> - the KPI "do-more-users-leave-our-platform-earlier-if-our-matching-algo-is-just-too-good" - I promise: In alle the years, this question WAS NEVER - NEVER!!!!!!! - raised, regardless wich Manager or which Exec. This metric isnt even debated.
What labels do they use for training their algos though? What is their definition of a successful match, is it a date, a recurring date, or something closer to a long-term relationship?
If matches predominantly result in "failure" they might just not have enough "long-term success" labels to go by, and their proxy labels will be biased towards short-term successes.
Wrong approach;
at least until latest, NONE of the standard apps does apply any kind of "real AI" stuff, maybe this changed through the last 3 - 4 years.
All thi matching stuff like "match with X%" is just bullshit.
The only platform having a useful approach here was OKC years ago. (but even for their scoring you would not need any type of sophisticated tech)
I've cancelled by NYTimes subscription just a few months ago. Didn't have to do any phone calls-- just answered a few questions in the chat, right there on their website.
They "generously" let you use an IM chat now, yes. But there's no excuse for it not being a simple button. The only reason is so they can raise the effort barrier just a bit more.
Had a similar experience when I left my previous ISP. I didn’t even have a choice to keep the service as I was moving. Upgrading the service was an easy click, but cancelling consisted of reading through their help pages to find a number to call, which of course was only open a few hours per week. It explicitly said that this was the only way to cancel.
Instead I simply sent them an email insisting that I wanted to cancel, and that I refused to talk to them on the phone to do it. They replied with a “great offer” to reduce the price. In my reply I simply insisted on cancelling. I didn’t hear from them again, except for a confirmation by post a while later that the service had been cancelled.
Of course, this is much more effort than it should be, especially an emotional effort because being treated like this is very upsetting. But it might be worth trying.
They wouldn't let me cancel until I answered the question "how are you doing today?" I said "I want to cancel" but they just repeated the question until I finally said "fine" and then the conversation could proceed. Fun times.
Note that you will want to own at least two and enroll both of them to properly lock down a service so that it doesn't need some plan B. The reason is that obviously if it's locked down to a single U2F Security Key and that key breaks or is lost you're screwed.
Google's programme aimed at high risk people (e.g. journalists covering government corruption) specifically aims to leave you in a position where so long as you have control over the physical devices your secrets are safe, and if the devices are destroyed then your account is irrevocably lost and too bad. Doing that with just one key is asking for trouble.
If you're just dipping your toe in the water, buying one key and having your plan B be a bunch of one time codes written in the back of a diary in your locked desk drawer makes sense, and if you're mostly just interested in the cool technology and not worried about security then going to a Key with Google Authenticator as plan B is fine too.
But if you want this to solve all your problems as advertised, buy two keys.
Thanks that’s good advice. Is the idea your token is synced across both devices? Or you have two separate tokens that allow you to authenticate? I’ve only set up a mobile based authenticator per account before...
Ordinary Security Keys can't be synchronised so to as to be interchangeable. A good U2F/ WebAuthn implementation lets you enroll several of them and use any each time you sign in. So yeah, separate tokens, any of them works.
It's pretty different from authenticator apps, it can be much more convenient (no trying to quickly type in six digit numbers) but it's kinda expensive for now.
Ordinary Security Keys only know how to do exactly one thing, prove that they're still the same Security Key that they were the last time. They can't even prove which one they are in particular (most can prove which model they are, because a bank or something might be like "Ooh, we like this technology but we insist you use Bank of America brand Keys..." but they don't even know like a "serial number" or anything). This is deliberate - it allows the strongest possible privacy guarantees while still delivering a useful security function. The Firefox implementation lets you pick "No" when sites ask which model it is - I always do, none of their bloody business, it's a Security Key, eat it.
When you "enroll" a Key the site gets back a "cookie" (not an HTTP cookie) that is only useful to identify that site to that Security Key; a Elliptic Curve public key; and a signature proving the Key knows the corresponding private key and was enrolling with this specific web site. The site puts those somewhere (presumably a database table) ready to use them when you log in subsequently.
When you need to log in, the site gets its list of all the keys you've enrolled and says "OK, here are the cookies for some keys you enrolled, prove you still have one". If you have one of these keys it can find its cookie among the set, and it knows the private key that goes with that cookie, so it can sign a new message saying "Hi, this is still me, signing in to $domain right $now" and the site verifies that with the public key.
In principle the Security Key could be keeping a big database of every site it has enrolled with and the cookies used versus private keys. In practice what it does is make a random new private key each time it enrolls, encrypt the private key and put the result in the cookie. Only it knows how to decrypt the cookie, so there's no danger from this approach.
Thanks, I'm going to buy more than one for backup and use the multi-key approach. Appreciate the long explanation. I've been meaning to set this up for years.
This should be standard practice taught to kids in school! Especially considering their whole life is digital now.
I saw that Linux's full disk encryption supports Yubikey as well as Gnome login screens which is neat.
This is a good quote. One similar way to state it is by using the System 1 / System 2 [1] terminology. Manipulation is when someone pushes the audience to System 1 reasoning, diverting them from using System 2.
It's not a perfect distinction (there probably are cases where appealing to System 2 might look a lot like manipulation), but it's succinct, ends-neutral and not too subjective to be useful in practice.
This is only true if you assume that System 2 is always better. But if someone has rationalized great cruelty, it is perfectly appropriate to appeal to their emotions if it will get them to stop (this is essentially equivalent to a point the author of the original piece makes).
Rationalizing is very different from actually making a rational decision, and is part of System 1 thinking.
Post hoc, ergo propter hoc.
A rational decision starts with defining the epistemic system in which it is made. Is it bayesian, frequentist, higher order fuzzy logic? How the weighting, if any, works? What is the data/proposition acquisition algorithm? Which heuristics are to be used to simplify decision and why?
Rationalization can mean rationalizing a system 1 decision. It can also just mean justifying a bad decision using explicit reasons, which is the sense in which I mean it.
Except for the detail that the article entirely credits the man with being more productive and in no way whatsoever suggests that his wife's work in any way whatsoever contributes to it:
Maybe marriage makes men work more hours; maybe it makes them work harder per hour; maybe it makes them control their tempers better; maybe all of these and more.
It gives zero credit to the wife somehow genuinely enhancing his life by cooking, cleaning, having sex with him, etc. It just suggests that married men choose to be more virtuous. Done.
Why is the parent being downvoted? Multiple codecs are a real problem for video-streaming services, and introducing a new one definitely has potential to bring more compatibility problems. Patent trolls are also a real obstacle (clean-room implementation provides no guarantees against patent infringement).
If you believe there're reasons that this specific codec will not cause such problems, please explain it in the comments instead of just downvoting.
> incredible track record of dead-wrong predictions
The Austrian School economics methodology does not allow for making predictions. See e.g. the note at the beginning of http://wiki.mises.org/wiki/Austrian_predictions It provides certain tools and models that one can use for explaining events and reasoning about policy options. These reasoning can be used as a basis of predictions (along with the other, non-economics-theoretic assumptions), but the predictions themselves will fall outside of the economics science in the Austrian School sense.
Basically, if your analysis manages to reason away uncertainty of the future, you're not following Austrian methodology. And no, the Austrian methodology won't even let you quantify that future uncertainty in probabilistic terms: that would only be possible if it used causal explanations of human action; instead, it relies on teleological, goal-oriented explanations.
Now, questioning and disputing the value, coherence and real-world applicability of the Austrian methodology is perfectly OK. (Personally, I find the Austrian economists' disregard for formal notations very unfortunate, as I've mentioned in another discussion: https://news.ycombinator.com/item?id=1647747) Your comment, however, seems to be simply misguided, based on a flawed understanding of the issue, and the sarcastic tone is unlikely to facilitate a constructive discussion.
You probably know more about economics than me (I'm just a mathematician), so I won't argue against you, but if I understand your two posts correctly, it seems like what you're describing is completely useless. If a model doesn't have predictive power then what is it good for?
I understand from your other post that the justification of Austrian "models"/"explanations" comes from their simplicity. Which I guess is some application of Occam's razor? But that makes no sense - the simplicity of a model on it's own is no indication of correctness. You always need some notion of the likelihood of the observed evidence under that given model. Otherwise, you can just say "all people make all decisions completely randomly" and that's the simplest model of all.
A completely non-predictive model has no real-world meaning, because it can't be used to effect reality. I suspect that Austrian models do effectively have some small predictive power - there is an implicit causal analysis that comes from the small element of qualitative evaluation of uncertainty, no doubt used instinctively by practitioners to decide whether an argument sounds plausible.
Anyway, it all sounds completely ridiculous. How is their method of choosing explanations any better than a witchdoctor deciding that thunder is a sign of the gods being angry?
Austrians argue that predictions can't, as a matter of epistemology, be made.
Of course, humans do that all the time. What's missing from the original Austrian concept in Mises is allowing for imprecision -- making accurate predictions with error bars for variability below the threshold of detection.
What von Mises didn't like about this and what Rothbard hated about it is that such predictions are statistical and probablistic in nature. They aren't deontological, they can't be nutted out a priori from first principles.
They have a real bee in their bonnet about it. It's a shadow (ha!) of Platonist/Aristotelian bunfights.
Out here in that's-nice-but-I-have-shit-to-do land, grownups accept that models are wrong. But even if the map is not the territory, most of the time it's still good to have a map.
Non-predictive analysis can't be used to make decisions, and therefore can't have any effect on the real world unless it is being used incorrectly. By their own argument, Austrian economics is a waste of time.
A dislike of stochastic models (however understandable in historical context) should not be taken seriously in the modern world. Complex chaotic systems are modelled through intensive probabilistic simulation all the time.
For me, the major idea that the Austrians bring to the table is that the study of economics is not a science and never can be. Attempts to create and support models by quantitive analysis of data are flawed because human economies are so astronomically complex that it's impossible to isolate variables, and therefore impossible to design valid economic experiments or draw sound conclusions from economic data.
There are many parallels to algorithms analysis. If we want to know how fast an algorithm runs, our approach isn't to try to run it on every processor in existence and compare the results. Instead we break down the algorithm logically to deduce its theoretical running time.
Austrian economics--in its good parts anyway--is an attempt to build a framework for economics that is more like the asymptotic analysis of algorithms. But this very compelling goal is frequently derailed by politics and polemics.
Algorithmic analysis is absolutely scientific. Asymptotic analysis is based on models of computer behaviour that are testable (and people do test them and do improve on them). Moreover, the asymptotic models make predictions and those predictions are testable (and people do test them and do improve on them).
The asymptotic analysis of algorithms would not be useful if it didn't make accurate predictions about real world phenomenon. I've written peer reviewed computer science papers where I have devised an algorithm, predicted its asymptotic behaviour and then validated through empirical testing.
What you're describing sounds completely different. If the major idea is that human economies cannot be scientifically analysed, then surely all analysis is a waste of time? An analysis that isn't based on the scientific method isn't more likely to be correct.
Well, pure mathematic models also don't have any real-world meaning, and no predictive power with regard to the real world events. Still, nobody thinks that the analytical proof of the Pythagorean theorem is somehow inferior to the empirical one (involving multiple reproducible experiments that measure sides of the actual triangles drawn on a piece of paper with a ruler). We are content with mathematics being consistent within itself.
However, if we (1) add certain assumptions about, say, physical laws, the structure of the reality, the way we can observe it, (2) propose a reasonable way to map some aspects of the world to our mathematical models, (3) calibrate the result according to the future experiments/observations -- we might get a quite useful natural science like physics or astronomy. The pure math will serve only as a low-level tool. That's how the Austrian economists see the place of the pure economics theory: just an analytical tool to be used along with the other disciplines/sciences for solving real-world problems.
For example: Austrian economists believe that it can be deduced from the first principles/axioms that enforcing minimal wage above existing market wages results in involuntary unemployment (most of the non-Austrian economists also believe this is the case, although their methodological basis is different). Assume there was a policy act that included, among other things, raising the minimal wage. A year passed, and it so happened that there have been changes in the economy indicators, including lower unemployment rate. The act was not the only thing that happened in that time; lots of factors may have contributed to the outcome. An Austrian researcher will definitely not count raise-the-minimal-wage factor as contributing in favor of the lower unemployment rate; the factor will be counted against the lower-unemployment-rate outcome, and will require further explanations, what other factors have outweighed its effect.
Using math (well, formal systems/notations) does not guarantee your whole research about the real world is correct, but it establishes a baseline for consistency and rigorous reasoning. It's the same with basic economics.
(My formal education was in mathematics / CS. I study economics, epistemology etc. only as a hobby.)
Unfortunately for the Austrian school, prediction is the heart and soul of science. You cannot claim to be doing science in any sense if you do not make predictions based on falsifiable hypotheses.
As you imply, this is why teleology is not a science, though large-scale statistical studies in fields like politics and sociology have made plenty of scientific headway by ignoring the core Austrian belief system and conducting empirical studies about human actions and preferences anyway.
Austrian School economists don't make predictions, but when they do they're right 100% of the time, except when they're wrong in which case I will remind you that Austrian School economists don't make predictions.
If you think I'm exaggerating, check out this unintentionally hilarious excerpt from the Ludwig von Mises Institute wiki. See if you can count how many times they flip-flop on whether they make predictions.
Austrian predictions
This page attempts to list various predictions made by
Austrian economists about important economic and other
developments.
Important note: Austrian economists, as Austrian
economists, or praxeologists, do not predict. They can
predict not as formal economists, or praxeologists, but,
rather, in their role as thymologists, or economic
historians. In praxeology, A causes B, other things
remaining the same. But, in the real world, other things
cannot be relied upon to always remain constant. Therefore,
predictions of the "A will necessarily lead to B" type are
strictly prohibited. Instead, praxeologists, but not
thymologists, must limit themselves to statements of the
if-A then-B variety.
With this in mind, it is interesting that Austrian
economists have been quite successful at predicting major
events.[1]
Great point, expect that no one is touching them. Deposits over EUR 100K are being hit /frozen, those up to EUR 100K are not touched
Yes, that's the current decision. The linked article was written at the time the Parliament was about to ratify haircut of the insured depositors along with the rest.
(There's no timestamp on the page, but I saw this text around 21th of March.)
Economics is a value-free science, in that it analyses real-life processes and policy implications _without_ prescribing what specific goals are to be pursued by the policy makers.
Politics (unlike _political science_, arguably) is all about promoting specific values and fulfilling special interests. It's not necessarily "bad", it is just not what economics is about. Economics provides the tools for analyzing what _means_ are appropriate (or not) for attaining some or other ends, but the latter lie outside of the scope of economics as a science.
I wish this distinction was so hard and true. It has unfortunately been my experience that to address the assumptions of most economic thinkers, political engagement is required. Bringing up heterodox or marxist _economic_ arguments tends to be met with political resistance. I will redouble my efforts to keep the distinction alive.
Austrian school covers many disciplines -- economics, history, law, politics etc. One can reasonably disagree with their views on, say, economics science, but calling it all just "politics" is simply an incorrect portrayal of what they do.
What labels do they use for training their algos though? What is their definition of a successful match, is it a date, a recurring date, or something closer to a long-term relationship?
If matches predominantly result in "failure" they might just not have enough "long-term success" labels to go by, and their proxy labels will be biased towards short-term successes.