WEBVTT NOTE This file was generated by Descript 00:00:05.000 --> 00:00:08.720 Raphaël: Hey folks, and welcome to this latest episode of 00:00:08.720 --> 00:00:11.213 The Small Tech Podcast by EC. 00:00:11.273 --> 00:00:12.333 I'm your host Raph. 00:00:12.353 --> 00:00:16.523 And today we're going to be talking about AI before we get going, though. 00:00:16.523 --> 00:00:19.433 Remember to like, and subscribe if you like the work that 00:00:19.433 --> 00:00:21.053 we do, we're a small team. 00:00:21.053 --> 00:00:22.763 We'd really appreciate the support. 00:00:22.763 --> 00:00:27.423 It helps a lot leave us a review and if you want to join us on an episode 00:00:27.473 --> 00:00:29.603 we just released our first interview. 00:00:29.913 --> 00:00:31.803 And we'd love to have you on, if you want to talk about 00:00:31.803 --> 00:00:33.723 building small tech products. 00:00:33.806 --> 00:00:36.356 So let's talk about AI. 00:00:36.356 --> 00:00:41.936 AI has been in the news a lot lately over the past year or two years. 00:00:42.416 --> 00:00:46.096 But something that I started to recognize recently. 00:00:46.636 --> 00:00:49.636 Is that as the hype grows around it. 00:00:49.726 --> 00:00:52.156 So do the misconceptions around it. 00:00:52.546 --> 00:00:59.986 And while I'm no expert in the mechanics of what goes on deep under the hood, I do 00:00:59.986 --> 00:01:04.846 recognize some things that people don't understand about what this technology is. 00:01:05.146 --> 00:01:09.646 And specifically the stuff that we're talking about these days. 00:01:10.006 --> 00:01:11.296 So a little bit of history. 00:01:11.696 --> 00:01:14.486 We've been talking about things that. 00:01:14.886 --> 00:01:19.536 Really are just different forms of statistical analyses. 00:01:20.126 --> 00:01:25.496 As AI or machine learning or whatever for a while now. 00:01:26.036 --> 00:01:29.276 And so what we're calling AI at the moment. 00:01:29.626 --> 00:01:35.626 Is An extension of what we were referring to as machine learning more 00:01:35.626 --> 00:01:38.086 recently and big data prior to that. 00:01:38.086 --> 00:01:42.766 And it's mixed in with other concepts like neural networks and Gans and. 00:01:43.516 --> 00:01:45.046 All kinds of other stuff. 00:01:45.096 --> 00:01:51.376 And I remember playing with Tools based off of generative adversarial networks 00:01:51.376 --> 00:01:54.136 way back many years ago at this point. 00:01:54.136 --> 00:01:58.496 But they didn't enter this sort of hype cycle that we've 00:01:58.496 --> 00:02:01.986 gotten into with AI more lately. 00:02:02.386 --> 00:02:04.606 And I think the reason for that is. 00:02:05.116 --> 00:02:10.006 The results, the outputs, the things that you got out of those tools. 00:02:10.406 --> 00:02:15.086 Either from a consumer perspective, weren't as fun and exciting. 00:02:15.486 --> 00:02:21.846 And from a business perspective were great, but they didn't tap into 00:02:21.846 --> 00:02:26.646 people's imagination in the same way that the tools that we have now do. 00:02:27.046 --> 00:02:33.136 So we've had AI being used to do all kinds of stuff from predicting trends from large 00:02:33.136 --> 00:02:39.926 datasets to parsing audio and generating text in different ways for a while now. 00:02:39.926 --> 00:02:45.086 The thing that's changed in the past couple of years specifically, is what 00:02:45.086 --> 00:02:47.426 we're now referring to as generative AI. 00:02:47.966 --> 00:02:52.346 And I think specifically there's two things in there that people 00:02:52.346 --> 00:02:56.996 are particularly enamored with and find fascinating to play with. 00:02:57.396 --> 00:03:00.426 One is the image generators. 00:03:00.486 --> 00:03:04.306 So being able to just describe something and have an image 00:03:04.336 --> 00:03:05.746 come out of that description. 00:03:05.746 --> 00:03:10.786 Until recently there wasn't a great way to generate decent 00:03:10.816 --> 00:03:12.706 outputs for something like that. 00:03:12.736 --> 00:03:16.406 But in the past few years, That the results have gotten way 00:03:16.406 --> 00:03:17.786 better than they were before. 00:03:18.186 --> 00:03:22.386 And of course with that comes a whole slew of controversies around 00:03:22.386 --> 00:03:26.706 intellectual property and more, we're not going to talk so much about that. 00:03:26.766 --> 00:03:30.136 In this episode, but we will talk a little bit about the mechanics 00:03:30.136 --> 00:03:31.756 of what's going on under the hood. 00:03:32.156 --> 00:03:36.296 The other thing is large language models, which kind of do the same thing. 00:03:36.296 --> 00:03:37.256 But with text. 00:03:37.556 --> 00:03:39.836 So you provide some sort of input. 00:03:40.236 --> 00:03:41.586 That might be a short prompt. 00:03:41.616 --> 00:03:43.026 It might be a long prompt. 00:03:43.116 --> 00:03:47.586 And basically you get something back that either continues a conversation. 00:03:47.586 --> 00:03:52.886 If you're using something in a chat style, interface with an LLM. 00:03:53.286 --> 00:03:54.726 Think of chat GPT. 00:03:55.156 --> 00:03:56.236 Or you. 00:03:56.896 --> 00:04:00.446 Or it might just complete something that you started. 00:04:00.576 --> 00:04:02.496 Though, we're seeing less of that at the moment. 00:04:02.896 --> 00:04:07.366 Functionally though under the hood, that's what, even the chat interfaces do. 00:04:07.766 --> 00:04:10.166 And here's where we'll start to dive into the mechanics 00:04:10.166 --> 00:04:11.996 because I've seen people and. 00:04:12.626 --> 00:04:16.196 I want to talk specifically to people who are not deeply technical. 00:04:16.526 --> 00:04:21.926 So if you're an AI expert or you're a developer, who's really deep into this 00:04:21.926 --> 00:04:23.996 stuff, this probably isn't for you. 00:04:23.996 --> 00:04:27.326 And if you do keep watching, you might find yourself shaking your 00:04:27.326 --> 00:04:32.096 fist at the screen if you're watching or in the air, if you're listening 00:04:32.306 --> 00:04:34.256 saying that's not quite right. 00:04:34.316 --> 00:04:35.966 And I think that's okay. 00:04:36.476 --> 00:04:41.156 I'm going to try and cover things in a way that I think makes sense for 00:04:41.156 --> 00:04:46.316 someone who is non-technical, but is interested in what can be done with 00:04:46.316 --> 00:04:49.886 these technologies or wants to better understand them, wants to understand how 00:04:49.886 --> 00:04:51.836 they might fit into a product roadmap. 00:04:51.836 --> 00:04:53.996 See what they can build with this technology. 00:04:54.396 --> 00:04:58.266 Or even somewhat technical folks who just haven't played with it yet. 00:04:58.666 --> 00:05:03.476 I don't actually understand very well what's going on with the image generators. 00:05:03.776 --> 00:05:08.966 On some level what I understand is essentially images are fed into the 00:05:08.966 --> 00:05:13.386 training systems and don't think of them as being stored in there. 00:05:13.416 --> 00:05:17.226 It's more like information about the images that's being stored. 00:05:17.626 --> 00:05:19.816 And correlated and. 00:05:20.216 --> 00:05:24.026 Mashed together into this system that can then. 00:05:24.426 --> 00:05:31.716 Correlate descriptions with those metadata about the images and spit out 00:05:31.716 --> 00:05:34.476 something that fits those correlations. 00:05:34.816 --> 00:05:41.456 So if you fed a lot of images by van Gogh into one of these systems, and 00:05:41.456 --> 00:05:46.316 they're all described as van Gogh and they might say oil painting as 00:05:46.316 --> 00:05:48.146 well, somewhere in the description. 00:05:48.516 --> 00:05:53.046 There might be descriptions of the types of colors and the brush strokes 00:05:53.046 --> 00:05:57.816 and all of those types of things that really define a van Gogh painting. 00:05:57.816 --> 00:06:01.746 During training, basically what happens is the systems learn to 00:06:01.746 --> 00:06:03.486 recognize those similarities. 00:06:03.516 --> 00:06:09.786 Oh, we see things about the quality of the brushstrokes and oil paint and van Gogh 00:06:10.356 --> 00:06:13.086 and it correlates that with those loose. 00:06:13.086 --> 00:06:17.466 With that style of painting with those with the way that the pixels 00:06:17.466 --> 00:06:21.506 are basically grouped next to each other, the contrasts the colors. 00:06:21.906 --> 00:06:26.646 The patterns, those types of things, and then it can apply that to something else. 00:06:26.646 --> 00:06:30.666 So if you have many images of dogs, when those are being fed in and it says. 00:06:31.346 --> 00:06:38.316 Dachshund or Husky or golden retriever and the shapes of the dogs are parsed 00:06:38.316 --> 00:06:41.676 and stored as metadata in the system. 00:06:42.076 --> 00:06:45.556 They're not storing the images of the dogs, but they're storing the 00:06:45.556 --> 00:06:51.226 sort of rough shapes and ideas of how a dog is, or those dog breeds are. 00:06:51.556 --> 00:06:53.176 And then you can match those together. 00:06:53.236 --> 00:06:56.926 And so if you have the shape of a dog, but the texture of a Vango. 00:06:57.196 --> 00:07:01.486 You get a painting of a golden retriever in the style of van Gogh. 00:07:01.886 --> 00:07:06.926 But of course there's limits to how we describe images and paintings and. 00:07:07.466 --> 00:07:08.366 All of these things. 00:07:08.366 --> 00:07:10.136 And so you'll find when you play these. 00:07:10.466 --> 00:07:13.976 So you'll find when you play with these tools that they come up with really 00:07:13.976 --> 00:07:17.576 weird stuff, and you can see that they don't really understand, and you can 00:07:17.576 --> 00:07:19.916 also see that they're clearly not copies. 00:07:20.396 --> 00:07:22.346 Because things come out weirdly. 00:07:22.346 --> 00:07:23.876 They don't make sense. 00:07:24.276 --> 00:07:29.076 And the tools are getting a lot better at that, but think of that as the way these 00:07:29.076 --> 00:07:31.956 systems work, they understand patterns. 00:07:31.956 --> 00:07:33.036 They understand. 00:07:33.436 --> 00:07:38.476 Metadata about these images, they bring in descriptions and keywords 00:07:38.476 --> 00:07:40.696 and they mash those things together to 00:07:41.096 --> 00:07:46.266 to deconstruct all of the training data and mash it into correlational machine. 00:07:46.666 --> 00:07:51.466 And then when you provide a prompt, It goes from that sort 00:07:51.466 --> 00:07:53.386 of loosey goosey something. 00:07:53.786 --> 00:07:58.376 Building backup on those correlations and spits out something that 00:07:58.436 --> 00:08:00.026 may or may not make sense. 00:08:00.446 --> 00:08:03.986 Of course, more and more, they are making sense and they're doing things that. 00:08:04.386 --> 00:08:06.366 They're generating outputs. 00:08:06.366 --> 00:08:07.836 That look really good. 00:08:08.236 --> 00:08:12.706 So do without what you will, maybe that helps you understand a bit more 00:08:12.706 --> 00:08:16.606 of the context for the controversies, or it may give you some context 00:08:16.606 --> 00:08:18.346 for how to work with these things. 00:08:18.746 --> 00:08:19.946 But hopefully it helps. 00:08:20.346 --> 00:08:22.896 The other thing is large language models. 00:08:23.296 --> 00:08:27.776 You can think of it very similarly to basically these systems. 00:08:28.316 --> 00:08:29.456 During training. 00:08:29.786 --> 00:08:30.386 Bring in. 00:08:30.896 --> 00:08:35.816 Tons and tons of data about all kinds of written texts from throughout history. 00:08:36.216 --> 00:08:40.026 Again, controversies around how those data are sourced. 00:08:40.296 --> 00:08:41.856 We're not going to dive into that. 00:08:42.336 --> 00:08:49.536 But basically you have massive data sets of written language in various languages. 00:08:49.936 --> 00:08:53.326 And it is really under the hood. 00:08:53.326 --> 00:08:55.216 Just fancy auto-complete. 00:08:55.916 --> 00:08:59.006 That is basically what these systems do. 00:08:59.366 --> 00:09:05.876 But on a really large scale when you're writing an essay. 00:09:06.326 --> 00:09:10.136 A paragraph your writing is based off of the previous paragraph you 00:09:10.136 --> 00:09:15.546 wrote and the previous chapter that you wrote or the research 00:09:15.546 --> 00:09:17.646 that you have put into that paper. 00:09:17.676 --> 00:09:22.926 Maybe you could imagine as being before that further up the chain, 00:09:22.956 --> 00:09:26.586 if you had all of your research and your writing in one document. 00:09:26.976 --> 00:09:28.236 You could think of it that way. 00:09:28.266 --> 00:09:33.066 It's just you have all of these inputs and then your next paragraph 00:09:33.096 --> 00:09:38.856 is predictable in a sense, based off of the inputs and the previous writing 00:09:38.856 --> 00:09:41.076 that you've done, what comes next? 00:09:41.736 --> 00:09:44.736 Essentially, that's what these systems are trained to do. 00:09:44.796 --> 00:09:47.766 They recognize patterns in language. 00:09:48.066 --> 00:09:51.846 And so they know that if you have a paragraph about. 00:09:52.246 --> 00:09:53.806 How playful dogs are. 00:09:54.206 --> 00:09:56.756 And then maybe your next paragraph is going to be about 00:09:56.756 --> 00:09:58.946 the joy that they bring to their. 00:09:59.276 --> 00:09:59.996 Humans. 00:10:00.446 --> 00:10:01.976 Something along those lines. 00:10:02.086 --> 00:10:05.656 And that would depend on the rest of the context of what you're writing. 00:10:06.056 --> 00:10:07.886 But they're not coming up with. 00:10:08.286 --> 00:10:10.446 They're not coming up with anything novel. 00:10:10.476 --> 00:10:12.696 There is some randomness in this system. 00:10:13.096 --> 00:10:16.226 That is intentionally placed there. 00:10:16.626 --> 00:10:18.606 But it is really just auto-complete. 00:10:18.636 --> 00:10:23.016 And even when you do the chat interfaces, Functionally, what they're 00:10:23.016 --> 00:10:25.956 doing is still doing auto-complete. 00:10:26.406 --> 00:10:30.846 If I have a conversation with someone and I say, hi, how are you? 00:10:31.176 --> 00:10:32.226 Predictably? 00:10:32.496 --> 00:10:36.306 The response will be something about how the person I'm talking to is 00:10:36.306 --> 00:10:39.036 doing, and probably asking me back. 00:10:39.726 --> 00:10:40.956 Hey, how are you doing? 00:10:41.356 --> 00:10:44.026 I'll respond and we carry on with the conversation. 00:10:44.426 --> 00:10:48.656 They're encoded in the language that we've left strewn around the internet 00:10:48.686 --> 00:10:51.566 and through books and other media. 00:10:51.966 --> 00:10:56.136 So these systems find those correlations, they understand the patterns 00:10:56.136 --> 00:10:58.116 they're trained on those patterns. 00:10:58.146 --> 00:11:01.746 If there's one word in front, then what's the next word that comes after. 00:11:02.146 --> 00:11:05.476 And fundamentally, they just do long chains of that. 00:11:05.926 --> 00:11:07.666 Based off of the training data. 00:11:08.066 --> 00:11:12.356 So when people talk about these hallucinations, for example, 00:11:12.756 --> 00:11:16.386 I think it's really important to understand that the outputs of these 00:11:16.386 --> 00:11:19.056 systems are just word predictions. 00:11:19.086 --> 00:11:20.706 They're not knowledge predictions. 00:11:20.706 --> 00:11:22.386 They're not logical predictions. 00:11:22.416 --> 00:11:23.826 They don't understand. 00:11:24.306 --> 00:11:26.046 What's going into them really? 00:11:26.446 --> 00:11:29.806 You could have philosophical debates about whether we understand 00:11:29.806 --> 00:11:31.696 what's going into our brains. 00:11:31.696 --> 00:11:33.916 As we hear words and we parse them. 00:11:34.316 --> 00:11:37.946 I'm not going to go into that, but the mechanics of it is here 00:11:37.946 --> 00:11:41.636 are some words, what are the next logical words that should follow? 00:11:42.036 --> 00:11:47.886 And that also means that these systems don't have real-time access to data. 00:11:48.286 --> 00:11:49.576 Now there's a caveat there. 00:11:49.636 --> 00:11:51.316 And we're going to talk about that in a second. 00:11:51.946 --> 00:11:56.626 But if you're just chatting with something like ChatGPT or Claude or any of the other 00:11:56.626 --> 00:12:01.766 ones, And you're not using a version of them that is connected to the internet. 00:12:02.366 --> 00:12:06.486 Their data that they're trained on is potentially old. 00:12:06.886 --> 00:12:08.506 And it doesn't have context. 00:12:08.906 --> 00:12:12.266 Now there are techniques that you can use and that are being used 00:12:12.266 --> 00:12:17.006 by all kinds of systems, including search systems that use AI's. 00:12:17.126 --> 00:12:21.446 So if you're thinking about using Bing chat or if you're 00:12:21.446 --> 00:12:23.226 on a paid tier of ChatGPT. 00:12:23.246 --> 00:12:26.096 You can also ask it to make sure that when it responds to 00:12:26.096 --> 00:12:27.626 you, sources from the internet. 00:12:28.256 --> 00:12:31.436 Or maybe you're using a search system like perplexity. 00:12:32.136 --> 00:12:35.856 All of those systems, use a technique called RAG retrieval, 00:12:35.856 --> 00:12:37.356 augmented generation. 00:12:37.756 --> 00:12:43.866 Now, what that does is it uses I'm going to say algorithms, basically methods to 00:12:43.866 --> 00:12:49.056 figure out what is most relevant based off of your query in what fundamentally, 00:12:49.056 --> 00:12:50.946 it looks like a traditional search index. 00:12:51.276 --> 00:12:53.466 If you just type into Google. 00:12:54.066 --> 00:12:55.926 Recipes for apple pie. 00:12:56.586 --> 00:13:00.096 It will go find things and it'll give you a list of results. 00:13:00.576 --> 00:13:00.936 Now what? 00:13:00.936 --> 00:13:04.746 Bing chat and ChatGPT, one connected to the web on the paid 00:13:04.746 --> 00:13:10.386 tier or perplexity do is the kind of just do a normal web search. 00:13:10.786 --> 00:13:16.096 And then they add that text into the frame, the context of your question. 00:13:16.096 --> 00:13:17.686 So when they respond. 00:13:18.166 --> 00:13:19.996 It might not be there literally. 00:13:20.396 --> 00:13:23.006 But under the hood what's happening is they're pulling 00:13:23.216 --> 00:13:25.436 paragraphs from those sources. 00:13:26.156 --> 00:13:28.016 And saying, okay. 00:13:28.286 --> 00:13:29.336 If you asked. 00:13:29.996 --> 00:13:31.436 How do I make apple pie? 00:13:32.186 --> 00:13:34.016 Let's go fetch these recipes. 00:13:34.416 --> 00:13:35.976 Add them into the context. 00:13:36.576 --> 00:13:40.116 In a way that's hidden and then respond with that context. 00:13:40.536 --> 00:13:46.386 So now when the fancy auto-complete keeps going, it has both the context of 00:13:46.416 --> 00:13:51.786 the question, but also those paragraphs about how to bake an apple by and so 00:13:51.786 --> 00:13:54.066 when it responds, it will give you stuff. 00:13:54.066 --> 00:13:55.296 That's a lot more accurate. 00:13:55.536 --> 00:13:59.456 Now you can also get fancy with your prompting and basically tell it. 00:13:59.536 --> 00:14:05.526 Do not use anything other than valid sources and then it should understand 00:14:05.526 --> 00:14:10.176 from the context that whatever comes back from a search is the valid 00:14:10.176 --> 00:14:11.526 source that you're looking for. 00:14:11.796 --> 00:14:15.216 You can also tell it only return results with. 00:14:15.616 --> 00:14:17.446 URLs that you can visit. 00:14:18.106 --> 00:14:22.036 So you can then go validate those sources yourself, but you can imagine 00:14:22.036 --> 00:14:25.336 how you can integrate those types of techniques into your own products. 00:14:25.636 --> 00:14:30.466 Let's say you've got a note taking app with a database of notes. 00:14:30.866 --> 00:14:33.416 You need a system to parse a user's input. 00:14:33.446 --> 00:14:34.406 Who's asking. 00:14:34.806 --> 00:14:36.576 What am I doing on a Thursday? 00:14:37.176 --> 00:14:42.186 So you might take that input, convert that into a more standard search algorithm. 00:14:42.216 --> 00:14:44.226 You say what date is Thursday? 00:14:44.826 --> 00:14:46.116 Filter out notes. 00:14:46.206 --> 00:14:47.886 Grab the text from those notes. 00:14:48.426 --> 00:14:52.656 And then tell the large language model that you're using. 00:14:52.986 --> 00:14:54.996 Hey, this person wants to know what they're doing. 00:14:54.996 --> 00:14:59.106 Here are some notes that are relevant to this coming Thursday. 00:14:59.886 --> 00:15:04.086 Respond to them only with that context, do not include anything else. 00:15:04.836 --> 00:15:07.326 And now you've returned that response to the user. 00:15:07.726 --> 00:15:11.746 So those are some ways to like, think about what's happening in large 00:15:11.746 --> 00:15:16.156 language models, how they can be used appropriately to reduce hallucinations. 00:15:16.546 --> 00:15:20.266 What is really happening when you interact with one of these models. 00:15:20.266 --> 00:15:24.016 And how they may or may not make sense in a product. 00:15:24.406 --> 00:15:30.226 Now there's so much that we're going to figure out as a society 00:15:30.256 --> 00:15:34.246 over the coming years about how training data should be sourced. 00:15:34.646 --> 00:15:36.686 How the system should be used. 00:15:37.086 --> 00:15:38.586 How they should be regulated. 00:15:38.986 --> 00:15:42.016 But I think personally that there's something really interesting and 00:15:42.016 --> 00:15:47.776 exciting about the way that these systems work, particularly large language 00:15:47.806 --> 00:15:52.726 models, because they fundamentally become a different way of interacting 00:15:52.756 --> 00:15:54.826 with information and knowledge. 00:15:55.226 --> 00:16:01.856 The way I see it is as long as they're paired with search or search like systems. 00:16:02.066 --> 00:16:06.296 They become a new interface for the data that we store, the stories that 00:16:06.296 --> 00:16:09.996 we have the knowledge that we share. 00:16:10.396 --> 00:16:14.176 There are valid debates to be had about how the training data should 00:16:14.176 --> 00:16:19.246 be sourced and how we should oversee the training of these systems. 00:16:19.646 --> 00:16:23.726 I think it's undeniable that there's value in being able to parse language 00:16:23.756 --> 00:16:26.666 better than we have been using. 00:16:27.206 --> 00:16:32.826 Computers, which fundamentally have not been able to parse language well, So far. 00:16:33.226 --> 00:16:35.836 And I think it just opens up so many possibilities for 00:16:35.836 --> 00:16:37.846 things that we can create. 00:16:38.246 --> 00:16:43.706 Because so much of what we are as a modern society is encoded in language 00:16:43.736 --> 00:16:49.346 and language that is stored in ways that can be used to build these systems. 00:16:49.796 --> 00:16:51.776 Yeah, I think we need to be careful. 00:16:51.836 --> 00:16:56.246 I think there does need to be regulation primarily about the outputs. 00:16:56.396 --> 00:16:58.166 I'm less worried about the inputs. 00:16:58.566 --> 00:17:03.906 And I think there need to be best practices in place for practitioners 00:17:03.906 --> 00:17:06.066 who want to use these systems. 00:17:06.066 --> 00:17:07.626 Within their own products. 00:17:07.626 --> 00:17:11.376 But hopefully this episode helped you understand what's going on under the hood. 00:17:11.406 --> 00:17:15.636 What are the possibilities, how you might want to engage with these tools? 00:17:16.036 --> 00:17:17.686 And yeah, I hope it was informative. 00:17:18.086 --> 00:17:18.596 Alrighty. 00:17:18.626 --> 00:17:20.036 Thanks for listening folks. 00:17:20.436 --> 00:17:23.256 If you enjoy this stuff, please subscribe on YouTube. 00:17:23.656 --> 00:17:26.776 And subscribe to the podcast in your podcast, app of choice. 00:17:26.806 --> 00:17:29.866 Also leave us a rating and review really helps us out. 00:17:30.266 --> 00:17:31.646 We'd love to hear what you think. 00:17:31.676 --> 00:17:34.706 And I'd really personally love to have you on the podcast. 00:17:34.706 --> 00:17:37.166 I'd love to talk to other people about this stuff. 00:17:37.566 --> 00:17:40.806 Also make sure to sign up for our newsletter where we'll be sending 00:17:40.806 --> 00:17:44.826 you all kinds of great info about how to build a small tech product. 00:17:45.276 --> 00:17:49.626 There's going to be videos, blog posts, episodes of the podcast you may have 00:17:49.626 --> 00:17:51.066 missed and plenty of other stuff. 00:17:51.066 --> 00:17:54.906 So head to small tech podcast.com and subscribe there. 00:17:55.306 --> 00:18:00.046 So that's it for this week's episode and we all want to do good in the world folks. 00:18:00.076 --> 00:18:02.776 So go out there and build something. 00:18:02.776 --> 00:18:03.256 Good. 00:18:03.526 --> 00:18:04.126 See ya.