About a month ago, I found Wikipedia's page on the concept "Thought-terminating cliché". I really liked it and it put a (new) name for me on to something that I - in my own little mental world - had previously called "mental roadblocks". Unfortunately the page has been deleted now recently in August 2013, so I retrieved the last copy of it. It is below and under the same license as the articles on Wikipedia. I have no idea if the contents meet the standards of Wikipedia, but I find the thoughts presented interesting:
A thought-terminating cliché is a commonly used phrase, sometimes passing as folk wisdom, used to propagate cognitive dissonance (discomfort experienced when one simultaneously holds two or more conflicting cognitions, e.g. ideas, beliefs, values or emotional reactions). Though the phrase in and of itself may be valid in certain contexts, its application as a means of dismissing dissent or justifying fallacious logic is what makes it thought-terminating.
The term was popularized by Robert Jay Lifton in his 1956 book Thought Reform and the Psychology of Totalism. Lifton said, “The language of the totalist environment is characterized by the thought-terminating cliché. The most far-reaching and complex of human problems are compressed into brief, highly reductive, definitive-sounding phrases, easily memorized and easily expressed. These become the start and finish of any ideological analysis.” [1][2]
In George Orwell’s novel Nineteen Eighty-Four, the fictional constructed language Newspeak is designed to reduce language entirely to a set of thought-terminating clichés. Aldous Huxley’s Brave New World society uses thought-terminating clichés in a more conventional manner, most notably in regard to the drug soma as well as modified versions of real-life platitudes, such as, “A doctor a day keeps the jim-jams away.”
|
|
Thought-terminating clichés are sometimes used during political discourse to enhance appeal or to shut down debate. In this setting, their usage can usually be classified as a logical fallacy.[citation needed]]]
Thought-terminating clichés are also present in religious discourse in order to define a clear border between good and evil, holiness and sacrilege, and other polar opposites.[citation needed] These are especially present in religious literature.[citation needed]
The religious or semi-religious ideas of cults, heretics, and infidels are also often used as thought-terminating clichés, e.g. "Do not listen to him, he is an infidel," (a guilt by association fallacy) or "That line of thought sounds like a cult" (also a guilt by association fallacy).
The statement "that is a thought-terminating cliché" can itself function as a thought-terminating cliché. Once the stator has identified a first statement as a thought-terminating cliché, they may feel absolved of needing to determine whether that first statement is indeed a thought-terminating cliché, or provides useful insight, in the context under discussion.
Update 2019-06-17: A recent thread on reddittalks about "Gish galloping" which is another strategy for shortcutting reasoned arguments:
The opposite is gish galloping, where you make an argument by throwing out a large number of individually easily countered false arguments and knowing that it requires the person more effort to disprove each than it takes you to keep throwing them out. So you will appear to win as they fail to logically address each individual argument and you can say "THought Terminating Cliche!" if they just accuse you of generally lying.
An exchange of opinions is happening on twitter, between J. Craig Venter and Nassim Nicholas Taleb, about whether genetically modified crops are good or not. I tried to summarize the case for why GMOs are dangerous like this:
Ok, this is how I understand why GMOs are dangerous:
Nature has a certain tolerance (hysteres, non linearity) for shenanigans, if you go outside that envelope you may end up in a wasteland. This is potentially lethal if you do it on a global scale
GMOs are modification of small self-replicating things, hence have the potential to go global, and out of control. Even if that was not your intention.
It's not about the *probability* of this happening but about the *pay-off* (consequences) if it happens
So self-replicating things can be dangerous, but I started thinking, how small does something need to be, to be out of control?
If you look at land based organisms: If we should get, say genetically modified evil elephants, it would really not be a a huge problem. We can simply locate all those elephants and kill them. With smaller things it gets harder.
Size | Locate and kill all | Stow away capacity |
Elephant (10 meters) | Easy | They are conspicuous, you know |
Dog (1 meter) | Hard, but maybe doable | None |
Rat, snake (1 decimeter) | Probably impossible | Some |
Insects (1 centimeter) | Forget it | Carried by winds, in little spaces on ships and airplanes |
Seeds (1 millimeter) | Forget it | ditto + on fur, clothes, some UV protection often |
Bacteria, fungi, viruses (many in micrometer to nanometer range) | Forget it | Immune systems can fight them actually, which is good since they are everywhere |
I have tinkered with this previously with pickle, jart/redisbayes and jamesls/fakeredis (real redis slows down redisbayes a lot, pickle used for saving the net) in python, but this TextBlob module seems to have it too. Untested by me.
Yesterday, TextBlob 0.6.0 was released (changelog), which introduces Naive Bayes classification. This tutorial shows how to use TextBlob to create your own text classification systems.
Read more: Link - Steven Loria | Tutorial: Simple Text Classification with Python and TextBlob
Untested by me:
You have almost certainly seen the kind of problem ftfy fixes. Here’s a shoutout from a developer who found that her database was full of place names such as “BucureÅŸti, Romania” because of someone else’s bug. That’s easy enough to fix:
Read more: Link - ftfy (fixes text for you) version 3.0 | Luminoso Blog
Photos from digital cameras nowadays may contain metadata about which side is actually up on the photo. All image viewers I have tried on Ubuntu heeds this and auto rotates the photos for you. But Imagemagick does not understand this auto-rotation stuff and will put your text in the wrong place on some pictured because from Imagemagick's point of view the image is not rotated.
So how do you rotate them so Imagemagick gets it right? The image viewers won't help you since they "lie" to you and will show the image corrected rotationally. One of my image viewers has a setting for disabling this auto rotate behavior, but the setting does not work.
Update 2013-10-30:
Use exiftran
It will rotate the pictures based on the exif orientation information, and update that information.
exiftran -ai *.JPG
Old solution:
What you need to do is strip the metadata from the photos, and then use the image viewer of your choice (e.g. Gwenview), to see which pictures in fact aren't correctly rotated and rotate them.
Then use Imagemagick. This command will strip all metadata from your photos:
exiftool -all= *.JPG
It can probably be configured to just strip some of it it, if need be.
Mathematics is high status, probably due to its success in physics, and because of this it is used in fields and applied to problems that are too unstructured and messy for it to make any sense.
In fact sometimes it is used to obscure: You simply take some maths that is really hard to understand and rely on people not being able to follow your train of thought through it and accept your conclusions at face value. It's like HC Andersen's story about the Emperor's new clothes.
One example is complex mathematic functions applied to finance, but now we can read about another case applied to psychology. I think this quote from the linked article speaks all by itself (my boldface):
Upon reading a 1963 paper on Lorenz equations “with some difficulty,” Brown realized that the equation Fredrickson and Losada used to calculate the critical positivity ratio had no connection to their emotion data: Regardless of the volunteers’ data points, the equation would simply generate the same, meaningless number.
And this was a paper that gained quite some status:
“What’s shocking is not just that this piece of pseudomathematical nonsense received 322 scholarly citations and 164,000 web mentions, but that no one criticized it publicly for eight years, not even supposed experts in the field,” Sokal says.
Read more: Link - Ratio for a good life exposed as 'nonsense' | Psychology | Science News
There is a lot of talk, especially in the U.S., of the middle-class squeeze. I wonder if it could be explained by increased unpredictability. It is said that the middle class is doing worse and worse, and that the differences within the middle class is growing too: Between the upper crust of it and the rest.
Nicholas Nassim Taleb talks about Mediocristan and Extremistan, two imaginary places where in one case your income is predictable and proportional to the amount of work you put in, while in Extremistan rags and riches are decided much more by chance. An example of a Mediocristan job that Taleb gives is a dentist.
Examples of Extremistan jobs could be I guess anything connected with speculation: Building an app for the Iphone app store, making bets on the stock market, relasing a video game, developing a new pharmaceutical and so on. If you do ten, or even a hundred of these projects it is not clear which one will make money, if any. At the same time there may be huge profits, somewhere, sometime.
Let's assume that the world has moved in the direction of Extremistan, and that the actual financial gain from work is much harder to predict. Let's then simplify a lot and say we can divide people into three groups, let's call them
The squeezed middle class are people who toil away in projects where it is unclear if they will give any profit and in that case how much. Because of this there is a risk premium associated with each project, and this is reflected in the salaries paid. The squeezed middle class are doing just as good work as a couple of decades ago, but it is much harder to predicted the actual value of their work, so salaries get depressed as a risk premium.
The upper middle class in this simplified model are then people that simply got lucky by doing a couple of successful projects.
The upper class has enough money to spread the risks and reap the rewards of a high risk/high reward environment
Untested by me at this point in time;
jQuery Combinators adds five very useful methods to every jQuery selection: tap, into, select, ergo and when. These allow you to use your own functions as if they were built-in jQuery methods, which makes your code cleaner and more "jQuery-like."
Read more: Link - jQuery Combinators by raganwald
There can be a lot of callbacks when handling web sql. These libraries aim to make it easier. Web SQL is important on mobile platforms such as IOS and Android.
How do you model what can and cannot be done and what should be done from a certain position in a game?One way is with state machines, but another way is with behavior trees:
Welcome to a series of blog articles about my experiment (read: stumbling around) of marrying data-oriented, memory-streamlined behavior trees with a second representation to ease creation and modification during development.
Read more: Link - #AltDevBlog » Introduction to Behavior Trees
"My liner notes for spore/Spore Behavior Tree Docs - Chris Hecker's Website"
Understanding Behavior Trees | AiGameDev.com
Björn Knafla's whole series:
Libraries in javascript found via Google:
BehaviorTree.js | JavascriptOO.com
This page has some speed tests:
I am not up to speed (so to speak) on what performance repercussions the different settings for AES has, as presented on the page, but the forge library stands out as being five times faster at encryption and 2,5 times faster at decryption than any other library (as tested on a Linux Firefox 22.0 on a one core Celeron) and ten times faster than CryptoJS.
A library not included in the test that may be worth checking out is cifre.
Python module. The lines seem to work in any order, quite entertaining!
Generates random text using real English - taken from random samplings of the entire catalog of dialog spoken by Commander William Riker in every episode of Star Trek: The Next Generation.
Read more: Link - RikerIpsum - Lorem Ipsum: The Next Generation
Just a bookmark:
Read more: Link - Intuitive Classification using KNN and Python
Ever notice how small flourishes and subtle transitions dramatically increases the value of the experience you enjoy with an app or site? Designing and developing UIs for the mobile web is tricky, but it's extremely difficult to do that while delivering something that performs at 60fps.
Read more: Link - Effeckt.css
Summary:
There is a jsperf comparing encryption/decryption speed in javascript:
http://jsperf.com/encryption-decryption-comparisons/6
The two libraries that come out neck and neck on my Linux workstation x86 Chrome browser are the Stanford Javascript Library (http://crypto.stanford.edu/sjcl/) and Crypto-js (https://code.google.com/p/crypto-js/). But how do they perform on the older browser component in Phonegap/Cordova 2.8, on different versions of Android?
I have tested with Phonegap 2.8 on two Android Virtual Devices with Intel Atom, actually running on my Intel Ubuntu 12.04 x86 workstation, with Intel's KVM trick installed (Speeding Up the Android* Emulator on Intel® Architecture | Intel® Developer Zone) to up the performance of the emulators. The Android Phonegap app was configured to open the jsperf url directly in the config.xml.
When running under Android 4.2.2, the two libraries are again neck and neck, but on Android 2.3.3 on an Atom AVD, Crypto-js is twice as fast. See this screen shot (Yeah, I should have run it on a tablet AVD so the screenshot would fit, but I didn't, sorry):
Crypto-js did 50.28, and sjcl.js did 24.56, the latter almost exactly half of the former.
However on a real ARM phone, ZTE Blade running Android 2.3.7, the difference is only about 20%, and on Android 4.0.4 on an Lenovo K1 the difference is about 16%.
Footnotes: The "Block tea" library also performs well on Android 2.3.3 on Atom (less on ARM), but it says on the jsperf page that it is less secure so it might be like comparing apples to oranges. Furthermore crypto is hard and I don't know if there might be differences in quality between the different libraries. I also wonder if the AVDs are correct in that Intel Atom influences javascript performance differently than ARMs.
jSlovo is among fastest database engines for dictionaries. It is available for any platform where Java can be used (Windows, Linux, Apple, Android). It is mainly designed for the use with free dictionaries and thesauruses.
Read more: Link - JSlovo multiplatform dictionary
device-dpi: Use the device's native dpi as target dpi. low-dpi: 120dpi medium-dpi: 160dpi, which is also the default as of today high-dpi: 240dpi : We take any number between 70 and 400 as a valid target dpi. Fix http://b/issue?id=2071943 So how do I use it? Like this, for example:
Read more: Link - Philosophical Games: Customize Android Browser Scaling with target-densityDpi
Go to:
Window->Preferences->Editor->File Associations
...to register your external editor for a specific file type or file types.
Then go to:
Window->Preferences->General->Workspace and check the checkbox "Refresh using native hooks or polling"
...to make Eclipse aware of changes in the files whie you are editing. This setting is important since otherwise you will have to close the external editor in order for Eclipse to pick up the changes which will make backing out of changes with the undo buffer impossible (since you've closed the editor once you realise something is wrong).
Read more: Link - Adding an external editor in Eclipse — VIM Zone