1 00:00:05,830 --> 00:00:14,200 In this lesson, we are going to talk about some consideration considerations to select the right orchestration 2 00:00:14,200 --> 00:00:15,490 framework. 3 00:00:21,570 --> 00:00:32,369 So as you remember, we have three main alternatives, uh, to, to use as orchestration frameworks. 4 00:00:32,369 --> 00:00:40,410 We have long chain, we have lemma index and we have the API of OpenAI. 5 00:00:40,560 --> 00:00:44,430 So let's talk a little bit about each of them. 6 00:00:44,430 --> 00:00:50,670 And uh we will give you our own conclusions and recommendations. 7 00:00:50,790 --> 00:01:03,000 So let's start talking about long chain in its first version launching what was a simple way to learn 8 00:01:03,000 --> 00:01:08,610 the basic concepts of LM applications and develop a mature applications. 9 00:01:09,000 --> 00:01:18,660 Long chain has many connectors and it is very interesting if ChatGPT is not hegemonic. 10 00:01:20,160 --> 00:01:26,760 In its second version, what is called the long chain expression language. 11 00:01:27,000 --> 00:01:29,760 Long chain is not so simple. 12 00:01:30,030 --> 00:01:36,360 It seems they are turning towards a solution more prepared for professional applications. 13 00:01:37,540 --> 00:01:41,650 Langton is still a very young and a small company. 14 00:01:41,830 --> 00:01:47,230 It still needs to find its strategic vision and business model. 15 00:01:47,530 --> 00:01:53,080 It is highly dependent on the evolution of ChatGPT. 16 00:01:56,410 --> 00:02:02,500 So luncheon is the most popular orchestration framework. 17 00:02:03,070 --> 00:02:12,160 It is a very interesting way to learn the basic concepts of Elm applications, and is a good way to 18 00:02:12,160 --> 00:02:14,830 develop a mature application store. 19 00:02:14,830 --> 00:02:15,580 Demos. 20 00:02:16,060 --> 00:02:17,620 It was simple to use. 21 00:02:17,620 --> 00:02:19,030 It was easy to learn. 22 00:02:19,390 --> 00:02:25,690 In the second version they are preparing what they call long chain expression languages. 23 00:02:25,690 --> 00:02:28,780 Things are not so simple anymore. 24 00:02:30,300 --> 00:02:32,730 So we. 25 00:02:34,010 --> 00:02:38,690 We'll tell you what we think about the evolution of land change later. 26 00:02:38,930 --> 00:02:42,380 Let's first talk about llama index. 27 00:02:42,410 --> 00:02:48,470 So llama index is the top alternative to a long chain right now. 28 00:02:49,160 --> 00:02:57,140 In its first version, Llama Index, uh is a less generalist framework than long chain. 29 00:02:58,000 --> 00:03:03,040 He wants to do fewer things but do them better. 30 00:03:03,130 --> 00:03:13,300 It specialises in generating possibilities for professional rack applications, so launching does less. 31 00:03:13,720 --> 00:03:20,380 Excuse me, llama index does less things than launching, but better. 32 00:03:20,860 --> 00:03:22,120 This is very important. 33 00:03:23,200 --> 00:03:34,780 So for professional LMDh applications, engineers are paying attention to Lambda index because in many 34 00:03:34,780 --> 00:03:40,060 cases long chain is a so simple approach for them. 35 00:03:40,480 --> 00:03:50,170 So they are seeing that lambda index is better prepared for more complex, more professional solutions. 36 00:03:50,170 --> 00:03:55,810 So they go to long chain to learn a toy demos. 37 00:03:55,810 --> 00:04:01,000 And after that they go to Lambda Index for professional solutions. 38 00:04:01,420 --> 00:04:08,590 This is changing lately because long Chain has seen that and is improving their framework as well. 39 00:04:08,590 --> 00:04:11,770 But this has been the trend we have observed. 40 00:04:11,770 --> 00:04:13,390 These are personal opinion. 41 00:04:14,450 --> 00:04:23,000 Llama index is also a very young and small company, and it also needs to find its strategic vision 42 00:04:23,000 --> 00:04:24,230 and business model. 43 00:04:24,320 --> 00:04:31,070 And right now, like Lange chain, it is highly dependent on the evolution of ChatGPT. 44 00:04:31,430 --> 00:04:38,240 What do we mean when we say that they are highly dependent on the evolution of ChatGPT is that they 45 00:04:38,240 --> 00:04:44,120 are moving after the chat GPT uh moves? 46 00:04:44,120 --> 00:04:44,630 Why? 47 00:04:44,660 --> 00:04:52,130 Because as I was telling you, during this first year of Elm applications, 2023 was the first year 48 00:04:52,130 --> 00:04:54,200 in the life of applications. 49 00:04:54,200 --> 00:04:58,760 ChatGPT was absolutely dominant in the market. 50 00:04:58,760 --> 00:05:07,100 So these two orchestration frameworks were extremely dependent on the evolution of ChatGPT. 51 00:05:07,880 --> 00:05:16,400 If you remember, Lang chain and lambda index are like universal languages to talk with different Elm 52 00:05:16,400 --> 00:05:17,060 models. 53 00:05:17,060 --> 00:05:23,750 So this is one of the main advantages of working with Lang chain and lambda index that you can use different 54 00:05:23,750 --> 00:05:29,630 Elm models, you can use ChatGPT or you can use anthropic or Lambda two, etc.. 55 00:05:29,630 --> 00:05:37,250 And with this universal approach of lambda index and lang chain, you don't have to learn, you know, 56 00:05:37,250 --> 00:05:43,760 the anthropic native language or the Lambda two native language, or the ChatGPT native language, because 57 00:05:43,760 --> 00:05:49,820 you have this universal language which is lang chain or a lambda index. 58 00:05:50,360 --> 00:05:53,450 But this was a theoretical approach. 59 00:05:53,450 --> 00:06:03,440 In reality, in the first year of Elm applications, ChatGPT was so dominant that a lang chain and lambda 60 00:06:03,440 --> 00:06:08,720 index had had to follow the path of ChatGPT. 61 00:06:08,870 --> 00:06:10,400 This was the first year. 62 00:06:10,400 --> 00:06:11,930 Let's see what happens in the. 63 00:06:11,960 --> 00:06:17,930 In the following years, we will talk a little bit about that, uh, later in this in this lesson. 64 00:06:17,930 --> 00:06:23,660 So what about OpenAI, the API of OpenAI. 65 00:06:23,690 --> 00:06:31,910 So some people during the first year of Elm applications, they so they, they, they thought okay, 66 00:06:31,910 --> 00:06:32,750 so. 67 00:06:34,080 --> 00:06:36,930 ChatGPT is totally dominant. 68 00:06:37,020 --> 00:06:44,160 Why should I use a middleman to work with the ChatGPT API? 69 00:06:44,190 --> 00:06:52,710 Because at the end of the day, long chain or lambda index are just middlemen between the engineer and 70 00:06:52,710 --> 00:06:54,720 the LM model API. 71 00:06:55,200 --> 00:07:03,660 So since ChatGPT was so dominant during the first year, some engineers said, okay, you know what? 72 00:07:03,660 --> 00:07:05,670 I'm going to learn directly? 73 00:07:05,670 --> 00:07:08,070 The API of ChatGPT. 74 00:07:08,220 --> 00:07:10,920 What we call OpenAI API. 75 00:07:10,950 --> 00:07:14,310 You know, OpenAI is the company behind ChatGPT. 76 00:07:14,640 --> 00:07:15,630 So. 77 00:07:16,520 --> 00:07:22,550 In its first version, the OpenAI API was not very simple. 78 00:07:23,720 --> 00:07:34,790 In the second version, a consequence of the changes presented at Dev Day in November 2023, this second 79 00:07:34,790 --> 00:07:39,080 version aims to be simpler and more versatile. 80 00:07:39,710 --> 00:07:47,450 OpenAI seems to have taken the path Apple took with the iMac and iPhone. 81 00:07:48,370 --> 00:07:52,540 So it does not open its technology. 82 00:07:52,780 --> 00:07:58,750 It reserves many areas that remain opaque to the developer. 83 00:07:59,880 --> 00:08:08,010 It controls a closed up market and sometimes it cannibalizes it. 84 00:08:08,770 --> 00:08:11,860 It happened with the first round of startups. 85 00:08:14,100 --> 00:08:21,990 It focuses on generating revenue, especially for from large companies, but also from developers and 86 00:08:21,990 --> 00:08:32,130 users, and it is conditioned by its alliance with Microsoft, favoring its products like Azure, Redis, 87 00:08:32,130 --> 00:08:37,919 etc., and favoring also Microsoft's revenue streams. 88 00:08:38,730 --> 00:08:46,200 So what we saw in November 2023 is that OpenAI. 89 00:08:47,490 --> 00:08:53,700 Wants to be more simple and more versatile for developers. 90 00:08:54,470 --> 00:09:02,960 And they have changed their API in order to, you know, talk directly to ChatGPT. 91 00:09:03,110 --> 00:09:10,340 They have changed their API in order to make it simpler, more simple, and more versatile. 92 00:09:10,340 --> 00:09:21,650 But for engineers, OpenAI has some problems, does not open its technology. 93 00:09:22,420 --> 00:09:26,560 It controls a closed up market. 94 00:09:26,590 --> 00:09:33,280 Sometimes it cannibalizes the startups in the in the closed app market. 95 00:09:33,880 --> 00:09:41,950 It focuses on generating revenue and it is conditioned by its alliance with Microsoft. 96 00:09:41,980 --> 00:09:44,470 All these are personal opinions. 97 00:09:44,560 --> 00:09:49,060 This is what we are perceiving in the market. 98 00:09:50,010 --> 00:09:54,390 So what are our conclusions as of today? 99 00:09:54,900 --> 00:09:57,630 It is very important to highlight this work. 100 00:09:57,660 --> 00:10:02,700 As of today, because things are changing very quickly. 101 00:10:02,700 --> 00:10:09,150 So in one year from now or six months from now, we can have a totally different reality. 102 00:10:09,150 --> 00:10:14,100 But as of today, what are we seeing in the market? 103 00:10:14,100 --> 00:10:27,930 First, OpenAI made a very strong statement at the Dev Day in November 2023, implicitly indicating 104 00:10:27,930 --> 00:10:34,170 that it wants to take market share from long chain and Lamar index. 105 00:10:36,010 --> 00:10:44,560 It facilitates the creation of basic Elm applications while maintaining many opaque areas. 106 00:10:45,040 --> 00:10:50,410 And it launches multimodality and many other new features. 107 00:10:51,190 --> 00:10:52,780 So OpenAI. 108 00:10:54,910 --> 00:10:59,800 Made a strong statement in November 2023. 109 00:11:00,460 --> 00:11:07,570 A and it seemed that it wants to take some market share from long chain and Laminex it is saying, you 110 00:11:07,570 --> 00:11:14,890 know what people is using ChatGPT ah as foundation model for their LM application. 111 00:11:14,890 --> 00:11:20,710 So I don't want to have middlemen to talk with engineers. 112 00:11:20,710 --> 00:11:24,310 I want the engineers to talk directly with my API. 113 00:11:24,310 --> 00:11:27,010 So I'm going to make it more simple. 114 00:11:27,220 --> 00:11:29,470 I'm going to make it more powerful. 115 00:11:30,290 --> 00:11:43,340 But since OpenAI is a private company, and since it is deciding to close some of its, uh, you know, 116 00:11:43,550 --> 00:11:53,390 uh, uh, coding operations and some operations with ChatGPT with the OpenAI API are opaque. 117 00:11:53,450 --> 00:11:56,930 We engineers do not have control. 118 00:11:57,820 --> 00:12:06,010 Do not have the same degree of control that we have with other open source alternatives like long chain 119 00:12:06,010 --> 00:12:07,510 and lambda index. 120 00:12:07,540 --> 00:12:13,840 And this is a huge problem for professional engineers. 121 00:12:13,840 --> 00:12:17,590 So for very amateur people. 122 00:12:18,550 --> 00:12:19,630 Air for students. 123 00:12:19,630 --> 00:12:22,870 For rookies, the API of OpenAI. 124 00:12:22,900 --> 00:12:29,110 The API of API of ChatGPT may be a good alternative to start experimenting, but. 125 00:12:29,110 --> 00:12:39,490 But the moment you want to build a professional application, you don't like the fact that many doors 126 00:12:39,490 --> 00:12:50,620 in the OpenAI API and the OpenAI way of doing things are closed for you, you don't have control a around 127 00:12:50,620 --> 00:12:54,610 many features of the OpenAI API. 128 00:12:54,640 --> 00:13:06,370 Does the a perception we got after the death day in November 2023, and after listening to the community 129 00:13:06,370 --> 00:13:09,790 of artificial intelligence engineers about that. 130 00:13:09,790 --> 00:13:11,320 So OpenAI. 131 00:13:12,170 --> 00:13:13,940 Second line chain. 132 00:13:14,330 --> 00:13:24,410 So for us, line chain is a recommendable platform to start and familiarize oneself with the basic concepts 133 00:13:24,410 --> 00:13:26,420 of LM applications. 134 00:13:27,730 --> 00:13:30,280 Good platform to learn. 135 00:13:30,280 --> 00:13:31,300 But. 136 00:13:33,000 --> 00:13:34,440 In our opinion. 137 00:13:35,240 --> 00:13:45,410 Luncheon runs the risk of losing direction and being left in no man's land with the second version. 138 00:13:46,010 --> 00:13:53,000 What is called long chain expression language because it loses the advantage of simplicity. 139 00:13:55,590 --> 00:13:59,850 So launching it was a very good way to learn. 140 00:13:59,850 --> 00:14:04,680 Very good way to start, you know, trying things and experiment. 141 00:14:04,860 --> 00:14:11,340 But with the second version, right now, you can use the first and the second version of Land Chain. 142 00:14:11,970 --> 00:14:17,340 And you will see that in this program, we will use the first version for you to learn the basics. 143 00:14:17,340 --> 00:14:19,590 And then we will use the second version. 144 00:14:20,220 --> 00:14:28,530 But eh, we still don't know if they are going to keep the first one and the second one, or they are 145 00:14:28,530 --> 00:14:30,060 going to change courses. 146 00:14:30,060 --> 00:14:32,850 I don't know, it's too early and it's too young. 147 00:14:32,850 --> 00:14:36,510 The company to to to see what they are going to to do. 148 00:14:36,510 --> 00:14:42,840 But in our opinion this evolution of land chain can be risky. 149 00:14:42,840 --> 00:14:48,390 And it reminds us what happened, for example with Angular and React. 150 00:14:48,540 --> 00:14:56,370 So angular at some point was was the leader of the market in the front end frameworks of uh, of uh, 151 00:14:56,370 --> 00:14:58,140 internet applications. 152 00:14:58,260 --> 00:15:06,420 But at some point angular decided to go for the enterprise level applications. 153 00:15:06,420 --> 00:15:18,390 So it became extremely complex and it lost most of their market share and react, uh, which was the 154 00:15:18,390 --> 00:15:28,920 second, uh, in the competition, uh, react took the whole market by storm, by keeping their framework 155 00:15:28,920 --> 00:15:32,370 very simple, you know, and easy to use. 156 00:15:32,370 --> 00:15:44,220 So react right now is used in simple mid and also complex application because it is de facto a standard 157 00:15:44,220 --> 00:15:49,650 in the front end, uh, community, front end developers community. 158 00:15:49,650 --> 00:16:02,610 So we are afraid about long chain following the path of angular getting more complex, losing traction, 159 00:16:02,610 --> 00:16:10,770 losing users, and losing the simplicity that made it so popular at the at the beginning. 160 00:16:10,770 --> 00:16:14,820 So we are waiting and see what is happening there. 161 00:16:14,820 --> 00:16:18,630 But we have a, uh, this feeling, you know, this perception. 162 00:16:18,630 --> 00:16:28,020 Finally, llama index is, in our opinion, a recommendable platform to delve into the possibilities 163 00:16:28,020 --> 00:16:32,550 of configuring rack application applications. 164 00:16:32,970 --> 00:16:41,400 Although the first version of llama index was not very user friendly, they are now striving to make 165 00:16:41,400 --> 00:16:42,840 the next version better. 166 00:16:42,840 --> 00:16:44,700 So llama index. 167 00:16:45,920 --> 00:16:49,370 Is following the opposite path of long chain. 168 00:16:49,610 --> 00:17:00,410 Long chain is going to from simple to complex, and lambda index is going from complex to simple. 169 00:17:01,340 --> 00:17:06,020 Long chain was a generalist framework. 170 00:17:06,020 --> 00:17:08,420 They wanted to do everything. 171 00:17:09,109 --> 00:17:19,339 Lambda index was much more focused on drag functionality, so they do less things, but better. 172 00:17:19,339 --> 00:17:27,140 Both companies are similar in size and similar also in the investment amount they have raised. 173 00:17:27,500 --> 00:17:29,480 They are both in Silicon Valley. 174 00:17:29,480 --> 00:17:32,660 They are both a created by. 175 00:17:33,600 --> 00:17:38,250 Aim founders that used to work together in the same previous company. 176 00:17:38,250 --> 00:17:39,510 This is interesting. 177 00:17:39,600 --> 00:17:42,030 So we don't know what is going to happen. 178 00:17:42,030 --> 00:17:52,140 But our perception, our opinion is that right now, Lamar Index is following a better track than Lange 179 00:17:52,140 --> 00:17:52,410 chain. 180 00:17:52,440 --> 00:17:54,510 Lange chain is still the leader. 181 00:17:54,510 --> 00:18:04,500 It has a huge community and we are seeing that Lange Chain is also reacting, trying to learn from the 182 00:18:04,500 --> 00:18:06,990 lessons that Lamar Index is showing. 183 00:18:06,990 --> 00:18:15,480 So we are seeing that the Lange chain is focusing more on rack solutions, and we will see what happens 184 00:18:15,480 --> 00:18:17,490 with the second version of Lange chain. 185 00:18:17,760 --> 00:18:27,900 But if in in our opinion, if Lange chain makes itself a more complex, it can lose their leadership 186 00:18:27,900 --> 00:18:29,610 in the in the community. 187 00:18:29,610 --> 00:18:31,620 But let's see what happens right now. 188 00:18:31,620 --> 00:18:33,810 As you can see, we have three players. 189 00:18:34,410 --> 00:18:36,270 Lantagne is the leader. 190 00:18:36,270 --> 00:18:39,960 Lama index is a very interesting follower there. 191 00:18:39,960 --> 00:18:48,930 And then we have OpenAI, who this year, at the end of the 2023 year, made a strong statement like, 192 00:18:48,930 --> 00:18:51,600 hey, I don't want to be out of this market. 193 00:18:51,600 --> 00:18:58,890 My LM model is the favorite a among the engineers. 194 00:18:58,890 --> 00:19:01,800 So I don't want to see middle mans there. 195 00:19:01,800 --> 00:19:04,350 I want to go directly with the engineers. 196 00:19:04,350 --> 00:19:07,890 I don't want lang chain or lama index in the middle. 197 00:19:07,890 --> 00:19:18,000 But in our opinion, the approach of OpenAI making some areas opaque not available for the engineers 198 00:19:18,000 --> 00:19:24,720 is a bad choice if they want to work with professional engineers, if they want to work with amateurs, 199 00:19:24,930 --> 00:19:29,400 uh, you know, developers that are looking for toy demos, it's okay. 200 00:19:29,400 --> 00:19:38,220 But if you want to work with a professional engineers, you need to give them the versatility that, 201 00:19:38,250 --> 00:19:43,500 uh, open source frameworks like Lang Chain or Lama Index are giving them. 202 00:19:43,500 --> 00:19:48,390 So I just wanted to give you our honest opinion. 203 00:19:49,350 --> 00:19:59,400 The conclusion we have is that we are going to use in this program long chain to start learning about 204 00:19:59,400 --> 00:20:00,720 LLM applications. 205 00:20:00,720 --> 00:20:06,360 We will use the first version of long chain in our projects, then the second version. 206 00:20:06,360 --> 00:20:12,180 Then we will use long chain for you to understand how long chain works. 207 00:20:12,180 --> 00:20:16,080 And we will also use the OpenAI API. 208 00:20:16,080 --> 00:20:18,990 We will see these three alternatives. 209 00:20:18,990 --> 00:20:27,750 You will see that our focus in this edition of the program is going to be more into long chain, but 210 00:20:27,750 --> 00:20:32,220 we also will look into Lambda Index and OpenAI. 211 00:20:32,220 --> 00:20:42,540 So during the 2024 year, let's see what happens with these three players and also with the alternatives 212 00:20:42,540 --> 00:20:44,130 we have in the in the market. 213 00:20:44,130 --> 00:20:47,070 As you know, there are some alternatives right now. 214 00:20:47,070 --> 00:20:48,870 They are very minoritarian. 215 00:20:48,870 --> 00:20:52,110 But let's see what happens in the following month. 216 00:20:52,530 --> 00:21:01,230 So in this lesson we have seen some considerations to select the right orchestration framework. 217 00:21:01,230 --> 00:21:12,750 In the next we are going to uh, to talk about the usage of programming languages in LM app development. 218 00:21:12,750 --> 00:21:22,140 Super interesting lesson for everybody, even if you are familiar with programming languages, but especially 219 00:21:22,140 --> 00:21:25,140 if you are new to programming languages.