WEBVTT 1 00:00:00.762 --> 00:00:02.610 Way earlier in the course, 2 00:00:02.610 --> 00:00:06.780 I mentioned that using open Large Language Models 3 00:00:06.780 --> 00:00:10.620 can be great for certain use cases. 4 00:00:10.620 --> 00:00:13.080 And indeed, you already saw some examples 5 00:00:13.080 --> 00:00:14.370 here in this section. 6 00:00:14.370 --> 00:00:16.680 For example, you could use a locally running 7 00:00:16.680 --> 00:00:20.400 Large Language Model like this Gemma 3 model here 8 00:00:20.400 --> 00:00:23.580 to extract text from an image. 9 00:00:23.580 --> 00:00:26.070 That's one thing you could do. 10 00:00:26.070 --> 00:00:30.180 You can also use these models to summarize documents 11 00:00:30.180 --> 00:00:35.180 or ask questions about documents to chat with your PDF. 12 00:00:35.370 --> 00:00:38.100 You don't need some paid service for that. 13 00:00:38.100 --> 00:00:40.230 You can just use LM Studio, 14 00:00:40.230 --> 00:00:43.620 upload or add your files here 15 00:00:43.620 --> 00:00:47.040 and then use your model, your loaded locally running model 16 00:00:47.040 --> 00:00:49.350 to interact with those documents. 17 00:00:49.350 --> 00:00:51.390 These are all amazing use cases 18 00:00:51.390 --> 00:00:53.850 because they are all about privacy. 19 00:00:53.850 --> 00:00:55.980 You might not wanna share your documents 20 00:00:55.980 --> 00:01:00.030 or your receipts with OpenAI or any other providers. 21 00:01:00.030 --> 00:01:02.940 Well, when running the models locally, you don't have to, 22 00:01:02.940 --> 00:01:05.820 and they're amazing for tasks like this. 23 00:01:05.820 --> 00:01:08.550 But it's not all they're useful for. 24 00:01:08.550 --> 00:01:11.280 You can, for example, also use these models 25 00:01:11.280 --> 00:01:13.230 if you don't have an internet connection 26 00:01:13.230 --> 00:01:16.830 and you want to generate some code, for example, 27 00:01:16.830 --> 00:01:20.190 or if you need to ask some questions about code. 28 00:01:20.190 --> 00:01:23.100 You might be used to using GitHub Copilot 29 00:01:23.100 --> 00:01:25.860 or other models when writing code, 30 00:01:25.860 --> 00:01:27.720 but what do you do if you're in a plane 31 00:01:27.720 --> 00:01:29.640 without an internet connection? 32 00:01:29.640 --> 00:01:32.340 Well, since these locally running models 33 00:01:32.340 --> 00:01:35.010 are still Large Language Models, 34 00:01:35.010 --> 00:01:38.250 they are, of course, still quite good at generating code. 35 00:01:38.250 --> 00:01:41.670 Of course, relatively small 36 00:01:41.670 --> 00:01:43.680 locally running Large Language Model 37 00:01:43.680 --> 00:01:45.971 like this 12 billion parameters model 38 00:01:45.971 --> 00:01:49.210 might not be as good as Claude 3.5 39 00:01:50.280 --> 00:01:53.520 or the latest OpenAI model. 40 00:01:53.520 --> 00:01:55.500 We have to be honest about that. 41 00:01:55.500 --> 00:01:58.170 And of course, the knowledge cutoff date 42 00:01:58.170 --> 00:02:00.030 of these problems may be a problem 43 00:02:00.030 --> 00:02:02.340 just as it is the case to some degree 44 00:02:02.340 --> 00:02:05.910 for these state-of-the art proprietary models though. 45 00:02:05.910 --> 00:02:08.370 But nonetheless, for a lot of tasks, 46 00:02:08.370 --> 00:02:10.860 the code they generate will be good enough 47 00:02:10.860 --> 00:02:14.700 because, of course, it's still valid code in many situations 48 00:02:14.700 --> 00:02:16.800 and if you got no internet connection, 49 00:02:16.800 --> 00:02:18.510 it's definitely better than having 50 00:02:18.510 --> 00:02:21.330 no backup and no help at all. 51 00:02:21.330 --> 00:02:23.430 And you cannot just use them for generating code. 52 00:02:23.430 --> 00:02:26.250 Of course, you can also ask questions about your code 53 00:02:26.250 --> 00:02:28.110 or use them for debugging, 54 00:02:28.110 --> 00:02:30.810 all without sending your data to anyone 55 00:02:30.810 --> 00:02:33.423 and all without having an internet connection.