WEBVTT NOTE This file was generated by Descript 00:00:04.683 --> 00:00:05.793 Colin: Welcome to Build and Learn. 00:00:05.793 --> 00:00:06.693 My name is Colin. 00:00:06.693 --> 00:00:10.263 CJ: And m cj, and today we're talking about large language models 00:00:10.263 --> 00:00:14.673 and all this AI stuff that has happened in the last four months and. 00:00:15.183 --> 00:00:19.773 We were just talking about how the episode we just released, we recorded 00:00:19.803 --> 00:00:24.203 four months ago, and so it'll be like entering a time warp and we're 00:00:24.203 --> 00:00:28.288 gonna just fast forward four months and an incredible amount has changed, 00:00:28.468 --> 00:00:31.348 Colin: Yeah, little how the sausage is made. 00:00:31.348 --> 00:00:34.743 I was doing the editing on that last episode and, I was listening to it. 00:00:34.743 --> 00:00:36.033 I'm like, oh my goodness. 00:00:36.033 --> 00:00:38.523 I know this was one of our pre-canned episodes that we 00:00:38.528 --> 00:00:40.493 recorded so that we had some buffer. 00:00:40.498 --> 00:00:42.713 And I'm just listening to, this was in November. 00:00:43.433 --> 00:00:48.883 Things like Twitter getting bought by Elon and ChatGPT came out and we're 00:00:48.883 --> 00:00:50.773 like, yeah, we use it to summarize stuff. 00:00:50.773 --> 00:00:55.463 So what it's, it's cool, but so much has happened in four months and I think 00:00:55.468 --> 00:01:00.508 there's a big conversation going on about whether or not this replaces jobs 00:01:00.513 --> 00:01:03.508 or does it just increase productivity? 00:01:03.508 --> 00:01:07.868 And, we've seen these similar cycles, back when the calculator came out and 00:01:07.868 --> 00:01:09.938 the computer and the digital camera. 00:01:10.838 --> 00:01:13.958 And so I think having a chat today about how we use these 00:01:13.958 --> 00:01:16.588 tools, will be pretty fun to do. 00:01:16.678 --> 00:01:17.318 We can dive in. 00:01:18.183 --> 00:01:18.423 CJ: Totally. 00:01:18.593 --> 00:01:25.013 GPT-3 was pretty good at generating text, but it is nowhere near as good as 00:01:25.013 --> 00:01:31.633 GPT-4 and I think when we recorded, maybe chat, GPT hadn't been released or, I 00:01:31.633 --> 00:01:33.253 can't remember if it had or hadn't, but 00:01:33.388 --> 00:01:34.888 Colin: was like the release 00:01:35.013 --> 00:01:35.503 CJ: Okay. 00:01:35.548 --> 00:01:35.878 Colin: November. 00:01:36.728 --> 00:01:36.973 CJ: Got it. 00:01:37.003 --> 00:01:37.393 Got it. 00:01:37.883 --> 00:01:42.623 so yeah, I think the leap between three and four has just been like mind 00:01:42.623 --> 00:01:44.843 blowing to everyone who touches it. 00:01:45.413 --> 00:01:48.268 and yeah, I think also like when you asked. 00:01:49.058 --> 00:01:50.508 GPT-3 things. 00:01:50.958 --> 00:01:53.778 Sometimes the content you got back was like, okay, whatever. 00:01:53.778 --> 00:01:56.588 Maybe that was like a, a response that you would expect from a 00:01:56.588 --> 00:01:58.118 third grader, fourth grader. 00:01:58.718 --> 00:02:02.618 And the stuff that you get back from GPT-4, I feel like is, 00:02:03.018 --> 00:02:08.708 senior in high school with a professor editing the responses. 00:02:08.718 --> 00:02:11.468 And it just makes it so much more useful. 00:02:13.098 --> 00:02:15.588 Colin: And the difference between them being that four has been 00:02:15.593 --> 00:02:18.438 trained on more data, right? 00:02:18.438 --> 00:02:19.878 More language, more data. 00:02:19.878 --> 00:02:26.213 And so you can see what happens when you add more layers and More just data sets to 00:02:26.213 --> 00:02:28.073 it, that you get a more complete answer. 00:02:28.443 --> 00:02:34.303 we'll leave a link to, the Wolf from Alpha website about how ChatGPT works under the 00:02:34.303 --> 00:02:38.323 hood, and it's pretty amazing that it's literally like statistically detecting 00:02:38.713 --> 00:02:40.993 the next most probabilistic word. 00:02:41.473 --> 00:02:43.453 And so it's just guessing word by word. 00:02:43.453 --> 00:02:47.893 And when you have more words to pull from and more examples of those words 00:02:47.893 --> 00:02:52.398 in sentences, Papers and things like that, it gets really interesting. 00:02:52.398 --> 00:02:56.268 The thing that I'm most excited to see is what happens as this gets trained on 00:02:56.268 --> 00:03:02.968 proprietary data sets or like specific verticals like legal docs, constitution, 00:03:02.973 --> 00:03:06.388 all these like different things like court precedents and stuff like that. 00:03:07.023 --> 00:03:10.563 CJ: I think companies are starting to just think about how they can use it to 00:03:10.563 --> 00:03:12.393 do weird things that you wouldn't expect. 00:03:12.393 --> 00:03:15.933 It's not just generating texts, but it's, yeah, making informed 00:03:15.933 --> 00:03:17.193 decisions about things and. 00:03:18.068 --> 00:03:18.479 I don't know. 00:03:18.484 --> 00:03:21.983 It's crazy, but I'm, it's, I'm excited to hear how you're using it cuz every 00:03:21.983 --> 00:03:27.563 time I talk to someone who is in tech, who is using it for their job 00:03:27.563 --> 00:03:30.228 or just using it to improve their life, I pick up something, I'm like, 00:03:30.228 --> 00:03:31.488 okay, I'm gonna do that for sure. 00:03:31.938 --> 00:03:32.638 So yeah. 00:03:32.693 --> 00:03:34.673 what maybe what is like a recent thing that you're like, 00:03:34.703 --> 00:03:36.503 okay, this is useful and. 00:03:38.088 --> 00:03:41.958 Colin: Yeah, so I've mostly been using it for brainstorming 00:03:41.958 --> 00:03:43.398 and kind of riffing on ideas. 00:03:43.448 --> 00:03:47.068 if I have, I don't know, like I think when you name something, you 00:03:47.073 --> 00:03:49.968 always have to go check a bunch of domains to see if they're available. 00:03:50.238 --> 00:03:53.118 I don't think ChatGPT can go check like whether or. 00:03:54.253 --> 00:03:57.823 something's available for domain registration yet, but that should 00:03:57.828 --> 00:03:59.383 be a thing it should be able to do. 00:03:59.903 --> 00:04:01.283 we need a plugin for that. 00:04:01.613 --> 00:04:04.223 But if you're coming up with a name for a project, you're always like looking 00:04:04.223 --> 00:04:08.413 for what's a good name, what's available as it as it.com, dot dev, whatever. 00:04:09.058 --> 00:04:12.298 and then you, there just might be words like, I'll go to the thesaurus 00:04:12.298 --> 00:04:14.798 or dictionary, but now I can just be like, all right, ChatGPT. 00:04:14.818 --> 00:04:17.518 Let's have a little back and forth whiteboard session 00:04:17.518 --> 00:04:19.228 on this is the app idea. 00:04:19.288 --> 00:04:21.298 These are a bunch of root words. 00:04:21.328 --> 00:04:23.068 Give me a bunch of tangential things. 00:04:23.508 --> 00:04:26.208 and it comes back with some stuff that you just wouldn't have found. 00:04:26.678 --> 00:04:33.488 In Code I, I use GitHub co-pilot, so I've that's become a little too normal for me. 00:04:34.208 --> 00:04:36.938 And I'm really excited to see what comes in the new, like 00:04:37.098 --> 00:04:38.598 when it's backed by a four. 00:04:38.898 --> 00:04:41.148 Cuz I think right now it's just three, 3.5. 00:04:41.668 --> 00:04:45.538 but the most specific thing that I was blown away by was in React 00:04:45.543 --> 00:04:47.818 when I was trying to do something. 00:04:47.998 --> 00:04:50.643 It was like a component with a bunch of child components. 00:04:50.973 --> 00:04:52.443 And I write in Ruby most days. 00:04:52.443 --> 00:04:53.823 So I'm not just like I don. 00:04:54.248 --> 00:04:57.698 Know all the ways that you would use react off the top of my head. 00:04:58.208 --> 00:05:00.278 And so I just explained what I was trying to do. 00:05:00.283 --> 00:05:03.128 I was like, I want that thing to know about all these other things. 00:05:03.518 --> 00:05:07.268 And it gave me such a good example where I was like, okay, that makes sense. 00:05:07.358 --> 00:05:08.378 Why didn't I think of that? 00:05:08.918 --> 00:05:11.528 And I was like, you know what, we're just gonna do something very similar to that. 00:05:11.528 --> 00:05:16.319 And it was like one of those Spotify style, Spotify wrapped, we generate a 00:05:16.324 --> 00:05:19.708 bunch of images and we want to be able to download those images as a zip file. 00:05:20.218 --> 00:05:24.598 And so I needed a reference to all of the images in order to loop over them and 00:05:24.598 --> 00:05:26.308 download them and put them into a zip. 00:05:26.818 --> 00:05:29.338 And it was able to do a lot of that for me. 00:05:29.338 --> 00:05:30.928 So it was pretty, pretty cool. 00:05:31.273 --> 00:05:32.143 CJ: That's super cool. 00:05:32.383 --> 00:05:35.563 Brainstorming ideas has definitely been a huge part for me too. 00:05:35.633 --> 00:05:41.273 one, a project I was tasked with recently was like, help prepare for a presentation 00:05:41.278 --> 00:05:47.063 that's gonna happen at a conference and tell a story that fits in lots of 00:05:47.063 --> 00:05:48.743 different Stripe products into the story. 00:05:49.253 --> 00:05:52.883 And so I was able to just give it a bunch of context. 00:05:53.813 --> 00:05:58.893 The presentation and say, it should highlight X, Y, and Z features, 00:05:58.893 --> 00:06:03.543 create a company that will illustrate how you can use all these different 00:06:03.543 --> 00:06:04.623 features to do these things. 00:06:05.043 --> 00:06:10.453 And it spit out these story arcs with company names and, character names 00:06:10.513 --> 00:06:14.173 of the, the employees and the CEO and how they came up with the story. 00:06:14.173 --> 00:06:17.233 And like all of this stuff that you're able to really just like 00:06:17.233 --> 00:06:19.513 flesh out into this beautiful story. 00:06:19.543 --> 00:06:21.223 And you could even tell it, or I was. 00:06:21.778 --> 00:06:24.118 I don't know what the common story arcs are. 00:06:24.118 --> 00:06:26.628 So it start out by saying what are the five common story arcs? 00:06:26.628 --> 00:06:28.878 And then say okay, what is a good common, what is a good 00:06:28.878 --> 00:06:30.258 story arc for a presentation? 00:06:30.618 --> 00:06:34.488 And then it tells you, one, you're like, okay, write a story about x where it 00:06:34.488 --> 00:06:36.888 includes these different characteristics. 00:06:36.888 --> 00:06:39.493 And it just, yeah, it helped so much. 00:06:39.493 --> 00:06:41.233 The brainstorming piece is really cool. 00:06:42.803 --> 00:06:46.883 When approaching a blog post that I need to write, I might come up with an 00:06:46.883 --> 00:06:49.868 outline that has three or four steps and. 00:06:50.853 --> 00:06:52.638 I'll say okay, this is what I think I'm gonna write for the 00:06:52.638 --> 00:06:55.478 outline, and then I'll just go ask ChatGPT, write the same outline. 00:06:55.628 --> 00:07:00.488 And it will sometimes think of other steps or like it will have different 00:07:00.488 --> 00:07:04.058 arguments for or against something or different pros and cons that I hadn't 00:07:04.058 --> 00:07:07.208 thought of that you might come up with if you're brainstorming with a group 00:07:07.208 --> 00:07:12.633 of 5, 6, 7 people with lots of diverse experience, from different areas and like 00:07:12.633 --> 00:07:14.463 really interesting diverse perspectives. 00:07:14.463 --> 00:07:17.933 And Yeah, it definitely helps fill in the gaps as just a single 00:07:17.933 --> 00:07:19.353 person who's creating content, 00:07:19.563 --> 00:07:23.808 Colin: Yeah, and I think that's the thing I keep hearing is, Especially 00:07:23.808 --> 00:07:28.128 now that companies are trying to do more with less and things like that. 00:07:28.128 --> 00:07:28.788 Is that it? 00:07:28.788 --> 00:07:30.043 It's the Ironman suit. 00:07:30.573 --> 00:07:34.593 that you're, that I think GitHub co-pilot was designed and why they 00:07:34.593 --> 00:07:38.283 called it GitHub co-pilot was that they still want the human in the seat. 00:07:38.353 --> 00:07:40.543 And that, that you're a director of sorts. 00:07:40.543 --> 00:07:44.283 So you're producing what the story is you're doing the Hero's journey 00:07:44.283 --> 00:07:45.543 of the Stripe Conference, right? 00:07:45.643 --> 00:07:46.273 what does that look. 00:07:47.098 --> 00:07:50.518 And, that you're giving it inputs and it's able to go look at all those kinds 00:07:50.518 --> 00:07:53.398 of things without replacing a person. 00:07:53.398 --> 00:07:56.218 And I think there's a lot of people, I think the media has mostly picked 00:07:56.218 --> 00:07:59.238 up on this of I think my mom called me and was like, oh, aren't you 00:07:59.238 --> 00:08:02.188 worried you're gonna get, your job's gonna get replaced by ChatGPT? 00:08:02.208 --> 00:08:04.753 And it's whew, we got a long way to go, thankfully. 00:08:04.853 --> 00:08:08.763 CJ: I think the fact that it can generate code is pretty, impressive. 00:08:09.958 --> 00:08:14.373 And one trend that I've noticed that's really interesting, if you go in discord, 00:08:14.373 --> 00:08:18.713 in the Stripe discord in particular, people will drop in there and say I don't 00:08:18.713 --> 00:08:20.273 understand why my code isn't working. 00:08:20.783 --> 00:08:23.933 Can you help me like figure out how to integrate this API or whatever. 00:08:23.933 --> 00:08:25.743 And then we'll say okay, can you show us your code? 00:08:26.163 --> 00:08:26.943 And they'll show it to us. 00:08:26.943 --> 00:08:28.568 And we're like, That is super weird. 00:08:28.568 --> 00:08:29.738 Like, where did you find this? 00:08:29.738 --> 00:08:31.238 That is like not correct. 00:08:31.268 --> 00:08:34.538 or it's like odd in a certain weird way, and they're like, oh yeah. 00:08:34.538 --> 00:08:37.958 I was just like using GPT like chat, GPT to write all of this. 00:08:38.048 --> 00:08:39.278 We're like, oh my God. 00:08:39.328 --> 00:08:43.828 like a lot of people are already just using it to write a ton 00:08:43.833 --> 00:08:46.228 of code and they don't fully understand all the details, but. 00:08:47.753 --> 00:08:51.023 That's like proof though that I, that you still need humans involved 00:08:51.083 --> 00:08:55.613 to check, check to make sure that it's behaving as you expect, 00:08:55.713 --> 00:09:00.153 Colin: Yeah, I am excited to see what the new GitHub co-pilot X stuff does, 00:09:00.153 --> 00:09:05.053 because it seems to take that, like the idea that we could point ChatGPT 00:09:05.073 --> 00:09:07.143 at the Stripe Docs intentionally. 00:09:07.713 --> 00:09:12.243 And say, answer these things for us means that it's not making things 00:09:12.243 --> 00:09:13.833 up from the rest of the internet. 00:09:13.833 --> 00:09:16.953 It's making it up based on what's in the docs. 00:09:17.033 --> 00:09:20.453 and it has context like even now, like when you talk, when you go 00:09:20.453 --> 00:09:24.353 to the ChatGPT website, it doesn't have the context of your code unless 00:09:24.353 --> 00:09:25.583 you copy and paste it in there. 00:09:25.583 --> 00:09:30.113 And so being able to highlight something and say, explain this specific thing, 00:09:30.113 --> 00:09:34.843 or, Iterate over all of these things and add them up together, things 00:09:34.843 --> 00:09:36.823 like that becomes really powerful. 00:09:37.193 --> 00:09:39.113 but you still need to know what you want to do. 00:09:39.143 --> 00:09:41.303 Like that part is still important. 00:09:41.303 --> 00:09:47.513 It's not go build 10, dog walking apps in 10 different languages. 00:09:48.693 --> 00:09:52.073 Because there's gonna be a whole lot of inputs that, I'd be curious to see what 00:09:52.073 --> 00:09:54.633 those apps do, generate from scratch. 00:09:54.663 --> 00:09:56.613 It's okay, do you assume user login? 00:09:56.613 --> 00:09:57.963 Do you know what a dog is? 00:09:57.963 --> 00:09:58.233 Do you 00:09:59.648 --> 00:10:03.123 CJ: Yeah, there's still like leaps that it needs to make in understanding 00:10:03.153 --> 00:10:06.243 in order to connect dots between different systems and things like that. 00:10:06.753 --> 00:10:10.323 one of the things I was using it for early on was helping to write video 00:10:10.328 --> 00:10:14.463 scripts for videos where we're teaching people how to use stripe basically. 00:10:14.653 --> 00:10:18.703 My normal process was start from the Stripe documentation and then go 00:10:18.703 --> 00:10:22.633 through, build out a demo, and then walk through and write out by hand what 00:10:22.633 --> 00:10:24.853 I, or like roughly what I plan to say. 00:10:25.593 --> 00:10:28.233 Now I can just take the whole thing, give it to ChatGPT and say 00:10:28.233 --> 00:10:31.683 write me a video script that helps teach this in X, y, and Z ways. 00:10:32.193 --> 00:10:36.613 But I just saw this really mind blowing demo by Siraj Raval on YouTube. 00:10:36.613 --> 00:10:37.603 So we'll link to it. 00:10:37.603 --> 00:10:43.243 But he basically built an entire YouTube channel with an AI influencer 00:10:43.633 --> 00:10:50.623 and so it like will go on Twitter and find trending AI books, then it will. 00:10:51.453 --> 00:10:55.443 write a script that will pitch the book. 00:10:55.833 --> 00:10:58.233 Then it will like generate video. 00:10:59.133 --> 00:11:04.608 Of a fake person saying the script, and it will u it'll generate audio. 00:11:04.608 --> 00:11:08.148 It like generates their voice, it generates animations of 00:11:08.148 --> 00:11:09.738 the concepts in the video. 00:11:10.068 --> 00:11:12.353 And, all of this isn't just like a one. 00:11:12.623 --> 00:11:16.393 He couldn't just say ChatGPT make me a automated YouTube channel. 00:11:16.453 --> 00:11:20.563 Instead, he needed to know okay, here are the like 25 tools 00:11:20.653 --> 00:11:22.993 throughout the video, he uses like tons and tons of different tools. 00:11:23.643 --> 00:11:26.833 And Python libraries and, APIs. 00:11:26.833 --> 00:11:30.133 So he is using like the shot stack API to compose all these different pieces 00:11:30.133 --> 00:11:31.513 so that they show up the right way. 00:11:31.933 --> 00:11:34.433 But at the end of the day, he was able to create these. 00:11:35.673 --> 00:11:40.803 These, videos that are being automatically published to YouTube and are based on 00:11:40.803 --> 00:11:43.593 trending books that are on Twitter. 00:11:43.593 --> 00:11:46.353 And the idea was like, okay, now I want to take my affiliate 00:11:46.353 --> 00:11:48.123 link for those books on Amazon. 00:11:48.153 --> 00:11:50.763 Put them in the description so that you can like maybe 00:11:50.763 --> 00:11:52.323 build up this passive thing. 00:11:52.653 --> 00:11:56.533 So from my little idea of help me write a script for a video. 00:11:57.253 --> 00:12:01.573 It's like people are able to use these tools in different ways to combine 00:12:01.573 --> 00:12:05.763 them and build like these really wild automations and systems and, yeah, I 00:12:05.763 --> 00:12:09.093 don't know, like basically like the, I feel like the way that I'm using it 00:12:09.093 --> 00:12:13.618 is still very much just scratching the surface of what you can do with even 00:12:13.618 --> 00:12:14.968 just the stuff that's available today. 00:12:15.068 --> 00:12:17.708 Colin: I'll say, I haven't seen the one that you just talked about, but 00:12:17.708 --> 00:12:20.618 I will say that the difference it sounds is that you use the tool like 00:12:20.618 --> 00:12:25.268 a research assistant, and then you're still the person who's doing the video. 00:12:25.268 --> 00:12:31.178 And I have to wonder, we talk about this a lot with our show, like people 00:12:31.183 --> 00:12:35.918 wanna follow people and I feel like computers and AI are not gonna make this. 00:12:36.398 --> 00:12:38.348 It's actually gonna make it more valuable. 00:12:38.353 --> 00:12:39.998 Real conversations by real people. 00:12:40.388 --> 00:12:40.898 Do I. 00:12:41.723 --> 00:12:46.193 And trust a book review from an AI robot that's just meant 00:12:46.193 --> 00:12:48.443 to be an affiliate link farm. 00:12:48.893 --> 00:12:50.333 Or do I want to hear what c? 00:12:50.933 --> 00:12:51.803 Thanks about this book. 00:12:51.803 --> 00:12:56.323 And you can still use this to do summaries and to do research and, come up with a 00:12:56.323 --> 00:12:57.853 bunch of other books that we should read. 00:12:57.853 --> 00:13:01.873 But, it's similar to when, wizards of the Coast and Dungeons and Dragons 00:13:01.873 --> 00:13:06.563 was talking about, they're already seeing people using their IP and NFTs. 00:13:07.353 --> 00:13:09.818 And AI dungeon mastering and things. 00:13:09.818 --> 00:13:14.288 And a lot of people were worried that they're gonna focus on creating some 00:13:14.288 --> 00:13:18.728 sort of AI DM and replace dms when I think people forget why we played 00:13:18.733 --> 00:13:20.318 Dungeons and Dragons in the first place. 00:13:20.323 --> 00:13:23.008 I think there's been a meme going around, we're doing all this work and 00:13:23.008 --> 00:13:27.508 we're giving AI the ability to do art and music and things like that, right? 00:13:27.508 --> 00:13:30.448 And it's we should be doing the art and music and let the AI do the work. 00:13:30.908 --> 00:13:32.048 we should flip that around. 00:13:32.388 --> 00:13:35.178 I think there's really cool things that, there was a really good blog 00:13:35.178 --> 00:13:41.028 post about someone who had chat, GPT Run a Dungeons and Dragons game for 00:13:41.028 --> 00:13:45.228 them, and it was cool, but I still don't think I would necessarily be 00:13:45.228 --> 00:13:49.138 threatened by that or feel like I'm replacing my hobby with an AI for that. 00:13:49.938 --> 00:13:52.638 CJ: I guess we're like slipping into doom and gloom territory a little bit, 00:13:52.638 --> 00:13:55.638 but like I, yeah, I'm so conflicted. 00:13:55.668 --> 00:13:58.368 Okay, so this is coming from, my perspective as a parent. 00:13:58.888 --> 00:14:02.878 my kids, they're super into D&D and I am really conflicted about 00:14:03.508 --> 00:14:09.418 letting them use ChatGPT, because I'm worried it will take away their. 00:14:11.188 --> 00:14:14.458 And they'll start to lean on it as like entertainment and they'll lose their 00:14:14.458 --> 00:14:20.648 creativity like, before seeing ChatGPT, they would sit down and have piles of 00:14:20.648 --> 00:14:24.548 books across the dining room table that they borrowed from the library and were 00:14:24.548 --> 00:14:26.048 building their own character sheets. 00:14:26.048 --> 00:14:27.218 And they have like dozens of these. 00:14:27.218 --> 00:14:28.118 They're floating around the house. 00:14:28.123 --> 00:14:29.498 Many of them make it to the recycle bin. 00:14:29.528 --> 00:14:33.553 But, we sat down and then, I was like, okay, let's, let me just show 00:14:33.553 --> 00:14:34.873 you some of this ChatGPT stuff. 00:14:34.873 --> 00:14:39.013 We did jokes, we did poems, lots of fart stuff, lots of Mario Brothers stuff. 00:14:39.373 --> 00:14:43.363 But then we like, we generated a D&D character sheet and we like gave it 00:14:43.363 --> 00:14:46.333 some really specific things and my son was like, oh yeah, I wanted to 00:14:46.338 --> 00:14:48.823 have these cool weapons and I want to have this and that, whatever. 00:14:48.943 --> 00:14:52.913 And it generated this really impressive character sheet with like custom. 00:14:53.393 --> 00:14:56.153 Weapons that have, like cus I dunno, it's like damage or something. 00:14:56.153 --> 00:14:59.383 I don't remember like all the details of what they do, but it was like super, 00:14:59.383 --> 00:15:05.473 super customized and really comprehensive and complete and so I think that's cool. 00:15:05.473 --> 00:15:10.873 But I also am like, okay, I want them to be able to use it and get ahead 00:15:10.963 --> 00:15:15.523 by using it, but I also don't want it to steal their creativity and I 00:15:15.523 --> 00:15:16.783 don't know what the answer is to that. 00:15:18.103 --> 00:15:23.323 I also like when it comes to art and all like the creative pieces of my work. 00:15:24.293 --> 00:15:25.073 at work, right? 00:15:25.073 --> 00:15:28.628 Like when I'm doing my day job, I actually don't mind if 00:15:28.658 --> 00:15:29.948 it's doing the creative work. 00:15:29.948 --> 00:15:33.128 Like I don't wanna have to think up images and I don't wanna have 00:15:33.128 --> 00:15:36.908 to think up like these story arcs and have to have all the details. 00:15:36.908 --> 00:15:39.788 Like I'd rather the, AI do that for me. 00:15:39.938 --> 00:15:42.368 And then on the weekends or like when I'm not. 00:15:42.588 --> 00:15:47.008 Working for money, then I can spend my time thinking about my creative 00:15:47.013 --> 00:15:50.328 stuff or drawing, or how do I want to have a different creative outlet. 00:15:50.328 --> 00:15:54.498 And so just the tension between those two things right now is really, yeah. 00:15:54.498 --> 00:15:58.098 It's something that I know like I, I'm having an internal battle about. 00:15:58.148 --> 00:16:02.548 Colin: Yeah, I think this is the, old man yells at cloud segment of the podcast, 00:16:02.548 --> 00:16:02.728 right? 00:16:02.733 --> 00:16:06.603 Where, we both are like, we can see how it can be used for good 00:16:06.603 --> 00:16:08.253 and see how it can be used for bad. 00:16:08.283 --> 00:16:10.593 I think that the other side of that would be, it could make you 00:16:10.593 --> 00:16:12.183 more creative because you get. 00:16:12.763 --> 00:16:15.463 To iterate through lots of different versions really quickly. 00:16:15.903 --> 00:16:20.493 I do feel for artists who have their work has been trained. 00:16:20.918 --> 00:16:24.548 Into these systems where it's if someone was gonna hire someone for 00:16:24.548 --> 00:16:29.288 a portrait before, the odds that is gonna happen in the future, go down. 00:16:29.848 --> 00:16:33.328 and I, there was an article going around from someone who couldn't, 00:16:33.378 --> 00:16:37.668 the time to create a character in a game went from weeks to days and 00:16:37.668 --> 00:16:39.108 they don't feel as creative anymore. 00:16:39.108 --> 00:16:40.688 But also now that means that you can. 00:16:41.508 --> 00:16:45.228 A whole bunch, like AI can create a whole bunch of crappy characters, but it still 00:16:45.228 --> 00:16:47.718 needs some input like we're talking about. 00:16:47.718 --> 00:16:52.728 And so you might not get, have enough time or skill to pull off 00:16:52.728 --> 00:16:55.848 something 10 different ways, but with this you might be able to, and then 00:16:56.358 --> 00:16:58.728 pick the best one, improve on it. 00:16:58.728 --> 00:17:00.888 So you get a lot of that iteration on it. 00:17:01.338 --> 00:17:04.158 I think the thing that I've been using it for the most, other than the 00:17:04.158 --> 00:17:09.453 brainstorming has been like having it teach me And the challenge there 00:17:09.453 --> 00:17:12.633 is the hallucinations are real in ai. 00:17:12.633 --> 00:17:17.573 So you can't take it at its word all the time, but if you keep your 00:17:17.573 --> 00:17:19.433 curiosity and you keep asking questions, 00:17:19.483 --> 00:17:21.313 it's very interesting how it behaves. 00:17:21.313 --> 00:17:26.903 So like I gave it, a prompt in 3.5 and in four just to see the difference. 00:17:26.903 --> 00:17:30.443 And I highly recommend sometimes running the exact same prompt through 00:17:30.443 --> 00:17:31.613 to see what the difference is. 00:17:32.693 --> 00:17:39.503 But I had it create an API in Ruby in 3.5 and in four, and with 3.5 you 00:17:39.503 --> 00:17:43.553 could see how it got to where it got, you could fill in the missing gaps. 00:17:43.553 --> 00:17:44.633 But there were definitely gaps. 00:17:44.633 --> 00:17:49.178 if you ran this, it would not work by, just by the steps it gave me. 00:17:49.638 --> 00:17:51.198 but there were a few things missing. 00:17:51.203 --> 00:17:51.568 Four. 00:17:53.343 --> 00:17:53.643 it. 00:17:53.853 --> 00:17:58.443 Like first try, it had a bunch of gems that it didn't need and wasn't using. 00:17:58.863 --> 00:18:01.833 So like I would ask it like, oh, what is, why'd you include this? 00:18:01.833 --> 00:18:03.033 And they're like, oh good. 00:18:03.083 --> 00:18:03.743 my bad. 00:18:04.493 --> 00:18:06.323 Obviously that's not being used. 00:18:06.323 --> 00:18:07.013 And it was like, okay. 00:18:07.013 --> 00:18:10.733 So you're clearly looking at, most apps that include. 00:18:11.683 --> 00:18:14.683 this gem also included this gem, so that was the next 00:18:14.683 --> 00:18:16.333 probabilistic word, and it was fine. 00:18:16.543 --> 00:18:19.063 It was able to figure out, and that, I think that's the thing that still 00:18:19.063 --> 00:18:22.363 messes with my brain is that even though I know it's probabilistically 00:18:22.423 --> 00:18:26.738 detecting words, how does it unders it doesn't understand, is the point. 00:18:27.128 --> 00:18:31.298 But like, when you ask it about a thing, it knows what the thing is because 00:18:31.568 --> 00:18:34.988 I'll tell it to remove that gym and it will rewrite the code without that. 00:18:35.963 --> 00:18:36.293 CJ: Yeah. 00:18:36.343 --> 00:18:36.633 I. 00:18:37.538 --> 00:18:40.868 can I take a swag at trying my best to explain stuff that I 00:18:40.868 --> 00:18:42.038 don't actually understand yet? 00:18:42.398 --> 00:18:42.878 Colin: do it. 00:18:43.048 --> 00:18:45.958 CJ: I th yeah, I th when we were in college, one of the 00:18:45.963 --> 00:18:49.318 projects that we did was building markoff chains, which was cool. 00:18:49.678 --> 00:18:52.768 You take all this Shakespearean text and you dump it in, and then you 00:18:52.768 --> 00:18:57.508 just break up the text into lots of, Element, tuples, basically thrus, I 00:18:57.508 --> 00:18:58.528 dunno if it's actually a throuple. 00:18:58.918 --> 00:19:03.688 and then you like, look at, you look at a throuple and you say okay, the last 00:19:03.688 --> 00:19:06.058 two words in this three word array. 00:19:06.658 --> 00:19:12.968 Now randomly grab one of the next arrays that also start with those two words. 00:19:13.763 --> 00:19:15.263 And that gives you your third word. 00:19:15.293 --> 00:19:16.763 So then you build this chain. 00:19:16.763 --> 00:19:18.923 So that's like whatever, a dumb markoff chain. 00:19:19.133 --> 00:19:22.613 But you could have it spit out like what looked like Shakespeare and this 00:19:22.613 --> 00:19:25.403 was like 2006 or something, 2008. 00:19:26.603 --> 00:19:30.113 If you imagine every single word or concept. 00:19:31.778 --> 00:19:34.958 Being mapped onto like a 2D grid, right? 00:19:34.958 --> 00:19:38.078 So we have the X coordinate and the Y coordinate, and we're just gonna 00:19:38.078 --> 00:19:41.318 throw every word and they're gonna splatter somewhere on that 2D grid. 00:19:41.918 --> 00:19:47.798 Now we can figure out exactly where each word lands on that 2D grid, and 00:19:48.188 --> 00:19:54.448 each word will have a vector that has like the, the length and the x 00:19:54.448 --> 00:19:59.338 y coordinates, or like the x and y that will get us to that point in 2d. 00:20:01.063 --> 00:20:06.433 each of these models has a certain size, so there's like a certain number of 00:20:06.823 --> 00:20:12.943 dimensions on the model, and I think GPT-3 was like 2048 or something like that. 00:20:12.943 --> 00:20:15.523 And if you look at these new, the newer models, they're much bigger. 00:20:15.523 --> 00:20:17.373 I think they go up to 30,000 or something. 00:20:17.743 --> 00:20:21.893 So if you take that concept of having every word mapped out in 00:20:21.893 --> 00:20:25.463 this space and you exploded out to. 00:20:25.913 --> 00:20:31.283 2000 D basically, or a thousand D, then you have a lot more different dimensions. 00:20:31.283 --> 00:20:33.383 So it's not just X and Y, it's not X, y, Z. 00:20:33.383 --> 00:20:37.763 It's like X, Y, Z, a, B, abc, all the way up to 2000 of those. 00:20:38.303 --> 00:20:42.233 And in that space, words that are semantically similar. 00:20:43.013 --> 00:20:44.363 Will be close together. 00:20:44.783 --> 00:20:49.703 So like hotdog and burger are gonna be like, if you look at this like giant 00:20:49.703 --> 00:20:53.873 space, those are gonna be close together in terms of like where their vector lands. 00:20:53.933 --> 00:20:58.463 And so like you can use that piece of information to get a better idea of what 00:20:58.463 --> 00:21:00.383 like the next word in the sequence is. 00:21:00.923 --> 00:21:03.063 And so that's I don't know if that makes any sense. 00:21:03.453 --> 00:21:08.943 And explaining over audio is tough, but my understanding is that's like you can 00:21:08.943 --> 00:21:16.423 use this semantic similarity between these concepts and even between like sentences 00:21:16.428 --> 00:21:21.603 or paragraphs or like entire blobs of text to figure out what is this thing like? 00:21:21.603 --> 00:21:23.543 And then use that as part of your chain. 00:21:23.873 --> 00:21:25.223 Colin: Yeah, and then we can ask it. 00:21:25.223 --> 00:21:27.813 The question of is a hot dog a sandwich with 00:21:27.903 --> 00:21:29.493 CJ: Yeah, exactly. 00:21:29.563 --> 00:21:34.823 So one of the things that I think is in interesting is using these, part 00:21:34.823 --> 00:21:40.213 of these tools called embeddings, where you can say here is a concept 00:21:40.213 --> 00:21:44.593 or a sentence or a paragraph, and you can send that to open AI in their api 00:21:44.593 --> 00:21:46.333 and it will give you back a vector. 00:21:47.173 --> 00:21:50.953 That is just literally like a giant array of numbers that represents 00:21:50.953 --> 00:21:53.083 that concept for that given model. 00:21:53.383 --> 00:21:56.773 So say you're using Da Vinci, whatever the Da Vinci model for 00:21:57.053 --> 00:22:01.483 GPT-3 and you give it hotdog, it'll give you back something that's like 00:22:01.573 --> 00:22:04.093 an array that's 2048 elements long. 00:22:04.333 --> 00:22:05.443 That's a bunch of numbers. 00:22:05.653 --> 00:22:09.043 And those numbers, if you like, map them out through 2048 space. 00:22:09.043 --> 00:22:10.633 That's like where the hotdog lands. 00:22:12.343 --> 00:22:14.038 And there are, yeah. 00:22:14.188 --> 00:22:17.458 one of the things I was trying to do was take all my blog posts on my site 00:22:18.328 --> 00:22:22.888 and then run them through and build up these vectors and stick them into 00:22:22.888 --> 00:22:26.458 Pine Cone, which is like a vector database, and then use pine cones like 00:22:26.488 --> 00:22:32.188 querying language to have it give me back like search results for my website. 00:22:32.278 --> 00:22:35.168 I think this is like the bits and pieces that were, that I know, like I'm 00:22:35.168 --> 00:22:37.268 trying to like, learn about for sure. 00:22:37.273 --> 00:22:39.428 And build, I don't know. 00:22:39.818 --> 00:22:43.208 Th there's so much that I think will be built on top of this that, 00:22:43.668 --> 00:22:46.538 it definitely feels like, early, early days of the internet in 00:22:46.538 --> 00:22:50.318 terms of accessibility, tooling. 00:22:50.748 --> 00:22:53.268 like all these crazy new concepts they have to pick up. 00:22:54.078 --> 00:22:54.288 Colin: Yeah. 00:22:54.288 --> 00:22:57.318 And you have to eventually learn how, like I, I've been learning 00:22:57.318 --> 00:23:00.258 how to give it prompts and that's like a skill in of itself. 00:23:00.258 --> 00:23:04.588 And some of them, especially these image ones, you can, the prompts 00:23:04.588 --> 00:23:11.038 become these paragraphs of describing the mood, the scene, the reflections. 00:23:11.088 --> 00:23:13.908 I'll put it in the show notes too, but a friend of mine is a 00:23:13.908 --> 00:23:17.148 photographer and he is using these. 00:23:17.748 --> 00:23:21.978 As well, but he generated something that I would not have guessed was not 00:23:21.978 --> 00:23:25.648 shot on his camera, up in Tahoe Fair. 00:23:25.678 --> 00:23:29.188 I'll have to find out what his prompt was to see what it was, but it was like 00:23:29.578 --> 00:23:33.628 he's a nature photographer and he's over here generating from scratch the 00:23:33.628 --> 00:23:37.828 same thing and isn't really threatened by it as much as trying to explore 00:23:37.828 --> 00:23:39.888 and push and see what he can do there. 00:23:40.308 --> 00:23:43.458 I think the challenge is going to be whether or not. 00:23:44.353 --> 00:23:47.053 for some of the applications that I've seen, they're not things 00:23:47.053 --> 00:23:48.643 that I was gonna do today anyway. 00:23:48.673 --> 00:23:55.903 it's cool, but like we went through a bot, a hype cycle a few years ago. 00:23:56.393 --> 00:23:58.253 then we went through the crypto hype cycle. 00:23:59.118 --> 00:24:02.653 And I have to wonder, I think that this feels like magic because it 00:24:02.653 --> 00:24:05.013 understands what, it remembers. 00:24:05.013 --> 00:24:08.823 if you're in a ChatGPT session, you can say oh, what does that mean? 00:24:08.823 --> 00:24:11.793 And it knows what the thing above was that you were talking about. 00:24:11.793 --> 00:24:14.103 So that shared memory is super interesting. 00:24:14.823 --> 00:24:17.013 But I think people have to remember that these are still. 00:24:17.613 --> 00:24:21.993 Models like you just gave us the whole pitch on how they work and 00:24:21.993 --> 00:24:23.973 not another person on the other end. 00:24:23.973 --> 00:24:28.383 Cuz there's starting to be some very scary stories about people falling for 00:24:28.383 --> 00:24:30.953 these, chat sessions as real people. 00:24:31.033 --> 00:24:34.103 you have some notes here about some of the doom and gloom, or Boone 00:24:34.103 --> 00:24:39.518 type of things where, there have been, scams where someone will call. 00:24:40.463 --> 00:24:45.193 Record your voice and then use that voice to do a voice clone, and use 00:24:45.193 --> 00:24:48.403 these models to be able to then call. 00:24:48.778 --> 00:24:56.228 A significant other or a child, and pose as you, and, they won't, especially 00:24:56.228 --> 00:25:00.278 on a phone where it doesn't have high fidelity, it can sound a little goofy or 00:25:00.278 --> 00:25:05.868 distressed or, say I've been, kidnapped as was the example that I hurt and please 00:25:05.868 --> 00:25:07.938 send money to this location type of thing. 00:25:07.938 --> 00:25:08.648 And it's whew. 00:25:08.708 --> 00:25:12.933 Like how do we get to a world where that exists and how do we protect our. 00:25:14.238 --> 00:25:19.458 CJ: And the same tools that these scammers are using, we can, we're 00:25:19.458 --> 00:25:23.268 also already using that for this podcast, so one of the tools we use 00:25:23.268 --> 00:25:26.598 is called Descript, and like right after the podcast, we drop it in there. 00:25:26.598 --> 00:25:32.598 It transcribes everything and has this feature called Overdub where we can train. 00:25:33.423 --> 00:25:37.603 Descript and say this is, I think it's four, it needs 45 minutes or something. 00:25:37.603 --> 00:25:42.793 So I can say here's 45 minutes of Colin talking and here's 45 minutes of me 00:25:42.798 --> 00:25:47.588 talking and okay, so this was a word that we, one of us, stumbled over, 00:25:47.588 --> 00:25:49.448 or we said the incorrect word here. 00:25:49.778 --> 00:25:53.408 Then you can just edit the word as text as if you're just editing a Google doc 00:25:53.498 --> 00:25:58.088 and say, overdub, and it will replace that word with that person's actual voice. 00:25:58.853 --> 00:26:01.793 And that becomes really useful in scenarios where you're doing 00:26:01.793 --> 00:26:02.813 something like podcasting. 00:26:02.818 --> 00:26:05.733 You don't need to like, jump back on the mic and tr do a bunch of 00:26:05.733 --> 00:26:09.363 different takes to say the word again and then splice it in or whatever. 00:26:09.723 --> 00:26:11.883 but yeah, it's definitely a huge risk. 00:26:11.943 --> 00:26:18.133 And one of the recommendations I heard from, Rachel Woods on TikTok was that 00:26:18.133 --> 00:26:21.853 to avoid these voice scams is to create some sort of safe word with your family 00:26:22.213 --> 00:26:25.973 so that if anyone calls and they're like pretending to be someone, you, yeah. 00:26:25.973 --> 00:26:30.143 You have a way to, I don't know, crack their scam or whatever. 00:26:30.323 --> 00:26:32.543 Colin: that's your, safety tip from Feld and learn this 00:26:32.863 --> 00:26:33.353 CJ: Yeah, 00:26:33.433 --> 00:26:35.308 Colin: go create a safe word for sure. 00:26:35.828 --> 00:26:39.543 and I would say definitely don't be feeding these things, any of 00:26:39.543 --> 00:26:41.943 your private or secured data. 00:26:42.423 --> 00:26:43.653 anything that you like. 00:26:43.703 --> 00:26:46.343 I think I've been hearing people putting in there like medical 00:26:46.643 --> 00:26:52.643 records and tests and stuff into this thing, and it is an impressively. 00:26:53.048 --> 00:26:56.168 Able to understand protein chains and things. 00:26:56.168 --> 00:26:59.918 And but that's probably because there's a bunch of research out there on protein 00:26:59.918 --> 00:27:03.888 chains and they show up in a certain way and they show up on the speced, map the 00:27:03.888 --> 00:27:05.478 same way that you were describing earlier. 00:27:05.483 --> 00:27:09.978 And so I am hopeful for the things that it's going to be able to teach us. 00:27:09.978 --> 00:27:13.968 Like it's better at spotting things on in radiology than a human. 00:27:14.583 --> 00:27:15.063 Awesome. 00:27:15.573 --> 00:27:18.693 That doesn't mean it replaces the radiologist, it just means that now they 00:27:18.693 --> 00:27:20.953 can detect and look at things faster. 00:27:21.523 --> 00:27:26.363 but I have to wonder, and this happens, happened ever since the industrial 00:27:26.363 --> 00:27:30.113 revolution, but as we get more productive with these tools, does that mean that 00:27:30.113 --> 00:27:34.873 now we're just trying to churn out more stuff and in 40 hours a week, or 00:27:34.873 --> 00:27:38.323 do we start to head towards that four day work week, three day work week? 00:27:38.938 --> 00:27:41.518 CJ: Oh, I hope that three day work week comes, I think it's 00:27:41.518 --> 00:27:44.888 right around the corner . So going back to could it be a hype cycle? 00:27:44.888 --> 00:27:45.848 Is it a fad? 00:27:46.678 --> 00:27:52.208 I think when you, reflect about search and Google in general and 00:27:52.213 --> 00:27:56.288 the job to be done there is I want to find an answer to a question, 00:27:56.388 --> 00:27:59.298 . The way that I search is I'll like type in my search query and then I open 00:27:59.298 --> 00:28:02.816 like the first nine links , just like command, click all of them and then 00:28:02.816 --> 00:28:04.616 dig through the tabs as fast as I can. 00:28:04.616 --> 00:28:08.041 Scanning and my, my brain has been trained over the last 15 years, 00:28:08.431 --> 00:28:12.181 how to just try to extract maybe a right answer out of these things. 00:28:12.541 --> 00:28:16.566 And now you can just get the right answer, almost immediately. 00:28:17.086 --> 00:28:17.101 ha. 00:28:17.131 --> 00:28:19.981 Have you had a chance to try out the Bing, like the Bing client? 00:28:20.851 --> 00:28:21.361 Colin: I haven't. 00:28:22.381 --> 00:28:23.971 CJ: So what were we doing? 00:28:23.976 --> 00:28:27.991 We were looking for a type of Star Wars character. 00:28:27.996 --> 00:28:30.366 It was like a class of Star Wars character that's like this super 00:28:30.366 --> 00:28:36.126 esoteric character type that is only documented like on one page on Wikipedia, 00:28:36.176 --> 00:28:37.676 buried somewhere in the archives. 00:28:37.826 --> 00:28:40.646 And so we went on Bing and used like the Bing. 00:28:40.986 --> 00:28:44.321 GPT search or whatever, and it spit out the right answer. 00:28:44.321 --> 00:28:47.681 It was like, I can't even remember what it was, but it was like some type 00:28:47.681 --> 00:28:53.166 of, dark sister or something that, that had, yeah, it was like a force, 00:28:53.406 --> 00:28:56.536 I don't know, force sensitive thing. 00:28:56.856 --> 00:29:00.276 but yeah, anyways, like there's lots of different interfaces to this and 00:29:00.276 --> 00:29:04.326 I think the jobs to be done about just getting an answer quickly. 00:29:05.171 --> 00:29:09.686 If we can start to trust and rely on the models, which, I think one 00:29:09.686 --> 00:29:14.096 of the co-founders, Greg, one of the co-founders of OpenAI, said that coming 00:29:14.096 --> 00:29:19.496 soon, the answers that you get back from the, these models will be accurate 00:29:19.496 --> 00:29:24.446 and not be hallucinating, which, when that starts to happen, I think that's, 00:29:24.626 --> 00:29:25.916 that becomes a huge game changer. 00:29:25.916 --> 00:29:29.696 But yeah, until then, we all have to be skeptical and take 00:29:29.701 --> 00:29:30.566 every answer with a grain. 00:29:31.856 --> 00:29:35.396 Colin: Yeah, there's actually a good book on this that's called Invisible 00:29:35.401 --> 00:29:39.396 Women, data Bias in a World Designed For Men, which I think is also 00:29:39.401 --> 00:29:42.696 something we have to keep in mind is there's been a bunch of conversation 00:29:42.696 --> 00:29:44.736 about ethics around all of this. 00:29:44.736 --> 00:29:49.506 And to be honest, most of the worlD&Data sets are very much 00:29:49.506 --> 00:29:51.306 skewed for as perspective. 00:29:51.311 --> 00:29:54.856 It tends to be, people who are building these things tend to be. 00:29:55.146 --> 00:29:58.896 white men and or, a few different, types of backgrounds. 00:29:59.226 --> 00:30:02.526 And so that we need to have representation in both the data 00:30:02.526 --> 00:30:04.506 sets and the people working on this. 00:30:04.996 --> 00:30:08.121 because otherwise, I even think about simple run of the mill 00:30:08.126 --> 00:30:09.711 things like recommendations. 00:30:10.451 --> 00:30:10.811 Sure. 00:30:10.811 --> 00:30:15.941 Like it's harmless to ask for a pizza recommendation in a certain town. 00:30:16.481 --> 00:30:20.141 Now, if it's generative AI though, is it based on Yelp reviews? 00:30:20.141 --> 00:30:22.961 Is it based on just like it knows what a pizza shop is and 00:30:22.961 --> 00:30:24.101 it gives you the first one? 00:30:24.441 --> 00:30:25.701 is it the closest one? 00:30:25.701 --> 00:30:28.426 there's a whole bunch of stuff there that we don't yet know. 00:30:29.421 --> 00:30:34.111 And so if you're asking for actual answers to things, it does seem to 00:30:34.111 --> 00:30:38.641 also have this sense of the way that it writes in chat, GPT specifically, 00:30:38.821 --> 00:30:41.311 it tries to agree with you. 00:30:41.841 --> 00:30:45.831 like the way that it presents things is very pleasing and agreeable. 00:30:46.371 --> 00:30:51.076 And what I'd be curious to see, and I've tried a little bit with poking some 00:30:51.076 --> 00:30:57.406 conspiracy theories and things at it, is wh how it replies because it can also. 00:30:58.081 --> 00:31:00.901 Reinforce things that you tell it. 00:31:01.321 --> 00:31:04.561 And so if you have a belief whether or not that belief is true or not, 00:31:05.041 --> 00:31:08.081 and you give it to it, and you have, you create this little, reality 00:31:08.086 --> 00:31:09.461 distortion field for yourself. 00:31:09.461 --> 00:31:12.641 And there's a whole bunch of books in sci-fi that would love to, to 00:31:12.641 --> 00:31:14.231 take us all for a ride on that. 00:31:14.601 --> 00:31:18.261 so we only have to look at our whole body of science fiction to, 00:31:18.261 --> 00:31:20.061 to see like where this could go. 00:31:20.436 --> 00:31:22.596 But, I'm excited to see where it goes. 00:31:22.926 --> 00:31:27.586 It is impressive and every time I use it, I'm like amazed and, not 00:31:27.586 --> 00:31:31.526 concerned for the future as much as, we gotta take a measured approach 00:31:31.676 --> 00:31:36.906 to safety and ethics here and figure out how we can use it for the better. 00:31:37.700 --> 00:31:43.270 CJ: Anthropic is one of the main competitors to open AI and. 00:31:43.975 --> 00:31:48.775 this is again, gonna be stuff that I learned from Rachel Woods, but the, 00:31:48.885 --> 00:31:53.865 the team at Anthropic was a bunch of former open AI researchers who wanted 00:31:53.865 --> 00:31:58.730 to break away and make something less corporate and, they have this 00:31:58.730 --> 00:32:03.050 blog post that we can link to in the show notes about their views on AI 00:32:03.050 --> 00:32:10.000 safety, and it's my understanding that open AI is using humans to create 00:32:10.000 --> 00:32:12.520 the guardrails around the models. 00:32:13.090 --> 00:32:18.700 And Anthropic is taking a different approach called constitutional ai, where 00:32:18.700 --> 00:32:20.470 it's like giving it these principles. 00:32:21.510 --> 00:32:27.505 That are, these like things that basically say, behave or please behave 00:32:27.535 --> 00:32:30.625 nicely based on these set of values. 00:32:30.955 --> 00:32:33.865 And then it goes and like trains itself and then tries to make 00:32:33.865 --> 00:32:36.355 sure that its decisions are consistent with those values. 00:32:37.075 --> 00:32:41.185 And, I, yeah, it's it's a really interesting approach and it also 00:32:41.185 --> 00:32:45.425 made me think about, this is gonna take a little bit of a turn, but the 00:32:45.665 --> 00:32:54.295 way that our current constitution is, resulting in less than ideal. 00:32:55.530 --> 00:32:57.450 Country like formation, right? 00:32:57.450 --> 00:33:01.290 Like here we are realizing that the Second Amendment is actually 00:33:01.770 --> 00:33:06.570 not the best formation of, what the founding fathers expected. 00:33:06.570 --> 00:33:09.900 And we have all of this gun violence all over the United States. 00:33:10.350 --> 00:33:15.700 And so there is a chance that you write a Constitution with values that 00:33:15.760 --> 00:33:17.450 are consistent with what you believe. 00:33:17.695 --> 00:33:22.045 Correct and good today, but that doesn't stand up to the test of time. 00:33:22.050 --> 00:33:26.325 So figuring out how to evolve those values and principles in a way 00:33:26.330 --> 00:33:29.590 that's measured and safe and ah, it's gonna be really interesting. 00:33:30.130 --> 00:33:30.400 But, 00:33:30.910 --> 00:33:34.570 Colin: it's that interpretation and context that's key. 00:33:34.570 --> 00:33:37.155 And even some of this reminds me of the three laws of. 00:33:37.275 --> 00:33:41.675 Robotics too, which are very dated and from sci-fi, but they're 00:33:42.245 --> 00:33:43.355 something we should look at. 00:33:44.200 --> 00:33:44.590 CJ: Yeah. 00:33:44.645 --> 00:33:47.535 it is, sci-fi is like basically here today. 00:33:47.895 --> 00:33:48.555 It's wild. 00:33:49.005 --> 00:33:51.925 Colin: It's very similar to some people have said, with all of this, 00:33:51.925 --> 00:33:57.655 the last 5%, the last 1% is so hard and hardware is gonna be a limiting factor. 00:33:57.655 --> 00:34:02.630 But also like when we think about full self-driving, if say we're at 90% of 00:34:02.630 --> 00:34:07.430 the way there today, like there's still a lot that's gotta happen before you'll 00:34:07.430 --> 00:34:09.260 let ChatGPT take the wheel literally. 00:34:09.265 --> 00:34:13.825 It's like obviously large language models are not what's powering full self-driving, 00:34:13.825 --> 00:34:18.380 but if we look at it similarly to the vectors of what's a person in a crosswalk 00:34:18.380 --> 00:34:22.460 and what's someone backing up when they're supposed to be going forward, right? 00:34:22.460 --> 00:34:26.960 Things like that, that a human can make those decisions, but if it doesn't have 00:34:26.960 --> 00:34:28.550 enough examples and it doesn't have. 00:34:29.310 --> 00:34:31.740 To reason around it's gonna be hard to get there. 00:34:31.740 --> 00:34:35.370 And then can you have enough of that on hardware that's in 00:34:35.370 --> 00:34:37.500 your pocket or in your car? 00:34:38.000 --> 00:34:41.215 we've got a long ways to go and it might be that hardware is a 00:34:41.215 --> 00:34:42.445 limiting factor for a little while. 00:34:43.740 --> 00:34:46.955 I think to take it back to the build and learn side of things, we've got 00:34:46.955 --> 00:34:48.385 two things that we can leave you with. 00:34:48.390 --> 00:34:52.785 I think one is a funny, thing I saw come across Ruby Weekly, which was that 00:34:52.785 --> 00:34:56.385 chat, GPT co-authored at PR to Rails. 00:34:56.945 --> 00:35:01.365 and this was a Link baity title, but one of the rails core team members, 00:35:01.765 --> 00:35:08.005 went ahead and gave it a prompt, created a poll, request, code, test, 00:35:08.005 --> 00:35:12.515 change, log everything, and then submitted it with, Disclaimer that it 00:35:12.515 --> 00:35:15.245 was created with guidance by ChatGPT. 00:35:15.645 --> 00:35:17.565 and so we'll leave a link to that in the show notes. 00:35:17.565 --> 00:35:20.025 You can read the pr, you can read the whole conversation. 00:35:20.025 --> 00:35:23.095 And, it was interesting to see like if you believe ChatGPT is 00:35:23.095 --> 00:35:25.735 infringing on your copyright, please let me know type of thing. 00:35:26.195 --> 00:35:27.690 and then you have a list here. 00:35:27.695 --> 00:35:29.400 What do we have down below? 00:35:29.960 --> 00:35:32.490 CJ: Oh yeah, so this was, I used Notion ai. 00:35:32.490 --> 00:35:35.500 So we plan all of our shows inside of Notion and I just, you 00:35:35.500 --> 00:35:37.930 just do slash AI or something. 00:35:37.935 --> 00:35:42.580 And then I said, tell me five social media influencers that people should 00:35:42.585 --> 00:35:44.080 follow to stay up to date about ai. 00:35:44.080 --> 00:35:46.330 And it's spit out these five Rachel Woods. 00:35:47.170 --> 00:35:52.240 Logan, GPT, swyx, Lex Friedman, and Siraj Raval. 00:35:52.750 --> 00:35:56.560 So we'll have to do a little bit of, vetting and research about who 00:35:56.560 --> 00:35:58.975 we're gonna put in the show notes, but it's just I don't know, it's 00:35:58.975 --> 00:36:00.625 wild that, that it can do this. 00:36:00.725 --> 00:36:03.455 Colin: Absolutely, and I think that is where we'll leave it. 00:36:03.455 --> 00:36:06.515 We've got a whole bunch of links that we'll put in the show notes to 00:36:06.905 --> 00:36:10.425 a lot of tools, a lot of different, models that we talked about today. 00:36:10.935 --> 00:36:14.895 And if you want to hear more about this, please tweet at us. 00:36:14.955 --> 00:36:17.535 @buildandlearn_ on Twitter. 00:36:17.965 --> 00:36:19.795 you can find all that in the show notes as well. 00:36:19.795 --> 00:36:22.115 Find us, and we'll see you next week. 00:36:23.240 --> 00:36:23.630 CJ: All right. 00:36:24.050 --> 00:36:24.560 Bye friends.