Goodhart&#x27;s law leads to: Obesity epidemic<p>I&#x27;m not sure I believe this one. Exactly who&#x27;s target is &quot;access to nutrient-rich food&quot; and how would removing that target fix the US obesity epidemic? Is &quot;nutrient-rich&quot; a euphemism for high-calorie? My understanding is that there are plenty of places with high-nutrient food but different norms and much better health (e.g. Japan).<p>We can and do measure population health across (including obesity), this isn&#x27;t a proxy for an unmeasurable thing.<p>--- Goal: Leaders that act in the best interests of the population Proxy: Leaders that have the most support in the population Strong version of Goodhart&#x27;s law leads to: Leaders whose expertise and passions center narrowly around manipulating public opinion at the expense of social outcomes<p>Is this really a case of &quot;overfitting from too much data&quot;? Or is this just a case of &quot;some things are hard to predict?&quot; Or even, &quot;it&#x27;s hard to give politicians incentives.&quot; It&#x27;d be interesting if we gave presidents huge prizes if the country was better 20 years after they left office.<p>--- Goal: An informed, thoughtful, and involved populace Proxy: The ease with which people can share and find ideas Strong version of Goodhart&#x27;s law leads to: Filter bubbles, conspiracy theories, parasitic memes, escalated tribalism<p>Is &quot;the goal&quot; really a thoughtful populace? Because every individual&#x27;s goal is pleasure, and the companies goals are selling ads. So I don&#x27;t know who&#x27;s working on that goal.
ponow8 个月前
&gt; Distribution of labor and resources based upon the needs of society<p>Not a goal for me, and not for evolution. Survival, health, prosperity, thriving and complexity rate higher. Not everyone makes it.
评论 #41687424 未加载
skramzy8 个月前
What gets measured gets optimized
_wire_8 个月前
Too much efficiency countering too much efficiency makes everything worse.<p>This whole thesis easily tips over into a semantic gobbledygook, as efficiency is not a property of the larger world, but an utter contrivance of thought.<p>Focus on anything to the exclusion of everything use and things are going wrong. How has the obviousness of this observation has turned into a breakthrough? AI is the perfect nexus for such a discovery: trying to optimize a system when you don&#x27;t understand how it works naturally has pitfalls.<p>So what can it mean to try to mathematically formalize a misunderstanding? Maybe there&#x27;s a true breakthrough lurking near this topic: that all understanding is incomplete, so look for guiding principles of approximation?<p>The author is right to call out the forest for the trees.<p>—<p>Web Design: The First 100 Years<p><a href="https:&#x2F;&#x2F;idlewords.com&#x2F;talks&#x2F;web_design_first_100_years.htm" rel="nofollow">https:&#x2F;&#x2F;idlewords.com&#x2F;talks&#x2F;web_design_first_100_years.htm</a><p>How the SR71 Blackbird Works<p><a href="https:&#x2F;&#x2F;m.youtube.com&#x2F;watch?v=gkyVZxtsubM" rel="nofollow">https:&#x2F;&#x2F;m.youtube.com&#x2F;watch?v=gkyVZxtsubM</a>
tonymet8 个月前
The author is right that we rely on metrics too much. But he&#x27;s biased against capitalism and his proposed cure is more socialism. What&#x27;s actually lacking is wisdom and integrity.
bbor8 个月前
IMO the theory at the start of the post is well written and <i>almost</i> there, but it needs to more substantively engage with the relevant philosophical concepts. As a result, the title &quot;efficiency is bad!&quot; is incorrect in my opinion.<p>That said, the post is still valuable and would work much better with a framing closer to &quot;some analogies between statistical analysis and public policy&quot; -- the rest of the post (all the political recommendations) is honestly really solid, even if I don&#x27;t see a lot of the particular examples&#x27; connections to their analogous ML approaches. The creativity is impressive, and overall I think it&#x27;s a productive, thought-provoking exercise. Thanks for posting OP!<p>Now, for any fellow pendants, the philosophical critique:<p><pre><code> more efficient centralized tracking of student progress by standardized testing </code></pre> The bad part of standardized testing isn&#x27;t at all that it&#x27;s &quot;too efficient&quot;, it&#x27;s that it doesn&#x27;t measure all the educational outcomes we desire. That&#x27;s just regular ol&#x27; flawed metrics.<p><pre><code> This same counterintuitive relationship between efficiency and outcome occurs in machine learning, where it is called overfitting. </code></pre> Again, overfitting isn&#x27;t an example of a model being too efficacious, much less too efficient (which IMO is, in technical contexts, a measure of speed&#x2F;resource consumption and not related to accuracy in the first place).<p>Overfitting on your dataset just means that you built a (virtual&#x2F;non-actual) model that doesn&#x27;t express the underlying (virtual) pattern you&#x27;re concerned with, but rather a subset of that pattern. That&#x27;s not even a problem necessarily, if you <i>know</i> what subset you&#x27;ve expressed -- words like &quot;under&quot;&#x2F;&quot;too close&quot; come into play when it&#x27;s a random or otherwise meaningless subset.<p><pre><code> I&#x27;m not allowed to train my model on the test dataset though (that would be cheating), so I instead train the model on a proxy dataset, called the training dataset. </code></pre> I&#x27;d say that both the training and test sets are actualized expressions of your targeted virtual pattern. 100% training accuracy means little if it breaks in online, real-world use.<p><pre><code> When a measure becomes a target, if it is effectively optimized, then the thing it is designed to measure will grow worse. </code></pre> I&#x27;d take this as proof that what we&#x27;re really talking about here is efficacy, not efficiency. This is cute and much better than the opening&#x2F;title, but my critique above tells me that this is just a wordy rephrasing of &quot;different things have differences&quot;. That certainly backs up their claim that the proposed law is universal, at least!
futuramaconarma8 个月前
Lol. Nothing to do with efficiency, just humans recognizing incentives and acting in self interest.
layer88 个月前
I propose to denote the worsening trajectory by the term “enshittification”.
nobrains8 个月前
It would be nice, if us on HN, can crowdsource, some good KPIs&#x2F;Proxies for the goals mentioned in the article.<p>These ones:<p>Goal: Educate children well Proxy: Measure student and school performance on standardized tests Strong version of Goodhart&#x27;s law leads to: Schools narrowly focus on teaching students to answer questions like those on the test, at the expense of the underlying skills the test is intended to measure<p>--- Goal: Rapid progress in science Proxy: Pay researchers a cash bonus for every publication Strong version of Goodhart&#x27;s law leads to: Publication of incorrect or incremental results, collusion between reviewers and authors, research paper mills<p>--- Goal: A well-lived life Proxy: Maximize the reward pathway in the brain Strong version of Goodhart&#x27;s law leads to: Substance addiction, gambling addiction, days lost to doomscrolling Twitter<p>--- Goal: Healthy population Proxy: Access to nutrient-rich food Strong version of Goodhart&#x27;s law leads to: Obesity epidemic<p>--- Goal: Leaders that act in the best interests of the population Proxy: Leaders that have the most support in the population Strong version of Goodhart&#x27;s law leads to: Leaders whose expertise and passions center narrowly around manipulating public opinion at the expense of social outcomes<p>--- Goal: An informed, thoughtful, and involved populace Proxy: The ease with which people can share and find ideas Strong version of Goodhart&#x27;s law leads to: Filter bubbles, conspiracy theories, parasitic memes, escalated tribalism<p>--- Goal: Distribution of labor and resources based upon the needs of society Proxy: Capitalism Strong version of Goodhart&#x27;s law leads to: Massive wealth disparities (with incomes ranging from hundreds of dollars per year to hundreds of dollars per second), with more than a billion people living in poverty<p>---<p>I will start:<p>Goal: Leaders that act in the best interests of the population<p>Good proxy: Mandate that local leaders can only send their kids to the schools in their precinct. They can only take their families to the hospitals in their precincts.
curious-tech-128 个月前
perfect reminder that when you focus too hard on the proxy, you might win the battle and lose the war
评论 #41684625 未加载
015a8 个月前
&quot;Efficiency&quot; meaning, given some input cost, reducing the loss of applying that cost, toward some measured outcomes. High efficiency implies something about each of those three stages, none of which are reasonable to apply in all situations:<p>1. That the only input to the system is cost&#x2F;money (or proxies of that, like compensated human time). Put another way: That the model you&#x27;re working with is perfectly liquid, and you don&#x27;t need to worry about fundamental supply constraints.<p>2. That the loss is truly loss, and there isn&#x27;t some knock-on effects from that loss which might range from generally beneficial and good, to actually being somewhat responsible for the output metric, and your model is measuring the wrong thing.<p>3. That the output metric correctly and holistically proxies for the real-world outcomes you desire.<p>Using the example from the article on standardized testing: A school administration might make an efficiency argument by comparing dollars spent to standardized test scores.<p>* Dollars isn&#x27;t the only input to this system, however; two major ones also include the quality of teachers and home life of the students. Increasing the spend of the system might do nothing to standardized test scores if these two qualities also can&#x27;t be improved (you might make the argument that increasing dollars attracts better teachers, and there&#x27;s some truth to this, but generally (even in tech) these two things just aren&#x27;t strongly correlated; many organizations have forgotten what it even means to be &quot;good at your job&quot; and how to screen for quality in interviews. When organizations lose that, no amount of money can generate good hires because the litmus test doing the hiring is bad).<p>* &quot;Loss&quot; in this system might be the increase of funding without seeing proportionally increasing test scores; which does not account for spending money in extracurriculars like music, art, and sports; all generally desirable things we believe money should be spent on (isn&#x27;t it interesting that we call these things &quot;extra&quot;curriculars?).<p>* Even if a school administration can apply this model to increase test scores, increasing test scores might not be an outcome anyone really wants. As the article says, all that guarantees is a generation of great test-takers. Increasing college acceptance rates? We&#x27;ve guaranteed a generation of debtors and bad degrees. Turns out, its impossible to proxy for the real world thing you want, in a way that can be measured on a societal level.<p>All of this is really just symptoms of the &quot;financialization of everything&quot;, which has been talked about endlessly. In particular to this discussion, society has broadly forgotten about what the word &quot;service&quot; means; that public transit in your city must be a capitalistic enterprise, it <i>itself</i> has an efficiency metric that must be internally positive, because the broader positive efficiency impact that public transit network has on the people and businesses in the city, and thus municipal tax income, is too complex to account for within a more unified economic model.