1 00:00:05,850 --> 00:00:11,190 This lesson we are going to talk about ChatGPT versus Elm applications. 2 00:00:12,150 --> 00:00:13,380 We will learn. 3 00:00:13,380 --> 00:00:15,060 What are the foundational. 4 00:00:16,550 --> 00:00:25,580 And we will see that LLM applications solve the limitations of these foundational models. 5 00:00:31,210 --> 00:00:34,570 So let's talk a little bit about ChatGPT. 6 00:00:35,740 --> 00:00:41,920 So ChatGPT is a private foundational model. 7 00:00:43,360 --> 00:00:43,870 This. 8 00:00:43,870 --> 00:00:50,560 Models are like the brains of artificial intelligence, let's call them that. 9 00:00:50,560 --> 00:00:55,000 But you will see that these brains are a little bit limited. 10 00:00:55,800 --> 00:01:05,129 You can find private foundational models like ChatGPT, and you can find open source foundational models 11 00:01:05,129 --> 00:01:08,970 like lamassu, Falcon, and others we will see later. 12 00:01:09,740 --> 00:01:17,600 In this case, ChatGPT is a private foundational model developed from a company called OpenAI. 13 00:01:18,500 --> 00:01:21,620 And as a private foundational model. 14 00:01:21,620 --> 00:01:23,300 It can be costly. 15 00:01:23,300 --> 00:01:31,040 And you will see that when we start developing LM applications, one of the most important thing we 16 00:01:31,040 --> 00:01:33,830 have to keep in mind is cost. 17 00:01:34,190 --> 00:01:43,430 So ChatGPT can be costly and as it is a private product, it can change at any time. 18 00:01:43,430 --> 00:01:51,830 So OpenAI, the company behind ChatGPT, can decide at any moment that you want to change the way ChatGPT 19 00:01:51,830 --> 00:01:52,580 works. 20 00:01:53,370 --> 00:02:00,180 On the contrary, LM applications are not foundational models. 21 00:02:00,180 --> 00:02:04,620 They are built on top of foundational models. 22 00:02:05,280 --> 00:02:13,710 So we can build an LM application on top of one or on top of several foundational models, and we can 23 00:02:13,710 --> 00:02:22,320 decide if we are going to use a private foundational model like ChatGPT or an open source one. 24 00:02:22,650 --> 00:02:30,270 So LM applications do not depend on an individual company like the case of ChatGPT. 25 00:02:31,050 --> 00:02:36,630 And with LM applications, we can control the cost. 26 00:02:36,630 --> 00:02:45,000 And we can control the changes that are, uh, being made in the application because it's our application. 27 00:02:46,460 --> 00:02:49,850 ChatGPT has some other interesting features. 28 00:02:49,850 --> 00:02:53,420 One of them is what we call context. 29 00:02:53,810 --> 00:03:02,420 We will talk more about, uh, context later, but right now, just think or think about context as 30 00:03:02,660 --> 00:03:05,390 let's say short time memory. 31 00:03:05,510 --> 00:03:11,240 So ChatGPT has a limited context, has a limited short time memory. 32 00:03:11,240 --> 00:03:19,100 And on the contrary, LM applications can have a broader context, can have more short memory. 33 00:03:20,030 --> 00:03:24,170 ChatGPT does not know your company's data. 34 00:03:24,170 --> 00:03:26,570 This is a very important thing. 35 00:03:26,990 --> 00:03:36,230 So ChatGPT doesn't know anything about the data of your company, but an LM application can be trained 36 00:03:36,470 --> 00:03:37,760 on your computer. 37 00:03:38,840 --> 00:03:46,520 And this is important because many companies, they don't want ChatGPT to have anything to do with their 38 00:03:46,520 --> 00:03:51,650 data because they don't trust ChatGPT or ChatGPT security. 39 00:03:52,070 --> 00:03:58,310 So when you are talking with ChatGPT, your conversations are in the cloud. 40 00:03:58,940 --> 00:04:05,300 And when you are talking with an LM application, your conversation can be in the cloud or it can be 41 00:04:05,300 --> 00:04:06,110 private. 42 00:04:06,320 --> 00:04:10,610 So you have more flexibility with LM applications. 43 00:04:11,300 --> 00:04:19,040 ChatGPT has a way to integrate it with external tools, what we call integrations. 44 00:04:19,040 --> 00:04:20,450 So external APIs. 45 00:04:20,450 --> 00:04:29,720 But it has a limited number of integrations, while an LM application has unlimited integrations possible. 46 00:04:30,050 --> 00:04:34,340 And finally, ChatGPT has a limited functionality. 47 00:04:34,370 --> 00:04:41,690 It has a huge, very interesting number of, uh, possibilities, but they are limited. 48 00:04:41,690 --> 00:04:44,900 You cannot do anything you want with ChatGPT. 49 00:04:44,990 --> 00:04:53,750 You are limited to a number of things you can do, but with LM applications you have unlimited functionality. 50 00:04:53,750 --> 00:04:58,160 So your imagination is the limit with LM applications. 51 00:04:58,160 --> 00:05:02,510 Let's learn a little bit more about that. 52 00:05:02,510 --> 00:05:05,150 So in short. 53 00:05:06,820 --> 00:05:13,210 Foundational models like ChatGPT or others like llama two. 54 00:05:13,240 --> 00:05:14,230 Falcon. 55 00:05:14,230 --> 00:05:15,220 Mistral. 56 00:05:15,250 --> 00:05:24,490 Are like the gasoline of like the gasoline for LM applications and LM applications would be like cars, 57 00:05:24,490 --> 00:05:33,040 motorcycles, boats, machines, etc. that run on the gasoline provided by foundational models like 58 00:05:33,040 --> 00:05:33,700 ChatGPT. 59 00:05:34,240 --> 00:05:44,980 So ChatGPT is the gasoline, and LM applications are the cars that are using the gasoline of ChatGPT. 60 00:05:46,640 --> 00:05:56,300 ChatGPT right now is currently the highest quality foundational model, but it is also the most expensive. 61 00:05:57,230 --> 00:06:05,930 Alternatives to ChatGPT are becoming increasingly numerous, higher in quality and less costly. 62 00:06:06,740 --> 00:06:08,780 Foundational models. 63 00:06:09,290 --> 00:06:10,700 They can do. 64 00:06:11,700 --> 00:06:12,630 A lot of things. 65 00:06:12,630 --> 00:06:16,350 You can create conversational conversational agents. 66 00:06:16,650 --> 00:06:18,300 You can summarize documents. 67 00:06:18,300 --> 00:06:19,860 You can generate content. 68 00:06:19,860 --> 00:06:27,210 You can do tasks like sentiment analysis, text classification, code creation, debugging, etc., 69 00:06:27,210 --> 00:06:32,370 etc. but they have many things they cannot do. 70 00:06:32,400 --> 00:06:34,740 They have many limitations. 71 00:06:34,740 --> 00:06:40,410 And this is where LM applications play a very important role. 72 00:06:40,410 --> 00:06:50,550 Because LM applications solve many limitations of ChatGPT and other foundational models. 73 00:06:51,720 --> 00:06:54,120 What kind of limitations has. 74 00:06:55,220 --> 00:06:57,440 Our foundational model like ChatGPT. 75 00:06:58,010 --> 00:06:59,180 The cost? 76 00:07:00,060 --> 00:07:01,710 The context. 77 00:07:02,830 --> 00:07:05,080 The regulatory challenges. 78 00:07:05,560 --> 00:07:07,150 Data privacy. 79 00:07:08,000 --> 00:07:08,690 Inference. 80 00:07:08,690 --> 00:07:09,650 Latency. 81 00:07:10,250 --> 00:07:10,970 Speed. 82 00:07:11,600 --> 00:07:13,700 Diversity of data sources. 83 00:07:13,700 --> 00:07:19,220 Whenever we are going to use different data sources, we are going to have a problem with ChatGPT. 84 00:07:21,280 --> 00:07:24,250 Lack of reproducibility. 85 00:07:24,280 --> 00:07:25,840 We'll see what it is later. 86 00:07:25,840 --> 00:07:27,220 Hallucinations. 87 00:07:27,220 --> 00:07:33,160 This is fake content made up by ChatGPT evaluation. 88 00:07:33,900 --> 00:07:37,350 Fragility of prompts, outdated knowledge. 89 00:07:37,740 --> 00:07:41,520 So, as I was telling you. 90 00:07:42,290 --> 00:07:47,090 ChatGPT is not as omnipotent as it seems. 91 00:07:47,090 --> 00:07:50,780 It is very powerful, but it is not as omnipotent as it seems. 92 00:07:50,780 --> 00:07:57,440 And L.l.m. applications are the key to use the gasoline. 93 00:07:57,440 --> 00:08:05,150 ChatGPT and other foundational models give you to do really amazing things. 94 00:08:05,150 --> 00:08:12,530 So his LLM application, the ones that are going to create the magic in the artificial intelligence 95 00:08:12,530 --> 00:08:13,400 world. 96 00:08:13,970 --> 00:08:19,430 So in this lesson we have seen we will talk more about that later. 97 00:08:19,430 --> 00:08:25,070 But we have seen a little bit about ChatGPT versus LLM applications. 98 00:08:25,190 --> 00:08:28,550 We saw what are the foundational models. 99 00:08:28,790 --> 00:08:39,140 And the most important thing we saw that LLM applications, LLM apps solve the limitations of the foundational 100 00:08:39,140 --> 00:08:48,500 models in the next lesson, we will tell you how to download the two books included with this bootcamp. 101 00:08:49,230 --> 00:08:51,720 Keys to artificial intelligence. 102 00:08:51,720 --> 00:08:57,390 This is a 300 page book and 100 AI startups. 103 00:08:57,390 --> 00:08:59,790 These are 200 pages book. 104 00:09:00,120 --> 00:09:06,030 Both of us are available in Amazon for more than $50. 105 00:09:06,790 --> 00:09:13,480 Combine, so it's a very good deal for you to have them included in this boot camp.