1
00:00:04,770 --> 00:00:13,770
In this lesson, we are going to talk about some real LM ops solution for you to to understand the kind

2
00:00:13,770 --> 00:00:18,990
of things you are going to find in the market when you are looking for solutions.

3
00:00:22,870 --> 00:00:25,300
Remember that we, uh.

4
00:00:26,540 --> 00:00:29,720
We do not work with any provider at all.

5
00:00:29,720 --> 00:00:36,620
We, when we recommend you uh, some tools, is because we think these are the most efficient tools.

6
00:00:36,620 --> 00:00:38,600
These are going to be the better tools for you.

7
00:00:38,600 --> 00:00:42,950
But as you know, this program is not sponsored by any company.

8
00:00:42,950 --> 00:00:46,370
So we are totally independent in our advice.

9
00:00:46,370 --> 00:00:55,250
Uh, when we when we, uh, talk about, uh, tools and alternatives in, in, in cases like this one.

10
00:00:55,250 --> 00:01:06,950
So the first thing, uh, you need to know about the LM ops market is that, uh, the LM tools, the

11
00:01:06,950 --> 00:01:17,450
LM ops tools market is that it is extremely recent, and it is in full expansion with new alternatives

12
00:01:17,450 --> 00:01:19,160
appearing frequently.

13
00:01:19,160 --> 00:01:25,340
So I will I will say it again, the market for LM ops tools is very recent.

14
00:01:25,340 --> 00:01:31,190
And in full expansion, with new alternatives appearing frequently.

15
00:01:31,190 --> 00:01:33,320
So market is booming.

16
00:01:33,980 --> 00:01:43,700
There are more comprehensive tools like Wildlife's, like Wild Labs, and more specific ones like guardrails.

17
00:01:43,730 --> 00:01:52,430
I in this lesson we will analyze two popular products from the company Wild Labs the product line kit

18
00:01:52,430 --> 00:01:55,910
and the product LM Security Management.

19
00:01:56,210 --> 00:02:04,400
So you will see that in the LM ops industry, uh, you find different kinds of tools.

20
00:02:04,610 --> 00:02:11,480
There are some very specific tools, like the one provided by the company, guardrails I.

21
00:02:12,490 --> 00:02:20,860
And you have other approaches that are more comprehensive that cover all the LM ops, uh, stages,

22
00:02:20,860 --> 00:02:25,150
like some of the products that the company Y labs is offering.

23
00:02:25,180 --> 00:02:32,290
So just as an example, we are going to talk about a couple of these solutions.

24
00:02:32,650 --> 00:02:39,970
Uh, these are solutions, uh, uh, provided by the company Y labs, which is one of the companies

25
00:02:39,970 --> 00:02:46,150
is one of them is one of the popular companies lately, but is not the best or the worst is just one

26
00:02:46,150 --> 00:02:46,420
of them.

27
00:02:46,420 --> 00:02:54,760
So we are going to talk about two products just for you to see the features they may offer, the things

28
00:02:54,760 --> 00:03:00,010
they are, uh, right now solving the things that are still not on the table, etc..

29
00:03:00,010 --> 00:03:09,880
So why, uh, the first of these products, uh, Lang Kit, focuses on extracting actionable insights

30
00:03:09,880 --> 00:03:12,310
for content moderation and observability.

31
00:03:12,310 --> 00:03:13,000
City.

32
00:03:13,000 --> 00:03:21,820
The other one, LM Security Management, focuses on protecting LM applications against a wider range

33
00:03:21,820 --> 00:03:29,230
of security risks, including prompt injections, data leaks, and misinformation.

34
00:03:29,230 --> 00:03:32,620
So let's talk a little bit about each of them.

35
00:03:33,960 --> 00:03:37,650
What are the main features of language?

36
00:03:37,980 --> 00:03:41,880
As you know, one of the products offered by Y labs.

37
00:03:42,780 --> 00:03:52,410
Language uses natural language techniques to extract actionable insights from prompts and responses.

38
00:03:53,040 --> 00:03:56,580
Identifying and mitigating malicious prompts.

39
00:03:56,580 --> 00:03:58,470
Sensitive data.

40
00:03:58,650 --> 00:04:00,810
Toxic responses.

41
00:04:00,810 --> 00:04:02,670
Problematic topics.

42
00:04:02,670 --> 00:04:07,110
Hallucinations, and jail jailbreak attempts.

43
00:04:08,670 --> 00:04:11,610
Language allows defining limits.

44
00:04:12,190 --> 00:04:19,570
And detecting problematic prompts and responses in real time, taking appropriate actions in case of

45
00:04:19,570 --> 00:04:20,470
failures.

46
00:04:22,089 --> 00:04:32,200
Blanket validates how LLM applications respond to known prompts, both continuously and adhoc, to ensure

47
00:04:32,200 --> 00:04:36,730
consistency when modifying prompts or changing models.

48
00:04:37,420 --> 00:04:47,500
It extracts key telemetry data and compares it with intelligent baselines over time, aiding in debugging

49
00:04:47,500 --> 00:04:50,950
and fine tuning of the LLM application.

50
00:04:52,310 --> 00:04:56,900
Blanket integrates easily with public APIs.

51
00:04:58,130 --> 00:05:00,290
And proprietary models.

52
00:05:01,220 --> 00:05:10,130
It provides over 50 telemetry signals to assess the quality, relevance, sentiment, and safety of

53
00:05:10,130 --> 00:05:11,930
prompts and responses.

54
00:05:12,980 --> 00:05:23,090
Okay, so as we said, Lang Kit is focused on extracting actionable insights for content moderation

55
00:05:23,090 --> 00:05:24,590
and observability.

56
00:05:24,620 --> 00:05:30,020
Okay, so one one of the areas of liveops.

57
00:05:30,910 --> 00:05:37,720
What about LM Security Management, which is the name of the second product of wildlife labs we are

58
00:05:37,720 --> 00:05:39,430
going to analyze.

59
00:05:39,460 --> 00:05:49,990
So this product includes protection against malicious attacks, prevention of data leaks, defense against

60
00:05:49,990 --> 00:05:54,280
prompt injections, mitigation of disinformation.

61
00:05:54,610 --> 00:05:58,600
And it adopts best security practices.

62
00:05:59,110 --> 00:06:02,800
It implements telemetry to capture security risks.

63
00:06:02,800 --> 00:06:11,260
Define it on all the you know here they are citing, you know, the name of the law that they are covering,

64
00:06:11,260 --> 00:06:17,440
you know, allowing inline guardrails, continuous evaluations and observability.

65
00:06:18,710 --> 00:06:23,390
This product handles various security risks like.

66
00:06:24,800 --> 00:06:25,550
Insecure.

67
00:06:25,550 --> 00:06:27,350
Handling of outputs.

68
00:06:27,350 --> 00:06:27,950
Training.

69
00:06:27,950 --> 00:06:29,150
Data poisoning.

70
00:06:29,150 --> 00:06:30,590
Denial of service.

71
00:06:31,100 --> 00:06:35,900
Supply chain issues and overreliance on LMS.

72
00:06:36,440 --> 00:06:41,750
And finally, it includes guardrails and customizable logging.

73
00:06:42,260 --> 00:06:50,180
It implements inline guardrails with customizable metrics, thresholds, and actions, and logs each

74
00:06:50,180 --> 00:06:51,800
prompt respond.

75
00:06:52,960 --> 00:06:55,660
Unlocks each prompt response pair.

76
00:06:56,410 --> 00:07:03,580
So this is the literature you are going to find when you are trying to learn about the different tools

77
00:07:03,580 --> 00:07:07,420
you have available in the Mops market.

78
00:07:07,840 --> 00:07:15,700
As you see, I think the interesting, the interesting part of this review is that even when LM applications

79
00:07:15,700 --> 00:07:27,190
are still so young, we are a finding a lot of areas where we can optimize them, where we can monetize

80
00:07:27,190 --> 00:07:31,450
them, and where we can solve problems.

81
00:07:31,780 --> 00:07:41,440
So you are going to find a large number of tools associated with LM application development and also

82
00:07:41,440 --> 00:07:43,660
with LM ops.

83
00:07:44,320 --> 00:07:49,660
So I think it was interesting to take a look at this couple of examples.

84
00:07:50,050 --> 00:08:00,070
In the next lesson we are going to talk about a very important question, which is once you are working

85
00:08:00,070 --> 00:08:10,270
as a artificial intelligence engineer or an LM application engineer, how do you, uh, stay up to date

86
00:08:10,270 --> 00:08:21,010
in an industry, uh, with so many changes, so many news, so many new tools, so many new things to

87
00:08:21,010 --> 00:08:21,670
learn.

88
00:08:21,730 --> 00:08:29,410
Okay, so in the next lesson, we are going to talk about the top information channels for artificial

89
00:08:29,410 --> 00:08:34,929
intelligence engineers in their own, uh, responses.

90
00:08:34,929 --> 00:08:43,480
So we are going to see the results of a survey, uh, among artificial intelligence engineers regarding

91
00:08:43,510 --> 00:08:46,270
top information channels in the next lesson.