One interesting study would be the correlation between the characteristics of the hiring manager and team members. In my experience the correlation is strong, especially in terms of race and ethnicity. At Bell Labs in the 90's, one out of the three research area organizations had a very high representation of Indian managers and researchers. At Sun, the same was true except that most managers and researchers were white (Americans, Europeans, Australians, etc.). I worked at another company where the manager was Serbian and two other researchers were Serbian.

This is a hypothesis that can be fairly easily corroborated with statistical studies. I'm fairly certain that the bias exists. Of course, whether that bias is good or bad is a separate question.

Two questions about this research:

(1) How did the researchers account for operational language profiles? Language A may have more negative words than positive words, but maybe the one happy word is used 80% of the time. To me, the incidence of positive vs. negative usage is much more important than the histogram of the available vocabulary.

(2) How did the researchers compare the same word in different languages? Is this comparison possible without the introduction of bias in the selection of words for each of the two languages. From the paper authors' website, "This is a comparison between the average user reported happiness scores between several languages. The "happiness" of each word is rated by 50 distinct users on a scale of 1(sad) to 9 (happy). Words from each row language are then translated into each column language and intersected with each other corpora."

So, how much are the results a reflection of the experimenter's biases and skills in translation to the 2nd language. I'm suspicious of this type of comparison. From the article (not the paper), "For example, on a scale of 1 to 9 with nine being the happiest, Germans rate the word “gift” as 3.54. That’s slightly negative. By contrast, English speakers rate “gift” as strongly positive at 7.72." As a somewhat fluent German speaker, I know that the German word "gift" means poison, and I would consider it not just slightly negative but extremely negative. If the experimenters actually presented the German speakers with the German translation of the English word "gift", e.g., something like "Geschenk", then I imagine the German response would have been very positive.

The problem is with the concept of a "legitimate" offense, as though some quintessential characteristic of an action or statement should outlaw personal feelings. How one feels is reality. For those of who are married, try telling your wife that she shouldn't feel offended because you didn't mean to hurt her, so that should make her feel better.

This entire discussion of "legitimate" offenses boils down to whether one cares about what others feels. If I care about the listener and that listener feels offended, then I would address their feelings, regardless of my personal feelings about the offense. If I don't care about the listener, then I either ignore the listener, or if I feel the sting of societal condemnation, I attack the "legitimacy" of the offense or the morality or intelligence of the listener.

It's much more nuanced than that. There's this incorrect aggregation of the notion of offense, as though something is either offensive to everyone or offensive to no one. A speaker can offend without intention to offend. It should be obvious that each individual is the only person who has not only the right but the ability to determine personally felt offense. The right and ability to determine offense for oneself should not be confused with the legal or moral right to determine the resulting societal actions.

That is, each person gets to decide if the term "redskin" is personally offensive. However, just because one or more people take offense doesn't necessarily grant that group or society as a whole to impose sanctions for that offense. But, likewise, the lack of societal sanction should not be extended to prohibit personal feelings or thoughts. In fact, in my opinion, hearing "You have no right to be offended!" is much more hurtful than the original offense.

Making any comments about the desirability or appropriateness of the percentage of women, blacks, hispanics, etc. in a company implies personal knowledge about the desired or appropriate range of percentages. But, that's where it gets difficult. It's easy to say that the numbers are too low, but it's hard to pin down what the target numbers should be. Should the percentage of women be 50%? But, if there are 10x more men applying for the job, is that appropriate? As others have discussed in this forum, a separate question is whether the 10:1 ratio of men:women applicants is appropriate.

This discussion also presents perhaps a hazy melding of two distinct, important questions: (1) Are there an appropriate percentage of a certain demographic group within a company, and (2) do the members of that demographic group have the same opportunities to get a job in that company? The first question deals with the aggregate end result, while the latter question deals with the probabilities for an individual. In my opinion, the second question is more important but difficult to measure. So, the first question is used as a lazy alternative.

Actually the paper says that the Google guys approximated power-on hours with a notion of age, which I assume was approximated by a knowledge of either the manufacture date of the delivery date. From the paper, annualized failure rate (AFR) is somewhat correlated with age, but not necessarily strongly enough to predict probability of failure. Even with their large drive population, the paper points out that the drive model mix is not consistent over time and therefore, not much can be made of the apparently weak correlation between AFR and age, which could be perhaps be more greatly influenced by drive model.

The negative correlation with very cold temperatures is interesting but hard to understand without further analysis. Perhaps some drive models didn't handle fly height adjustments well at low temperatures. It's hard to figure out without more data. It should also be pointed out the temperatures were obtained via SMART, and the SMART standard doesn't mandate how temperature is reported. So, different manufacturers could report temperatures in different ways, i.e., different locations (which can easily vary by up to 30 degrees C), different aggregation methods (time windows, sampling frequency), etc. So, the aggregate data is probably not as useful as the data per drive model.

I actually prefer the built-in system. For example, I have a limited data plan because my non-wifi data needs are low, so using an online map like Google Maps or Waze costs extra on a recurring basis. With a phone, I would have to go through the motions of removing it from my pocket, attaching it to the bracket, and plugging it in, as well as the reverse motions when leaving the car. I would also need a special bracket to hold the phone, and even with this bracket, the screen would shake more than an integrated system. The phone screen is significantly smaller. The phone audio is not integrated into the car speakers, so audio levels have to be adjusted, and navigation directions sometimes compete with music. The built-in system also has a physical joystick and dial with voice feedback (not to mention voice recognition) so that I can control it more easily than the using the phone's touchscreen.

I do use my phone for navigation sometimes when I drive my older car. But, given a choice, I always prefer the built-in system.

The goal of the upgrade was an easy performance upgrade, i.e., pop it in and enjoy the speed.

What my experience shows is that DRAM costs less, yields a better performance increase, and most importantly is plug-and-play. Obviously, different machines and operational workloads will affect results, but if SSD performance is not plug-and-play, the value proposition for many users is greatly diminished.

Hmm, if you assume there is no seek time, then ideally IOPS will be 120 if your request size is exactly a full track, which is almost never the case. IOPS will be 1s / (seek_time + rotational_latency + access_time + overhead). How many IOPS you can get from a 7200rpm drive depends on a lot of factors. Are the requests uniformly randomly distributed? Is NCQ enabled? What is the max track span of the requests? If the max track span is narrow and on the outer part of the disk, then you can get a lot IOPS, quite a bit more than 120.

Whether or not an SSD is worthwhile depends on the use case. I just put a low-end SSD in an old desktop to replace an HDD. The performance difference was basically negligible. I was quite disappointed. The system bootup was a bit faster, but application startup wasn't any better. And, since I always put my machine to sleep instead of shutting down, bootup times are unimportant. Fortunately, I could live with the dramatically smaller storage capacity because my main storage is on another machine. In contrast, I doubled the DRAM on my other machine, and the performance boost was amazing. The $30 I spent on the DRAM was a much better investment than the $70 for the SSD.

This is an example of correlation not necessarily indicating causation. There's nothing inherently wrong with a temporary structure. Does the fact that some temporary structure have bad ventilation mean that all temporary structures have bad ventilation? If you take the same temporary structure and move it to a rich neighborhood, do the rich kids have similar levels of absenteeism?

Of course, a poorly designed or erected temporary structure is bad, just as a poorly designed or erected permanent structure is bad. However, the underlying problem is funding (and the associated problems arising from neighborhoods with less educated and poorer parents).

Artificial intelligence and neural networks are a hot topic, so this is piggy-backing on that trend. It's not a surprise that Andrew Ng's work is referenced quite a bit.

While the modeling is interesting, it's seems to be just modeling at this point. The main claim of the white paper is high PUE prediction accuracy by the model. While that's academically interesting, the real use is in feedback for optimization. The white paper author realized that and included that optimization problem as one of the examples in the paper. However, the optimization was achieved "through a combination of PUE simulations and local expertise." I'm guessing that the local expertise part was relatively significant because there is basically no discussion of this even though it is the one application that would really make this work practical and really interesting. The paper claimed that this neural network-based optimization reduced PUE "by ~0.02 compared to the previous configuration." But, I have no idea how that would have compared to optimization using just local expertise without the benefit of neural network modeling.

While it's true that skin color and wealth cannot be used as independent indicators or predictors of academic performance, the correlation is nonzero. I would even venture to guess that the correlation is more than weak. Yes, correlation is not causation, but correlation is a definite indication of, well, correlation, i.e., there's a relationship.

It's an excuse when we don't like the idea, but a truism when we do. Racism may not be a relevant cause of poor academic performance, but it's statistically clear that racial factors have significant correlation to academic performance. [Please no correlation is not causation garbage -- that type of thinking is just an excuse to avoid further discourse.]

I agree. I only use my phone for situations where a PC is unavailable because the PC user interface is almost always vastly preferable. Yes, Angry Birds is better with a touchscreen, but it's the exception. A physical keyboard is so much better than a virtual keyboard in terms of speed and annoyance. A mouse with buttons has much greater speed and utility than a touchscreen. A 24" screen is much more readable than a 5" or 10" screen, and web page formatting is more usable.

The sole redeeming value of the phone is its mobility.

Absolutely. The original commentary that it's the disc-based DRM that's offensive is illogical. Streaming is the always on, immediately revocable, non-bypassable version of DRM. It's the perfect DRM.

Apple is mucking around with accounting numbers to argue that iPad sales haven't decreased. They may actually be correct. However, that's a moot point. What is inescapable is that iPad sales are not growing. And, this trend is not a single quarter phenomenon. What should be troubling for Apple is that the iPad doesn't appear to be following the iPhone growth in sales and that this growth slowdown has come at a much earlier product age.