1
00:00:02,860 --> 00:00:12,190
Okay, so let's talk about how language myth solves the main challenge that you will face during the

2
00:00:12,190 --> 00:00:13,750
prototyping phase.

3
00:00:13,750 --> 00:00:20,200
So let's go to the platform and we can see how all this applies in the platform.

4
00:00:20,350 --> 00:00:29,080
So the first thing is easy to have language myth enabled since day one is just doing what we, uh,

5
00:00:29,080 --> 00:00:32,020
we saw in the in the previous lesson.

6
00:00:32,020 --> 00:00:40,810
So you just include the language Smith variables in the dot m file of your Lang Lang chain project,

7
00:00:40,810 --> 00:00:41,680
and that's it.

8
00:00:41,680 --> 00:00:47,680
Lang Lang Smith is going to be working as you advance in your project.

9
00:00:47,680 --> 00:00:48,250
Okay.

10
00:00:48,340 --> 00:00:52,270
So this is very interesting in one side.

11
00:00:52,660 --> 00:00:58,180
But you also have to, uh, be aware of the cost.

12
00:00:58,180 --> 00:00:58,750
Okay.

13
00:00:58,750 --> 00:01:07,480
So as you know, your free account with Lang Smith has, I think, 8000, uh, traces included in the

14
00:01:07,480 --> 00:01:08,260
first month.

15
00:01:08,260 --> 00:01:11,560
So, I mean, have that in mind.

16
00:01:11,560 --> 00:01:16,480
If your project is a small project, probably you are going to be totally fine.

17
00:01:16,480 --> 00:01:21,880
But be careful, because if your project is a large project and you have a lot of traces in the first

18
00:01:21,880 --> 00:01:24,010
month, it can be costly for you.

19
00:01:24,010 --> 00:01:25,930
So pay attention to that.

20
00:01:25,930 --> 00:01:32,890
But it is going to be very interesting in principle to have Lang Smith enabled since day one.

21
00:01:33,460 --> 00:01:39,640
Second, how to use the Lang Smith Playground to iterate and experiment.

22
00:01:39,640 --> 00:01:46,690
How to experiment with prompts in the Lang Smith Hub prompt playground environment.

23
00:01:47,020 --> 00:01:55,780
This is a very long name, so, uh, if you remember, we have what they call the hub, which is where

24
00:01:55,780 --> 00:02:01,570
you can find a prompt examples that you can you can reuse.

25
00:02:01,570 --> 00:02:07,870
And once you select one of these prompt examples, you can experiment with it.

26
00:02:07,870 --> 00:02:10,720
Going to the playground okay.

27
00:02:10,720 --> 00:02:13,000
So you go to the try it button.

28
00:02:13,000 --> 00:02:15,040
You come to the playground.

29
00:02:15,040 --> 00:02:20,710
So they have uh and the playground are not exactly the same.

30
00:02:20,860 --> 00:02:24,700
But you can say they go together.

31
00:02:24,700 --> 00:02:25,360
Right.

32
00:02:25,360 --> 00:02:32,290
So one way to use the the playground is and this way we show you it's like, okay, you go to the hub,

33
00:02:32,320 --> 00:02:36,940
you select one prompt and then you click on the try it button.

34
00:02:36,940 --> 00:02:38,860
And you go to the playground.

35
00:02:38,860 --> 00:02:45,700
And as you can see in this screen, in this playground, you can play with different things.

36
00:02:45,700 --> 00:02:49,030
You can change, you know, the prompt.

37
00:02:49,180 --> 00:02:57,850
Uh, so the system, uh, template, the human, uh, template, you can uh, play around with uh,

38
00:02:57,850 --> 00:03:02,470
inputs etc. and you can also play around with models.

39
00:03:02,470 --> 00:03:03,040
Okay.

40
00:03:03,460 --> 00:03:12,580
So interestingly they are providing two free models, A and if you want to use any of the paying models

41
00:03:12,580 --> 00:03:15,370
you will have to include your API key.

42
00:03:15,400 --> 00:03:19,000
We are going to see how to do that later.

43
00:03:20,480 --> 00:03:24,650
And then you also can play around with the features of the model.

44
00:03:24,680 --> 00:03:29,780
Okay, so this is one way of using the playground.

45
00:03:29,990 --> 00:03:34,880
So we will we will we will get to the to the second way of using it.

46
00:03:34,880 --> 00:03:38,300
So you can use it just by going to the to hub.

47
00:03:39,020 --> 00:03:43,850
In the Language Myth hub you will find examples of prompts.

48
00:03:44,000 --> 00:03:46,280
You go to the language Myth Hub.

49
00:03:47,380 --> 00:03:49,420
You will see example.

50
00:03:49,420 --> 00:03:55,960
You will find examples of prompts that other developers are using for many different cases purpose,

51
00:03:55,960 --> 00:03:57,460
model, etc..

52
00:03:57,670 --> 00:04:02,980
Clicking on a prompt example will open the prompt playground environment.

53
00:04:03,010 --> 00:04:07,030
Not exactly because it's you need a second click.

54
00:04:07,030 --> 00:04:11,050
You need to click in the try it, but more or less.

55
00:04:11,800 --> 00:04:15,670
One interesting feature here is prompt versioning.

56
00:04:15,700 --> 00:04:22,270
With this you can you can see how the initial prompt has evolved with the contributions of different

57
00:04:22,270 --> 00:04:22,870
people.

58
00:04:22,870 --> 00:04:29,530
This can be an interesting way of collaborating between developers and subject matter experts, product

59
00:04:29,530 --> 00:04:38,950
managers, managers, marketing people, etc. so this is interesting, uh, because sometimes the LM

60
00:04:38,950 --> 00:04:49,450
app developer doesn't have the proper knowledge in order to write a correct prompt, and it is going

61
00:04:49,450 --> 00:04:58,510
to be the feedback from the product manager or marketing people, or subject matter expert who help

62
00:04:58,510 --> 00:05:03,580
the LM app developer to use the correct prompt.

63
00:05:03,580 --> 00:05:09,580
Do you remember the example we use a few lessons ago when when we were talking about, for example,

64
00:05:09,580 --> 00:05:11,920
a legal application, right?

65
00:05:11,920 --> 00:05:19,000
So we are going to develop an application that is going to help lawyers doing whatever.

66
00:05:19,000 --> 00:05:27,610
So in this particular field is going to be very interesting to have one expert, one subject matter

67
00:05:27,610 --> 00:05:34,630
expert included in your team, because he is going to help you not just to design the solution, but

68
00:05:34,630 --> 00:05:39,580
also to interpret properly the feedback that you are getting from your users.

69
00:05:39,580 --> 00:05:46,000
Because since you are not an expert in the field, you can be, uh, wrong in your, uh, conclusions.

70
00:05:46,000 --> 00:05:50,080
Okay, so this is an interesting feature.

71
00:05:50,080 --> 00:05:54,730
The prompt, a prompt versioning.

72
00:05:55,480 --> 00:06:03,580
You can import a prompt from the Lang Smith have into your long chain application without having to

73
00:06:03,580 --> 00:06:05,650
copy the entire prompt.

74
00:06:05,650 --> 00:06:11,410
And here you have a link, uh, to to see detailed instructions on how to do this.

75
00:06:11,410 --> 00:06:11,920
Okay.

76
00:06:11,920 --> 00:06:19,210
This this is interesting, but the interesting thing is that you can use it from any trace.

77
00:06:19,210 --> 00:06:27,520
So as I was telling you, you can go to one of your projects, you can open any of your traces.

78
00:06:27,520 --> 00:06:31,570
And from the trace you can go to the playground.

79
00:06:31,570 --> 00:06:34,270
And in the playground you can start playing around.

80
00:06:34,570 --> 00:06:41,440
So you can open any trace in this prompt playground environment and change the prompt, the LM model

81
00:06:41,440 --> 00:06:46,090
or the LM model features like temperature, etc..

82
00:06:46,660 --> 00:06:50,500
There are two LM models you can use here for free.

83
00:06:51,490 --> 00:06:52,210
Okay.

84
00:06:52,690 --> 00:06:57,040
And yeah, fireworks and Google Pan.

85
00:06:57,250 --> 00:06:57,880
Okay.

86
00:06:57,880 --> 00:07:03,940
To use ChatGPT, you will need to enter your OpenAI API key.

87
00:07:04,480 --> 00:07:12,610
When you are in the playground, there is a button in the right top corner called secrets and API keys.

88
00:07:12,610 --> 00:07:14,440
To do that you see this button.

89
00:07:14,440 --> 00:07:17,950
This is where you will enter your OpenAI key.

90
00:07:18,010 --> 00:07:18,640
Okay.

91
00:07:20,230 --> 00:07:21,880
Next solution.

92
00:07:21,880 --> 00:07:29,740
How to use Lang Smith comparison view to compare the performance of alternative approaches.

93
00:07:30,160 --> 00:07:36,700
So if you go to the data set dashboard go to one data set.

94
00:07:37,390 --> 00:07:39,220
Select several tests.

95
00:07:39,220 --> 00:07:45,190
In this case we only have one, but you would select several tests that you perform using that data

96
00:07:45,190 --> 00:07:48,160
set and click on the compare button.

97
00:07:48,160 --> 00:07:49,780
You see here the compare button.

98
00:07:49,780 --> 00:07:51,670
To see the comparison view.

99
00:07:51,670 --> 00:07:58,630
We can use this to compare several tests, compare outputs, compare performance, etc. so the best

100
00:07:58,630 --> 00:08:05,020
way to see this in action is where we look at one professional project.

101
00:08:05,020 --> 00:08:13,090
Okay, so with one professional project you are going to see the full potential of this platform because

102
00:08:13,090 --> 00:08:18,190
you will see all the different examples and it's going to be super interesting for you.

103
00:08:18,190 --> 00:08:21,580
But you you get an initial idea now.

104
00:08:21,940 --> 00:08:28,930
And finally, the four way that Lang Smith is going to help us during the prototyping phase is.

105
00:08:29,820 --> 00:08:36,750
Creating a test data set and include trace include tracing trace examples in it.

106
00:08:36,750 --> 00:08:40,409
So this is this is a very easy.

107
00:08:40,409 --> 00:08:49,110
So in the data set uh dashboard we can create a new data set with this button.

108
00:08:49,110 --> 00:08:55,710
And once we have a new data set we can include any trace.

109
00:09:00,200 --> 00:09:04,700
In the data set just by clicking here.

110
00:09:05,490 --> 00:09:06,000
Okay.

111
00:09:06,000 --> 00:09:08,610
You can click here in any trace.

112
00:09:08,610 --> 00:09:12,450
And the trace will be included in the test data set.

113
00:09:12,450 --> 00:09:19,260
Or when you are reviewing the the particular trace you can also use the button here.

114
00:09:19,260 --> 00:09:21,750
So remember why.

115
00:09:23,700 --> 00:09:25,680
Do we use a test data set?

116
00:09:25,680 --> 00:09:32,760
Remember that in the test data set, we have the correct inputs and outputs that we use in order to

117
00:09:32,760 --> 00:09:40,560
test different prototypes or different versions to check their accuracy, their, uh, latency, their

118
00:09:40,560 --> 00:09:43,380
cost, the number of tokens, etc., etc..

119
00:09:43,380 --> 00:09:43,830
Right.

120
00:09:43,830 --> 00:09:53,040
So here is this is how you can use language myth to create a test data set and include trace examples

121
00:09:53,040 --> 00:09:53,640
in it.

122
00:09:54,180 --> 00:10:00,240
So in the next lesson we are going to see some very interesting advanced tips.

123
00:10:00,270 --> 00:10:04,020
Uh, regarding language myth data sets.

124
00:10:04,020 --> 00:10:05,580
We are going to see this in the next lesson.

125
00:10:05,580 --> 00:10:07,950
It's going to be also very interesting.