WEBVTT 00:00.320 --> 00:06.230 I'm going to walk you through notebook LM, which is a really useful tool that Google has released. 00:06.260 --> 00:09.770 It feels like almost a ChatGPT moment. 00:09.800 --> 00:12.050 It's pretty amazing how well it works. 00:12.080 --> 00:17.960 And so at its core, what it is is just a way of sharing context with the LM, its free tool, which 00:17.960 --> 00:18.860 is really nice. 00:18.890 --> 00:24.980 And it has a bunch of crazy features which you'll see, but by and large, what I'm using it for is 00:25.010 --> 00:27.140 a very quick to upload sources. 00:27.140 --> 00:27.590 Let me see. 00:27.590 --> 00:30.620 I have here we go, a bunch of famous AI papers. 00:30.800 --> 00:36.740 So say, for example, I wanted to understand a little bit more about DSP. 00:37.190 --> 00:40.400 So this is a paper that was tough to read potentially. 00:40.430 --> 00:42.260 Right if you're not very technical. 00:42.260 --> 00:48.470 And and it could be, even if you are to be honest, it could take a while to really understand the 00:48.470 --> 00:49.880 things that you need. 00:49.910 --> 00:52.220 There's a few different things that we've just done here. 00:52.250 --> 00:56.630 First of all is like all we did was just upload that PDF and then boom, we suddenly have a summary, 00:56.660 --> 00:56.870 right? 00:56.900 --> 00:59.060 Like we have a pretty straightforward summary. 00:59.210 --> 01:05.770 We also have some suggested questions of like things we could ask, but we could ask our own stuff so 01:05.770 --> 01:11.650 we could say, what is the firm using DSP given paper? 01:13.300 --> 01:17.650 And it makes it into a chat like you can chat with your documents. 01:17.650 --> 01:19.960 This is not groundbreaking, right? 01:19.990 --> 01:26.320 There are things like this exist, but it's pretty cool that you can do this for free and it works pretty 01:26.320 --> 01:26.650 well. 01:26.650 --> 01:28.900 But we'll get around to get into the really interesting features. 01:28.900 --> 01:29.380 Here we go. 01:29.380 --> 01:30.910 So this is really cool. 01:31.150 --> 01:33.580 Now you have some answers here. 01:33.580 --> 01:41.980 You can click and it's going to find specifically where in that PDF it's talking about these improvements. 01:41.980 --> 01:43.960 So you can sense check it. 01:44.110 --> 01:46.600 You know yourself which is really helpful. 01:46.600 --> 01:50.230 And if you find this really useful you can click save. 01:50.590 --> 01:52.780 And now you have that saved somewhere. 01:53.020 --> 01:55.930 You can also delete the note or add more notes. 01:55.960 --> 01:59.020 Right now a note could be a prompt. 01:59.050 --> 02:00.550 You can think about it like this. 02:00.910 --> 02:01.600 So you can. 02:01.780 --> 02:06.430 Or it could be a prompt like a chat, or it could just be something that you want to know or like something 02:06.430 --> 02:07.750 that you want to keep to yourself. 02:07.750 --> 02:16.240 So just say this Fury could be a great way to summarize my prompts. 02:16.390 --> 02:19.000 Or client X didn't have to. 02:19.030 --> 02:20.470 It doesn't have a mixed mode, right? 02:20.500 --> 02:23.440 Like you don't have to use LMS for everything. 02:23.440 --> 02:25.630 It can be a note taking app as well. 02:25.630 --> 02:31.210 For this project, there's a few different really cool extra features that I want to show you, and 02:31.210 --> 02:36.070 these are the things that actually make it really powerful because even this is useful in itself. 02:36.070 --> 02:42.310 But if you click Notebook Guide, you can do things like set up a timeline. 02:42.610 --> 02:48.160 So if you want to kind of see how different things have been developed or whatever, then you can generate 02:48.160 --> 02:50.380 that from the prompt, which is really cool. 02:50.410 --> 02:53.890 You can go, you can just create like a table of contents or a study guide. 02:53.890 --> 03:00.240 So say, for example, you wanted to study the spy and get really good at it, then you could, you 03:00.240 --> 03:02.910 could do generate that and then take a look at it. 03:02.940 --> 03:03.420 Here we go. 03:03.450 --> 03:07.050 The timeline wasn't applicable for this one, so it's smart enough to know when it doesn't make sense. 03:07.050 --> 03:11.220 But it has talked about the different people involved, which is quite nice and who they are. 03:11.250 --> 03:14.160 It's extracted these entities, which is quite helpful. 03:14.250 --> 03:15.840 I find that really useful. 03:15.870 --> 03:20.340 You can also generate a briefing doc or FAQ or whatever it is. 03:20.340 --> 03:25.830 You have the suggested questions, but the really cool thing is you can click here and generate. 03:25.860 --> 03:32.160 You can actually generate a podcast of people talking about DSP. 03:32.190 --> 03:34.170 So they're talking about this paper here. 03:34.170 --> 03:37.230 And if you haven't heard one of these yet, it's pretty powerful. 03:37.260 --> 03:42.600 I'm going to share the recording of this one in the in the module. 03:42.600 --> 03:44.700 So you'll be able to download that and listen to it. 03:44.730 --> 03:47.250 But you could also go check this out yourself. 03:47.280 --> 03:52.650 I'm just going to call this the as Why Paper podcast. 03:53.190 --> 03:55.440 And it takes a few minutes. 03:55.860 --> 03:57.500 You don't have to stick around to like it says. 03:57.530 --> 04:04.130 And that's the nice thing here, is that you can just let it go and it will come up with different things. 04:04.190 --> 04:06.980 By the way, this is the study guide. 04:07.160 --> 04:13.940 So the prompt engineering challenge is their approach to figure out what the key abstractions are. 04:14.300 --> 04:19.070 Signatures versus prompts, teleprompters versus manual optimization. 04:19.100 --> 04:23.570 The advantages talked about which what data sets they do tests on. 04:23.600 --> 04:24.800 Yeah pretty interesting stuff. 04:24.830 --> 04:27.650 It even come up with some FAQs like quizzes. 04:27.860 --> 04:28.100 Yeah. 04:28.130 --> 04:29.120 Really powerful. 04:29.150 --> 04:36.260 This is, I think, already like the most powerful way to learn something that I've come across. 04:36.290 --> 04:37.970 It's genuinely amazing. 04:37.970 --> 04:42.260 When you listen to these, this conversation, you're going to be blown away at the quality. 04:42.290 --> 04:42.560 Yeah. 04:42.590 --> 04:45.350 If you want to check this is experimental. 04:45.380 --> 04:46.550 It's relatively new. 04:46.580 --> 04:48.860 I'm sure they're going to be making a bunch of changes. 04:49.010 --> 04:51.170 They already made a bunch of changes recently. 04:51.170 --> 04:55.310 One of the things you can do now is you can upload a YouTube video into here. 04:55.430 --> 04:57.230 You can upload planned websites. 04:57.470 --> 05:02.930 So you could actually have multiple sources in here and get it to summarize different things. 05:02.930 --> 05:09.020 So for example, let's see if I've got another DSP paper in here or something. 05:09.110 --> 05:10.190 Oh yeah here's Oprah. 05:11.270 --> 05:14.840 This is one of the papers that influenced DSP. 05:15.170 --> 05:20.420 Once that's uploaded I could just ask it like how is Oprah's cited in the DSP paper? 05:20.450 --> 05:23.240 Or you know, what are the similarities? 05:24.740 --> 05:33.950 Similarities or differences between and for DSP approaches. 05:36.110 --> 05:40.520 And that you can make those comparisons between the two, which is pretty amazing. 05:45.710 --> 05:50.870 The conversation's going to be generated soon, so I'll put that up with the with the link to this guide. 05:50.900 --> 05:52.670 But yeah, pretty cool stuff.