daddyo

That's why BTC-e does not show a giant FBI logo with a notice.

The law cares. Like I said, in many jurisdictions this was only recently amended with special clauses -- clarifying the distinction between physical and virtual goods. In some jurisdictions theft requires fraudulently taking a physical tangible good: virtual goods can not be stolen (but you can still be charged with computer intrusion). Remember also the debate about downloading a cam movie vs. stealing property of movie studios.

> If you are using something that is not equally known to both parties, and equally disclosed, then it is fraud.

But the contract is out there for both parties and their lawyers to have a look at it, before agreeing to it. If Google indexes my /admin directory because I made a typo error in our crawling contract (robots.txt), who is ultimately to blame? Judging by my actions and panic, the directory was clearly meant to be excluded. If we end up with the "smell test" in court for smart crypto contracts, we should just go back to "dumb" paper contracts and signatures.

> So if I forget to lock my door, then my house should be free game for everyone?

Non-sequitur. If you publish an article on Wikipedia then it is free game for everyone to visit it, edit it, and you can not retro-actively say: you are not supposed to be here.

It's a crime of computer / network intrusion. Not a crime of property law (you can't own a record in a database as property, and therefor I can not steal your property).

> if you're abusing a fault in the code, that's very clearly fraudulent behavior.

Another way to put this is that you are using the contract in a manner how it was defined by the author. Compare with a misconfigured web server showing open directories of files, or a robots.txt with a typo in it (ignore: /adminn). What is a fault and what is a feature? Who decides this? Solely the author of the contract? The parties involved (who splits the ties)? A majority of 3rd party volunteers? If everything is decentralized and open to anyone, whose computer network are you intruding/disturbing?

> doesn't mean that it doesn't fall into existing laws

If law was a software product, we are definitely a few pull requests behind its intended use. Look at how long it took to update authorship/copyright laws with the evolution of the internet, and how ugly things are when wrestled into the old framework of: I create it, I forever own it.

I think in my jurisdiction we have a law against pawning stolen goods: If the price is too good to be true (100$ macbook), and you still buy it, you can get your goods confiscated. But how does this translate to cryptocurrency and its volatile pricing (a 50% drop or increase in price is not extremely rare)? Is it illegal to set a buy order for 50% of the price? Especially if you set this before the hack, just hoping to cash in on a flash crash, I can't see which law you break.

About stealing coins, of course this (should) be against the law. But then again, data is not a good. For many jurisdictions, data isn't anything at all. You can not own data in the legal sense, because it only applies to tangible goods.

As to "stealing" coins by manipulating a smart contract, its a grey area. Of course in the real world, contracts can be breached in spirit, not only by the letter. But with smart contracts, you only have the letter of the contract: The code is law.

Fairly giving them a time investment of 30 minutes (reading CV, tailoring technical interview, answering questions) would mean 200 hours of productivity loss. There is no way to defend this to a company, especially when hiring more than 1 a year.

If no coding challenge then it all comes down to the CV. You'll miss out on promising humble candidates who lack the CV buzzword bingo, and get burned on mediocre candidates whose parents paid for them to go through a top university. That's not fair to the talent either.

Of course if someone completes the coding challenge, you give them the time investment they deserve (8-16 hours spread over multiple employees). You save this time by declining those that do very poorly on the challenge, or refuse to do it out right.

You may not have an idea how awkward and depressing a technical phone interview is with a candidate that is not suited to the role. And delegating this to HR/Recruiters is a surefire way to increase noise and crash the hopes of people who pass the screen, but fail badly on future interviews.

I'll always look for curiosity, passion, intellectual honesty.

Depending on the job I'd further look at what they did to distinguish themselves (extra-curricular, self-study, Github projects, etc.)

- Since Cryptocurrency is international, the US SEC does not have jurisdiction everywhere in the world. When there is millions on the line, you could just move to another country and try a scheme, or direct a foreign lackey to do it.

- "The Federal Reserve simply does not have authority to supervise or regulate bitcoin in any way. To the best of my knowledge, there is no intersection at all in any way between Bitcoin and banks that the Federal Reserve has the ability to supervise and regulate."

- To count for an exchange you have to issue shares. Not everybody does this.

- Is cryptocurrency a token or a security?

- How to distinguish between nouveau riche BTC millionaires trying out their luck with an ICO and a criminal organization using it to launder money?

- Who is the single legal entity to target when the ICOs are distributed, and no single entity issues coins?

- What to do with those that profit from future illegal activity, as a 3rd party? Right now there is a lot of obvious market manipulation going on. Whales banding together to influence and set prices. Pumping up interest with bots and 5-cent army trolls. Selling stolen coins for 50% of market value. Sharing upcoming announcements with a small group of investors, devs, and supporters, allowing them to speculate on insider knowledge. How do they prove I must have known about the stolen coins, when the news hasn't even broken yet and I already put out a buy order of 50% of the price in case of a flash crash?

Now if you believe the price will bounce back, you can make a lot of profit on stolen coins. Immoral? Quite possibly. Against the law? Not this year.

https://arxiv.org/abs/1706.04964 "Learning Deep ResNet Blocks Sequentially using Boosting Theory"

(As for the lay-man description: I thought boosting performed better out-of-the-box on dense data than on sparse data, because most feature sub-selections for bagging are on zero'd features)

I wonder how close we are to running these "excessive" ensembles in a production environment.

Like how we went from using decision trees to random forests, it seems to me only a natural progression to move from random forests, to a random forest of random forests.

Some Kaggle competitors use over a 1000 RF estimators in their ensemble, but this is not yet possible/pragmatic to put in production for most use cases. But an ensemble of 10 complex base estimators is already within reach for applications that demand the highest accuracy.

About the Netflix prize, the engineers said:

> This is a truly impressive compilation and culmination of years of work, blending hundreds of predictive models to finally cross the finish line. We evaluated some of the new methods offline but the additional accuracy gains that we measured did not seem to justify the engineering effort needed to bring them into a production environment.

So it also depends on the additional gains, if going the route of complex ensembles makes any business sense. But the next 20 years can make a lot of difference.

Anyone have experience putting complex ensemble models in production?

Another progress I find really interesting is the https://arxiv.org/abs/1701.06538 "Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer". It seems possible to learn how to selectively prune a giant ensemble, to select a handful of experts that do well on particular samples. This makes it computationally feasible to get predictions from a giant ensemble. In the paper they solely use neural nets, but I guess there is no reason to not try this with other models, like SVM's or gradient boosted decision trees.

https://arxiv.org/abs/1601.01705v4 (Learning to Compose Neural Networks for Question Answering) comes close to breaking this barrier.

> I haven’t found a way to properly articulate this yet but somehow everything we do in deep learning is memorization (interpolation, pattern recognition, etc) instead of thinking (extrapolation, induction, etc). I haven’t seen a single compelling example of a neural network that I would say “thinks”, in a very abstract and hard-to-define feeling of what properties that would have and what that would look like.

> All the while I'm thinking: this thinking process this person goes through as he analyzes this data: THAT is what Machine Learning SHOULD do

-- Andrej Karpathy

Deep learning for image recognition works because our visual world is made up of structured hierarchical features: Dark/Light, Texture, Edge, Part of Object, Object, Scene. Deep learning layers create increasingly higher-level features in a computationally feasible way.

Random Forests can give feature importance, but that does not account for interactions between features. So, in the end, you don't know how a model made a decision (it could be because there is a feature with high importance, but it could also be because there is an informative interaction between lower importance features).

If you want to compare deep learning with linear models, you should leave image data out of it. Compare them on structured data and bag of words.

MLP's and boosted decision trees, in my experience, definitely beat decision tree and linear models, on structured data. But they lack longterm robustness (complex forecasting models need constant retraining, which can hamper their adoption by business units) and don't pass regulation (it is not enough to say "has_asthma" is a high-importance feature).

In finance and health care, interpretability is enormously valued. It is a constant trade-off between accuracy and interpretability.

A long time ago, Caruana made hospital triage models, with neural networks being the clear winner in generalization performance. Instead, they opted for a simple logistic regression when productionizing. Why?

> [...] patients with pneumonia who have a history of asthma have lower risk of dying from pneumonia than the general population. Needless to say, this rule is counterintuitive. But it reflected a true pattern in the training data: patients with a history of asthma who presented with pneumonia usually were admitted not only to the hospital but directly to the ICU (Intensive Care Unit). The good news is that the aggressive care received by asthmatic pneumonia patients was so effective that it lowered their risk of dying from pneumonia compared to the general population. The bad news is that because the prognosis for these patients is better than average, models trained on the data incorrectly learn that asthma lowers risk, when in fact asthmatics have much higher risk (if not hospitalized).

http://people.dbmi.columbia.edu/noemie/papers/15kdd.pdf

Though there is nothing holding you back from using both simple linear, and complex non-linear models at the same time: Only when the models severely disagree do you pick the interpretable model. Or use the linear model to find data issues, like those mentioned above, that are tremendously obscured (if not impossible to identify) when only using deep learning in a train-test framework.

There are increasingly more ways to distribute wealth unevenly, than evenly.

Reduce the problem to the simplest case of 3 persons: `a`, `b`, and `c`. Person `a` has decision to give to either `b` or `c`. Then use combinatorics:

    from itertools import product

    decisions_a = ['ab', 'ac']
    decisions_b = ['ba', 'bc'] 
    decisions_c = ['ca', 'cb']

    for combination in product(decisions_a, decisions_b, decisions_c):
        print combination

    >>> ('ab', 'ba', 'ca') # uneven
    >>> ('ab', 'ba', 'cb') # uneven
    >>> ('ab', 'bc', 'ca') # even
    >>> ('ab', 'bc', 'cb') # uneven
    >>> ('ac', 'ba', 'ca') # uneven
    >>> ('ac', 'ba', 'cb') # even
    >>> ('ac', 'bc', 'ca') # uneven
    >>> ('ac', 'bc', 'cb') # uneven

Mathematically, closest to that would be Hilbert's program.

Though neural nets can paint like Van Gogh nowadays, asking them to come up with Hilbert's program may be a bit too much of an ask. Yet I would not deeply mind if researchers would revisit papers like http://www.ics.uci.edu/~rickl/publications/1996-icml.pdf "On the Learnability of the Uncomputable".

If a student wants to learn how to play the guitar, you show them 3 chords so they can play Bob Marley or Oasis.

You don't require them to first study consonance, dissonance, rhythm, melody, timbre, dynamics, articulation, texture, form, expression, notation, song writing, Schenkerian analysis, harmonic identity, semiotics, and musical set theory.

Someone who can play the guitar with a passion, can be taught to learn musical notation. The other way around is not guaranteed.

Your suggestion is not necessarily bad: It's good to learn the maths about the Wasserstein metric if you are using GAN's. But for effective teaching your suggestion is archaic, and part of the mindset that makes student's eyes glaze over when being taught mathematics. Can you point to a success story of a student to neural network researcher that did not start with a practical application?

Using 4chan vs. Tay as an example of the limit of chat support bots. Using Netflix's 2012 competition ensembles developed over 1.5 years as an example of AI being too expensive. Telling us that AI will fail when you don't have any data to input it.

Just replace AI with IT/Computer Science and see what remains of this fluff piece.

- hidden text,

- doorway pages,

- cloaking, or

- sneaky redirects

They just show a big popover nagging you to log-in. But you can click this away.

If certain Facebook content pages rank low, or do not rank at all, it is because Facebook actively blocks Googlebot from accessing the content, not because Facebook is trying to deceive Google (or the user).

Though Facebook does not need Google, it could get quite a lot more visitors if it lowered the wall of its garden a bit. As is, Facebook is an inaccessible social echo chamber, and I don't lose any sleep over this.

- You have no/little sites or forums linking to your content.

- Your page titles are uninformative. Biggest offender is probably the homepage, with a page title of "index". But even for your reference page, it it is just "Syntax Reference" (way too general), and Google actually uses your page headings to repair this to "SGML Syntax Reference". Try inverse breadcrumb style "SGML Syntax Reference | Docs | SGML.js". BTW: you ranked 2nd for "SGML Syntax Reference".

- Suspicion: Content not visible (like those in the content slider) is ranked lower than always visible content. Chrome headless crawler can detect this. Add this slider content as regular text to your homepage, and also try to expand content there. Include links to your latest blogs.

- I prefer hierarchical headings, not just sections and <h1> for everything. This, because hierarchical headings can not hurt, but non-hierarchical headings could hurt.

- Finally, SGML being a standard, there are simply a lot of competitors for this keyword. These competitors are not commercial competitors, but authoritative websites with lots of informative content. Exactly the sites that Google likes to rank high. If you want to rank for SGML, you may be fighting an uphill battle.

> Webspam pages try to get better placement in Google's search results by using various tricks such as hidden text, doorway pages, cloaking, or sneaky redirects. These techniques attempt to compromise the quality of our results and degrade the search experience for everyone.

Though I trust Google to make a decision based on user satisfaction, not unrelated meta-politics like this: Any website with a paywall should (and probably is) painted with the same brush. WSJ would benefit from pushing the angle that "Google punished us for doing critical journalism".

* I assume tracking becomes more aggressive when you have an account. For instance, Facebook could connect you to everyone viewing your account page to create a shadow social network for you.

* Skirting real-name policy is against TOS. Using real-name opens you up to crawlers from governments, trolls, collection agencies, and data brokers/analytics companies. Besides governments can request all your data. You also have another account to keep separate /track of, when doing proper OPSEC.

* Even if for you your account is just used once a month, people you are connected to may have different expectations ("you have an account, why didn't you reply?"), and you have to micro-manage this.

* You miss events, while still having an account or invite, and it is bad social form.

* You are forced to combat social persuasion tactics: Facebook is carefully design to maximize clicks and time-on-site. It's like trying to quit drinking while going to the bar once a week: your determination is actively attacked.

* People can tag your account in photos. A high school teacher I know got in trouble because she was tagged in a photo where people around her were drinking alcohol.

As for Youtube: seems like you fell into the trap of planning too far ahead. Most channels never work out. Even if you stick to a weekly schedule, viewership will only steadily rise. Why not release your current videos, try a (bi-)monthly schedule and see where it goes?

Fear of commitment is a rationalized excuse your brain makes to avoid burning calories :). Or it's to avoid a setback: you put in all this effort and it did not work out. People avoid relationships for fear of commitment too, without ever giving it a shot, or asking if the other person also wants commitment.

For open source or research I give myself very small timeframes. You can think about it the rest of the week, and have a very productive hour or two coding or writing.

As for procrastination, check out "productive procrastination". It helped me get stuff done, when even more important stuff needed to be done :).

http://www.lifehack.org/articles/productivity/10-ways-for-pr...

The summary gives the example of securing your neighbors roof when a tornado is about to hit. Possible laws to break to do this, are "breaking and entering" or "trespassing".

Note that a lot of these laws state that care must be taken not to break laws unnecessarily. Bricking IOT devices that can be used for DDOS-attacks may be a step too far.

And strictly, in the case of patching a server under negotiorum gestio, you have not broken any laws: It is not unlawful computer intrusion when you have implicit permission of the owner of a device (the same goes for entering your neighbors house when they are on vacation, and have accidentally left a pot of milk to boil on the stove).

But I guess such far-reaching Good Samaritan laws are very foreign to the US, since there, off-duty doctors are sued for performing a painful Heimlich maneuver.

> Negotiorum gestio (Latin for "management of business") is a form of spontaneous voluntary agency in which an intervenor or intermeddler, the gestor, acts on behalf and for the benefit of a principal (dominus negotii), but without the latter's prior consent. The gestor is only entitled to reimbursement for expenses and not to remuneration, the underlying principle being that negotiorum gestio is intended as an act of generosity and friendship and not to allow the gestor to profit from his intermeddling. This form of intervention is classified as a quasi-contract and found in civil-law jurisdictions and in mixed systems (e.g. Scots, South African, and Philippine laws).

> For example, while you are traveling abroad, a typhoon hits your home town and the roofing of your house is in danger. To avoid the catastrophic situation, your neighbour does something urgently necessary. You are the 'principal' and your neighbour here is the 'gestor', the act of which saved your house is the negotiorum gestio.