252 karma · joined August 24, 2020
Having said that, I think its ideal use cases at the current time are probably less of a general messaging solution with friends, and more for specific scenarios where off-internet and security is maybe most salient.
To me it causes some mild confusion (if that's the right word — maybe too strong a word) with paragraph breaks. It's just too big a gap. For me it decreases readability if anything.
In those cases, I might want or need to use the web for some reason before I get a new one (to locate or purchase a new one if nothing else). I definitely have appreciated sites that take keyboard navigation into consideration at those times.
The keyboard is a common UI mechanism. Why not try to design the site to increase access via diversity of mechanisms instead of dismissing them due to low use rates?
I agree with some of the other posters that something has gone awry with UI testing and design. A lot of it is good, but there's these huge holes that pop up all the time. Usually it feels like something that's a trend that's being applied inappropriately... Other times obvious considerations seem completely ignored.
In general I think there's problems with UI trying to be too clever and novel, or too "fresh" without focusing critically on functionally. I'm all for clever, novel, and fresh, as long as it's actually driven by function and not using the UI to implicitly assert some kind of status of taste or ability.
The two other institutions I taught at used to have very strict policies about it and then at some point they became laxer in a similar way. It used to be a no warnings policy, and then became more wishy washy.
I always assumed it was to be more realistic, in that many students were being granted exceptions of sorts. There also can be a lot of grey area with cheating. Overall my impression was that some lawyers recommended the change to avoid liability. But I really don't know.
The author talks about reactions of physicians dismissing convincing data — the befuddlement and dismissals — and it's the same reaction athletes and coaches give when they're part of some of a controlled trial.
https://www.seriouseats.com/2014/06/cheese-101-washed-rind.h...
Most washed rind cheeses share fundamental characteristics that resemble Limburger (to me, old barn), in case you're interested in trying others.
It's like realizing this childhood world wasn't that way just because you have changed from a child to an adult, but because the world has changed.
I'm starting to become worried over what my daughter might experience in that regard. Things that were normal for me or that she might encounter that will be gone instead. Like, will fireflies still be around for her as an adult? What about walking around with piles of snow higher than you everywhere? Will she ever see a natural unlit clear night sky? It's difficult to convey what that many stars looks like, or seeing the milky way stretched across it.
More to the point of the article, I remember freshwater bivalves walking upstream across stream beds. They aren't mussels, but still — I haven't seen any in the last several years where I used to.
I'm not suggesting Pfizer is up to something nefarious, just that the biomedical literature is littered with studies with N < 100 for the outcomes of interest and they don't replicate. The N > 40k in some ways is irrelevant if none of those participants have the processes in play. It's more like recruiting N >40k to get N=94 to study.
I'm also not suggesting with 90% effectiveness that there's probably no effect, just that my guess is a true value of 60% isn't outside the realm of possibility once you consider all the factors that could be in play besides the random sampling variation that's assumed by their tests.
https://write.as/ccgaoox9alf3hbv4.md
I think the key is to not necessarily have state-run ISPs completely OR private-run ISPs completely. Something like the courier/delivery market where there's access to multiple providers is key.
Any kind of local monopoly is bad.
The other elephant in the room is the right-of-way that's granted to these telecom companies over private and public land. In exchange for what?
Lots of room here for lots of progress.
I think to some extent this maps onto the idea of rent-seeking monopolies in current society and undercompensation of individuals. It's hard to go to something like this
https://mkorostoff.github.io/1-pixel-wealth/'
and not conclude that many people are being undercompensated grossly, and a small number of others are being overcompensated grossly.
For what it's worth, I agree with others that I don't think this maps onto "braininess" per se. I just think that's one of the character traits that happens to be valued the most in Western societies at the moment, so it's projected by the elites the most. In the past it might have been valor or nobility or courage or whatever it was -- the thing that is presented as justification for the excessive income, and protection of that income.
In general I think individuals are misvalued, not in the sense that persons don't matter, but the economic value placed on single individuals can be (and usually are) grossly overstated or understated. I think this transcends many many fields.
I swear somewhere here or elsewhere I read a story that one of the major quantum computing contributions that accelerated the field was essentially the result of a faculty member spending several years doing nothing but teaching and thinking about the problem, with no grants or anything but salary. Maybe it's a myth or I'm misremembering it though; I've tried searching for it (to confirm or disconfirm my memory) with no luck because of the volumes of irrelevant results.
Also, history is an interesting example, as it seems one of the first sets of experts everyone seemed to turn to with all the events of the last few years are historians. "What happened with previous pandemics?" "How did past pandemics effect economics of the time?" "What did people do then?" "Is this the first time this situation occurred?" So forth and so on. The more "unprecedented" the problem the further back in history people will look for precedents and guidance.
It also seems implausible to me that the same sorts of history research could happen by hobbyists as happens at universities, just due to time and funding constraints.
There's a lot of problems in academics at the moment, at least in the US, but I think 2/3 of it is actually coming from outside those in the field (pick your field). Administration problems, federal incentives, state incentives, private incentives (yes, corporate management requiring useless degrees or offloading employee training elsewhere to decrease investment costs and its consequences) etc etc etc
A better example than face masks maybe is the recent controversy over Twitter's AI and Obama images (https://www.theverge.com/2020/9/20/21447998/twitter-photo-pr...).
A lot was made of racial issues, which is fine, but the broader issue is why subtle changes in photos, like cropping, should confuse things so completely.
The target piece (the focus of this HN thread) sums itself up this way:
"There’s a good set of params somewhere nearby. When we start walking to it, we can’t ever get stuck along the way, because there are no local optima. Once we’ve stumbled upon a good set of parameters, we’ll know it and we can just stop."
I think there's some useful insights there, but this is in many ways the definition of local optima. What I might argue is that because there's so many locations in high-dimensional space that will satisfy some classification goal, it's "easy" to find one that works with regard to some population that defines the model development space (training + test). However, that model development space/population is itself implicitly defined by a certain set of constraints -- it's a subpopulation of some broader population. What you want to generalize to to define overfitting is broader. You can still not overfit to your model development space, but be overfitting with regard to some broader set of possible inputs.
Whether or not the constraints of the model development population/space are important and reasonable considerations -- e.g., in your argument, not having access to things like body language etc -- is maybe a little variable. In some cases the implicit defining characteristics of the model development population are meaningful, but in other cases they're hidden.
In Twitter's case, you end up finding out later that there's weird things that probably defined the space of their images that they didn't intend. It's only in the adversarial case that you learn about this.
In classical statistics, you talk about generalization and overfitting, but there's an implicit population you're sampling from that defines both of those things. That is, you have a training/fitting/initial sample, and you ask yourself how well your model would perform on a test/validation sample. But implicit in that is some assumption about what it means to be a random sample from the same population.
I think lots of times with DL, the cross-validation/test sample is also implicitly defined as coming from some population. But the population isn't some model, it's some source. Some image dataset, something like that. There will be things about that source that are "of interest", but other things that are idiosyncratic about it, and unrepresentative of the "real" population of interest. In this way, I'm not sure that held-out samples from some source are really the right way to think of generalization and overfitting -- I think the adversarial setting is the generalization setting.
https://www.sciencemag.org/news/2020/05/eye-catching-advance...
Along the way from the classical to the DL regime I think there have been some overlooked issues about what it means to generalize, what you're really sampling from, and what your "population" actually is. It parallels tensions about theory versus experimentation because having a population in the classical case that you're sampling from requires a certain data-generating theory, which is often lacking in DL. The closest thing in the classical regime to DL generalization theory is maybe a sort of ultra-high-dimensional bootstrapping with random effects: showing that your bootstrap samples are close to your observed sample isn't the same thing as showing they're close to the population, or to other samples drawn from that population, especially in the presence of random effects.
Many adversarial cases are good examples of this: a DL model being completely thrown off by something very incidental, that a human would instantly recognize as not being within a class. Not just something a human would instantly recognize as not being within a class, but something a human would be perplexed by as an adversarial case.
The point isn't that humans are better or worse, it's that the models do often seem to be overfitting, but overfitting in a way that isn't evident until the inputs are generalized beyond whatever is in the development samples. Put another way, they might be learning something about your development datasets more so than the actual features of interest, which is the whole idea of overfitting. It's just that what it means to "generalize" is much broader.
It's a really interesting piece but I think there's lots more to the story.
Numerical computing is something where there's always been something that works. Before Python it was C and Fortran.
The problem people ran into is that when everything is wrapped around low-level libraries for speed, you eventually run into the catch 22 of using the slower language that you prefer for clarity, or the faster one that makes it acceptable performance-wise.
In other words, with Python, to get the performance you would have got in C or Fortran, you have to code in C or Fortran. Then you're not using Python anymore. The idea (in theory, and a lot in practice) with Julia (or Nim, or other LLVM-targeting languages) is that you don't have this penalty.
So in that sense Julia is providing something that isn't working in Python or Matlab.
I think Julia's not quite what it's cracked up to be, but mostly it is, and I'd probably prefer working with it over Python or Matlab for numerical stuff.
I bring this up mainly because one of my first thoughts in discussing, e.g., US corn agriculture versus Subsaharan agriculture is whether or not you'd even be able to compare them well because the crops optimally suited to each would often be entirely different. I think often probably so, but at the same time I wonder if things would look different if the same volume of money and resources were put into things that assume different consumption preferences.
I'm not opposed to conventional agriculture, and don't believe in pointing fingers when it comes to food sustainability (in an ecological as well as humanistic sense). I do sometimes wonder, though, if conventional practices are often driven by assumptions or factors that are unwarranted or problematic in themselves.
Sure, it takes the same amount of time to mark an essay. But it takes much less time to score exams and to administer them.
Writing research papers takes much, much, much less time because of word processors, online submission systems, copyediting changes, and so forth. A lot of research itself takes less time to do because of statistical analysis libraries, computerized administration and recruitment, etc etc etc
This is actually something I've heard older, very professors lament -- some of them have claimed that when, say, papers took longer to produce, you had more time to think when things were being mailed back and forth, and there wasn't so much of a factory approach to paper writing where they were treated as widgets. I'm not saying I agree or disagree with this necessarily but I'm really skeptical of claims that productivity hasn't risen in certain fields.
Now you can get into a discussion of "real" versus "false" productivity gains but that's a little different.
E.g., you could have A -> B, and A -> C, and B != C. Then C is not B, but implies A just as much as B might (in the very least it doesn't imply not A per se, as A might be true). It seems like there's some implicit assumptions going on.
It's great to be using quantitative measures, and almost a necessity, but as someone in this area who has done clinical work as well as research, I can say from personal experience EHR in mental healthcare has fumbled repeatedly with handling the complexities of behavioral information.
Good to see attention in this area though.
This is true of a lot of languages, python included, but I think when part of the language's selling point is something like "c-like speed with python-like syntax" it can be a little (although not entirely) misleading.
Having said that, I still prefer Julia over python (at least for numerical computing, not so sure about other things). I just like the language more. I also think, with some exceptions, that I don't have the same dependency hell problems that I've run into with python. Even now, I'm in the process of switching over to Julia from python for a project because the python library I'm using depends on about 6 different other libraries, but only specific versions, that you have to run in a specific standalone conda environment to avoid using the wrong combination of packages, all of which are pre-python3, and so forth and so on. Even then, when you manage to thread the needle, it still falls apart later for unknown reasons. This is surely this particular package, but my experience with Julia is that things are much cleaner (R is similarly problem-free usually but it's a lot slower and Julia as a language is more coherent to me).
I'd really prefer something like Nim to be seeing the attention that Julia is getting, something more general-purpose, but there's no consensus of momentum around something like that at the moment. Maybe in the near future ocaml will pick up steam, or maybe the next version of C++ will essentially make it look like python, or maybe there will be something not quite on people's radar at the moment, but at the moment it is what it is.
Discussion of credit for CRISPR goes back years. The first controversy began with patents. It's been obvious to anyone following this that credit for it is difficult to attribute. Everyone agrees CRISPR is in need of recognition, and yet the major systems for doing so are still based on an erroneous "lone genius" paradigm (with modifications).
Here, on HN and on Scott's post, there's an implicit equating of "meritocracy" with "test score."
Do people here really think test score and ability — to say the least of merit per se — are equivalent?
Go find a scatterplot of some data correlated 0.80 (e.g., https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3576830/). That's a very generous estimate of how correlated that test score will be with actual ability, whatever that means. The correlation probably is lower, like 0.50-0.70. Look at the y axis. Now cut it and take the upper fraction of your choice. See how wide a range of the x axis is represented?
That's the problem.
The reason why test scores are a problem is because when you select on an imperfect indicator, at some level you are randomly selecting with regard to actual ability, which is even further removed from merit. Throw in any kind of bias of any sort associated with anything, and the problem gets worse.
The truth is, it was already a lottery before. It was just a lottery that everyone pretended wasn't a lottery, and that's the problem people have with it.
The problem with meritocracy isn't the idea of selecting people based on merit, it's the idea that test scores or other such quantitative metrics are equivalent to merit.
The analogy with sports is fundamentally flawed because, among other things, the metrics are the thing of interest in sports. If you are interested in who can swim 50m the fastest, then your speed on the 50m swim is the metric of interest. In school, though, presumably you're not interested in test score, you're interested in someone's ability to solve problems, or create things, or be productive, or whatever. And we're not talking about sport, we're talking about gateways into careers in general. If we are selecting into schools to maximize ability to achieve high test scores on standardized tests per se, then at least be frank and honest about that, and accept that what you are arguing is that the school is an institution whose purpose is to achieve high standardized test scores, and nothing more. But if you want to argue that the school should produce people who are solving problems, innovating, and so forth, then you need to be honest about selection based on test scores, because the scores aren't what they intend to measure.
Scott laments the loss of the ability of some kid to work their way into the school based on some test score. But shouldn't he be screaming about a system that doesn't provide resources to kids where they're at? That assumes the kid's ability is equivalent to their score on some entrance exam? What about those problems?
I agree that throwing out test score and relying on GPA per se is a problem; relying on GPA alone is just as much of a problem as relying on test score alone. You could, for example, have a lottery for everyone whose score is above some generous threshold, or for everyone whose GPA or score is above some threshold. But staying with the system as it is seems at worst like turning a blind eye to systemic injustice, and at best like missing the forest for the trees.
Another problem is that academics is full of dead ends, and it has to be if it's being done right. But no one likes that either. So rather than working on changing expectations, we do the opposite and bean count.
Yet another problem is the classic fan fiction / fan community issue: if you get into enough expertise, the experts have to become the audience, because no one else understands it, but then that leads to increasingly narrow fields of view because of increasingly narrow interests. You could broaden the audience, which is increasingly a demand, but that has downsides too, because often what's really the way forward is incomprehensible and boring to the average person. People love their iphones; not so much all the incremental computer science, physics, and engineering that went into every little part.
Something I think this essay maybe misses is that there are a lot of senior academics who would say publishing doesn't matter, that a paper is a dime a dozen, and what does matter are grants, and lots of them. This is how this dynamic has shifted in a lot of places, from the science -> publishing -> money. You might argue that this is better but it has its own set of incentive problems.
I do think there's something to be said for liability for creating a port that is so similar to USB-C, but also causing damage, akin to copyright laws, based on consumer confusion. I.e., if a reasonable person might think it is a variant of USB-C, and USB devices seem to work for long periods of time without apparent damage, then Nintendo is liable by virtue of resulting damage to the consumer's property (not to the USB organization). There's a certain liability for negligence in that case. But I could also see reasonable arguments that if Nintendo were explicitly saying it is not a USB port, that they shouldn't be liable (I don't agree but see it as a reasonable argument).
But if Nintendo is advertising it in anyway like that, they should be held liable. I just don't see a reasonable argument for why that wouldn't be the case. You can't have your cake and eat it too.