WEBVTT 00:00.440 --> 00:04.160 Let me first import the prompt template class. 00:05.160 --> 00:06.560 What are prompt templates? 00:06.640 --> 00:11.240 Well, large language models receive as input something which is called a prompt. 00:11.560 --> 00:18.240 What is a prompt is simply text input that we give the LLM and the LLM processes it and returns us an 00:18.240 --> 00:18.880 output. 00:19.360 --> 00:25.480 Now, if you want to learn the formal definition of prompt and what elements it comprised of, then 00:25.480 --> 00:31.200 feel free to check out this video where I elaborately break down all the elements of a prompt formally. 00:31.560 --> 00:33.480 Okay, that's pretty straightforward. 00:33.720 --> 00:40.760 However, when we're dealing with programmatic access to the LLM, then we might want to take a prompt 00:40.760 --> 00:42.120 and to give it parameters. 00:42.520 --> 00:46.440 So one time I want to run this prompt where the product is cat food. 00:46.480 --> 00:49.160 The other time where the product is purchase. 00:49.200 --> 00:54.560 And for the last time, let's say I want to run it where the product is a piano. 00:55.000 --> 00:55.320 Okay. 00:55.360 --> 01:01.100 So for each different prompt I will get different output because the input was different. 01:01.780 --> 01:05.620 So this is the first abstraction that blockchain is introducing us. 01:05.660 --> 01:10.540 It's a prompt template and it's simply a wrapper class around a prompt. 01:10.660 --> 01:14.260 So it adds functionality to prompt to receive an input. 01:14.260 --> 01:18.380 For example, we can run this prompt now many times with different inputs. 01:18.580 --> 01:25.300 And the gist of it is that it simply helps us format things into strings, which are eventually being 01:25.300 --> 01:26.460 sent to llms. 01:27.060 --> 01:32.740 It also provides a lot more functionality than that which we are going to explore and check out in the 01:32.740 --> 01:33.220 course. 01:34.100 --> 01:34.540 Cool. 01:34.580 --> 01:44.940 So let's now import the chat OpenAI class, which is a chat model and a wrapper over the OpenAI API. 01:45.300 --> 01:52.820 The chat model object is often going to be our primary interface for interacting with large language 01:52.820 --> 01:53.460 models. 01:53.740 --> 02:00.920 And it is a standard way that LinkedIn help us to talk to those llms like GPT four. 02:01.360 --> 02:09.880 Anthropic lad, Google, Gemini, and even open source models like Llama by Facebook via O llama. 02:11.840 --> 02:20.480 So historically, many Llms just took a single string of text as an input and returned as a single string 02:20.480 --> 02:21.440 of text. 02:21.640 --> 02:29.560 However, modern llms are designed for conversation and interaction between a user, so they work best 02:29.560 --> 02:36.480 when we provide them with a list of messages representing the dialogue and the history, and they return 02:36.480 --> 02:37.880 us a message back. 02:38.120 --> 02:41.160 And this is the core of the chat model interface. 02:41.440 --> 02:48.560 The input is going to be a list of structured messages like a system instruction, user questions or 02:48.600 --> 02:49.680 AI responses. 02:49.920 --> 02:56.080 And the output is going to be an AI message representing the response of the large language model. 02:56.480 --> 03:05.180 However, chat model objects are very powerful and beyond generating human like text, they have very 03:05.180 --> 03:08.940 cool capabilities which a lot of them will be using in the course. 03:08.940 --> 03:14.100 And if you're curious about those capabilities, you can check out this video in the glossary section. 03:14.100 --> 03:19.220 But don't worry about it too much because we will be covering this hands on in the course. 03:19.580 --> 03:21.500 All right, let's go back to the code. 03:21.860 --> 03:30.260 And one of the best habits you can develop as a developer is to look inside the actual source code of 03:30.540 --> 03:33.660 the frameworks you're working with, and in this case Linkchain. 03:33.860 --> 03:39.260 And exploring the implementations directly is a great way to really understand what's going on behind 03:39.260 --> 03:39.900 the scenes. 03:39.900 --> 03:42.300 And we're going to be doing a lot of this in the course. 03:42.300 --> 03:44.260 So let me just show you how to do it. 03:44.260 --> 03:48.420 So I'm simply going to click command and on the class name. 03:48.420 --> 03:50.700 And then we can see the implementation of the class. 03:55.260 --> 03:57.720 Let's do the same for chat OpenAI. 04:00.960 --> 04:08.280 And here we can check out all the documentation, which is hardcoded in the source code itself and the 04:08.280 --> 04:09.480 functionality itself. 04:09.480 --> 04:11.280 But we're not going to cover this right now. 04:11.280 --> 04:13.320 This is a very high level overview. 04:13.360 --> 04:17.960 And of course, I'm not expecting you to know the inside implementation of all of those classes and 04:18.000 --> 04:18.600 objects. 04:18.760 --> 04:21.600 We're here to write our first link chain chain. 04:21.960 --> 04:30.480 And a link chain chain is a workflow that connects multiple components in linked chain together. 04:30.480 --> 04:37.360 In a sequence where the output of one step becomes the input of the next step. 04:37.680 --> 04:45.880 And just to give you a bit more detail here, each step can be an LM call, a prompt, a plate data 04:45.880 --> 04:50.440 transformation or tool call, and we'll get to tool calls. 04:50.440 --> 04:51.560 Don't worry about it. 04:51.560 --> 04:53.880 And it can even be another chain. 04:54.200 --> 05:00.220 And a chain Let us go beyond just making a single LLM prompt in response. 05:00.500 --> 05:07.220 So, for example, instead of asking a model one big question, we might want to format the user query 05:07.220 --> 05:09.020 into a structured prompt. 05:09.300 --> 05:11.940 We want to send this prompt to the LLM. 05:12.260 --> 05:19.540 Then we want to parse the llms output into some structured data, and then use this data to call to 05:19.580 --> 05:25.020 an external API and feed the API response into another LLM prompt. 05:25.180 --> 05:30.100 And with this composition idea, we can build some very advanced tech here. 05:30.300 --> 05:37.580 And this is why LinkedIn became so popular and so widely adopted and used, because it was the first 05:37.900 --> 05:45.420 framework that really allowed us to do those really complex things and to build on top of those llms 05:45.460 --> 05:52.820 some very cool stuff like agents, and we'll be diving very deep into agents, so don't worry about 05:52.820 --> 05:53.100 it.