1
00:00:03,670 --> 00:00:07,180
Okay, so the Lang Smith team.

2
00:00:08,039 --> 00:00:16,890
Has prepared some very interesting guides and recommendations that we have linked there here in the

3
00:00:16,890 --> 00:00:17,970
in the notebook.

4
00:00:17,970 --> 00:00:18,510
Okay.

5
00:00:18,510 --> 00:00:27,210
So you have a several very interesting how to guides regarding lamb-smith tracing, how to specify the

6
00:00:27,210 --> 00:00:30,720
percentage of traces you send to Lamb-smith.

7
00:00:30,720 --> 00:00:36,840
This is very important because this has to do with your cost.

8
00:00:36,840 --> 00:00:41,010
So we highlighted this bold as you can see.

9
00:00:41,010 --> 00:00:45,930
So how to specify the percentage of traces you send to Lamb-smith.

10
00:00:46,110 --> 00:00:50,460
How to add metadata and tags to traces.

11
00:00:50,460 --> 00:00:54,750
How to add an annotation to one particular trace.

12
00:00:56,070 --> 00:00:59,640
Here you can see all the how to guides in this section.

13
00:00:59,640 --> 00:01:05,730
And here you can see the Lang chain specific guides for this section.

14
00:01:06,460 --> 00:01:16,480
Then we have some use cases guides for Lang Smith tracing to monitor application sentiment, to summarize

15
00:01:16,480 --> 00:01:22,540
application usage and for few shot prompting with a Lang Smith datasets.

16
00:01:22,540 --> 00:01:24,190
Very interesting as well.

17
00:01:24,910 --> 00:01:30,580
We have some, uh, interesting guides regarding Lang Smith evaluation.

18
00:01:30,580 --> 00:01:32,470
You have one quick start guide.

19
00:01:32,500 --> 00:01:33,400
Very interesting.

20
00:01:33,400 --> 00:01:41,620
And also a detail a document about Lang Smith evaluation concepts.

21
00:01:41,620 --> 00:01:50,950
And finally you have several very interesting how to guides regarding Lang Smith evaluation, how to

22
00:01:50,950 --> 00:01:59,950
create custom evaluators, how to paraphrase examples to expand your data set, how to create new examples

23
00:01:59,950 --> 00:02:03,760
using prompting and other how to guides.

24
00:02:05,550 --> 00:02:16,170
The Lang Smith team has also included a list of recommendations for us when we are working on evaluation.

25
00:02:16,500 --> 00:02:18,120
This is what they say.

26
00:02:18,240 --> 00:02:20,670
Test early and often.

27
00:02:21,510 --> 00:02:25,260
Create domain specific evaluators.

28
00:02:25,650 --> 00:02:28,740
Use labels where possible.

29
00:02:29,460 --> 00:02:32,220
Use aggregate levels.

30
00:02:33,060 --> 00:02:35,580
Measure model stability.

31
00:02:36,290 --> 00:02:39,620
Measure performance of subsets.

32
00:02:40,470 --> 00:02:42,930
Evaluate production data.

33
00:02:44,000 --> 00:02:47,390
Don't train on this data sets.

34
00:02:48,140 --> 00:02:50,480
Test the model yourself.

35
00:02:51,350 --> 00:02:58,580
Ask appropriate questions and here you have interesting resources to learn more.

36
00:03:02,240 --> 00:03:05,630
Well, I think these are general recommendations.

37
00:03:05,630 --> 00:03:12,560
Probably they don't apply to every case, but I think they are very good general recommendations.

38
00:03:13,280 --> 00:03:24,620
Then you have a interesting a information, a link to interesting information about monitoring, a link

39
00:03:24,620 --> 00:03:35,840
to a quick start guide about the prompt hub, and a couple of interesting links with more information.

40
00:03:35,960 --> 00:03:41,360
This is the blog post of the announcement of a.

41
00:03:41,360 --> 00:03:47,990
The general availability of Lang Smith, in my opinion, is a little bit difficult to read because they

42
00:03:47,990 --> 00:03:55,310
have included so many gifts there that this is really not easy, but a well, there you have it.

43
00:03:55,310 --> 00:04:03,020
And remember that we have based many of our previous lessons in the information contained here.

44
00:04:03,020 --> 00:04:06,830
So probably right now you are very familiar with the content.

45
00:04:07,190 --> 00:04:12,890
And this second a video is very interesting.

46
00:04:12,890 --> 00:04:17,390
This is the Lang Smith in-depth by the Lang Smith team.

47
00:04:17,690 --> 00:04:20,480
Also very interesting resource.

48
00:04:20,480 --> 00:04:31,010
So in the next lesson we are going to see a detailed view of the next features of Lang Smith.

49
00:04:31,010 --> 00:04:33,290
What are they preparing for us.

50
00:04:33,290 --> 00:04:35,240
So this is going to be very interesting.

51
00:04:35,240 --> 00:04:38,810
We will see that in the next lesson.