1
00:00:11,040 --> 00:00:16,050
Now, one thing you may be thinking at this point is this article spinner doesn't seem to work that

2
00:00:16,050 --> 00:00:16,470
well.

3
00:00:17,370 --> 00:00:23,070
Of course, this should not be a surprise, given that we are only using simple programs without considering

4
00:00:23,070 --> 00:00:25,050
parts of speech or other context.

5
00:00:26,190 --> 00:00:31,140
Recall that other techniques, especially those involving deep learning, can be applied to the same

6
00:00:31,140 --> 00:00:35,190
idea of replacing a middle word given surrounding words.

7
00:00:35,790 --> 00:00:41,370
Unlike programs, deep learning techniques can account for more context, as well as potentially learn

8
00:00:41,370 --> 00:00:44,040
ahead and structures such as grammar and parts of speech.

9
00:00:45,180 --> 00:00:49,980
However, since we are not yet at that point in the course, you will just have to wait until you're

10
00:00:49,980 --> 00:00:51,690
able to increase your knowledge.

11
00:00:52,080 --> 00:00:57,390
That is, you must further gather the prerequisites you need in order to progress to those more advanced

12
00:00:57,390 --> 00:00:58,170
techniques.

13
00:00:59,280 --> 00:01:04,379
That being said, it's worth comparing these results to the screenshot of the article spinner program

14
00:01:04,379 --> 00:01:06,060
I showed you earlier in the section.

15
00:01:06,660 --> 00:01:11,490
In fact, what you have seen is on par with applications which people actually pay for.

16
00:01:12,180 --> 00:01:17,160
Remember that these spinner programs don't just replace words automatically, they give you a list of

17
00:01:17,160 --> 00:01:18,630
suggestions to choose from.

18
00:01:19,290 --> 00:01:24,960
So even just knowing what you know so far, this is enough to build a product that people will happily

19
00:01:24,960 --> 00:01:25,950
spend their money on.

20
00:01:26,640 --> 00:01:31,620
And that brings us to the topic of this lecture, which is all about article spinning gone wrong.

21
00:01:36,350 --> 00:01:41,480
So I won't be naming any names, but I'm sure if you watch machine learning videos on YouTube, you've

22
00:01:41,480 --> 00:01:43,130
come across this story yourself.

23
00:01:43,940 --> 00:01:48,530
So this story is about a pretty popular YouTuber who made a videos about machine learning.

24
00:01:49,790 --> 00:01:54,380
Now it's important to realize that these weren't technical videos they just pretended to be.

25
00:01:55,190 --> 00:01:59,630
For instance, he would make a video like Watch me build a stock trading bomb in five minutes.

26
00:01:59,870 --> 00:02:02,810
But it was really code that he took from someone else's GitHub.

27
00:02:03,380 --> 00:02:07,490
It was more for entertainment than it was about increasing your technical ability.

28
00:02:08,860 --> 00:02:14,110
Of course, he would never be able to explain the code because he didn't write it himself, nor even

29
00:02:14,110 --> 00:02:15,190
try to understand it.

30
00:02:15,820 --> 00:02:20,500
It was not a machine learning expert, but rather only someone who played one on YouTube.

31
00:02:21,400 --> 00:02:26,710
As a side note, whenever I make videos like common beginner mistakes when predicting stock prices without

32
00:02:26,760 --> 00:02:27,370
stems.

33
00:02:27,790 --> 00:02:30,430
These are some of the kinds of videos I'm referring to.

34
00:02:31,330 --> 00:02:36,970
So anyway, it wasn't long until this fellow decided to publish his own paper on the topic of neural

35
00:02:36,970 --> 00:02:37,720
qubits.

36
00:02:38,590 --> 00:02:40,000
Sounds pretty advanced.

37
00:02:44,590 --> 00:02:47,590
Unfortunately, the entire paper was plagiarized.

38
00:02:48,190 --> 00:02:51,580
As mentioned, this was not an actual machine learning practitioner.

39
00:02:51,880 --> 00:02:54,010
Only one who pretended to be on YouTube.

40
00:02:55,060 --> 00:02:59,680
Sadly, some people still believe this YouTuber actually knows about machine learning.

41
00:03:01,550 --> 00:03:06,740
Luckily, this paper wasn't published anywhere reputable, but rather just a weird website called the

42
00:03:06,740 --> 00:03:10,190
VIX Rob, which is known for publishing works by crazy people.

43
00:03:11,090 --> 00:03:16,580
So the main highlight of this paper is that it really makes no attempt to hide that it is completely

44
00:03:16,580 --> 00:03:17,450
plagiarized.

45
00:03:17,990 --> 00:03:22,160
Both the images and math formulas were simply screenshots of the original.

46
00:03:23,030 --> 00:03:26,740
And this would be highly unusual if it were a true academic paper.

47
00:03:31,400 --> 00:03:36,860
For the text portion, the basic strategy was to do something similar to what we described in the section,

48
00:03:37,400 --> 00:03:42,920
simply replace random words with synonyms, hoping that they would still make sense in context.

49
00:03:43,670 --> 00:03:45,020
So what was the result?

50
00:03:45,710 --> 00:03:48,050
Well, there were some very strange replacements.

51
00:03:48,470 --> 00:03:51,750
For example, quantum gate was changed to quantum door.

52
00:03:52,760 --> 00:03:56,780
Of course, in everyday language, gate can certainly be a synonym for door.

53
00:03:57,410 --> 00:04:01,800
The problem is that was not the case in this context, in this context.

54
00:04:01,820 --> 00:04:07,100
Gate refers to a logic gate, for example, and or not XOR and so forth.

55
00:04:08,750 --> 00:04:14,720
Another popular example is complex Hilbert space, which was changed to complicated Hilbert space.

56
00:04:15,320 --> 00:04:20,029
And of course, in everyday language, complex can be a synonym too complicated.

57
00:04:20,660 --> 00:04:26,330
For example, beginners who don't meet the prerequisites to my courses think of my courses are too complex.

58
00:04:26,750 --> 00:04:30,620
This has the same meaning as beginners think my courses are too complicated.

59
00:04:31,310 --> 00:04:37,520
But in this context, complex does not mean complicated, but rather it's referring to complex numbers.

60
00:04:38,150 --> 00:04:44,300
So in fact, one downside to this approach is that it did not take context into account like we did

61
00:04:44,300 --> 00:04:45,200
in this course.

62
00:04:46,280 --> 00:04:51,560
Thus, one could say that if this YouTuber had used the techniques in this course, he wouldn't have

63
00:04:51,560 --> 00:04:53,300
made such silly mistakes.

64
00:04:57,780 --> 00:05:02,640
Now, my goal in this lecture is not to make fun of this YouTuber, although it is a very interesting

65
00:05:02,640 --> 00:05:05,280
story, but there are many lessons here.

66
00:05:06,090 --> 00:05:11,460
Firstly, notice that this method of content spinning really is in use in the real world.

67
00:05:12,000 --> 00:05:17,160
It may sound like I've just made this up for the purpose of this course, but it really is a real world

68
00:05:17,160 --> 00:05:18,420
practical use case.

69
00:05:19,920 --> 00:05:24,150
Secondly, and most importantly, notice how it is not hard to detect.

70
00:05:24,660 --> 00:05:30,030
So if you want to be taken seriously or to publish content that people actually enjoy and respect.

71
00:05:30,360 --> 00:05:31,830
This is not the way to do it.

72
00:05:32,340 --> 00:05:37,620
Not only was this easily detected by people, but it would be just as easy for machines as well.