WEBVTT 00:00.040 --> 00:01.040 Hey there Eden here. 00:01.040 --> 00:05.920 And in this video I'll be showing you what we're going to be building in this section. 00:06.120 --> 00:11.680 And we'll be implementing a search agent, which is an agent which has searching capabilities. 00:11.960 --> 00:13.680 So let's start with a quick demo. 00:14.240 --> 00:16.360 So I'm in ChatGPT right now. 00:17.040 --> 00:20.160 Let's now go and click the plus button. 00:20.160 --> 00:23.280 And here we can go and select Web Search. 00:23.320 --> 00:30.000 Now when we do this we provide ChatGPT with web search capabilities. 00:30.040 --> 00:36.720 And if our query will require a web search it will go and use this tool, get the results, chew it 00:36.720 --> 00:41.440 up and then return us the response grounded in the search results. 00:41.440 --> 00:43.120 So let's try with this query. 00:43.280 --> 00:50.960 Search for three job postings for an AI engineer using LinkedIn in the Bay area on LinkedIn and lists 00:50.960 --> 00:52.000 their details. 00:52.400 --> 00:58.320 Let's now send this query to ChatGPT and we'll follow along to see what it's going to do. 00:58.680 --> 01:02.360 So you can see right now that now it's going to search the web. 01:02.360 --> 01:04.800 And we can see we have some LinkedIn icons. 01:04.800 --> 01:09.020 So this is some cool generative UI which we're going to discuss in the course. 01:09.020 --> 01:12.220 So the application is reflecting what the agent is doing. 01:12.620 --> 01:15.020 And it's now going to output the response. 01:15.020 --> 01:22.860 And we can see that for each position we get we have here also a URL which is going to be where this 01:22.860 --> 01:23.780 answer came from. 01:23.780 --> 01:29.460 So we can see here this a engineering position for example which is hiring. 01:29.460 --> 01:32.460 Now let's go and check out another position. 01:35.540 --> 01:40.300 And here we can see we have a link chain index rag. 01:40.340 --> 01:42.860 We have this description in the title. 01:43.380 --> 01:45.500 Let's go and check the last one here. 01:48.620 --> 01:51.940 And yeah here we can see generative AI Software Engineer. 01:52.620 --> 01:57.260 Let's go and search here for link chain or. 02:00.020 --> 02:00.380 Yeah. 02:00.380 --> 02:02.580 So we can see we have here link chain. 02:03.180 --> 02:07.940 And this looks like indeed like a solid job positioning. 02:08.500 --> 02:09.300 Alrighty. 02:09.340 --> 02:12.490 So again we can see we have now the summary. 02:12.490 --> 02:14.850 And we have here the next steps. 02:14.850 --> 02:19.890 And the important thing to note here is that for each response here we had the source. 02:20.130 --> 02:26.970 Now this source, this grounding actually is very important to create the trust between the user and 02:26.970 --> 02:28.290 the system. 02:28.290 --> 02:34.690 Because if we only had the answers, then we couldn't decide as the users of this application whether 02:34.690 --> 02:36.490 they're trustworthy or not. 02:36.650 --> 02:40.810 And we know that Llms can hallucinate, they can generate a lot of garbage. 02:41.010 --> 02:46.930 So when we have those sources URLs, we can go, we can check them, we can browse them, we can cross-check 02:46.930 --> 02:48.570 what is returned from the LLM. 02:48.730 --> 02:52.370 And that is indeed grounded in the source that it's used. 02:52.370 --> 02:55.810 And we can even decide if we want to trust this source or not. 02:55.970 --> 03:01.610 So this is a quick preview for something which is called generative UI, which we're going to discuss 03:01.610 --> 03:02.770 in this course as well. 03:02.970 --> 03:06.850 But the important thing to note here is the searching capabilities. 03:07.050 --> 03:09.490 Large language models by themselves. 03:09.530 --> 03:11.930 They are simply text in, text out. 03:11.930 --> 03:17.730 Or if they're multimodal, they can accept also images and videos and audios and even produce them. 03:17.750 --> 03:22.870 But the important thing to note is that they don't have real time information. 03:23.110 --> 03:26.670 They're static in time because they're trained on a large corpus. 03:26.670 --> 03:31.590 And if you're interested about that, feel free to check out the theory section where I discuss this. 03:31.710 --> 03:34.630 But in general, they don't have access to the internet. 03:34.630 --> 03:40.070 And if you want to give them this kind of access or access to external capabilities, we need to provide 03:40.070 --> 03:41.350 them with tools. 03:41.350 --> 03:43.710 And this is what we're going to do in this section. 03:43.790 --> 03:48.230 So this section is going to be using learning agents for the first time. 03:49.070 --> 03:55.150 And an interesting story is that when I created this course around 2022, ChatGPT and all the other 03:55.150 --> 03:58.630 chat applications, they didn't have built in search. 03:58.630 --> 04:05.350 And today all of those chat applications have built in search, and it's very useful and it's quite 04:05.390 --> 04:06.710 a standard right now. 04:06.750 --> 04:09.990 However, back in the day, this was not the case. 04:10.110 --> 04:13.190 And it's interesting to see how those chat applications evolved. 04:13.390 --> 04:19.070 So in this section we'll be implementing with Linkchain a search engine which is an LLM that's going 04:19.070 --> 04:24.670 to have access to the exact same kind of search that we're seeing right now in this demo.