karmacondon

I think it would be interesting to find a way to put the study projects to practical use. For example, look for incomplete feature requests on open source projects in various languages and critique the students on their implementations. Or find people who need small website work done and set up a dev environment where the students can learn on a real project, and then get feedback on their work. A kind of practical learn by doing, with a teacher.

The approach you describe is one of many. Traditional collaborative filtering using matrix factorization will "discover" the features you listed, or other non-intuitive features, from patterns in the data. They don't always need to be listed explicitly, and training individual asymmetric models (like your one NN per person) isn't always ideal. Per person models won't make a lot of sense unless each person has rated a large number of songs, which usually isn't the case.

In general, methods like this require an epic amount of data to work well for consumers. The Netflix Prize made a dataset with 110 million ratings available, and that was barely enough to make impressive predictions. The issue here isn't the statistical model being used, but the paucity of training data. If you trained this or any other model on one billion rated songs, the quality of the recommendations could blow your mind. They would however still skew toward the average musical tastes.

I'm starting to lose faith in the concept of online voting. You get Pinboard, Boaty McBoatface and white house petitions about the Death Star. Maybe it's how easily links can be shared to reach a wider audience. Or just something about the nature of using the internet to vote. For some reason, nonsense seems to be more likely when people make decisions over http.

There's a fine line between "wisdom of the crowds" and "American Idol for startups". I'm not sure exactly where Apply HN falls. It doesn't seem like any individual investment will make or break YC, and this is an interesting idea for an experiment. But I don't come away from this feeling upbeat about the democratic process.

The only way I can see UBI working is if most people get no benefit. So someone making the average US income of $50,000/yr would get a check for $20k from the government and then see their taxes go up by ~$20k. Rich people would pay much more, effectively transferring their wealth to the poor. People who are solidly middle class wouldn't see much change, except having a new class of potential customers (the people that used to be poor).

This is why we can't have nice things. Protest votes are fun, but they also undermine the system. "If people don't take voting seriously, then what's the point of voting in the first place?", people in power will ask themselves. So yeah, it's cool to vote for Donald Trump or Boaty McBoatface or whatever other silly thing. But the protest itself isn't accomplishing anything. Choosing the silliest option available doesn't make you Ghandi. It's just a waste.

Making it into one of the two slots available for YCF could be a really big deal for someone. The founder of Pinboard is already very successful by any standards, and the product itself probably isn't going change drastically. Can't we let someone else catch a break?

Going to use HN for some quick market research... would anyone be willing to pay for a human curated news feed? Say $10 or $20 a month for someone to go through news sources and pick out the stories that will be interesting to you, and potentially summarize them?

I'm thinking of curation based on a personal relationship instead of an algorithm. You would indicate that you're interested in machine learning, libertarian politics and cooking (or whatever), and people would check hn, new york times, etc and compile lists of those items. I know it's not the same thing as this post, but it's a problem and I'm wondering if anyone else feels it strongly enough to pay for a solution.

So many comments here are in the form of "People won't trust autonomous vehicles." Well I'm people, and I trust them. And it's for one simple reason: Whatever bugs they have can be fixed for every autonomous car on the road at the same time. That just can't be said for human beings. Every one of us has to learn how to idle icy roads, fog and other hazards individually, with wildly varying results. A computer program does it once, and is constantly getting feedback from the millions of cars on the road all the time.

Sure, people want someone to blame for accidents. And people might not like the idea of cars without drivers, or worry about contrived "Does the car kill the driver or the pedestrian" situations. But at the end of the day, statistics is overwhelmingly on the side of driverless cars. Saying "Yes, but people aren't reasonable enough to see that" isn't helpful. The question is, how quickly can we get people to do what's best after the technology has been perfected?

The company making the vehicles will be responsible for any damages, and paying a massive insurance premium I'm sure. There will be legal consequences for gross negligence and anyone damaged will be made whole. I don't see how that's any different than the system we have now.

And the emotional appeals cut both ways. How long before the parents of a child killed by a drunk or sleeping truck driver say "Why was he driving anyway? Why wasn't he replaced by a machine that wouldn't drink a six pack and nod off while at the wheel?"

As an aside, I have thought this same thing. The non-inverted pyramid style is horrible and pervasive. I would pay $$/month to have technology or people summarize these things for me.

The US government isn't going to kidnap Edward Snowden at any point. He's too high profile and the press would eventually find out. And I don't think Putin will let him leave Russia unless it would hurt the US in some way. If Snowden really wants to leave, he can probably leave. But the Russian government won't make it easy for him.

Snowden should just come home and face the music. He'll probably get a pro bono all-star legal team and have 50/50 odds with a jury of his peers. With sufficient lawyering he could probably get away with five years at worst, and then go back to life in America as even more of a hero.

No phonograph or light bulb on the original list? How odd.

This is either insane, or pure genius. Has anyone else tried this?

YC includes a vesting clause as a standard part of their incorporation documents, in part for just this situation.

I've heard that a corporation was formed BEFORE the YC application. If shares were issued in that corporation without a vesting clause, then that's that.

If shares were not issued and the YC application is the only evidence of a 50/50 split, then why wouldn't one assume that vesting was part of that agreement? The question on the application was "describe the planned equity ownership breakdown...". Someone could reasonably argue that they agreed to a 50/50 split given the assumption of a vesting schedule, and would not have agreed to it otherwise. Which is to say, if the YC application is the proof of ownership, then how can you ignore everything that is also implicit in that application?

I think it's a "rule" as opposed to a strategy like "don't sacrifice your pieces in exchange for nothing" is a rule in chess. It's possible to succeed without following that guide line, but it's very difficult.

VC firms produce modest returns for a very high risk. You've probably only heard of one out every hundred companies that have been venture funded. Low risk investments just don't produce enough return.

The only way I can see Basic Income working is if the average person has a net income change of $0. That means someone making the average income of ~$45k/yr gets a $10k check from the government every year, but also has their taxes go up by ~$10k. I believe others have referred to this as a negative income tax.

From what I can tell the average salary of a hackernews reader is above $100k. Each of us would probably end up paying ~$800+ per month in additional taxes, maybe up to $1,500/mo for those making that BigCo money. [1]

The first question is, does that sound reasonable? Do you want to subtract $400 from each paycheck for the sake of total strangers? And I guess the second question is, why aren't we doing that now? How much do each of us currently give to support other people in various ways?

I think Basic Income is a good idea. If people have more money, they'll be able to click on internet ads and buy things, and that will help the companies that many of us work for. But it won't be an even exchange and many people here will end up with less disposable income. I think that it will be worth it in the long run, to ensure that the economic engine that produces our paychecks continues to function. But I don't know if every UBI proponent is prepared to deal with the short term costs.

[1] I completely made these numbers up. Could be more, could be less.

I think the fundamental purpose of a business is to accomplish a task, not to employ people. For example, I don't care how many people are employed to keep hackernews running. I'm thankful for their efforts, but I don't care if it's one person, ten people, or one bot. Or how much any of them get paid. The only thing that consumers and businesses care about is the end result.

If a job can be done by a machine, that's fine. People will find new forms of employment. Or they won't. But we can't expect businesses or consumers to misallocate their resources for the sake of someone's job.

Fortunately, fully automating the process of driving or running a restaurant is decades away. We have time to implement a basic income. Or just come up with more ways to employ people. The number of available jobs is limited only by imagination. "Instagram Photo Producer", "Online Forum Discussion Facilitator", "3D Printed Object Finisher". Someone can pay someone else to do any given thing, and now that's an occupation. "Potential Job Titles" isn't a finite list, and we can't ever run out of them.

Why do you think minimum wage laws exist? It's because at one point in our nation's history, businesses implicitly colluded to set wages low. They have more power in wage negotiations than workers do, because there are generally people who can afford to do unskilled labor for very small amounts. Businesses got greedy and took that too far, offering people wages far beyond a minimum standard of living. So in response, the people voted to set a minimum that would prevent corporations from being outright exploitative.

The same thing could happen today. Google engineers will always be paid a lot of money, but the average McDonalds workers could see their income fall way below the national average cost of living. Whether the minimum should be $5/$10/$15 is up for debate, but without businesses would have an unfair amount of bargaining power.

This is what passes for humor in the ml community? Ouch.

I would normally flag this as being April Fool's Day nonsense, but I like the idea too much. The title is an homage a karpathy blog post that demonstrated how an RNN could write code that didn't necessarily compile, but looked reasonable. It would be pretty cool if someone took that and tried to develop it into something, though the StackOverflow bit might have been a bit much.

Sooner or later one of these April Fool's Day things is going to turn into a serious project.

The problem with not intervening is that it doesn't work unless every other actor in the region agrees not to intervene as well. Russia likes Assad because Syria gives them influence in the region. They're going to give the Syrian army aid and resources whether the US is involved or not. The US got involved in this ethnic conflict in part because they generally don't want Russia to gain any more influence in any part of the world. So they try to give equal amounts of aid and resources to the Syrian opposition. One side makes a move, the other side makes a counter move.

Whether or not you agree with those actions, or the whole system, is a matter of politics. I'm not condoning anything, just trying to explain the potential reasoning behind the decisions. I do hope that radical islamism runs its course in the same way that communism did. One big complication is that there are a lot more foreign powers involved in the Middle East than there were in the Soviet Union. It could be harder for them to find their own way while being manipulated by so many other hands.

So what happens to the other 200+ iphones that law enforcement needed opened?

It seems like there are at least certain iphones that can be opened using this method and similar hardware based approaches. Are the feds going to move forward with that? It seems like accessing this particular phone wasn't trivial or cheap. So I'd imagine that it will come down to whether or not it's "worth" the cost of flashing individual chips for each of the other devices.

And what happens if there is potentially exonerating evidence on one of those phones? Does the defense team have to come up with the money to pay for a lab or outside company to open the phone? And also, is an encrypted phone going to be the new 'dna evidence'? Like, will a brand new iphone that's linked to a major crime be held away for years until the security community can hack it, potentially setting people free or sending them to jail?

Sorry for all the interrogatives. Even though this particular case has been settled, there are still a lot of questions surrounding default device level encryption.

Well, if you want to be technical it doesn't appear that she was told NOT to use her personal BlackBerry and email server in any official way. This article mentions a memo that said "Any unclassified Blackberry is highly vulnerable in any setting to remotely and covertly monitoring conversations, retrieving e-mails, and exploiting calendars." This isn't the same thing as an official policy.

I know this seems like nitpicking, but you said "She went against protocol because she wanted to and she could, and she should own the consequences". The fact is that she did not go against protocol. If the president or congress wanted to prohibit the use of certain communication devices, they would have. Just like if someone's supervisor didn't want them to use a blackberry, they wouldn't write a memo pointing out problems with the device. They would just say "Don't use a blackberry."

The issue is did she break any laws or policies regarding the handling of classified information. That's for the courts to decide. But if there was any evidence I'd imagine they would have brought an indictment by now.

The data show no such thing. Correlation != causation.

People self-select where they live. It could easily be that happy people choose to live in the country instead of the city, or people in cities have more stressful jobs, etc

I still think that Uber's biggest innovation was charging 30% less than cab companies. Given a smart phone, I have 0 problem looking up a cab company and calling them (in most cases). And New York City cabs have had apps for years.

The Uber app has a lot of details that are nice, but not essential. The minute they no longer have the cheapest fares, I'll use any other solution.

I would really really like to pay per article for content from the New York Times, the Wall St Journal and a few other news sites with annoying paywalls. Whenever I go to subscribe I think, "This isn't worth it because I don't read publication X frequently enough to justify a bill of any size on my credit card". But if this lets me just pay $20/mo to this service and get access to N articles from a number of subscription based publications, then this is a good deal. Sign me up.

But I'm just not clear on how this works from looking at the home page. Is this all paywalled content? Or will I be paying $0.25 for something I could have read for free, like a sucker?

And also, is the NYT coming out ahead here? Does it make sense to split ~$0.30 per article with blendle instead of getting a $1/week or whatever from a regular subscription? Because I would hate to get used to using this and then have it disappear in six months because the basic economics of the business don't make sense.

If it works out I think it does, this is great. If not, hopefully it can get there soon.

The obvious answer here is that basic income won't be evenly distributed to all citizens. A household earning the average national income will get $1,000/month from the government, and see their taxes go up by about $1,000/month. So it's difficult to raise rent on everyone when only a fraction of your customers have true extra income.

Imagine you're a landlord who runs a property for those of normal income. How much do you raise rent because people who currently earn below the poverty line suddenly have $1,000 extra per month? You can raise rent on everyone, but then most of your middle class tenants will move out to seek better prices. The same is true for the price of groceries and other staples.

A smart business will maintain their prices and benefit from a larger customer base. Since everyone doesn't experience an increase in income, raising rents of all forms will only open those businesses up to price competition.

There have been multiple tests of the ideas behind BI:

* https://en.wikipedia.org/wiki/Mincome

* https://decorrespondent.nl/541/Why-we-should-give-free-money...

* http://freakonomics.com/podcast/fighting-poverty-with-actual...

The problem is that it often isn't possible to falsify or confirm ideas based on limited tests. You can set up a minimum income for one town, or give money with no strings attached to several groups of people, but it won't answer a lot of the long term societal questions. It's like asking if you can test social security or universal healthcare. Those tests might not illuminate problems that could occur on a large scale, 50 years down the line.

Sometimes the only way to test something is to do it.

I'm ashamed to admit that I'd never heard of Andy Grove until today. I did find a great discussion of High Output Management by Ben Horowitz that may be useful to fellow uninitiates: http://a16z.com/2015/11/13/high-output-management/

The slight danger of this "post your salary" trend is that it's really difficult to make comparisons out of context. Someone is going to go to their supervisor and say, "Hey, it says that a senior dev is getting paid $200k/yr at Company X. I want $200k too!" and then their supervisor is going to say, "Well, we don't feel that you're worth that much money."

And now what? The supervisor could be 100% right, and maybe the employee just isn't worth that amount of money. Or maybe it doesn't make for the company to pay that amount of money. The employee could claim discrimination for any of a number of reasons, and it's almost impossible to prove that person X isn't worth as much to the company as person Y is to a different company. It seems like it could just create acrimony in what was previously a good relationship.

Transparency in pay is important, but it's almost impossible to evaluate in an empirical way, much less a scientific one. Not to mention that there is wide discretion in compensation. It's illegal to pay someone less because of their race or gender, but employers can pay someone an extra $5k a year just because they liked the color of their tie during the last negotiation.

Get paid what you're worth, but don't depend entirely on others to find out what that is.

What would be the advantage of binarized gradient descent?

I have a hard time believing that Google sold Boston Dynamics because they didn't want to be associated with military applications or with "evil" AI. Boston Dynamics had deep relationships with the government and clear military applications when Google decided to acquire them. I don't see how the PR situation would have changed between now and then.

This seems like a case of a merger/acquisition that just didn't work out. Hopefully it's for the best. The military robots that BD makes have the potential to save a lot of lives and do good, in combat and non-combat situations. Ultimately, Google/Alphabet is an advertising company. Maybe it's better for Boston Dynamics to go their own way. They can still maintain a relationship with Google and other companies to share AI knowledge and technology.