I know that the series of articles in media seem to make it appear so. But there really aren't any stats to support the claim that India has a "rape problem". The case being built is highly anecdotal, which is easy when you have a country of 1.2 BILLION to pick cases from. There are about 25K rapes reported in India annually. For the second most populous country that amounts to 2 per 100K. That is pretty low.
Granted, only a minority of rapes get reported in relatively conservative societies. But under-reporting occurs everywhere at different levels. 1 in 5 or 6 women report rape in surveys in US. Only a fraction of those get reported. Even if you reasonably adjust for the fact that rape is much more under-reported than say US, it is a very difficult case to make, that India has much more per capita rape than US. For a country, one fourth the size of India, CDC counted 1.3 *million* *reported* rapes in US in 2010. With a more strict definition, FBI counted 86K.
It is "possible" that India has a lot of rape. But the case has not yet been made in data. It also needs to be shown that rape in India is somehow specially higher than other countries in the region with similar development indices. For police recorded offenses, the stats are:
India: 0.4 US: 27.3 Pakistan 28.8 (where I assume much more conservative in reporting than India) Nepal: 0.8 (which can be assumed to be about the same as India) Sri Lanka: 7.3
I feel that the real story was how people came to the streets angry about the crime. Which other country had its capital shutdown by the public in response to a rape case? The people do care.
I agree about the outlet argument of course. But in Inda, bans come and go. It isn't China. The state does not really control people. People don't take these seriously. *All* of Indian press condemned the bans as attempts at a nanny state. I doubt that it will last, when subjected to scrutiny. At the end of the day, India is still a noisy democracy.
Thanks. That's useful feedback. Obviously, this topic is complicated with several factors to muddy the waters.
> 90% of your time is usually spent in 5% of your code, so it's really the ability to optimize easily when you need to without resorting to convoluted tricks and hooks into other languages.
Interestingly, the Python/glue-language philosophy is just the opposite, for exactly the same reasons. Since it is just the 5% of the code that needs optimization, it says: why not write that in C/C++ and the rest in something easier?
These days, you can get GPU power outside C++. For instance, Theano brings GPU math to Python. I don't have experience with comparing the performance differences between say PyCUDA/jcuda vs straight C/C++ CUDA. The net difference may not be 100x, but C/C++ programmers will certainly tend to do deeper optimizations.
You can bundle JRE with your jar/exe, if you don't want the user to separately download a runtime. Many Eclipse Rich Client Apps do this. I have done it myself. As someone who used Delphi/C++ Builder and is very used to static linking to a 1-2 MB distributables with no further dependencies, I am not that happy with Qt dll sizes for client apps. Its certainly better than Java, but is quite bloated for native code. All other native solutions are much more compact (Qt of course wins by feature set). Static linking with Qt is possible, but not recommended.
I was looking for info on Qt server apps. On the server, runtimes aren't a concern. Startup times are also not important for HPC apps with long execution times. What is important is how things go after the startup. JVM does need to set more memory aside for the GC. Micro benchmarks typically put C++ at 1.5-2.0x faster than Java. I wondered how real world performance differs.
I understand all that. I used Qt myself, but just for simple GUIs. I was just wondering what his real world numbers were. I was not challenging him for his choice. I know the micro-benchmarks between native and VM code. I was curious of how things fared in his larger apps. I normally do JVM for long-running server code, but C++11 and onwards is increasingly attractive, although JVM is still a bit simpler to work with, on the whole. I am exclusively interested in his (or anyone else's) real world experiences on the performance differences between JVM HPC code (server VM config) and C++ server code that uses Qt as the main library.
I don't disagree with anything, but I am just curious: what kind of improvements did you get with C++/Qt stack over say, Java or Scala? (I am assuming you evaluated alternatives) Say, over 2x improvement in performance? Much lower memory utilization?
> but I suspect it would be a giant mash up of "if mobile do X else do Y," which isn't really "universal" in my mind
Well, it would link against different runtime libs, either static or dynamic, with the same interface - it would not bundle the binaries of all platforms (closer to Apple's use of the term Universal). Universal API is just cross-platform API, like Qt or anything similar.... except with support for a lot more disparate architectures... like FireMonkey. Or it is just better Java. Java, Qt (QtQuick is much closer), wxWidgets, LCL... all support multiple platforms. But they never really grew beyond the traditional x86/64 desktop/server (Java Mobile and Android Java were not seamless). This would just add mobile platforms to the mix with near complete API.
> They used to provide a library (skypekit) to do that but they decided to cut it off
Sure. I am not arguing about their business choices, just the technical ones. It is fairly easy to port the protocol code across different architectures. I used to use SIP, in preference to Skype. It worked well enough. So the problems are not technical, if they try.
You are not arguing against the Universal API. You are arguing against mobile specific UIs on the desktop.
Skype UI isn't that complicated to replicate. In mobile development, it is typical to maintain multiple UIs. That's not considered the main challenge. The original Skype was developed in Delphi. AFAIK, it still is. Delphi today supports multi-platform apps and UIs (with platform specific behaviors), with a UI designer tailor-made for that role, although it uses its own FireMonkey framework (not sure about the extent to which it supports Universal API) which supports both Desktop and Mobile apps. Skype UI was written on VCL. It is supposed to be fairly easy to port to FireMonkey. Modern frameworks, especially the modern mobile frameworks, make it quite easy to scale to a wide range of resolutions and devices, as long as devs keep that in mind. The Skype dev team might have gone with MS API since FireMonkey was perhaps not mature enough then.
I recall that the protocol code was written in C++, which I assume is portable C++, since Skype is already available on all platforms and I doubt that they have multiple code bases for protocol code. I am not sure how good the native code interop is in Universal API.
I don't use the Universal App API. So I have to ask. How is it worse than the model used by the Android and iOS API? Why wouldn't it be adequate for an app like Skype.
> Makes me wonder how I managed a 16 hour surgery the other day without ever getting bored or distracted (kind of hard to do when the patient is trying so hard to die on your table).
Clearly, without your patient's help in keeping you focused, you would have gotten distracted and wandered away from the OP after seeing a squirrel in the window:-).
Seriously, you know enough Statistics to know that your circumstance does not make a case against the study in any way, even if it was a comparable task. Your surgery task is a compound, concurrently distributed team task. It is not anything like the candidate tasks under consideration in this study. You don't need to have special powers of concentration to not be distracted in a surgery (at least no more than passing boards).
I mean the full Python stack (IPython notebook + Spyder with IPython, PyLab, Pandas, statsmodels).
For almost everything in stats, I prefer the RStudio experience. The flow feels much better, even though my Python is much better than my R. Machine Learning is one stats topic though, where I still prefer Python - I just like Scikit-learn.
If I was doing linear algebra directly, I would have preferred the Python stack with NumPy. PyLab stack is more for Matlab users than R users. On the stats side, Pandas and statsmodels are still not yet an R replacement for me. They are a great start though and seem to have gotten everything right so far.
I use both R and Python. R itself is actually quite nice and more efficient for interactive use, once you get used to it. For interactive exploration with statistics, I actually prefer it to Python (and I have been using Python for ~15 years). Lots of helper functions. Everything uses the DataFrame datastructure. Good, concise and consistent documentation.
Unless you are a R library dev, for most users, its best to see R as a shell for statistics, rather than a programming language. So its language horribleness does not matter much.
I use Python to process data and R to explore it. Once I settle on something, if I need to put it into a larger pipeline, I either find an implementation in Python or link R to Python via rpy2.
> A python statistics library with some funky C linkage to the R library would take over in milliseconds
> there's nothing else out there that is even attempting to solve the same problems.
You mean move semantics? That would be the main innovation of Rust. C++ also has them now. Perhaps Rust has them better, but it would be inaccurate to say that no one is even attempting to solve these problems.
I do indeed prefer mplayer over VLC since the CPU utilization is better. However, my mplayer does not do VP9. VLC nightly was suggested and it worked. But I would like to switch back to mplayer as soon as I can.
I did update my ffmpeg. The one that comes with Trusty did not work with youtube-dl.
I am aware..NET Native just removes some JIT costs and improves load times (60% is the claim). It does not make actual program execution much faster. I don't expect the benchmarks to change much. Although, MS CLR was probably a bit better in performance than the Mono implementation. Let's see how this code merge fares. There are free and commercial native compilers for Java. They don't help all that much. I imagine things to be similar. No idea about how good the disk and memory improvements will play out.
I use youtube-dl to download presentations from Youtube. I have been getting VP9 webms for months from Youtube. If you type youtube -F , you can see all the DASH webm streams, which are encoded by VP9. The non-DASH webms are VP8 videos. With youtube-dl, you can select the DASH video and audio streams and combine them with ffmpeg. The file sizes are indeed much better.
Short Test Video: youtube-dl --prefer-ffmpeg -f 247+171 https://www.youtube.com/watch?... 39 secs of this 720p clip comes out to 5.6 MB. With H264, it would 10.8 MB.
The only problem I have is that I have to play them by dropping them in Firefox. I have not managed to get any of my desktop media players to get the codecs (Ubuntu 14.04). If any of you solved this, let me know.
But programs written on it have memory requirements that are higher than ones written in plain systems languages. The runtime footprint on the disk is also massive. I don't think you can really make a case that C# is a low-level language. It is not that much more CPU efficient than Java. Mono performance is worse than Java.
I don't understand art very much and I certainly don't understand modern art. I don't have a taste for Picasso. But it cannot be denied that he introduced/developed several novel art forms and was a talented painter.
> I've cleaned a paintbrush on a piece of cloth and created something better than Picasso. I've seen preschoolers with finger paints do better.
That's just a cheap shot with no basis. You should at least make an effort to find out why people who venerate him as one of the greatest ever, do so. You and I are just not qualified to judge it. Your critiques are no different from a religious fundamentalist who pooh poohs Science without making an effort to study it beyond high school.
Well, of course, there are good and bad lectures and lecture videos.
> Well done presentations are the exception -- don't try to build a rule on them.
I am not sure I agree. I have been satisfied with the quality of video lectures in MOOCs. I expect MOOC videos (I just use Coursera) to be better than simple lecture videos that I was accustomed to in the pre-MOOC era. M is for Massive. So I do expect that better care is taken in their production.
> which brings us back to the thrust of the article -- doing video properly takes more time than it's worth
We have an article because these lecturers are the exception (IPython Notebooks are quite good teaching tools though). If I wanted a simple presentation with no expectation of effort on media, I'd normally just go download some course lectures from iTunesU.
A good presentation does not need a whole lot of effort. A screen cast format is not bad at all. It can involve slides, live code building, refer to web resources, screen drawing etc. That's pretty multi-modal and does not need a complex set up.
Hah. I have seen a bit of what you mean and I cringed myself. I am from a developing country (and studied in the West). Its just that the educational culture is a bit different over here. Students can get rather needy. Project work expectations are pretty low here (unlilke test performance) and they might be having a harder time to adjust. Its probably not a bad idea to have regional realms of some sort, so that students of similar cultures can participate, without stepping on other's toes.
Harsh. Tell a student to do Linux from scratch, he will find it intimidating. I assume most failed in the first few attempts, back when they was no video option. Show him a video of it once, he will find it much less intimidating. Video has its place.
Another thing is: you need "good written instructions", as you say. Not everyone can write good instructions. But just about anyone can show. Creating install videos does not require as much skill because a lot of information is informally and implicitly encoded in the demo.
> Perhaps if you can't appreciate a mathematical subject as it is presented in its dry text form, then it isn't something you are likely to ever understand
I dunno. I find animations of mathematical concepts to be quite effective in communicating the intuition behind them, much better than text.
Perhaps, you just haven't seen good use of multimedia.
> I think one of the problems with the video format is that it entices you into being passive
I prefer videos over lectures. The reason is that I can pause them, replay them, for technical stuff, try things out.
You might say: Well, you can do that with a book. For me, the lecture uses a more approachable language than the more formal format of the book (good for further exploration and lookups). A video demonstration is just more compact and more effective because it is multi-modal, than the full description in text.
> because the three forms crowd each other out.
In a well-done presentation, they are complementary... multi-modal.
Slashdot Mirror
> with a pervasive rape problem
I know that the series of articles in media seem to make it appear so. But there really aren't any stats to support the claim that India has a "rape problem". The case being built is highly anecdotal, which is easy when you have a country of 1.2 BILLION to pick cases from. There are about 25K rapes reported in India annually. For the second most populous country that amounts to 2 per 100K. That is pretty low.
Granted, only a minority of rapes get reported in relatively conservative societies. But under-reporting occurs everywhere at different levels. 1 in 5 or 6 women report rape in surveys in US. Only a fraction of those get reported. Even if you reasonably adjust for the fact that rape is much more under-reported than say US, it is a very difficult case to make, that India has much more per capita rape than US. For a country, one fourth the size of India, CDC counted 1.3 *million* *reported* rapes in US in 2010. With a more strict definition, FBI counted 86K.
It is "possible" that India has a lot of rape. But the case has not yet been made in data. It also needs to be shown that rape in India is somehow specially higher than other countries in the region with similar development indices. For police recorded offenses, the stats are:
India: 0.4
US: 27.3
Pakistan 28.8 (where I assume much more conservative in reporting than India)
Nepal: 0.8 (which can be assumed to be about the same as India)
Sri Lanka: 7.3
I feel that the real story was how people came to the streets angry about the crime. Which other country had its capital shutdown by the public in response to a rape case? The people do care.
The justice for harassment is sometimes weird :-).
https://www.youtube.com/watch?...
I agree about the outlet argument of course. But in Inda, bans come and go. It isn't China. The state does not really control people. People don't take these seriously. *All* of Indian press condemned the bans as attempts at a nanny state. I doubt that it will last, when subjected to scrutiny. At the end of the day, India is still a noisy democracy.
Here are Indian comedians at the previous attempt at the same ban
https://www.youtube.com/watch?...
Thanks. That's useful feedback. Obviously, this topic is complicated with several factors to muddy the waters.
> 90% of your time is usually spent in 5% of your code, so it's really the ability to optimize easily when you need to without resorting to convoluted tricks and hooks into other languages.
Interestingly, the Python/glue-language philosophy is just the opposite, for exactly the same reasons. Since it is just the 5% of the code that needs optimization, it says: why not write that in C/C++ and the rest in something easier?
These days, you can get GPU power outside C++. For instance, Theano brings GPU math to Python. I don't have experience with comparing the performance differences between say PyCUDA/jcuda vs straight C/C++ CUDA. The net difference may not be 100x, but C/C++ programmers will certainly tend to do deeper optimizations.
You can bundle JRE with your jar/exe, if you don't want the user to separately download a runtime. Many Eclipse Rich Client Apps do this. I have done it myself. As someone who used Delphi/C++ Builder and is very used to static linking to a 1-2 MB distributables with no further dependencies, I am not that happy with Qt dll sizes for client apps. Its certainly better than Java, but is quite bloated for native code. All other native solutions are much more compact (Qt of course wins by feature set). Static linking with Qt is possible, but not recommended.
I was looking for info on Qt server apps. On the server, runtimes aren't a concern. Startup times are also not important for HPC apps with long execution times. What is important is how things go after the startup. JVM does need to set more memory aside for the GC. Micro benchmarks typically put C++ at 1.5-2.0x faster than Java. I wondered how real world performance differs.
I understand all that. I used Qt myself, but just for simple GUIs. I was just wondering what his real world numbers were. I was not challenging him for his choice. I know the micro-benchmarks between native and VM code. I was curious of how things fared in his larger apps. I normally do JVM for long-running server code, but C++11 and onwards is increasingly attractive, although JVM is still a bit simpler to work with, on the whole. I am exclusively interested in his (or anyone else's) real world experiences on the performance differences between JVM HPC code (server VM config) and C++ server code that uses Qt as the main library.
I don't disagree with anything, but I am just curious: what kind of improvements did you get with C++/Qt stack over say, Java or Scala? (I am assuming you evaluated alternatives) Say, over 2x improvement in performance? Much lower memory utilization?
> but I suspect it would be a giant mash up of "if mobile do X else do Y," which isn't really "universal" in my mind
Well, it would link against different runtime libs, either static or dynamic, with the same interface - it would not bundle the binaries of all platforms (closer to Apple's use of the term Universal). Universal API is just cross-platform API, like Qt or anything similar.... except with support for a lot more disparate architectures... like FireMonkey. Or it is just better Java. Java, Qt (QtQuick is much closer), wxWidgets, LCL... all support multiple platforms. But they never really grew beyond the traditional x86/64 desktop/server (Java Mobile and Android Java were not seamless). This would just add mobile platforms to the mix with near complete API.
> They used to provide a library (skypekit) to do that but they decided to cut it off
Sure. I am not arguing about their business choices, just the technical ones. It is fairly easy to port the protocol code across different architectures. I used to use SIP, in preference to Skype. It worked well enough. So the problems are not technical, if they try.
You are not arguing against the Universal API. You are arguing against mobile specific UIs on the desktop.
Skype UI isn't that complicated to replicate. In mobile development, it is typical to maintain multiple UIs. That's not considered the main challenge. The original Skype was developed in Delphi. AFAIK, it still is. Delphi today supports multi-platform apps and UIs (with platform specific behaviors), with a UI designer tailor-made for that role, although it uses its own FireMonkey framework (not sure about the extent to which it supports Universal API) which supports both Desktop and Mobile apps. Skype UI was written on VCL. It is supposed to be fairly easy to port to FireMonkey. Modern frameworks, especially the modern mobile frameworks, make it quite easy to scale to a wide range of resolutions and devices, as long as devs keep that in mind. The Skype dev team might have gone with MS API since FireMonkey was perhaps not mature enough then.
I recall that the protocol code was written in C++, which I assume is portable C++, since Skype is already available on all platforms and I doubt that they have multiple code bases for protocol code. I am not sure how good the native code interop is in Universal API.
I don't use the Universal App API. So I have to ask. How is it worse than the model used by the Android and iOS API? Why wouldn't it be adequate for an app like Skype.
> Makes me wonder how I managed a 16 hour surgery the other day without ever getting bored or distracted (kind of hard to do when the patient is trying so hard to die on your table).
Clearly, without your patient's help in keeping you focused, you would have gotten distracted and wandered away from the OP after seeing a squirrel in the window :-).
Seriously, you know enough Statistics to know that your circumstance does not make a case against the study in any way, even if it was a comparable task. Your surgery task is a compound, concurrently distributed team task. It is not anything like the candidate tasks under consideration in this study. You don't need to have special powers of concentration to not be distracted in a surgery (at least no more than passing boards).
> in a state were Asians and blacks make up a similar portion of the population.
A quick peek at Wikipedia says that Asians (14.9%) are over twice the population of African-Americans (7.2%) in California. Just nitpicking.
I mean the full Python stack (IPython notebook + Spyder with IPython, PyLab, Pandas, statsmodels).
For almost everything in stats, I prefer the RStudio experience. The flow feels much better, even though my Python is much better than my R. Machine Learning is one stats topic though, where I still prefer Python - I just like Scikit-learn.
If I was doing linear algebra directly, I would have preferred the Python stack with NumPy. PyLab stack is more for Matlab users than R users. On the stats side, Pandas and statsmodels are still not yet an R replacement for me. They are a great start though and seem to have gotten everything right so far.
I use both R and Python. R itself is actually quite nice and more efficient for interactive use, once you get used to it. For interactive exploration with statistics, I actually prefer it to Python (and I have been using Python for ~15 years). Lots of helper functions. Everything uses the DataFrame datastructure. Good, concise and consistent documentation.
Unless you are a R library dev, for most users, its best to see R as a shell for statistics, rather than a programming language. So its language horribleness does not matter much.
I use Python to process data and R to explore it. Once I settle on something, if I need to put it into a larger pipeline, I either find an implementation in Python or link R to Python via rpy2.
> A python statistics library with some funky C linkage to the R library would take over in milliseconds
That's what rpy2 is.
> there's nothing else out there that is even attempting to solve the same problems.
You mean move semantics? That would be the main innovation of Rust. C++ also has them now. Perhaps Rust has them better, but it would be inaccurate to say that no one is even attempting to solve these problems.
Yes, worked for me too. Thanks.
"intergrate video in interface" isn't working though. Had the same problem with nightlies too.
I do indeed prefer mplayer over VLC since the CPU utilization is better. However, my mplayer does not do VP9. VLC nightly was suggested and it worked. But I would like to switch back to mplayer as soon as I can.
I did update my ffmpeg. The one that comes with Trusty did not work with youtube-dl.
That worked. Thanks.
I am aware. .NET Native just removes some JIT costs and improves load times (60% is the claim). It does not make actual program execution much faster. I don't expect the benchmarks to change much. Although, MS CLR was probably a bit better in performance than the Mono implementation. Let's see how this code merge fares. There are free and commercial native compilers for Java. They don't help all that much. I imagine things to be similar. No idea about how good the disk and memory improvements will play out.
> Does it mean the video has been encoded in VP9 if the nerd stats say DASH in Flash?
Not all DASH streams are VP9. From what I have seen: The mp4 streams (DASH and non-DASH) are h264. webm DASH streams are VP9. webm non-DASH is VP8.
I use youtube-dl to download presentations from Youtube. I have been getting VP9 webms for months from Youtube. If you type youtube -F , you can see all the DASH webm streams, which are encoded by VP9. The non-DASH webms are VP8 videos. With youtube-dl, you can select the DASH video and audio streams and combine them with ffmpeg. The file sizes are indeed much better.
Short Test Video:
youtube-dl --prefer-ffmpeg -f 247+171 https://www.youtube.com/watch?...
39 secs of this 720p clip comes out to 5.6 MB. With H264, it would 10.8 MB.
The only problem I have is that I have to play them by dropping them in Firefox. I have not managed to get any of my desktop media players to get the codecs (Ubuntu 14.04). If any of you solved this, let me know.
> Both C# and C++ offer low level functionality
Not really. Can you write a device driver in C#? How about a plain DLL? CLR is a VM. Its CPU performance is OK (2-8 times slower than C).
http://benchmarksgame.alioth.d...
But programs written on it have memory requirements that are higher than ones written in plain systems languages. The runtime footprint on the disk is also massive. I don't think you can really make a case that C# is a low-level language. It is not that much more CPU efficient than Java. Mono performance is worse than Java.
http://benchmarksgame.alioth.d...
Of course, CLR is better than dynamic language aka scripting language runtimes. But that's about it.
I don't understand art very much and I certainly don't understand modern art. I don't have a taste for Picasso. But it cannot be denied that he introduced/developed several novel art forms and was a talented painter.
> I've cleaned a paintbrush on a piece of cloth and created something better than Picasso. I've seen preschoolers with finger paints do better.
That's just a cheap shot with no basis. You should at least make an effort to find out why people who venerate him as one of the greatest ever, do so. You and I are just not qualified to judge it. Your critiques are no different from a religious fundamentalist who pooh poohs Science without making an effort to study it beyond high school.
Well, of course, there are good and bad lectures and lecture videos.
> Well done presentations are the exception -- don't try to build a rule on them.
I am not sure I agree. I have been satisfied with the quality of video lectures in MOOCs. I expect MOOC videos (I just use Coursera) to be better than simple lecture videos that I was accustomed to in the pre-MOOC era. M is for Massive. So I do expect that better care is taken in their production.
> which brings us back to the thrust of the article -- doing video properly takes more time than it's worth
We have an article because these lecturers are the exception (IPython Notebooks are quite good teaching tools though). If I wanted a simple presentation with no expectation of effort on media, I'd normally just go download some course lectures from iTunesU.
A good presentation does not need a whole lot of effort. A screen cast format is not bad at all. It can involve slides, live code building, refer to web resources, screen drawing etc. That's pretty multi-modal and does not need a complex set up.
Hah. I have seen a bit of what you mean and I cringed myself. I am from a developing country (and studied in the West). Its just that the educational culture is a bit different over here. Students can get rather needy. Project work expectations are pretty low here (unlilke test performance) and they might be having a harder time to adjust. Its probably not a bad idea to have regional realms of some sort, so that students of similar cultures can participate, without stepping on other's toes.
Harsh. Tell a student to do Linux from scratch, he will find it intimidating. I assume most failed in the first few attempts, back when they was no video option. Show him a video of it once, he will find it much less intimidating. Video has its place.
Another thing is: you need "good written instructions", as you say. Not everyone can write good instructions. But just about anyone can show. Creating install videos does not require as much skill because a lot of information is informally and implicitly encoded in the demo.
> Perhaps if you can't appreciate a mathematical subject as it is presented in its dry text form, then it isn't something you are likely to ever understand
I dunno. I find animations of mathematical concepts to be quite effective in communicating the intuition behind them, much better than text.
Perhaps, you just haven't seen good use of multimedia.
> I think one of the problems with the video format is that it entices you into being passive
I prefer videos over lectures. The reason is that I can pause them, replay them, for technical stuff, try things out.
You might say: Well, you can do that with a book. For me, the lecture uses a more approachable language than the more formal format of the book (good for further exploration and lookups). A video demonstration is just more compact and more effective because it is multi-modal, than the full description in text.
> because the three forms crowd each other out.
In a well-done presentation, they are complementary... multi-modal.