1) Every advertisement is a recommendation.
2) Some content is an advertisement.
3) The recommendation algorithms usually assume the user wants only to view similar content which can result in an echo chamber and is more likely to result in indoctrination.
4) The recommendation algorithms could assume the user wants to view all sides of a topic which is not an echo chamber and is less likely to result in indoctrination.
5) The case is being made that doing (3) is aiding and abetting the terrorists.
6) Had they done (4) instead of (3), IANAL but I doubt they could make the case for aiding and abetting the terrorists.
I would be happier if the Supreme Court said that you are protected by section 230 if you do (4) but you are not protected by section 230 if you do (3).

"The only thing that can ruin an entire economy is a government on a misguided quest for economic equality [cnn.com] -"

This is clearly not the only thing that can ruin an entire economy.
For instance, the government could be on a misguided quest for economic inequality. If all of the wealth/rewards went to one individual or a small group of individuals there would be no incentive for anyone to work hard since hard work would not equate to improved living conditions. Fortunately, in this day and age, one can modify the misguided government by voting the current people running it out of power.

If it can make the words sound like someone, then it must be able to get the emotions and emphasis correct. I imagine this would be useful in translating someone's voice into a foreign language with all of the correct inflections, emphasis, and emotional impact. If it could be used in real time, you could talk out loud to someone and they could hear the translated version through their ear buds. It might even sound like the person if they learned the language themselves. Of course, your gestures and facial expressions may be out of synchronization. This might be like a "babel fish" for audio. It might make travel and international communication easier. For instance, the people in call-centers would not need to know the language of the person they are helping.

I think Dean Kamen might have solved this problem already.

http://www.redferret.net/?p=10...

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

Taking things to the extreme, subsistence wages means zero market for manufactured goods. And taking things to the other extreme, the alternative of zero employment in manufacturing means zero market for manufactured goods. So manufacturing companies in the process of automating will have to produce less and less at either a lower and lower price or for fewer and fewer people making it harder and harder to pay off their investment in automation. In other words, they will lose their economy of scale making capital investments harder to pay off. So the extreme cases imply that manufacturing companies, or more correctly all luxury suppliers, are eventually doomed to shrink unless people can add value in a non-manufacturing capacity. Seems unlikely. More likely is that extreme cases don't generally produce the correct answer when the system has non-reinforcing feedback, and the economy definitely has non-reinforcing feedback. Which implies that since both supply side economics and demand side economics are based upon extreme cases, they will probably not give a correct answer either.

Also, unless each robot has to file an income tax return, the government will lose tax base and either will have to increase taxes on the rich, will have to live within a shrinking budget, will have to print more money, or will incur a ever increasing debt. Incidentally, a similar thing happens when a job is outsourced to a different country. Unless each outsourced employee has to file an income tax return, the government will lose tax base and either will have to increase taxes on the rich, will have to live within a shrinking budget, will have to print more money, or will incur a ever increasing debt.

It doesn't measure "if the programmer was over or under-qualified to write the particular code" either. (IDNRTFA) but sounds like it measures biometrics such as heart rate relative to the code module shown on screen. I am guessing that the researcher assumes a cause and effect relationship between code difficulty and heart rate, presumably, since they detected a correlation when testing in controlled environment. I assume that they then assumed the "real world" functions the same as their "controlled environment" and propose using it in the real world to have a real cause and effect on some poor programmers, presumably, so the researchers can make a lot of money selling/licensing their product which, if true, might be a conflict of interest.

Also, the US and foreign companies will move all of their high technology work overseas as it will be the only way to secure their trade secrets, proprietary information, and intellectual property that has not yet been patented. So, there will be no US technology companies. The bad guys don't have to destroy the United States if they can manipulate the US government into doing it for them.

At the beginning of every downturn in the economy, the companies make do with fewer employees by either not filling open positions or by laying people off. So the productivity goes higher since measurable-output/employee = productivity and measurable-output takes awhile to be impacted by the loss of the employees. We say that changes to the measurable-output lag behind changes in the number of employees. This is partially because other employees, who know their current job well, put in extra effort to do the job of the missing employees. Eventually, the employees doing 2 jobs reach burn out or find another job and leave the company. At that point, the measurable-output drops dramatically resulting in a corresponding drop in productivity. The company then hires two people to replace the employee they lost who was doing the two jobs resulting another large drop in productivity. I am not an economist, but I don't think productivity is a very useful indicator.

Productivity may also drop when you hire two new college graduates or two H1B visa employees instead of one older experienced programmer.

As I understand the HFT (High Frequency Trading) (Flash Boys book) often saw an order on one exchange and beat the order to other exchanges that were needed to fill the order thus driving up prices. Liquidity is fine but not at the expense of something similar to "front running". Each exchange should list the number of shares they can fill. Purchase orders should be made for fewer shares than the number they can fill rather than relying on the exchange to fill the order from other exchanges. The person making the order should place orders at several exchanges so the arrival time of the orders is simultaneous. So the exchanges should allow pinging to determine the arrival time. Of course, this has to be over a dedicated line as the internet can send packets by different routes.

This case seems to be about someone purchasing options to manipulate the stock price. So if a source creates a lot of orders which are never executed, why does it affect the price at all? It seems the market does not rate the credibility of the orders that are posted. If someone buys an option to purchase at some time in the future at some price but has a history of cancelling 99% of the their options, why would the market give any credibility and consequently a corresponding price change in response to the purchase of the option? The new or unknown people should have limits on the value of the orders until their credibility is determined at that level of order. So they would have to work their way up a tiered system to affect the market. The higher the value of the order the higher the impact on their credibility. All options are not equally credible.

"Those patents disclose algorithms. Basically, applied math. Which should have never, ever been allowed as claims in a patent since they are antithetical to the compromise between the inventor's and society's benefit the patent system was designed to facilitate."

False, algorithms are not math. Math refers to proofs of the basic relationships of mathematical constructs. Algorithms are original procedures to do a certain thing. I believe the patent office allows patents on algorithms since there is more than one algorithm to accomplish the same thing and it does not prevent anyone from doing a proof. It also promotes the development of better algorithms which is what the patent system is for.

"So, yes, they are pretty much what one would call "software patents"."

False, algorithms are not what one would call "software patents". Software patents refer to patents on software/business practices which consist of unoriginal procedures to do common things such as "putting products in a shopping cart".

"That being said, under no circumstances would I recommend a client to hire you if I caught wind that you owned patents applicable to the field in which you would be working. That simply screams "conflict of interest", "subsequent lawsuit", and "humongous liability.""

One can normalize the perceived "conflict of interest". Before becoming employed by a company, start your own company (LLC or small corporation) and assign the patents to your company. You will own your company and your company will own the patents. You will not be able to give away, sell, or license the patents unless you sign as CEO of your company or sell them ownership of your company. A company is an entity independent of you even though you own it. No company should complain about this as you probably own stock in many corporations which hold intellectual property and they know they must deal with the company if they want access to the intellectual property. I am not sure how they will like having you be an employee of their company and the maintaining CEO of your own company but you can separate "newly generated intellectual property" from "intellectual property related to supporting the application of patents by your company".

I think the correct question is "Is prior-to-publishing peer review necessary?". Something like arxiv could publish everything and then scientists and/or engineers could have logins with their real names and qualifications under which they could read and review the papers after they are published. A person could search for papers using various filters. (ie. unreviewed, reviewed by individual, reviewed by 10 people with these credentials, reviewed and notation of prior publishment or related articles, notation of articles contradicting the results, retracted etc . . .) Then a person could quantify the quality and quantity of the review and better guage the quality of the work. Also, if a researcher was giving false reviews to "game" the system, all of their reviews could be deleted from the system in short order. If someone was rediscovering or redoing the work of lesser known people who published first, then the lesser know person could note the prior publishment in their work. Also, one could get a timeline tree of the development of work in the area. So everyone gets published in a peer reviewed journal. There would be no peer-review-induced delay in publication. The level of peer review quantity and quality is quantifiable. Gaming the system could be detected and removed. Work would no longer be presented as standalone as future papers contradicting the work could be linked to it. Tenure committees would have a more refined tool than mere quantity as a person could note who was reading their work to indicate its importance.

Rather than ask if Intellectual Property is important. Is brand name recognition important to your business? Is marketing important to your business? Are brand names important to you as a customer? When you work, is it important for you to get paid? Is your company name important for your business? Should a company be paid more for higher quality products? Should a company be paid more for innovative products? Should a competing company be allowed to reverse engineer your product and market the exact same thing at the cost of manufacturing until you go out of business?

If you have not tried it, try to inline your simple functions. The cost may be your function call overhead. I am assuming you are using C or C++. Also, it may help to look at my post on the same level as this one.

Information for information works as well as money. Thanks for the iterative CORDIC algorithm information and clarifying reformulating the problem to not use trig was an option. I was using an Intel 386 processor, and I may have purchased a coprocessor, but I don't recall for sure. If I did, my measurement may be of the coprocessor time rather than the algorithm. I am also not sure if I compiled with coprocessor support or not. Thanks for pointing out my possible error. The method in the tan() implementation was to use Chebyshev polynomials (See http://en.wikipedia.org/wiki/Chebyshev_polynomials#Trigonometric_definition) to make the Taylor's series converge faster (See http://en.wikipedia.org/wiki/Approximation_theory). It required some rescaling as well. First, find the Taylor's series of the tan(x) function to the number of terms needed and then choose the Chebyshev polynomials T_i(x) with values smaller than the approximation accuracy you need. Use the largest T_i(x) first since T_(n+1)(x) T_n(x).. Solve for the highest power of "x" and use it to eliminate the power of "x" in the Taylors series. You may have to add up the errors of each Chebyshev polynomial you use. For instance, T_4(x) = 8x^4 -8x^2 + 1 so x^4 = (1/8) * (T_4(x) + 8 x^2 - 1). Use x^4 = 8 x^2 - 1/8 to eliminate x^4. This reduces the number of multiplies and additions needed for the evaluation of the Taylors series to a certain approximation. I did also see a case where someone used fractional powers of "x" to possibly reduce the complexity still further. I am recalling this from memory so a I may have got a few details wrong. One of the 3 books that mentioned this method was a Math book by Korn and Korn. There is a Dover edition.

Back in the late 80's/ early 90's, I considered using a lookup table for some trig calculations. The first thing I did was check the computation time for a "multiply" and for a tan() calculation in a loop dividing by the number of iterations. I was supprised to find the cost of a tan() computation was only the cost of 8 multiplies. I am curious as to how you are beating this using vector math. Are you are using normalized 3 vectors and the dot and cross products to compute sine and cosine? And maybe sin(x) = sqrt( 1 - cos(x) * cos(x)). While I was using closed source code, a friend had access to the open source implementation of tan() and printed it out for me. They might have been different. I might still have it. It was an interesting technique. I eventually found a book that described it. I believe I still have the book. How much would you pay for the information? I like people who don't like math, they pay well!