1
00:00:05,040 --> 00:00:13,050
In this lesson, we are going to talk about the selection of a stack of tools, uh, when building an

2
00:00:13,050 --> 00:00:20,010
LMS application before that, just very quickly, a, I want to tell you that in the attached materials,

3
00:00:20,010 --> 00:00:29,490
you are going to find a document talking about one of the alternative LMS to ChatGPT, which is palm

4
00:00:30,210 --> 00:00:30,420
A.

5
00:00:30,420 --> 00:00:39,960
I think that is a little bit early to, uh, to show this information for you, but we just wanted to

6
00:00:39,960 --> 00:00:43,080
a tell you about that.

7
00:00:43,080 --> 00:00:45,570
Of course, there are alternatives to ChatGPT.

8
00:00:45,570 --> 00:00:52,470
And these alternatives are, uh, becoming more and more, uh, strong.

9
00:00:52,470 --> 00:01:03,090
So our prediction is that in 2024 and the coming years, uh, we are going to have more and more alternatives

10
00:01:03,090 --> 00:01:08,850
to ChatGPT in the open source and also in the private, uh, field.

11
00:01:08,850 --> 00:01:14,430
So just just a quick note about this document you have in the attached materials.

12
00:01:14,550 --> 00:01:20,580
Now in this lesson, we are going to talk about the selection of, um, the stack of tools.

13
00:01:20,580 --> 00:01:26,730
You have also a document, uh, about that in the attached materials.

14
00:01:26,730 --> 00:01:29,430
And the interesting thing.

15
00:01:30,360 --> 00:01:31,830
About this.

16
00:01:31,830 --> 00:01:35,640
Is that a you will see that.

17
00:01:36,340 --> 00:01:43,660
Some of the tools that are most popular among the artificial intelligence engineers.

18
00:01:45,040 --> 00:01:54,640
I'm not going to be the tools we recommend you because a in some cases these, uh, um, artificial

19
00:01:54,640 --> 00:02:04,930
intelligence engineers responding the, the, the 2023 survey, they are working in very, very big

20
00:02:04,930 --> 00:02:05,770
companies.

21
00:02:06,340 --> 00:02:15,400
So these are engineers all over the world that are coming to an international summit in the city of

22
00:02:15,400 --> 00:02:18,790
San Francisco for one whole week.

23
00:02:19,090 --> 00:02:27,460
So, as you can imagine, these kind of professionals usually come from big multinationals.

24
00:02:27,460 --> 00:02:33,220
And because of that, they are going to use very large tools.

25
00:02:33,310 --> 00:02:42,700
And in some cases, the reason for using these large tools are that they are, you know, inside very

26
00:02:42,700 --> 00:02:49,030
big IT teams that are using other, you know, compatible frameworks, etc..

27
00:02:49,030 --> 00:02:59,230
So our recommendations are going to be based not just in the popularity of the tool, but in the relationship,

28
00:02:59,230 --> 00:03:02,260
you know, cost benefit of the tool.

29
00:03:02,260 --> 00:03:10,660
So we are not a sponsor by any company or any brand like any other, like some other courses you can

30
00:03:10,660 --> 00:03:15,550
find there, they are sponsored by some tools and they are recommending those tools.

31
00:03:15,550 --> 00:03:17,620
It's not the case of this program.

32
00:03:17,920 --> 00:03:24,370
We are going to make honest recommendations based on our own experience.

33
00:03:24,370 --> 00:03:38,110
So in the case of the stack used when the artificial intelligence engineers responded the 2023 survey,

34
00:03:38,500 --> 00:03:43,150
they highlighted some, uh, tools.

35
00:03:43,300 --> 00:03:44,320
So.

36
00:03:45,280 --> 00:03:55,480
In the orchestration framework, for example, they responded a that long chain and lambda index are

37
00:03:55,480 --> 00:03:57,700
practically tied.

38
00:03:58,390 --> 00:04:06,010
This is not the reality we are seeing, uh, in the community from our point of view.

39
00:04:06,040 --> 00:04:13,330
We see that long chain is still far more popular than Lambda index, uh, in the community of users.

40
00:04:13,330 --> 00:04:23,860
But the more you go to the professional level user, the more you see lambda index, uh, increasing

41
00:04:23,860 --> 00:04:25,150
in in use.

42
00:04:25,150 --> 00:04:35,860
So it is a reasonable to see that in this survey, uh, from professional artificial intelligence engineers,

43
00:04:35,860 --> 00:04:43,270
especially engineers that are working in big companies launching an lambda index are practically tied.

44
00:04:44,060 --> 00:04:47,450
As the orchestration framework most used.

45
00:04:47,450 --> 00:04:55,520
So this is interesting because if you extrapolate this survey to the general public, probably land

46
00:04:55,520 --> 00:04:58,910
chain is going to be by far the most used.

47
00:04:58,910 --> 00:05:05,660
But when you go to the professional, uh, segment, you see that lambda index is more used.

48
00:05:05,660 --> 00:05:10,580
We will see why, uh, in a, in a, in a future, uh, lesson.

49
00:05:12,030 --> 00:05:19,590
In the orchestration frameworks you would see, you will see that there are some other frameworks,

50
00:05:19,590 --> 00:05:23,040
uh, highlighted as well in the survey.

51
00:05:23,460 --> 00:05:33,720
In the database you will see a lot of a different alternatives, uh, highlighted in the survey.

52
00:05:34,780 --> 00:05:37,810
You will see that pine cone is the top one.

53
00:05:38,050 --> 00:05:43,810
Postgres the second one, then comma, supabase, redis, etc. etc..

54
00:05:43,930 --> 00:05:55,150
In our case you will see that in the program we are going to make use of Postgres, comma, a face and

55
00:05:55,150 --> 00:05:56,710
some others as well.

56
00:05:58,720 --> 00:06:03,850
In the monitoring and observability frameworks.

57
00:06:03,850 --> 00:06:05,860
You will see that in the survey.

58
00:06:05,860 --> 00:06:16,240
We have also a large number of frameworks, uh, uh, highlighted by the by the artificial intelligence

59
00:06:16,240 --> 00:06:20,860
engineers that responded this uh 2023 survey.

60
00:06:21,040 --> 00:06:32,140
In our case, we will highlight Lang Smith and we will explain why, uh, in in the next lesson, a

61
00:06:32,170 --> 00:06:37,570
for model serving and hosting uh, tools.

62
00:06:38,270 --> 00:06:38,780
You will.

63
00:06:38,780 --> 00:06:44,960
See also a large number of, uh, alternatives highlighted in the survey.

64
00:06:45,110 --> 00:06:54,110
This is something that you are going to see, uh, every time you are studying a what is happening in

65
00:06:54,110 --> 00:07:03,650
a very early stage field like the one we are, uh, we are, uh, in the LM applications field is still

66
00:07:03,650 --> 00:07:05,450
in a very early stage.

67
00:07:05,450 --> 00:07:13,280
So you have a lot of, uh, tools, a lot of frameworks that are trying to position themselves as leaders.

68
00:07:13,280 --> 00:07:19,400
But it they are not there yet.

69
00:07:19,400 --> 00:07:19,850
Right?

70
00:07:19,850 --> 00:07:28,910
So we have a lot of competition, a lot of alternatives, but we still don't have a solid, uh, positioning

71
00:07:28,910 --> 00:07:29,450
of tools.

72
00:07:29,450 --> 00:07:33,770
So in the model serving and hosting, you have a lot of them.

73
00:07:34,790 --> 00:07:40,100
In my opinion, the most interesting is OpenAI directly.

74
00:07:40,100 --> 00:07:48,080
But you, you have, you know, alternatives like hugging face like replicate, for example, like a

75
00:07:48,920 --> 00:07:55,730
Amazon Web Services SageMaker, Google Vertex as well a Azure.

76
00:07:56,240 --> 00:08:02,660
Uh, but in my opinion, the most interesting and the most use for the general public is going to be

77
00:08:02,660 --> 00:08:06,890
OpenAI directly, especially in 2023.

78
00:08:06,890 --> 00:08:13,280
Let's see what happens in the in the future for model training and fine tuning.

79
00:08:13,280 --> 00:08:16,760
This is a category included in the service in the survey.

80
00:08:18,110 --> 00:08:27,350
I don't think it is very relevant in the real world for companies mid-size, small size, startup,

81
00:08:27,350 --> 00:08:33,049
etc. because fine tuning is something very expensive and very complex.

82
00:08:33,049 --> 00:08:41,809
And you will see that most, uh, AI engineers are not using fine tuning, are using LM applications

83
00:08:42,500 --> 00:08:45,380
or other things like prompt management.

84
00:08:45,380 --> 00:08:48,620
You have some, uh, tools highlighted here.

85
00:08:48,620 --> 00:08:51,980
Not very important in this in this moment for us.

86
00:08:52,660 --> 00:09:00,370
So I just wanted to show you in these lessons, in this lesson, what are the most popular tools you

87
00:09:00,370 --> 00:09:05,200
highlighted by professional artificial intelligence engineers?

88
00:09:05,200 --> 00:09:12,700
And to tell you that in some cases, the most popular are not the ones we are going to recommend.

89
00:09:12,700 --> 00:09:22,750
We are going to make recommendations based on cost benefit relationship, based on our own experience

90
00:09:22,750 --> 00:09:31,150
and the tools we think are going to be more interesting for you from a cost point of view, from an

91
00:09:31,150 --> 00:09:33,640
efficiency point of view, etc..

92
00:09:34,480 --> 00:09:42,220
In the next lesson, we are going to talk a little bit more about the selection of the orchestration

93
00:09:42,220 --> 00:09:43,450
framework.