WEBVTT 1 00:00:04.679 --> 00:00:09.480 Hey, folks, welcome back to another episode of the Ruby Rogues podcast. 2 00:00:09.759 --> 00:00:14.119 This week, on our panel we have Valentino Stole. Hey Now, I'm 3 00:00:14.199 --> 00:00:17.839 Charles Maxwood from Top End Devs. We have a special guest this week and 4 00:00:17.960 --> 00:00:21.760 it is Obi Fernandez. Obi, it's been a while. How are things 5 00:00:21.800 --> 00:00:25.839 going? Pretty good? Man? Thanks for the invite. It's an honor 6 00:00:25.879 --> 00:00:30.239 to be back. Yeah, So do you want to just I know you're 7 00:00:30.239 --> 00:00:36.039 writing an AI book and I see you active in the Ruby AI discord. 8 00:00:36.119 --> 00:00:39.719 But what else are you up to? What else should people know about you? 9 00:00:42.079 --> 00:00:45.560 Well, I've been I've been living mostly in Mexico City since twenty seventeen, 10 00:00:45.640 --> 00:00:51.119 So people usually find that interesting and asking what that's about because I grew 11 00:00:51.200 --> 00:00:55.799 up outside of New York City and spent twenty years in Atlanta, you know, 12 00:00:55.840 --> 00:00:59.119 in the south east US. But yeah, Mexico City is super cool. 13 00:01:00.000 --> 00:01:04.879 It's a great place for expats to live, great lifestyle. And I've 14 00:01:04.879 --> 00:01:11.519 done a couple of startups in recent years. One was an NFT platform that 15 00:01:11.519 --> 00:01:15.519 that kind of went took me, took me in one hundred other employees on 16 00:01:15.359 --> 00:01:18.159 a on a big roller coaster ride with venture backing, and then crashed and 17 00:01:18.200 --> 00:01:21.920 burn along with the rest of the crypto market a couple of years ago, 18 00:01:23.799 --> 00:01:29.239 so that was interesting. And yeah, for the last year, I've been 19 00:01:29.280 --> 00:01:37.599 working on another startup called Olympia, which is a platform that gives you AI 20 00:01:37.799 --> 00:01:45.519 powered consultants to add to your team. Is kind of humanizes AI. So 21 00:01:47.599 --> 00:01:49.760 yeah, that that that's been a lot. That's been a lot of fun. 22 00:01:49.760 --> 00:01:53.280 And it's doing well, growing about ten twenty percent month a month. 23 00:01:53.400 --> 00:01:59.280 So oh wow, that doesn't seem to be letting up. Good deal. 24 00:02:00.280 --> 00:02:04.599 I love Olympia. Oh, I'm happy. I'm a happy user here. 25 00:02:05.400 --> 00:02:09.159 Super cool. I didn't actually realize that. That's a surprise. That's awesome. 26 00:02:12.719 --> 00:02:15.879 So where where do you find it? Just Olympia dot com or something 27 00:02:15.439 --> 00:02:22.280 Olympia dot chat that chat Okay, yeah, I need to I need to 28 00:02:22.280 --> 00:02:27.879 bite the bullet and spend on the un the AI or dot com writing there, 29 00:02:27.879 --> 00:02:32.159 but they're expensive. Yeah, we'll have to check it out. In 30 00:02:32.199 --> 00:02:36.319 the meantime, we're I kind of want to get back to the book. 31 00:02:36.520 --> 00:02:40.919 You know that you're writing about AI and Ruby and I'm kind of curious. 32 00:02:40.960 --> 00:02:46.319 It sounds like you've been working in this for a while, but how did 33 00:02:46.319 --> 00:02:50.800 you get started with this and you know, what, what have you done 34 00:02:50.840 --> 00:02:57.360 that makes you the guy to write this book? Well? This this isn't 35 00:02:57.400 --> 00:03:04.280 my first startup that I've tried to do in AI. It's like back in 36 00:03:04.319 --> 00:03:08.919 twenty eighteen, twenty nineteen, as a side project that kick Us Partners, 37 00:03:08.960 --> 00:03:15.680 which was my consulting company that I merged with Magma Labs. Back in twenty 38 00:03:15.719 --> 00:03:23.639 twenty, I did a project called Demos Guru and the Idea. So I'm 39 00:03:23.680 --> 00:03:29.560 also involved in music and music production, and so I'm and was running a 40 00:03:29.560 --> 00:03:34.479 somewhat active record label at the time as a hobby. So one of the 41 00:03:34.520 --> 00:03:39.960 problems that identified for myself and other people that run record labels is that you 42 00:03:40.000 --> 00:03:44.479 get this flood of demos that have nothing to do with the kind of music 43 00:03:44.520 --> 00:03:51.520 you put out, or they're just not up to par in different ways, 44 00:03:51.560 --> 00:03:58.319 you know, not mixed properly and whatnot. And I didn't actually realize that 45 00:03:58.800 --> 00:04:01.800 machine learning stuff, it was a craze back in that era twenty eighteen, 46 00:04:01.840 --> 00:04:08.000 twenty nineteen. I was just reading some analysis of it and and I don't 47 00:04:08.000 --> 00:04:12.280 know, maybe COVID scrambled my brain, but I thought it was pretty unique. 48 00:04:12.280 --> 00:04:15.280 What I what I was thinking of doing back then, is like basically 49 00:04:15.240 --> 00:04:20.600 trying to use machine learning to detect different qualities of a piece of music, 50 00:04:20.639 --> 00:04:27.279 including its genre mastering, you know, quality of its mastering and mixing it, 51 00:04:27.600 --> 00:04:31.079 and then other things that you could use to identify music. So worked 52 00:04:31.079 --> 00:04:36.920 on that for nearly a year. I didn't get anywhere really really really hard 53 00:04:38.000 --> 00:04:43.279 math and you know, just kind of difficult problem that wasn't cracked yet. 54 00:04:45.839 --> 00:04:47.519 I'm not sure that it's cracked yet right now. Actually, no, I 55 00:04:47.519 --> 00:04:51.120 think about it. I'm really thought about it in a while. But the 56 00:04:51.160 --> 00:04:54.839 people are doing the reverse, which is taking text, you know, and 57 00:04:54.879 --> 00:04:58.319 turning it into music, and I was trying to take music and turn into 58 00:04:58.360 --> 00:05:00.720 text, which I haven't seen yet. So if any was working on that, 59 00:05:00.839 --> 00:05:04.920 reach out to me. I'm super interested. But yeah, like a 60 00:05:05.560 --> 00:05:09.920 we were talking about Olympia right like like a year ago I started working on 61 00:05:09.959 --> 00:05:14.759 this. This startup actually actually goes before that. So the last Rails comp 62 00:05:15.000 --> 00:05:16.959 I went to, which was the first Rails comp in a while, was 63 00:05:17.000 --> 00:05:24.360 in Atlanta last year, and I was super excited about launching like a chat 64 00:05:24.360 --> 00:05:30.800 TBT clone in Rails at the time, so so I scrambled and I made 65 00:05:30.839 --> 00:05:34.399 a little platform called Magma Chat and I opened sourced it during Rails comp and 66 00:05:34.480 --> 00:05:41.439 kind of announced that it showed it to a bunch of people even at that 67 00:05:41.560 --> 00:05:44.160 comp. Even at the Rails comp, I mean a lot of people. 68 00:05:44.199 --> 00:05:46.319 I was showing it too. Had never seen chat to BT yet because it 69 00:05:46.399 --> 00:05:50.959 was still recently, you know, fairly recent didn't know what's going on. 70 00:05:51.959 --> 00:05:57.120 So blew some minds just in terms of showing people what you know you could 71 00:05:57.160 --> 00:06:04.959 do with GPT four. And then my girlfriend kept complaining about her copywriter at 72 00:06:05.000 --> 00:06:09.600 work. So she was she was running partnerships for a large conference series, 73 00:06:10.360 --> 00:06:15.639 and she had this copywriter in India who was paid very little money and produced 74 00:06:15.800 --> 00:06:21.759 very low quality copywriting, so she was rewriting everything. And I said, 75 00:06:21.759 --> 00:06:25.680 well, hey, I got this Magma Chat thing deployed. Can I make 76 00:06:25.720 --> 00:06:30.000 you a copywriter? But and she said sure, and then I gave it 77 00:06:30.040 --> 00:06:33.519 to her and I watched, I watched her use it, and she's just 78 00:06:33.560 --> 00:06:38.439 started using this thing for hours a day and basically started doing all her copywriting 79 00:06:38.439 --> 00:06:43.759 herself. And then very soon thereafter she said, can you can you make 80 00:06:43.800 --> 00:06:46.439 me a whole editorial team? And I said, well, what do you 81 00:06:46.480 --> 00:06:49.079 need And she said, give me a copy editor, give me an se 82 00:06:49.120 --> 00:06:54.560 O expert, give me a marketing expert, you know, give me a 83 00:06:54.600 --> 00:06:58.240 PR expert. So so I put it together and very quickly realized, Okay, 84 00:06:58.240 --> 00:07:01.000 there's got to be a startup here, because she was willing to pay 85 00:07:01.040 --> 00:07:04.079 for it, and I started showing it to other people and they were interested 86 00:07:04.079 --> 00:07:08.000 in paying for it. So I said, do you want to you want 87 00:07:08.000 --> 00:07:11.560 to be my CEO? Because I was CEO of my last startup and I 88 00:07:11.600 --> 00:07:15.920 hated it. I'm more of a natural CTO as you might imagine, so 89 00:07:15.920 --> 00:07:19.240 so yeah, so I brought her on board and we started Olympic together, 90 00:07:20.199 --> 00:07:26.279 and I had a beta product ready in a couple of months. We launched 91 00:07:26.279 --> 00:07:32.240 it in July into close beta, and then we launched to the public in 92 00:07:32.279 --> 00:07:41.120 September. And it's all it's not using Python or typescript or any of the 93 00:07:41.120 --> 00:07:45.319 typical, you know, usual sub suspects you'd want to use for something like 94 00:07:45.360 --> 00:07:50.680 this, or you'd expect to use. Brought in rails right from the beginning, 95 00:07:51.360 --> 00:07:56.639 kind of forked Magna chat when I started, and then eventually ended up 96 00:07:56.680 --> 00:08:00.560 replacing almost all of it. Have not done a good job of keeping magma 97 00:08:00.639 --> 00:08:05.079 Chat up to date. Sorry whoever picked that up really screwed you over on 98 00:08:05.120 --> 00:08:11.360 that. But my partners at Magna Lab have you know, also kind of 99 00:08:11.399 --> 00:08:15.439 fork from where I started, and they have a commercial version of Magma Chat 100 00:08:16.160 --> 00:08:20.000 that they'll implement for people, you know, behind the firewall kind of thing. 101 00:08:20.959 --> 00:08:24.879 So, yeah, you were asking about the book. I mean essentially 102 00:08:24.920 --> 00:08:30.079 I took you know, I took notes. Given I've written technical books before, 103 00:08:31.279 --> 00:08:35.000 it's just a few, so you know, I was taking notes along 104 00:08:35.039 --> 00:08:39.080 the way. And right from the beginning, I was like, Wow, 105 00:08:39.120 --> 00:08:45.559 this is comp to me completely novel kinds of programming, you know, it 106 00:08:45.679 --> 00:08:50.440 just it's just totally new paradigm in the way, just things I haven't seen 107 00:08:50.440 --> 00:08:54.080 before. So I was like, okay, I started documenting what were essentially 108 00:08:54.120 --> 00:08:58.240 patterns and approaches and things, and I guess I did a little bit of 109 00:08:58.240 --> 00:09:01.080 blogging about it as well. I mean some of the early blogging and did 110 00:09:01.080 --> 00:09:07.559 on the subject went viral. When something when the topic is so hot, 111 00:09:07.600 --> 00:09:09.919 like the D and D post I did for a I got hundreds of thousands 112 00:09:11.000 --> 00:09:18.399 of the reads on medium, so that was kind of insane. And yeah, 113 00:09:18.519 --> 00:09:20.960 so that was that was what first took beyond it. So like when 114 00:09:22.000 --> 00:09:30.600 I first got access to the open AI API, my kids were actually here. 115 00:09:30.879 --> 00:09:35.759 So I have a twenty seven year old daughter and a twenty four year 116 00:09:35.799 --> 00:09:39.480 old son and now sixteen year old son, and the sixteen year old is 117 00:09:39.519 --> 00:09:46.120 really into Dungeons and Dragons, and as a family, we did a lot 118 00:09:46.159 --> 00:09:50.919 of gaming growing up as well. So they were here for spring break or 119 00:09:52.320 --> 00:09:54.679 I guess it wasn't spring break. Oh, I guess it was like a 120 00:09:54.720 --> 00:09:58.799 late holiday break. I can't remember exactly, but they were all here in 121 00:09:58.799 --> 00:10:01.200 Mexico City. We're all here at the same time. We were playing Dungeon 122 00:10:01.240 --> 00:10:05.159 and Dragons, and I said, hey, I bet you that this this 123 00:10:05.279 --> 00:10:13.120 AI thing could be a good DM for a role playing game. And then 124 00:10:13.240 --> 00:10:15.559 so so I started playing around with it, and I tried to make an 125 00:10:15.559 --> 00:10:18.519 app for it, and then I described it in the blog post, and 126 00:10:18.639 --> 00:10:22.159 that blog post blew up really big, and I started getting that like you 127 00:10:22.159 --> 00:10:28.320 know, Spidey sense of like, okay, here's another big wave to ride, 128 00:10:28.559 --> 00:10:35.919 you know, so go go ahead, go ahead, Valentino. Oh 129 00:10:37.039 --> 00:10:39.360 no, I mean I have a lot to say here. Yeah, I 130 00:10:41.120 --> 00:10:45.639 mean, I love the dungeons and dragons use case, which is kind of 131 00:10:46.279 --> 00:10:52.120 interesting now with chat GPTs like multimodetdality, like just talking with it. I 132 00:10:52.159 --> 00:10:56.000 had like a long ca car ride recently where I was just like talking through 133 00:10:56.000 --> 00:11:01.799 a business idea and it's pretty remarkable. But yeah, I mean the whole 134 00:11:01.840 --> 00:11:05.639 Magma chat stuff. I love that. It was actually really inspirational, like 135 00:11:07.399 --> 00:11:11.919 for how to use AI in a lot of ways. So you're out there 136 00:11:11.000 --> 00:11:15.639 wondering, like that's definitely like a great place to start to just like pick 137 00:11:15.679 --> 00:11:20.320 apart a you know, how to do X Y z uh, you know, 138 00:11:20.879 --> 00:11:22.559 until your book comes out, because a lot of that is conveyed in 139 00:11:22.600 --> 00:11:26.519 there. Well the books available, we'll talk. You can get it right 140 00:11:26.559 --> 00:11:31.759 now on lean pup. And there's actually much better code out there now than 141 00:11:31.799 --> 00:11:35.720 Magma chats. So if you're listening to this, don't don't don't go use 142 00:11:35.759 --> 00:11:41.600 Magma chat. I would use ray for Rails or one of the other newer, 143 00:11:41.759 --> 00:11:45.679 you know kind of things for this, But yeah, I mean I'd 144 00:11:45.679 --> 00:11:48.960 love to chat through some of the just the concepts in there. We don't 145 00:11:48.960 --> 00:11:52.279 have to talk about the details of it, but like you know, kind 146 00:11:52.320 --> 00:11:58.159 of just like giving memory to like agents, right, which I thought was 147 00:11:58.399 --> 00:12:03.799 really interesting idea. And you know how because a lot of people think, 148 00:12:03.919 --> 00:12:07.519 like, you know, to your point of like making a copywriter or Dungeon 149 00:12:07.639 --> 00:12:11.519 dragons master, right, like these are all kind of like just roles, 150 00:12:11.639 --> 00:12:13.000 and a lot of people think, oh, like you could just do that 151 00:12:13.039 --> 00:12:18.320 with chat GPD, give it a system message or precluding user messages to say 152 00:12:18.519 --> 00:12:22.000 hey, you're now this role like and a lot of people just say, 153 00:12:22.000 --> 00:12:26.559 okay, well, why would you need any other service or any other like 154 00:12:26.159 --> 00:12:31.559 you know, advanced thing where CHATBD is just going to do that for you, 155 00:12:31.679 --> 00:12:33.879 Like, you know, what are you doing on top of that? 156 00:12:35.000 --> 00:12:39.039 Right? Like what is because there's like a lot of like yeah, the 157 00:12:39.159 --> 00:12:45.080 key insight there and which is really the foundation of why me and many other 158 00:12:45.080 --> 00:12:52.159 people can do startups on this stuff. Is that prompt engineering. To us 159 00:12:52.200 --> 00:12:56.320 as programmers seems like no big deal. It's just how you use the tool. 160 00:12:58.519 --> 00:13:03.600 But to a person is not a programmer, it really really seems like 161 00:13:03.679 --> 00:13:09.240 programming, which means they're not interested in doing it or not very good at 162 00:13:09.279 --> 00:13:11.600 it. You know, it's just not the way their brain works. So 163 00:13:11.960 --> 00:13:18.679 when you give them an option, it's like our our profile at Olympia. 164 00:13:18.720 --> 00:13:22.399 Like our heaviest power users that we've interviewed, a lot of them are non 165 00:13:22.399 --> 00:13:26.320 technical. They're not they're not programmers. They are early adopters. They're very 166 00:13:26.360 --> 00:13:33.039 excited by the technology and they follow religiously, but they're not so interested in 167 00:13:33.120 --> 00:13:37.960 learning the intricacies in and outs of a prompt engineering And I mean, yeah, 168 00:13:39.960 --> 00:13:41.200 no, I'm just I kind of want to bring it up a level 169 00:13:41.279 --> 00:13:46.519 or two because you know, we went right into kind of the the GPT 170 00:13:46.279 --> 00:13:52.000 largely model and you know and chat GT GPT like things and prompt engineering, 171 00:13:54.720 --> 00:13:58.320 and so I'm just wondering, as far as you know your book and some 172 00:13:58.360 --> 00:14:01.799 of these other things, is that the hope of your book is covering that 173 00:14:01.840 --> 00:14:07.000 stuff and maybe you don't go into things like transcriptions with whisper or you know, 174 00:14:07.240 --> 00:14:09.879 computer vision or things like that. And the other question that I have 175 00:14:11.039 --> 00:14:13.639 is is is that where people can jump in and how do they do that 176 00:14:15.360 --> 00:14:18.919 right? Because because I want to make it very approachable for people to go, 177 00:14:18.000 --> 00:14:20.559 huh, this is the Okay, I'm in a new way. I 178 00:14:20.720 --> 00:14:24.840 this is this is where I jump in. And yeah, I don't know 179 00:14:24.879 --> 00:14:31.480 if prompt engineering is the way to go and anyway, so here's my answer 180 00:14:31.519 --> 00:14:37.919 to that and the read. The thing that kind of tipped me over into 181 00:14:37.000 --> 00:14:43.639 like, Okay, I should start this book now was that an old friend 182 00:14:43.639 --> 00:14:48.320 of mine, Tom Mornini, who might be listening to the show. We 183 00:14:48.559 --> 00:14:50.279 were at a conference. I was giving a talk on the eye, and 184 00:14:50.519 --> 00:14:52.960 he was in the audience and raised his hand at the end of the talk 185 00:14:54.039 --> 00:14:56.120 and said, where do I get started? Like this is also overwhelming, 186 00:14:58.039 --> 00:15:01.440 and that that was a spark where I was like, Okay, I got 187 00:15:01.440 --> 00:15:05.039 to start working on this, and I literally, I literally start working on 188 00:15:05.080 --> 00:15:11.279 the way home, on the flight, and then just dedicated a huge amount 189 00:15:11.279 --> 00:15:15.320 of hours to try and get as much done as possible on it as quickly 190 00:15:15.320 --> 00:15:24.080 as possible. The reason is my belief, and I'm not sure how many 191 00:15:24.120 --> 00:15:28.240 other people necessarily share this, but I mean think it's trying to spread is 192 00:15:28.279 --> 00:15:33.759 that you don't have to start an AI project in order to use this stuff, 193 00:15:33.879 --> 00:15:35.240 Like you don't have to have like an idea for an AI startup or 194 00:15:35.279 --> 00:15:43.480 something like that. Like almost everyone who's currently maintaining or developing an application on 195 00:15:43.279 --> 00:15:52.200 Ruby unrails can probably get some benefit from taking a particularly complex part of their 196 00:15:52.679 --> 00:16:02.720 code and dropping in a discrete AI component to handle it so that you have 197 00:16:02.720 --> 00:16:15.759 an example. So the example that the example that I like using. So, 198 00:16:15.759 --> 00:16:22.840 so okay, anything that involves user generated content, for instance, you 199 00:16:22.919 --> 00:16:26.919 mean like a blog, like a blog with comments and whatnot. And I 200 00:16:26.960 --> 00:16:30.120 realized that there's services that that you can use to to do what I'm about 201 00:16:30.159 --> 00:16:36.000 to describe. But I mean anything involves evaluation of user generated content. Right 202 00:16:36.000 --> 00:16:42.440 now you can do very very easily with using large language models, and you 203 00:16:42.480 --> 00:16:47.559 know existing APIs that that's kind of a no brainer use case. But but 204 00:16:47.679 --> 00:16:51.440 when I go into the book and I've been I've been using it as an 205 00:16:51.480 --> 00:16:56.559 example and talks and whatnot, is that you can actually take parts your code 206 00:16:57.840 --> 00:17:02.720 that are implemented code. But if you had a little person that you could 207 00:17:02.759 --> 00:17:07.599 stick into your code, then you might want to use you a little person 208 00:17:07.640 --> 00:17:11.480 instead. So in Olympia, the example I give is the example of the 209 00:17:11.519 --> 00:17:21.720 account manager class, and the account manager class handles a few different operations involving 210 00:17:21.960 --> 00:17:30.799 changing a customer's account, most notably adding an assistant to your account. So 211 00:17:30.000 --> 00:17:36.480 internally, so in the user interface, this is this works with a button, 212 00:17:36.880 --> 00:17:41.000 so like you you look through a catalog of users that are available to 213 00:17:41.039 --> 00:17:45.160 add. We call them clones. YadA YadA. But when the user presses 214 00:17:45.160 --> 00:17:49.119 that button, internally, the internal API uses plane plane language. The internal 215 00:17:49.200 --> 00:17:56.079 API tells the account manager user with this ID would like to add this bot 216 00:17:56.079 --> 00:18:03.839 config with this ID to their account. And if I here, I'll just 217 00:18:04.000 --> 00:18:10.200 I'll just pull up the the actual prompts. I can read it to you. 218 00:18:12.160 --> 00:18:18.519 So it sounds like though you've got some form of interaction that I, 219 00:18:18.720 --> 00:18:25.720 as your user, do and that triggers essentially a prompt to be sent, 220 00:18:25.799 --> 00:18:30.960 Right, it's not this explicit function or method call that there's a Yeah, 221 00:18:30.960 --> 00:18:37.559 there's a component in my app which has the role of account manager, and 222 00:18:37.119 --> 00:18:45.240 it has a selection of functions that it can call, and those functions can 223 00:18:45.559 --> 00:18:52.799 mutate the customer's account. Right. It also has functions for communicating with the 224 00:18:52.839 --> 00:18:59.079 customer or with an account service rep or both. And it also has a 225 00:18:59.160 --> 00:19:04.240 finalized because the request is encapsulated in a change request object, which is tracked 226 00:19:04.240 --> 00:19:08.680 in the database. So when it's done handling the change request calls finalized which 227 00:19:10.319 --> 00:19:14.039 persists that along with the transcript of what happened. But it gives you a 228 00:19:14.079 --> 00:19:19.799 tremendous amount of flexibility and also it saves you time development time. So I 229 00:19:19.839 --> 00:19:23.680 wrote this very very quickly. I mean, once you identify the pattern, 230 00:19:25.480 --> 00:19:30.759 you start wanting to look for places to use it. And then the prompt 231 00:19:32.160 --> 00:19:38.559 reads like the requirements for the code. Okay, So the way that you 232 00:19:38.680 --> 00:19:41.880 described this, I think it's okay to do this on a podcasts. I 233 00:19:41.920 --> 00:19:44.440 think people are listening. So this is the this is the prompt for my 234 00:19:44.519 --> 00:19:51.079 account manager component. Okay. You are an account manager for Olympia, a 235 00:19:51.119 --> 00:19:56.359 platform that provides AI powered consultants and personal AI clones. Your role is to 236 00:19:56.400 --> 00:20:02.000 assist Olympia Clone owners by answering their questions, help them configure and manage their 237 00:20:02.039 --> 00:20:06.359 e clones, and executing tasks related to the clones and Olympia platform. That 238 00:20:06.440 --> 00:20:10.119 itself is an approach or pattern that's described in the book. You're basically giving 239 00:20:10.119 --> 00:20:15.279 a role to the AI. So and then now here come the instructions. 240 00:20:15.319 --> 00:20:18.880 When interacting with clone owners. You should provide clear, concise, helpful responses 241 00:20:18.920 --> 00:20:23.440 to the inquiries about account account management, building platform features in clone configuration. 242 00:20:25.160 --> 00:20:26.599 The reason I can do that is because I give it functions to be able 243 00:20:26.640 --> 00:20:33.160 to look up that information and convey it to the customer assist with tasks such 244 00:20:33.200 --> 00:20:37.400 as initiating client conversations with their I clones, modifying. Oh, I think 245 00:20:37.440 --> 00:20:42.559 I picked up there. I'm bringing the prop from the clone manager, not 246 00:20:42.599 --> 00:20:48.000 the account manager point. Let me let me take a step back for a 247 00:20:48.039 --> 00:20:49.799 second. I think if you're listening to what I'm just saying, you kind 248 00:20:49.839 --> 00:20:52.640 of get the gist of it. You can give instructions to these things. 249 00:20:52.839 --> 00:20:57.920 There's another one which has smaller scope. This one's to help specifically with people 250 00:20:59.000 --> 00:21:03.240 that run clones on Olympia. There's one that is an account manager for regular 251 00:21:03.240 --> 00:21:07.240 customers, and one thing it can do is add a particular botcom fig to 252 00:21:07.319 --> 00:21:12.839 your account. We rolled this out and within a week week and a half, 253 00:21:14.240 --> 00:21:18.720 a JPR customer service guy goes, hey, we have a problem with 254 00:21:18.920 --> 00:21:25.920 the clone adding. Someone added a clone and their account is inactive. Would 255 00:21:25.960 --> 00:21:30.279 you like me to create a notion story for it? And I opened the 256 00:21:30.359 --> 00:21:33.160 code while I was chatting with him and Slack about it, and I looked 257 00:21:33.160 --> 00:21:36.279 at it and I said, no, I think I got it. And 258 00:21:36.359 --> 00:21:48.960 I added in a statement to the prompt which said only allow addition of botcomfigs 259 00:21:49.000 --> 00:21:53.519 if your account is active, I literally just type that into the prompt of 260 00:21:53.519 --> 00:22:00.640 the account. Oh you get what I'm saying. You didn't have to add 261 00:22:00.640 --> 00:22:03.559 a conditional to the code, you know, extra checks or unit tests or 262 00:22:03.839 --> 00:22:11.480 yeah anything, modify your pundit scope or your pundit rules. Yeah. I 263 00:22:11.559 --> 00:22:19.839 literally just enhanced the instructions to the account manager and did a quick manual test 264 00:22:19.880 --> 00:22:25.000 and yeah, it worked, and we're good to go. And in fact, 265 00:22:25.880 --> 00:22:27.240 you know, if you think about what you would do traditionally with that, 266 00:22:27.440 --> 00:22:30.000 like you might you know, if you have a state which is invalid 267 00:22:30.000 --> 00:22:33.440 like that, then you might have to actually like write an email template for 268 00:22:33.640 --> 00:22:37.480 it, you know, in email or and whatnot. And in this case, 269 00:22:37.519 --> 00:22:42.279 you don't because it's instructed to communicate with the user and let it know 270 00:22:42.319 --> 00:22:47.839 what happened with the operation. So if their account is not active and it 271 00:22:47.880 --> 00:22:49.720 doesn't add the bout configura, we'll send them an email and say, sorry, 272 00:22:51.119 --> 00:22:52.720 your account's not active, soul, I couldn't do the thing you asked 273 00:22:52.720 --> 00:23:02.599 me to do. So going back to the idea of getting started, then, 274 00:23:03.000 --> 00:23:04.920 I mean, is that is is this the kind of thing you're encouraging 275 00:23:04.920 --> 00:23:07.680 people to get started with is just a I don't know. I mean, 276 00:23:07.680 --> 00:23:14.119 this is at GBT and start writing prompts and it doesn't use chat g BT 277 00:23:14.319 --> 00:23:18.920 uses it uses uh, you know, the API, and there are there 278 00:23:18.960 --> 00:23:25.799 are I have my own framework called Ray and there's other you know, frameworks 279 00:23:25.839 --> 00:23:30.279 for for how to use this. You could you could go at it at 280 00:23:30.319 --> 00:23:33.519 different abstraction levels. Right, So I mean, but for the most part, 281 00:23:33.599 --> 00:23:40.519 you need to understand the chat completion operation. So this this concept of 282 00:23:41.400 --> 00:23:45.680 you you give a system directive that kind of sets the stage for what you 283 00:23:45.680 --> 00:23:48.839 want the AI to do, and then you have a conversation with it where 284 00:23:49.119 --> 00:23:52.240 where you say something and then they say something and then you respond so and 285 00:23:52.160 --> 00:23:56.119 and so forth. So you have to understand that modality. You can engage 286 00:23:56.119 --> 00:24:03.680 with that modality at a couple different levels of traction. Those abstraction layers are 287 00:24:03.680 --> 00:24:07.759 provided to you by a library. So for instance, right, the one 288 00:24:07.799 --> 00:24:11.599 that describe in the book, And I mean you can go whole hog, 289 00:24:11.640 --> 00:24:15.079 which is how I would characterize the account manager example I've been talking about. 290 00:24:15.119 --> 00:24:21.759 But you can also do much much smaller things. So like one of the 291 00:24:21.759 --> 00:24:25.680 patterns in the book is the predicate pattern, and that's just a small component 292 00:24:25.680 --> 00:24:30.079 that answers a question yes or no. We answer yes no questions all the 293 00:24:30.079 --> 00:24:38.200 time in our code, but they're usually very very narrow scope, and if 294 00:24:38.200 --> 00:24:41.440 they're not narrow scope, then they tend to be brutal or difficult to maintain 295 00:24:41.640 --> 00:24:45.920 because pretty much any unanticipated state will break it. Right, And one of 296 00:24:45.960 --> 00:24:51.640 the great things about AI is that if you implement a component like this with 297 00:24:51.640 --> 00:24:56.720 with an l M back end, it can sometimes handle unexpected situations very very 298 00:24:57.079 --> 00:25:06.400 smoothly. You know. Yes, I think you bring up a great abstract 299 00:25:06.440 --> 00:25:10.559 point in it's a lot See, there's a lot of people that are like, 300 00:25:10.559 --> 00:25:12.240 oh, no, like, you know, all this AI stuff is 301 00:25:12.240 --> 00:25:17.359 coming from my job in X right like, and programmers for sure, like 302 00:25:17.880 --> 00:25:18.920 like you see this and you're like, oh, why don't we just have 303 00:25:18.960 --> 00:25:23.880 a programmer bot and it takes requests and we just you know, chat to 304 00:25:23.960 --> 00:25:27.200 it and it'll you know, make our application for us, right like, 305 00:25:27.440 --> 00:25:32.519 which is not totally unreasonable, but is unreasonable in a lot of ways. 306 00:25:33.559 --> 00:25:36.599 Uh, you know, hallucinations come to mind, like how do you make 307 00:25:36.599 --> 00:25:41.920 sure that the account manager right like just doesn't start deleting things? And right 308 00:25:41.119 --> 00:25:45.039 like that there's like a rogue function that it's created along the way, and 309 00:25:45.920 --> 00:25:48.599 now it's part of that tool chain. Like I know, there are a 310 00:25:48.640 --> 00:25:52.880 lot of checks that you can put in place, you know, to prevent 311 00:25:52.920 --> 00:25:57.200 that kind of stuff. But you know, a lot of it is just 312 00:25:57.240 --> 00:26:02.720 like getting natural language, though brought to whatever that you are doing, to 313 00:26:02.799 --> 00:26:07.599 compliment. It's how I've seen it worked most effectively, which seems to be 314 00:26:07.680 --> 00:26:11.160 how you've kind of created your framework here for rey, right, whereas it's 315 00:26:11.240 --> 00:26:17.519 complimenting the experience that you already have as a real developer, just making it 316 00:26:17.559 --> 00:26:22.559 easier to take advantage of this new tool, which is really AI. Do 317 00:26:22.599 --> 00:26:26.119 you see it that way or do you also see like maybe it is coming 318 00:26:26.119 --> 00:26:29.200 from your job. No, I don't think well, I don't think it's 319 00:26:29.240 --> 00:26:33.720 coming for any of our jobs anytime soon. And it most recently I got 320 00:26:33.759 --> 00:26:42.960 I got access to something called GitHub Copilot workspace. You got to sign up 321 00:26:42.960 --> 00:26:48.519 for it. There's a waiting list. It's a preview product where the interface 322 00:26:48.599 --> 00:26:52.960 is super cool and like kind of the promise of it is super cool. 323 00:26:52.880 --> 00:27:02.000 Essentially, you attach it to your repo and then you start a session and 324 00:27:02.039 --> 00:27:06.359 in the session, you go I need to fix this issue, you know, 325 00:27:06.480 --> 00:27:07.799 or I need to add this feature. You describe what it is in 326 00:27:08.079 --> 00:27:12.519 plain language, and it scans the code and then it comes up with a 327 00:27:12.519 --> 00:27:18.160 plan of action, so a checklist of like I need to do this and 328 00:27:18.200 --> 00:27:22.680 this and this and this and changes this class and so on and so forth. 329 00:27:22.920 --> 00:27:26.960 It kind of looks like what a discipline developer would do when you sit 330 00:27:27.039 --> 00:27:32.799 down because you picked up a feature or an issue. So far, so 331 00:27:32.880 --> 00:27:36.279 good. I was really impressed with that part, and then I said, 332 00:27:36.480 --> 00:27:42.359 okay, go right, Like I asked it to change a particular class, 333 00:27:42.359 --> 00:27:48.400 and maybe I was too ambitious. I wanted it to refactor, you know, 334 00:27:48.519 --> 00:27:56.079 three different, three different, very similar implementations into one thing that I 335 00:27:56.160 --> 00:28:00.000 want to get into bloody details. But so I had go. And then 336 00:28:00.000 --> 00:28:03.920 I thought about it for a while, and then it pulled up the code 337 00:28:03.000 --> 00:28:07.640 and I saw the changes in action, so that part's super cool. And 338 00:28:07.680 --> 00:28:11.079 then it highlighted a bunch of the code, the existing code, and it 339 00:28:11.160 --> 00:28:17.960 deleted it, and then it put in a bunch of nonsensical methods to replace 340 00:28:18.000 --> 00:28:21.960 it that had nothing to do with what it was supposed to do. And 341 00:28:22.640 --> 00:28:27.839 I was most definitely underwhelmed. And now it is very very possible that this 342 00:28:27.880 --> 00:28:37.000 works much better with Python and typescript, but it doesn't know the rails way, 343 00:28:37.359 --> 00:28:42.880 that's for sure. Because it was garbage. It was hot garbage. 344 00:28:42.880 --> 00:28:47.839 I tried to work with it a little bit, you know, And so 345 00:28:48.119 --> 00:28:52.559 now this needs to get much better, and there are I think very credible 346 00:28:52.599 --> 00:28:57.920 reasons to think that it might not get that much better anytime soon. I 347 00:28:57.920 --> 00:29:00.559 mean, I think I think right now. One of the lessons I've learned 348 00:29:00.599 --> 00:29:07.640 that Olympia is you have to throw a lot of brute force to get to 349 00:29:07.680 --> 00:29:11.799 get a usable result, you know, in terms of these things being quote 350 00:29:11.880 --> 00:29:15.400 unquote intelligent and having memory and figuring out what you want and being able to 351 00:29:15.400 --> 00:29:19.000 recalibrate. And the reason I'm able to do that is thanks to the race 352 00:29:19.039 --> 00:29:23.759 to the bottom and the price of commodity AI models. So a lot of 353 00:29:23.759 --> 00:29:27.440 these open source models that you hear about how mix Droll and Lama three and 354 00:29:27.799 --> 00:29:30.920 you know, Wizard LM from Microsoft, so on and so forth. So 355 00:29:30.960 --> 00:29:40.559 there's all there's all these model providers competing two you know, to in this 356 00:29:40.680 --> 00:29:45.000 market for AI services, and they're driving the price to zero and the performance 357 00:29:45.079 --> 00:29:48.079 up. So that's great. You can access a lot of this stuff, 358 00:29:48.519 --> 00:29:52.960 and I do think that the price will keep dropping for a lot of these 359 00:29:52.960 --> 00:29:57.160 things to the point where it'll start to make sense actually incorporate it into your 360 00:29:57.240 --> 00:30:02.319 run time, know, like at the rails level, and then maybe in 361 00:30:02.319 --> 00:30:07.359 the future at the Ruby level or even the virtual machine model for doing things 362 00:30:07.359 --> 00:30:11.759 like error recovery, or you know, like if you want to crank the 363 00:30:11.799 --> 00:30:15.680 dial up, Like if you're okay with uncertainty and nondeterministic behavior with some of 364 00:30:15.759 --> 00:30:18.799 us, are, you know, you can crank the dial up and have 365 00:30:18.880 --> 00:30:23.240 it do things like you know, cover for you if you give it the 366 00:30:23.279 --> 00:30:29.359 wrong type, or you know, try to automatically handle moll pointers and you 367 00:30:29.400 --> 00:30:33.319 know, you know, things like that. There's a lot of really interesting, 368 00:30:33.559 --> 00:30:38.240 magical, dangerous ship that I think we'll start getting into, you know, 369 00:30:38.279 --> 00:30:42.839 as we get more comfortable with these technologies and as the cost comes down, 370 00:30:42.880 --> 00:30:47.920 because right now the cost tends to be somewhat prohibitive, especially especially if 371 00:30:48.039 --> 00:30:52.319 using frontier models, which is what they call GPT for or or clad. 372 00:30:52.759 --> 00:30:59.480 But but the smaller faster models, man, they're so cheap. I mean, 373 00:30:59.680 --> 00:31:06.200 it's like point zero zero four cents you know per request, which for 374 00:31:06.559 --> 00:31:10.920 Olympia. Uh, you know the scale that we're at is okay, you 375 00:31:10.960 --> 00:31:12.880 know, that's a that's a manageable bill. You know, if you're if 376 00:31:12.880 --> 00:31:18.359 you're a much bigger company than maybe some of the things that I'm that I'm 377 00:31:18.400 --> 00:31:22.519 suggesting literally don't scale. I mean they end up costing you a thousand dollars 378 00:31:22.880 --> 00:31:27.839 for more on the topic of model choice, Like, I guess I have 379 00:31:27.880 --> 00:31:32.880 two questions around that, like how do you go about finding the right model 380 00:31:33.160 --> 00:31:37.839 for what you're trying to do? And how do you like, what is 381 00:31:37.960 --> 00:31:45.640 that exploration process? Like what is the what process? Like what's the exploration 382 00:31:45.720 --> 00:31:52.079 process for exploration? Yeah? Yeah, Well I'm a big fan of a 383 00:31:52.079 --> 00:31:56.119 service called open Router, and I wish that iniquity stake because I'm evangelizing them 384 00:31:56.119 --> 00:32:04.759 all the time. But they, uh they power the power number really popular 385 00:32:05.000 --> 00:32:12.440 AI based services like Silly Tavern and uh a new product that I'm using which 386 00:32:12.480 --> 00:32:20.839 i'll plug called novel Crafter. Uh. They're super awesome. And the expiration 387 00:32:20.920 --> 00:32:24.880 process is basically going to open routers playground and being able to compare uh you 388 00:32:24.880 --> 00:32:32.319 know, run prompts and parallel and see what happens with different and AI models, 389 00:32:32.599 --> 00:32:37.079 you know, like basically saying, okay, here's GPT four as my 390 00:32:37.519 --> 00:32:42.839 benchmark, and then try to run the same thing against SLAMA three and mix 391 00:32:42.839 --> 00:32:49.440 stroll and whatever. So I'm down the line. And the main reason he's 392 00:32:49.480 --> 00:32:52.519 open router and I featured in my pro and most of the open source I've 393 00:32:52.519 --> 00:32:55.200 done, you know, a half dozen open source things I've done, is 394 00:32:58.079 --> 00:33:01.519 that it's one a p I key, you know, one end point one 395 00:33:01.559 --> 00:33:07.079 account to manage and pay for, and it gives you access to everything open 396 00:33:07.119 --> 00:33:14.839 Ai, Claude, Gemini Pro, you know cohere, like all all the 397 00:33:14.880 --> 00:33:17.599 major models, and then a whole bunch of open source models including once you 398 00:33:17.720 --> 00:33:23.039 never heard are available there. It's probably have a couple of hundred now and 399 00:33:23.720 --> 00:33:29.160 for the ones that are offered by multiple providers, so so like the example 400 00:33:29.200 --> 00:33:34.000 of provider would be like DPT four is offered by open ai itself, but 401 00:33:34.079 --> 00:33:39.240 it's also open offered by a zor. So one of the things I really 402 00:33:39.279 --> 00:33:45.200 like about open router is that if GPT four is down for whatever reason, 403 00:33:45.799 --> 00:33:50.920 it can automatically fail over to a zor, which means I don't have to 404 00:33:52.039 --> 00:33:55.599 have a Zoro account, which is great because I don't like premature aging, 405 00:33:55.759 --> 00:34:01.680 so the you know, Zor is very very difficult to set up and work 406 00:34:01.720 --> 00:34:07.359 with and everything like that. So yeah, but you were asking the selection 407 00:34:07.440 --> 00:34:12.360 process is trial and error. I mean, I think there's some there's someone 408 00:34:12.400 --> 00:34:15.119 in the Rubai discord that's working on some I think it's called prompt Garage or 409 00:34:15.519 --> 00:34:22.639 something like that, which gives you some sort of framework for for doing this 410 00:34:22.719 --> 00:34:27.280 in a methodical fashion. But it's pretty easy to like once, once you 411 00:34:27.400 --> 00:34:30.800 got it going, it's pretty easy to just plug in a model name and 412 00:34:30.800 --> 00:34:32.639 see if it works. And I do tend to TDD these things. I 413 00:34:32.639 --> 00:34:36.639 don't leave them in my suite, or if I do leave them in my 414 00:34:36.719 --> 00:34:40.519 suite, they're backed by BCRs so that they run quick. But I do 415 00:34:40.559 --> 00:34:47.239 tend to TDD my my AI components. So I'll just sit there and iterate 416 00:34:47.880 --> 00:34:52.440 you know, red green until I get what I need. And one of 417 00:34:52.440 --> 00:34:54.960 those things that you change while you're trying to get it to work is the 418 00:34:55.000 --> 00:35:00.519 model. Trying to get the cheapest, fastest model that will consistently work. 419 00:35:02.159 --> 00:35:07.000 So do you walk through this process in your book? Yeah? Yeah, 420 00:35:07.000 --> 00:35:13.840 that's fair. Yeah, the whole first half of the book is kind of 421 00:35:13.880 --> 00:35:17.199 approaches, and then the second half of the book is like concrete. This 422 00:35:17.360 --> 00:35:21.239 is you know, this is a pattern, this is the name of it, 423 00:35:21.360 --> 00:35:24.960 this is how you use it. Here's example code. Yeah. I 424 00:35:24.960 --> 00:35:30.400 will say I haven't made my whole way through the book, but I've made 425 00:35:30.440 --> 00:35:35.960 a decent amount. And I do like how you keep reiterating on like the 426 00:35:36.000 --> 00:35:38.800 discrete components aspect of it, which I think is reflected well. And you're 427 00:35:38.960 --> 00:35:45.960 like ray framework right where you're just trying to, you know, encapsulate whatever 428 00:35:46.000 --> 00:35:50.800 the task at hand is and keep things small. And I think that's kind 429 00:35:50.800 --> 00:35:53.159 of reflective of how the models work in general, is like you give it 430 00:35:53.199 --> 00:35:57.519 too much and it's going to make more mistakes, which I think is true 431 00:35:57.519 --> 00:36:01.119 of people too. Yeah. Yeah, I think of them like little people. 432 00:36:01.159 --> 00:36:05.320 There's a chapter in the book called Multitude of Workers, because that that's 433 00:36:05.360 --> 00:36:07.440 the approach that I advocate. It's one way of thinking about it. It's 434 00:36:07.519 --> 00:36:12.800 like, you know, if you had little people that you could put not 435 00:36:12.920 --> 00:36:16.079 little people, but you know, if you had you know, human little 436 00:36:16.119 --> 00:36:22.599 humanized bits of code, you know that you could give jobs, but they 437 00:36:22.599 --> 00:36:27.760 have to be narrow jobs at the moment. You're absolutely right, Valentino. 438 00:36:27.800 --> 00:36:30.840 You give these things too much surface area, you know, too much to 439 00:36:30.880 --> 00:36:36.039 think about, and they mess up. The only model that I know right 440 00:36:36.079 --> 00:36:42.239 now that can reliably handle more than one or two tasks or functions or you 441 00:36:42.239 --> 00:36:46.800 know, just that kind of latitude, that kind of discretionary behavior is GPT 442 00:36:46.960 --> 00:36:55.840 four. Supposedly Opus from Claude can do it. But their tool their their 443 00:36:55.880 --> 00:37:00.960 framework for invoking functions is more immature. But like some of our some of 444 00:37:00.960 --> 00:37:08.840 our bots in Olympia have twenty twenty four functions available to them for every single 445 00:37:08.840 --> 00:37:12.760 request. So anytime you're talking to them, anytime you send them a prompt, 446 00:37:12.800 --> 00:37:17.840 they have twenty or more functions that they can call for things like checking 447 00:37:17.880 --> 00:37:25.719 system status, for sending email, for recalling things from memory, real time 448 00:37:25.800 --> 00:37:30.800 search, Google search, image search, image generation, you know, so 449 00:37:30.840 --> 00:37:32.599 on and so forth. This is kind of like their palette, their tool 450 00:37:32.599 --> 00:37:40.679 belt of things that they can use to in book and there's yeah good. 451 00:37:42.320 --> 00:37:45.400 I'm curious like the testing approach here, right, Like, is it does 452 00:37:45.440 --> 00:37:51.000 it start to like take a different shape than traditional web development? Right? 453 00:37:51.199 --> 00:37:54.559 Like? Is it more like because once you get to the multitude of workers, 454 00:37:54.639 --> 00:37:58.719 right, and you have more than one. I could see like encapsulating 455 00:37:58.760 --> 00:38:01.800 one as like a thing like maybe like a rails model or something like that. 456 00:38:02.519 --> 00:38:06.440 And you have like unit tests that make sure that it says what you 457 00:38:06.519 --> 00:38:09.039 wanted to in a certain format or whatever. But how does like the orchestration 458 00:38:09.119 --> 00:38:14.480 aspect, does that also translate the same way to like the controller flow, 459 00:38:14.920 --> 00:38:17.199 like to you know, feature tests or something like that. Is there a 460 00:38:17.239 --> 00:38:22.320 parallel to that multitude of workers working together or do you kind of just like 461 00:38:22.199 --> 00:38:30.039 is that not really something worth doing? I haven't gotten too far down the 462 00:38:30.079 --> 00:38:37.559 rabbit hole in terms of trying to do automated tests for intelligent workflows. There's 463 00:38:37.559 --> 00:38:44.840 no reason that you wouldn't, you know, if you were so inclined trying 464 00:38:44.880 --> 00:38:52.760 to think of how to intelligently answer that question. Speaking of a rabbit hole, 465 00:38:52.800 --> 00:38:57.639 I mean, yeah, you can start combining these things and it can 466 00:38:57.679 --> 00:39:06.360 start getting pretty micro service feeling. I realized that a couple of months ago. 467 00:39:06.880 --> 00:39:10.559 It started reminding me of some of the servilests approaches. You know, 468 00:39:12.039 --> 00:39:17.480 like I was really really hot on micro services powered with Lambda when the Services 469 00:39:17.599 --> 00:39:22.760 framework came out. So this is back like in twenty fifteen time frame twenty 470 00:39:22.760 --> 00:39:30.079 sixteen. I started writing a book on it, and it starts to it 471 00:39:30.119 --> 00:39:35.440 starts to feel like you have little micro services inside of your monolith when you 472 00:39:35.440 --> 00:39:40.559 incorporate these and rails, because they have you know, they have a very 473 00:39:43.480 --> 00:39:47.519 what I want to call it, the way you interact with them, like 474 00:39:47.559 --> 00:39:52.559 they're well encapsulated because they're black boxes. You know, you literally can only 475 00:39:52.800 --> 00:40:01.039 do input and output of messages with them. So I think it's cool. 476 00:40:01.119 --> 00:40:05.119 I mean, I love that approach, and I don't I I'm a hacker 477 00:40:05.159 --> 00:40:08.199 at heart. I started as a hacker when I was young, so I 478 00:40:08.239 --> 00:40:12.760 don't have a lot of hang ups about like this has to be absolutely correct. 479 00:40:13.280 --> 00:40:15.440 And actually I think you're probably a little too worried about hallucinations, Valentino, 480 00:40:15.559 --> 00:40:20.960 just from what what it sounds like, because for a lot of these 481 00:40:21.079 --> 00:40:23.320 use cases for discrete AI components, I mean, it is, well, 482 00:40:23.639 --> 00:40:30.280 since since you've since you've put a lot of constrains, constraints on what you 483 00:40:30.440 --> 00:40:35.159 want it to be. You know what you wanted to do, what kind 484 00:40:35.159 --> 00:40:39.239 of output you wanted to give, and you you crank the temperature parameter down 485 00:40:39.280 --> 00:40:45.119 to zero, which reduces the amount of creativity that the model, it like 486 00:40:45.239 --> 00:40:51.880 constrains the possible choices of the model down significantly. That brings me to another 487 00:40:52.159 --> 00:40:55.559 question that I feel like I always get, is like, you know, 488 00:40:55.880 --> 00:41:00.880 how do you find out what it's good at and what it is like it? 489 00:41:00.199 --> 00:41:04.679 Are there like definitive things or is it kind of just like learn as 490 00:41:04.719 --> 00:41:09.000 you go aspects of it. Well, I've definitely learned as I go. 491 00:41:09.239 --> 00:41:16.000 I mean there's there's some things which are I think a little too meta, 492 00:41:17.320 --> 00:41:24.280 so like to maybe I just haven't hit on the right way to do this. 493 00:41:24.400 --> 00:41:30.559 But like we do automate automated evaluations of the conversations in Olympia, and 494 00:41:30.920 --> 00:41:35.599 you said you're Olympia user. So basically there's a process that kicks off after 495 00:41:35.760 --> 00:41:42.360 conversation idols for like fifteen twenty minutes, and there's a component that will look 496 00:41:42.360 --> 00:41:46.639 at the conversation and give it a numerical grade up to one hundred and say 497 00:41:46.679 --> 00:41:51.000 this is what the bot did well and this is what did not so well, 498 00:41:51.360 --> 00:41:55.840 And if it drops below a certain number, then it gets flagged and 499 00:41:57.199 --> 00:42:00.519 that goes to a dashboard. So we can see kind of like the overall 500 00:42:00.519 --> 00:42:04.480 performance of the system and also so we can be aware in case that person 501 00:42:05.679 --> 00:42:08.039 that user is having a lot of issues, then we can reach out to 502 00:42:08.039 --> 00:42:10.960 them and say, hey, it seems like you're having some issues with your 503 00:42:10.960 --> 00:42:15.960 Olympia assistance. You know, can we help. And what we find with 504 00:42:15.000 --> 00:42:20.159 these evaluations is that the model is very, very generous. It's like the 505 00:42:20.239 --> 00:42:22.599 nicest teacher you ever had. I mean, almost everyone gets ninety five, 506 00:42:22.599 --> 00:42:29.199 it doesn't matter how how bad. Like the worst I've seen again is seventy, 507 00:42:29.599 --> 00:42:34.000 you know, and that's where everything was falling apart completely, And it's 508 00:42:34.280 --> 00:42:38.920 very kind and generous to itself, which is like kind of a weird thing 509 00:42:39.000 --> 00:42:44.599 to say, but you know, it's like the bot really really tried. 510 00:42:50.079 --> 00:42:53.760 Yeah, so that I couldn't get to work right, I mean, there 511 00:42:53.760 --> 00:42:59.119 were other things used. There used to be a lot of this has gotten 512 00:42:59.159 --> 00:43:01.199 better, I think too, just mostly because GPT four has gotten better. 513 00:43:01.239 --> 00:43:07.320 But there used to be a lot of hallucinations around what we call stalling. 514 00:43:07.519 --> 00:43:14.960 So you'd ask your Olympia as system to help you research something and they'll say, 515 00:43:15.960 --> 00:43:19.639 on it, I'll get back to you in three hours. That's a 516 00:43:19.719 --> 00:43:22.960 hallucination. They have no ability to go do a sync work yet. I 517 00:43:22.039 --> 00:43:24.800 mean, we're giving it to them, but six months ago they did not. 518 00:43:27.079 --> 00:43:30.800 And then the person would be like, great, how will I know 519 00:43:30.840 --> 00:43:34.079 when you're done? They're like, I'll get back to you, or sometimes 520 00:43:34.159 --> 00:43:37.719 say i'll email you what is done. So they're really taking this human nice 521 00:43:37.760 --> 00:43:43.159 thing to heart, right, But that's not good behavior because then you're sitting 522 00:43:43.239 --> 00:43:46.760 there And we used to joke around sometimes that your Olympias system could turn into 523 00:43:46.800 --> 00:43:51.360 the worst employee you ever had, because I mean you we had some users 524 00:43:51.360 --> 00:43:53.159 that would just keep asking them is that done yet? No, I'm still 525 00:43:53.159 --> 00:43:59.519 working on It's been a week, are you what are you going to be 526 00:43:59.559 --> 00:44:02.880 done with it? So thankfully a lot of stuff has gone away. But 527 00:44:04.039 --> 00:44:12.760 but I wrote a bunch of components to try to detect it. And it 528 00:44:12.800 --> 00:44:20.519 gets into it gets to the limitations of these statistical models with regards to detecting 529 00:44:20.639 --> 00:44:24.119 patterns, you know, because what you're really asking to do is to to 530 00:44:24.239 --> 00:44:30.519 detect a certain pattern. But the past it's you can't describe the pattern concretely. 531 00:44:30.559 --> 00:44:36.920 You have to describe it very subjectively, you know, like and it 532 00:44:37.000 --> 00:44:39.400 just doesn't do well. But in fact, this there's this arc challenge, 533 00:44:42.000 --> 00:44:47.159 right, now, which is to try to squeeze a state of the art 534 00:44:47.440 --> 00:44:54.440 pattern matching you know, behavior out of AI. And there's there's like big 535 00:44:54.480 --> 00:44:58.800 prizes for whoever can get it because they suck at it. And it's really 536 00:44:58.800 --> 00:45:02.719 really interesting to see. If if you think GPT four is super intelligent and 537 00:45:02.719 --> 00:45:06.239 can reason and stuff like that, and you want to you want you to 538 00:45:06.320 --> 00:45:09.559 knock down your expectations a bit, give it one of these ARC challenges. 539 00:45:10.559 --> 00:45:14.320 And I was gonna say I was, I was messing around with this ARC 540 00:45:14.440 --> 00:45:17.480 challenge uh the other day and it yeah, I was like, oh yeah, 541 00:45:17.719 --> 00:45:21.360 like the large language models, they should be able to handle this, 542 00:45:21.440 --> 00:45:25.039 no problem, uh, And it's it's remarkable they can't. Like they fail 543 00:45:25.320 --> 00:45:34.320 like kryptonite you see, Superman like, which is funny because like you give 544 00:45:34.360 --> 00:45:38.079 it like the whole thing is like you can give it arbitrary structured data and 545 00:45:38.320 --> 00:45:42.000 like it has no problem like reason or not reason. I don't want to 546 00:45:42.119 --> 00:45:45.400 use that word, but like you know, pulling out the pieces that fit 547 00:45:45.480 --> 00:45:50.239 together of that structured data, but anything where it's like it has no reference 548 00:45:50.360 --> 00:45:53.840 to which I guess makes sense because that's how it's built, is based on 549 00:45:53.920 --> 00:45:58.719 reference material. But you give it anything, like you know that it has 550 00:45:58.800 --> 00:46:04.119 no understanding previous knowledge of and it can't really fit the pieces together, even 551 00:46:04.159 --> 00:46:07.559 though like to us it just seems so obvious, right, which, Yeah, 552 00:46:07.760 --> 00:46:12.960 you know, maybe that we need another modality of like pattern matching. 553 00:46:15.800 --> 00:46:22.800 Yeah, I mean there's theoretical modalities that include like real understanding of the language, 554 00:46:24.320 --> 00:46:30.039 but I don't see anyone working on that yet. So all statistical models 555 00:46:30.039 --> 00:46:37.119 are the flavorite that you are. I would like to just quickly like dive 556 00:46:37.639 --> 00:46:42.920 into uh like cause you mentioned a lot of it's just like you know, 557 00:46:43.440 --> 00:46:46.519 you learn as you go and it's trial and error. Like, is there 558 00:46:46.559 --> 00:46:52.239 any like standout pitfalls that people should avoid, like as they're diving in, 559 00:46:52.599 --> 00:47:01.920 like to just stay steer clear from doing X. Right. Yes, absolutely, 560 00:47:04.239 --> 00:47:13.840 I have a perfect example of something you absolutely should not do in your 561 00:47:13.880 --> 00:47:22.800 AI journeys. In your programming using AI journeys, you will inevitably write something 562 00:47:22.840 --> 00:47:30.679 that loops. So basically it does a check completion, and the check completion 563 00:47:30.800 --> 00:47:37.559 will probably call a function right to do some behavior external to the model, 564 00:47:37.599 --> 00:47:40.280 and then you have to loop so that the model can take the results of 565 00:47:40.320 --> 00:47:46.199 the function call and incorporate it into its final answer to the user or to 566 00:47:46.519 --> 00:47:52.599 the rest of your system in some way. You definitely want to put some 567 00:47:52.639 --> 00:48:00.039 sort of limit on that loop aka circuit breaker, because you can run, 568 00:48:00.280 --> 00:48:06.159 you can rack up a huge bill very very fast if if you make a 569 00:48:06.199 --> 00:48:09.519 mistake there where it just continues calling, you know, calling again and again, 570 00:48:09.639 --> 00:48:15.159 especially if you have large contexts like we do at Olympia. I had 571 00:48:15.199 --> 00:48:23.920 a embarrassing failure like that where some subset of our conversations would uh just keep 572 00:48:23.960 --> 00:48:28.800 looping and the retry the circuit breaker was broken, which is always said, 573 00:48:29.880 --> 00:48:36.320 test your circuit breakers really well. So we had all of a sudden, 574 00:48:36.360 --> 00:48:43.559 my my consumption shut up from you know, like two deviations, and I 575 00:48:43.599 --> 00:48:46.559 was like, looking at that, gone oh, and I go into the 576 00:48:46.599 --> 00:48:52.000 logs an open router open rider shows you kind of like tokens into out and 577 00:48:52.039 --> 00:48:58.440 the cost associated with it, and it's just the same number of tokens in 578 00:48:58.480 --> 00:49:01.079 and out just boom, just like multiple per second, all the way down, 579 00:49:01.159 --> 00:49:07.840 just filling the logs. Yeah, with GPT four it's like, oh 580 00:49:08.079 --> 00:49:14.000 no, Luckily the open router guys kind of helped me out a little bit, 581 00:49:14.119 --> 00:49:20.719 but yeah, that was bad. So, yeah, these things are 582 00:49:20.719 --> 00:49:22.800 expensive. I mean you're using you when're using the expensive models, they are 583 00:49:22.840 --> 00:49:29.039 expensive, HM, in a way that many of us are not used to, 584 00:49:31.480 --> 00:49:35.320 you know, because we're used to you know, heroicodyno costs an x 585 00:49:35.320 --> 00:49:38.760 amount and you kind of know what you're getting. You know a lot of 586 00:49:38.800 --> 00:49:44.719 other SaaS APIs you know, kind of costs a certain amount, like you 587 00:49:44.760 --> 00:49:47.719 get it. This thing kind of has like it's like a it's like a 588 00:49:47.760 --> 00:49:52.599 short squeeze. I mean, it's gonna like unlimited boom, Like you're down 589 00:49:52.639 --> 00:49:55.320 a few thousand dollars just because you've made a bug, you know, your 590 00:49:55.320 --> 00:49:59.079 code. So a lot of us are not used to that, So yeah, 591 00:49:59.159 --> 00:50:04.519 be careful of that. Yeah. One thing that I was so You've 592 00:50:04.559 --> 00:50:10.559 mentioned journeys and context, and I've heard people talk about training on top of 593 00:50:10.679 --> 00:50:14.960 kind of the large language model, which I'm assuming is sort of the same 594 00:50:15.000 --> 00:50:19.079 thing. Maybe I'm missing the boat on this, but yeah, if I 595 00:50:19.119 --> 00:50:22.079 have a specific use case where I wanted to, I don't know, refine 596 00:50:22.199 --> 00:50:27.480 or summarize you know, some of the information that it pulled in or things 597 00:50:27.519 --> 00:50:32.599 like that. I mean, how do I get it to be more specific? 598 00:50:32.639 --> 00:50:37.880 So let's say that I had a text that I was routinely putting into 599 00:50:37.920 --> 00:50:42.480 it, and you know, I wanted it to summarize. You know, 600 00:50:43.320 --> 00:50:45.400 I'm thinking, here's what I'm thinking. So I was talking to some friends 601 00:50:45.400 --> 00:50:52.639 of mine and they were somebody mentioned that it would be nice if there was 602 00:50:52.679 --> 00:50:58.639 something that would summarize the bills that go through the legislature every year in Utah. 603 00:50:59.119 --> 00:51:02.320 Right, And there's usually like stricken out language, a new language put 604 00:51:02.360 --> 00:51:07.920 in and it shows you the context of the Utah Code that it's going to 605 00:51:07.039 --> 00:51:14.440 modify. And so you know, it looks like the models that you've built 606 00:51:14.480 --> 00:51:16.280 on Olympia. Right. It can go and it can read this stuff off 607 00:51:16.320 --> 00:51:20.800 the internet, and you know, and there's a way to consume it. 608 00:51:21.320 --> 00:51:25.599 So how do I get it to be proficient at getting through that stuff and 609 00:51:25.599 --> 00:51:30.599 then giving me what I want to know without having to go and read the 610 00:51:30.719 --> 00:51:40.880 entire bill? So you might you might be thinking of a technique called fine 611 00:51:40.920 --> 00:51:45.800 tuning. So a lot of people think of a lot of people think of 612 00:51:45.840 --> 00:51:50.639 training, and there's different ways of making models better at a specific purpose. 613 00:51:51.320 --> 00:51:59.000 Right, And I experimented enough with it to kind of know how it works 614 00:51:59.039 --> 00:52:02.760 and what it's for. But I've found that you don't really need fine tuning 615 00:52:02.800 --> 00:52:06.719 for most use cases, like what you just mentioned, you wouldn't need fine 616 00:52:06.760 --> 00:52:09.400 tuning at all. Okay. I wouldn't have to teach it how to read 617 00:52:09.440 --> 00:52:14.880 a bill. I could just give it the language and it could say, 618 00:52:15.119 --> 00:52:20.719 oh, takes this out and puts this in DPT. Yeah, GPT four 619 00:52:20.880 --> 00:52:27.000 and Claude and Gemini pro like what they call frontier models and I call expensive 620 00:52:27.000 --> 00:52:35.159 ones. They know all this stuff already. In fact, what in the 621 00:52:35.199 --> 00:52:43.119 book I call the I call the process of applying a large language model to 622 00:52:43.360 --> 00:52:46.880 your problem domain, narrowing the path. So when you're when you're thinking about 623 00:52:47.119 --> 00:52:53.920 the the power of your eye model, one very useful thing to have clear, 624 00:52:54.000 --> 00:52:59.199 especially if you're completely new to this, is that it literally knows everything 625 00:52:59.320 --> 00:53:04.400 for all intents purposes. Okay. So the space of its knowledge, the 626 00:53:04.519 --> 00:53:09.719 universe of its knowledge, is so so big that if you present it with 627 00:53:09.880 --> 00:53:15.119 something that's which is your prompt or what you're asking to do, it had 628 00:53:16.440 --> 00:53:22.519 unless you constrain the scope. It is literally drawing from all of that knowledge. 629 00:53:23.280 --> 00:53:27.639 Okay. And if you think about all that knowledge, it not only 630 00:53:27.679 --> 00:53:35.400 includes knowledge of the Utah legislature and legislatures in general, and procedures of you 631 00:53:35.440 --> 00:53:42.199 know, governing bodies. It also includes legislative bodies and governing procedures of science 632 00:53:42.239 --> 00:53:47.159 fiction novels. Just to give you just kind of a clear example of why 633 00:53:47.239 --> 00:53:53.840 there's a problem here, right, So multiply that to times, you know, 634 00:53:54.199 --> 00:53:59.639 a near infinite amount of different you know, depending on how general what 635 00:53:59.639 --> 00:54:01.480 you're asking to do, it could draw upon way, way more than what 636 00:54:01.519 --> 00:54:06.320 you wanted to Okay, So so what I call narrowing the path is what 637 00:54:06.519 --> 00:54:12.400 probably essentially what you mean by training. It's but that's you, you know, 638 00:54:12.480 --> 00:54:16.639 figuring out what constraints to give it. And one of the main one 639 00:54:16.639 --> 00:54:21.079 of the main ways you can give it a constraint, which is the most 640 00:54:21.079 --> 00:54:25.719 popular way right now, is called retrieval augmented generation. So if you're dealing 641 00:54:25.760 --> 00:54:30.199 in factual data, that's called nonfiction. You know, if you're in a 642 00:54:30.239 --> 00:54:34.719 nonfiction domain, so you know, your legislature example is the case of that 643 00:54:36.000 --> 00:54:37.119 you could make it And I'm just kidding, Yeah, I guess, so 644 00:54:39.480 --> 00:54:46.480 I didn't say non fantasy, but not you know, not so the you 645 00:54:46.480 --> 00:54:52.639 you can tell the model like, hey, you're going to operate on. 646 00:54:53.039 --> 00:55:00.440 You're going to to generate output based on this specific content that I give you. 647 00:55:00.360 --> 00:55:04.599 You add that content dynamically at runtime. That's what retrieval augmented generation means. 648 00:55:05.280 --> 00:55:08.480 Retrieval means you're retrieving it from somewhere other than the latent space of the 649 00:55:08.480 --> 00:55:14.320 models, other outside of the model, so you're you're going out, you're 650 00:55:14.320 --> 00:55:20.360 retrieving it. So that might be like the existing legislation and then the revision, 651 00:55:21.159 --> 00:55:25.320 and then you you structure data pattern which is covering the book to present 652 00:55:25.400 --> 00:55:32.840 that in the prompt. And then because of their training, the way that 653 00:55:32.880 --> 00:55:37.639 these models are trained, they understand that one of the main use cases that 654 00:55:37.679 --> 00:55:42.880 they're used for is to operate on data that they're provided to from the prompt, 655 00:55:43.880 --> 00:55:47.360 and therefore they don't hallucinate it. I gotcha, you get it. 656 00:55:47.920 --> 00:55:51.760 In fact, there there's I haven't played around with this yet, but but 657 00:55:51.840 --> 00:55:57.039 your listeners might find it really interesting and if I can't find it in the 658 00:55:57.079 --> 00:56:00.920 next five seconds, I'll definitely and give it to you to put in the 659 00:56:00.920 --> 00:56:06.320 show notes. But essentially, there's a notebook. There's a notebook product that 660 00:56:06.400 --> 00:56:09.360 my team was was kind of geeking out over. I'm gonna have to find 661 00:56:09.360 --> 00:56:13.400 it and give you the link, but but I can describe what it does. 662 00:56:13.880 --> 00:56:17.280 It was super cool. Like you, you start out with a blank 663 00:56:17.280 --> 00:56:20.880 slate, and when I say blank slate, I mean literally blank slate. 664 00:56:21.320 --> 00:56:25.360 So you can you give it a prompt. You say, I want you 665 00:56:25.400 --> 00:56:30.199 to help me make a high school curriculum, and it goes, what's a 666 00:56:30.280 --> 00:56:37.000 high school? So right, you you have you have to provide that information. 667 00:56:37.119 --> 00:56:39.000 This is what I mean by high school? And it goes, great, 668 00:56:39.039 --> 00:56:44.400 what's a curriculum? And then you you have to describe. Okay, 669 00:56:44.400 --> 00:56:49.880 when I say curriculum, this is what I mean by curriculum. And I 670 00:56:49.920 --> 00:56:53.800 and I actually think that, uh this I hadn't seen anything like this until 671 00:56:53.800 --> 00:56:57.119 a week ago, and I think it might be new and novel. So 672 00:56:57.199 --> 00:57:00.119 it's interesting to think about whether this will take off. It's actually really really 673 00:57:00.119 --> 00:57:05.800 super useful. In my opinion. You're constraining it all the way down to 674 00:57:05.880 --> 00:57:10.199 zero and then starting from zero so that the possibility of pulling in other things 675 00:57:10.199 --> 00:57:14.639 that you don't care about or that will mess it up are different. So, 676 00:57:14.679 --> 00:57:15.079 like I said, I'll find the link can give it to you, 677 00:57:15.079 --> 00:57:21.719 but I forget what it's called but right, So I guess the other question, 678 00:57:21.760 --> 00:57:24.320 and I think I've kind of inferred this from what we've talked about and 679 00:57:24.320 --> 00:57:29.000 some of the other things I've read. But you can have a long running 680 00:57:29.519 --> 00:57:32.920 context, and I know context means something in this world, but you know, 681 00:57:34.039 --> 00:57:37.880 a long running context that says, Okay, you know, I've restricted 682 00:57:37.880 --> 00:57:43.920 what you're going to run on and so I'm just going to run everything else 683 00:57:44.079 --> 00:57:47.239 through this context, right, And that's effectively what you're doing with your AI 684 00:57:47.360 --> 00:57:53.719 assistance that you're providing to people with olympia. Yeah, their context includes everything 685 00:57:53.760 --> 00:58:04.719 you've ever talked about, okay with them, so they're able to most of 686 00:58:04.760 --> 00:58:07.199 the time. You know, someonet know what you're talking about. I mean, 687 00:58:07.239 --> 00:58:12.320 this is still limit has fundamental limited the technology, but it keeps getting 688 00:58:12.360 --> 00:58:19.000 better and worse. Like we're like so in the infancy. It's so obvious 689 00:58:19.039 --> 00:58:21.480 that we're in the infancy of this and like you know, those of us 690 00:58:21.559 --> 00:58:23.840 that have been around long enough and remember IBM, PCs and you know, 691 00:58:23.920 --> 00:58:30.199 common or sixty years stuff like I do you know, actually live through the 692 00:58:30.280 --> 00:58:35.079 expansion of RAHM and whatnot, So you know, right, and it feels 693 00:58:35.360 --> 00:58:38.719 like there's parallels there like for like. I had a lot of trouble with 694 00:58:38.840 --> 00:58:44.440 UH with uploads in the early days of Olympia because the context was too small 695 00:58:44.440 --> 00:58:47.880 and for the most part, no matter how good my my text extraction was, 696 00:58:47.960 --> 00:58:51.559 let's say, from a PDF. Some cases I was using like a 697 00:58:51.679 --> 00:58:54.920 Zoro document intelligence, which is a really really good way of extracting structure and 698 00:58:54.960 --> 00:58:59.880 info from a PDF, even if it includes tables and diagrams and stuff like 699 00:58:59.880 --> 00:59:04.480 that. But then I couldn't fit it into the darn context of the model. 700 00:59:04.559 --> 00:59:07.079 It was like it together with everything else I was trying to put in 701 00:59:07.079 --> 00:59:09.480 the context, it would blow up the context and then it would fail, 702 00:59:09.880 --> 00:59:14.639 kind of like running out a ram. And then somewhere along the way, 703 00:59:14.639 --> 00:59:16.800 I don't remember exactly when Google Pro came out and it had like a million 704 00:59:16.920 --> 00:59:21.519 tokens of context and now it has two million or more, I can't remember. 705 00:59:21.599 --> 00:59:24.280 So in the matter of an afternoon, I switched it over so that 706 00:59:24.480 --> 00:59:31.000 now when you're referring to a document, it uses Google Pro and it just 707 00:59:31.039 --> 00:59:35.800 shoves the contents of the document, which could be like a full length novel 708 00:59:35.880 --> 00:59:37.719 or more. I mean, two million tokens is a lot, right, 709 00:59:38.079 --> 00:59:44.119 Yeah, So that it just gives that to Google Pro and then it asked 710 00:59:44.159 --> 00:59:47.320 Google pro the question that it wants to know, So it offloads to a 711 00:59:47.400 --> 00:59:52.239 model that's really really good at doing that sort of thing. And I've seen 712 00:59:52.239 --> 01:00:00.960 it do incredible stuff. Incredible I've seen it be able to extract figures from 713 01:00:00.119 --> 01:00:07.719 from a payroll report but like that that had a couple hundred employees, you 714 01:00:07.760 --> 01:00:12.639 know, and like basically help me do some ledger entries related to moving these 715 01:00:12.639 --> 01:00:15.599 employees, you know, expense from one category to another, and it goes 716 01:00:15.639 --> 01:00:23.280 into and pulls it out correctly, which is something gpt ford doesn't do reliably. 717 01:00:23.599 --> 01:00:27.360 And as we go back to trimal error, you kind of have you 718 01:00:27.360 --> 01:00:30.800 know, you start. It is its own field, and it's its own 719 01:00:31.239 --> 01:00:36.039 as its nuances. It has its nooks and crannies of expertise and things you 720 01:00:36.079 --> 01:00:38.360 have to know to do or not do. Like the Google thing. I 721 01:00:38.400 --> 01:00:42.800 couldn't have told you before I figured it out right that it would be good 722 01:00:42.800 --> 01:00:47.280 at that right. That was a trial error. But you were you were 723 01:00:47.280 --> 01:00:51.199 asking about context. I don't know if I answered your question your question correctly, 724 01:00:51.199 --> 01:00:55.199 but yeah, it operates on context context keys. So one of the 725 01:00:55.199 --> 01:00:59.159 main things you need to learn is just how to build that context the books 726 01:00:59.239 --> 01:01:02.119 into that pot. There's different ways of doing it. Actually, X. 727 01:01:02.199 --> 01:01:06.559 I think XML will make a comeback to some degree. I posted on Twitter 728 01:01:06.599 --> 01:01:10.559 the other day and I talk about it in the book because XML happens to 729 01:01:10.599 --> 01:01:15.880 be a really really good way of representing data to the to the model because 730 01:01:15.880 --> 01:01:21.760 it's really easy to parse out afterwards. You know, XML in XML out 731 01:01:22.159 --> 01:01:25.639 I can't even you know, OBI from ten to twenty years ago is cringing, 732 01:01:25.760 --> 01:01:30.920 but you know, it's like it's really really easy to pull information out 733 01:01:30.920 --> 01:01:37.360 of XML compared to other things that it can give you. Yeah, you 734 01:01:37.400 --> 01:01:42.599 know, it's really funny. You know, everybody always asks, you know, 735 01:01:43.519 --> 01:01:46.719 how does how does it adapt and to learn new things? Or you 736 01:01:46.760 --> 01:01:52.960 know, how does it learn X? It doesn't. It doesn't. You 737 01:01:52.039 --> 01:01:55.880 have to you have to tell it and then it can say it's It's kind 738 01:01:55.880 --> 01:01:59.840 of funny because like the process is like, you know, okay, you 739 01:02:00.119 --> 01:02:02.880 with the prompt to do this, and then you have a prompt that just 740 01:02:04.000 --> 01:02:07.000 like matches it with the data that you already have that can then give it 741 01:02:07.039 --> 01:02:10.159 to the prompt again and then be like, oh yeah, it's this thing 742 01:02:10.199 --> 01:02:15.519 that you told me that it is. Yeah, I don't think we'll get 743 01:02:15.519 --> 01:02:17.639 to AGI personally with the current approaches. I mean, I couldn't. I 744 01:02:20.079 --> 01:02:28.079 could be wrong, but I mean, the the the way I see it, 745 01:02:27.199 --> 01:02:30.199 it really really like, in order to be a full human replacement, 746 01:02:31.519 --> 01:02:36.719 it has to really learn on the fly, Like there has to be a 747 01:02:36.800 --> 01:02:45.840 modification to the way that inference works based on new information. And you know, 748 01:02:46.000 --> 01:02:50.079 some some people think you can just brute force your way out of that 749 01:02:50.159 --> 01:02:55.719 problem. So like, if your context is long enough, you can just 750 01:02:55.760 --> 01:03:02.639 add everything you've ever learned to the context. I would happily do that approach 751 01:03:02.679 --> 01:03:10.239 in Olympia, except that it's so damn expensive that it's unfeasible, you know, 752 01:03:10.320 --> 01:03:15.280 to do it, Like I could probably like I don't know this for 753 01:03:15.320 --> 01:03:17.800 sure, I haven't analyzed it very well, but but just kind of based 754 01:03:17.840 --> 01:03:24.239 on my gut and the amount of information that I know that that customers put 755 01:03:24.280 --> 01:03:30.599 into Olympia, I could probably fit all everything I know about a customer into 756 01:03:30.599 --> 01:03:38.199 a single Google Pro Google Gemini Pro context, which would be awesome, except 757 01:03:38.280 --> 01:03:43.400 every time I ran that it would cost me like four bucks, and you 758 01:03:43.400 --> 01:03:45.679 know, and a monthly plan is you know, ranges from twenty to sixty 759 01:03:45.679 --> 01:03:50.079 seven dollars, So we're still orders of magnitude away from being able to do 760 01:03:50.119 --> 01:03:53.800 that. When we get to the point where you can do that, where 761 01:03:53.800 --> 01:03:59.360 you can fit all the email correspondence you've ever may done, or maybe just 762 01:03:59.400 --> 01:04:03.440 all the sentiencesages, let's say, into a single context, then yeah, 763 01:04:03.480 --> 01:04:11.079 you got something, But at the moment it's too expensive, makes sense. 764 01:04:14.000 --> 01:04:17.239 So I think we're kind of getting towards the end of our time, and 765 01:04:17.360 --> 01:04:24.599 it feels like we're kind of whining down. Anyway, I'm trying to ramble. 766 01:04:24.880 --> 01:04:29.760 Uh just I could have you on for another hour if you want. 767 01:04:29.840 --> 01:04:33.039 I know you're going. I mean, I definitely have more things to ask. 768 01:04:33.800 --> 01:04:38.800 But I mean one of the one of the things we didn't cover, 769 01:04:38.920 --> 01:04:43.639 which you know, which uh in retrospect, I wanted to cover because it's 770 01:04:43.679 --> 01:04:50.079 it's for this audience, is I have occasionally been discouraged by the amount of 771 01:04:50.119 --> 01:04:57.960 Python. You know how strong Python is field right, uh huh, And 772 01:04:57.960 --> 01:05:01.760 and like fundamentally there's there's no reason, you know, there's no like technical 773 01:05:01.800 --> 01:05:10.880 reason why it couldn't have been Ruby. And you know, I hear people 774 01:05:10.880 --> 01:05:15.800 talk about the math libraries and the data libraries, and you know how those 775 01:05:15.840 --> 01:05:19.519 feed into training some of the systems at the lower level. But where we're 776 01:05:19.559 --> 01:05:24.559 talking about some of these other models, you're not necessarily building those models. 777 01:05:24.599 --> 01:05:28.800 So I mean, is there a reason why it's especially when it comes to 778 01:05:28.840 --> 01:05:31.320 you, No, there's not. It's just cultural. It's just it's just 779 01:05:31.480 --> 01:05:43.519 the inertia. So while sometimes I found it discouraging, what what I've managed 780 01:05:43.519 --> 01:05:47.280 to flip that around to as the eternal optimist, is that it's a positive 781 01:05:47.360 --> 01:05:53.320 because there's a wide open greenfield to do all sorts of cool new gems, 782 01:05:54.239 --> 01:06:00.119 and you can in some cases draw inspiration directly from an existing project and the 783 01:06:00.159 --> 01:06:05.719 other world. Like the most i mean super clear case of that is lang 784 01:06:05.800 --> 01:06:10.320 chain RB. So like one of the main libraries that people used to do 785 01:06:10.360 --> 01:06:14.360 this kind of stuff and Python and JavaScript is called lang chain and there's even 786 01:06:14.400 --> 01:06:17.440 a startup around it that's made a lot of money. And then this Ruby 787 01:06:17.519 --> 01:06:24.679 is named Andrea created lang chain RB and just kind of lifted the whole approach 788 01:06:24.760 --> 01:06:29.280 and wrote it in Ruby, and that's become a thriving project of its own. 789 01:06:30.079 --> 01:06:34.920 I I personally don't use lang chain RB because I I want it to 790 01:06:34.960 --> 01:06:40.519 feel like rails. I don't want it to feel like lang chain like when 791 01:06:40.559 --> 01:06:43.480 I'm when I'm when I'm using these things in my app. I don't want 792 01:06:43.519 --> 01:06:46.320 a whole new paradigm, a whole new architecture, all new set of things 793 01:06:46.320 --> 01:06:49.719 to learn, right Like, I want to apply it to the paradigm I'm 794 01:06:49.719 --> 01:06:55.159 already familiar with. But nonetheless, I mean, that's a pretty good example 795 01:06:55.280 --> 01:07:00.519 of a project that got very big and very popular very fast because they just 796 01:07:00.519 --> 01:07:03.079 said, Okay, well that's popular over there, let's make it in Ruby. 797 01:07:03.079 --> 01:07:06.719 I think that there's dozens, if not hundreds of things that could be 798 01:07:06.760 --> 01:07:11.199 done in that way. Uh, the smallest, you know, kind of 799 01:07:11.199 --> 01:07:18.000 lowest hanging fruit is just client libraries and SDKs for popular providers and SaaS apps 800 01:07:18.039 --> 01:07:25.199 and things like that. So like Alex the owner of the Rubai discord, 801 01:07:25.400 --> 01:07:29.800 got really famous with his Ruby open Ai library, which is you know, 802 01:07:29.920 --> 01:07:32.639 used by millions of people. I would say is probably not a stretch at 803 01:07:32.639 --> 01:07:39.679 this point. So super super popular library, very fast. I've done a 804 01:07:39.679 --> 01:07:42.880 bunch of little, you know, client libraries for like for open router, 805 01:07:43.000 --> 01:07:49.119 for anthropic et cetera. Like there's so much greenfield opportunity to just come in 806 01:07:49.159 --> 01:07:53.840 and do cool stuff. And that that's why I would reckon you know, 807 01:07:53.840 --> 01:07:57.920 if you're interested in that, especially if you're starting out your career, you 808 01:07:57.960 --> 01:08:00.199 know, and you want a popular open source project under your belt or you 809 01:08:00.199 --> 01:08:03.599 know, to be associated with. I think the timing couldn't be better. 810 01:08:04.400 --> 01:08:08.599 That really really couldn't be better. And the other thing too, is that 811 01:08:08.679 --> 01:08:15.760 it's not like it's not like years past where the scope of innovation was so 812 01:08:15.920 --> 01:08:18.600 narrow that if you like you missed the boat, you know, doing one 813 01:08:18.640 --> 01:08:23.800 of these open source projects or gems, you know, oh, well that's 814 01:08:23.800 --> 01:08:26.720 it got to wait five years for like the next big thing to come along. 815 01:08:27.039 --> 01:08:29.600 I mean there's literally a big thing happening every two weeks, two or 816 01:08:29.600 --> 01:08:35.199 three weeks, so you know, that also just makes it even more wide 817 01:08:35.239 --> 01:08:40.520 open, you know. I think it's the coolest place to be. I 818 01:08:40.520 --> 01:08:43.000 mean, it's not the end all be all, it's not a silver bullet. 819 01:08:43.079 --> 01:08:45.600 It's definitely not a hammer you should hit every you know, nail and 820 01:08:45.640 --> 01:08:50.960 your application with. But but it's incredibly cool. And again I go back 821 01:08:50.960 --> 01:08:53.880 to this, like, you know, most of the people I've seen that 822 01:08:53.920 --> 01:08:58.720 have objections to using AI in their projects or whatever. The you know, 823 01:08:58.800 --> 01:09:04.000 they say like it's a stochastic parrot, it's not actually intelligent, or I 824 01:09:04.039 --> 01:09:09.439 can't get deterministic results, or it hallucinates things like that. And I can 825 01:09:09.479 --> 01:09:13.079 tell you, as a practitioner with IT for last year that a lot of 826 01:09:13.079 --> 01:09:15.560 these a lot of these objections don't really make sense to me, right, 827 01:09:15.640 --> 01:09:18.239 Like I do come at it from hacker mentality. So it's kind of like 828 01:09:18.239 --> 01:09:23.319 whatever's clever and works, you know, it's cool with me, doesn't necessarily 829 01:09:23.319 --> 01:09:27.800 have to scale, doesn't necessarily have to be the most pure or you know, 830 01:09:27.920 --> 01:09:33.199 correct quote unquote way to do things, you know, But when you 831 01:09:33.199 --> 01:09:35.720 come at it from that mentality of like, hey, what can I do 832 01:09:35.760 --> 01:09:40.399 with this that's cool or magical or amazing or you know, really impress my 833 01:09:40.439 --> 01:09:46.279 friends and my grandma, you know, like this is I can't hype it 834 01:09:46.359 --> 01:09:51.680 enough. In other words, nice, So tell us about your book a 835 01:09:51.680 --> 01:09:57.279 little bit, because I mean I found patterns of application development using AI. 836 01:09:57.479 --> 01:10:01.680 So is this not real specific or is this not the right book or what? 837 01:10:01.920 --> 01:10:04.680 No, it's not it's not supposed to be real specific. I mean, 838 01:10:04.720 --> 01:10:10.119 I I've been doing rails day and day out for almost decades at this 839 01:10:10.199 --> 01:10:14.920 point, so that's what I chose to do my examples. And also I 840 01:10:15.800 --> 01:10:21.479 was very very happy to like lift code directly out of Olympia, because I 841 01:10:21.479 --> 01:10:28.039 mean, one thing I personally hate in tech books is like, you know, 842 01:10:28.279 --> 01:10:31.159 the car example. If you're going out to you know, or whatever, 843 01:10:31.239 --> 01:10:38.479 it's like non realistic examples or stuff that's super academic or just doesn't doesn't 844 01:10:38.520 --> 01:10:42.039 have practical applications. It's like, okay, well, one way to prove 845 01:10:42.079 --> 01:10:45.800 that something as a practical application is like, here's literally code that I work 846 01:10:45.199 --> 01:10:49.399 on that's in production. Here you go, here's an example, you know, 847 01:10:49.479 --> 01:10:54.279 especially for something like patterns, instead of being all handwavy and like, 848 01:10:54.520 --> 01:10:58.319 you know, like providing a concrete, specific explanation. So I try to 849 01:10:58.359 --> 01:11:02.319 do it as much as possible in the book. You could probably rebuild Olympia 850 01:11:02.359 --> 01:11:04.840 with With what's in the book, you know, I kind of give away 851 01:11:04.840 --> 01:11:09.920 a lot of our approaches and secrets and things like that, which is cool. 852 01:11:10.840 --> 01:11:16.920 And I'm planning a Python edition and a JavaScript edition down the line for 853 01:11:17.039 --> 01:11:20.640 people who just can't get over the fact that the examples are in Ruby but 854 01:11:20.840 --> 01:11:27.159 the patterns are not Ruby specific the patterns are patterns for how to use AI 855 01:11:27.800 --> 01:11:31.840 in your normal application development. You could use those same patterns in Ndjengo or 856 01:11:31.920 --> 01:11:38.640 lamorel or whatever you want, Java, et cetera. They're they're not specific 857 01:11:38.760 --> 01:11:41.199 to Ruby, and that's why there's no Ruby in the title or anything like 858 01:11:41.239 --> 01:11:49.239 that. And I mean, I'm I'm very influenced by the Gang of Four 859 01:11:49.279 --> 01:11:59.039 Patterns book by Martin Feller's book on Patterns of Enterprise Applicant UH Application Architecture by 860 01:11:59.279 --> 01:12:05.800 Greg Hope, and Dave Rice's book on Enterprise Messaging you know, or integration. 861 01:12:05.960 --> 01:12:13.840 Rather that those were like the formative books of my career coming up doing 862 01:12:13.960 --> 01:12:18.119 what I did, So they were very influential on me, and I like 863 01:12:18.199 --> 01:12:23.920 the approach. I think that they were invaluable for a couple of generations of 864 01:12:24.600 --> 01:12:28.560 people doing what we do. So that's what I went for, you know, 865 01:12:28.640 --> 01:12:30.720 I said, I think there's an opportunity here to write an evergreen book 866 01:12:31.079 --> 01:12:35.680 that really kind of covers the fundamentals of how you integrate this technology with your 867 01:12:35.920 --> 01:12:42.039 normal application. So things I don't go into in the book are will AI 868 01:12:42.199 --> 01:12:45.960 replace you as a programmer, and like all the philosophical stuff, not like 869 01:12:46.119 --> 01:12:49.960 not at all. I just give one sentence mentioned in the preface, say 870 01:12:50.000 --> 01:12:53.680 I don't talk about this. The other thing I don't talk about is AI 871 01:12:53.760 --> 01:13:00.800 generation of code. It's cool whatever. There's awesome projects like Scott Werner's Blueprints 872 01:13:00.840 --> 01:13:04.840 is cool, so plug him. Copilot is cool. I use it all 873 01:13:04.880 --> 01:13:09.720 the time. I've been experimenting with other things. I think copil it's probably 874 01:13:09.720 --> 01:13:14.319 still the best. But I don't get into it because to me, it's 875 01:13:15.039 --> 01:13:18.359 not interesting and it's not evergreen for a book. You know, it's literally 876 01:13:18.399 --> 01:13:20.439 just a tool. You install it, you get good at using it, 877 01:13:20.520 --> 01:13:24.159 and you're done, and they're going to continue changing it every year, right, 878 01:13:24.199 --> 01:13:28.359 so there's no point right book. The stuff that's in the book is 879 01:13:28.399 --> 01:13:31.119 truly meant to be evergreen, like it should be good five years from now, 880 01:13:32.279 --> 01:13:38.439 which is why we're actually going to go ahead and make a print copy 881 01:13:38.479 --> 01:13:43.159 of it so you can get it on Lean pub right now. It's one 882 01:13:43.159 --> 01:13:46.560 of the best selling books on Lean pub right now. Yeah, put a 883 01:13:46.600 --> 01:13:53.720 link in the comment so if you're on Twitch or Facebook or YouTube you can 884 01:13:53.800 --> 01:13:58.640 see it. And uh and yeah, I mean it's like four hundred pages. 885 01:13:58.880 --> 01:14:00.520 I mean it's meant to be a good, good, solid reference book 886 01:14:00.520 --> 01:14:02.880 that you that you refer to, and you should be able to pop it 887 01:14:02.880 --> 01:14:06.840 open on any page and find it interesting. I mean it's not. Uh, 888 01:14:09.840 --> 01:14:13.119 the first half of it is meant for people who are relatively new. 889 01:14:13.239 --> 01:14:16.279 I would say, you know, the second half is really, like I 890 01:14:16.319 --> 01:14:19.479 said before, the concrete patterns of things you'll refer two times time again. 891 01:14:19.560 --> 01:14:26.000 I mean you know when I say I'm influenced by those books, I mean 892 01:14:26.000 --> 01:14:30.199 they're literally sitting here next to Yeah. Uh, you know, it's kind 893 01:14:30.199 --> 01:14:33.800 of my go to for I thought there might be some overlap. Actually, 894 01:14:33.840 --> 01:14:40.079 and and I if I remember correctly, the only one this is uh, 895 01:14:40.840 --> 01:14:49.680 the Follower book, but that there's there's a pattern for proxying uh your API. 896 01:14:50.279 --> 01:14:58.680 Uh it is called uh I'm I gonna be able to find it real 897 01:14:58.720 --> 01:15:02.439 quick, yeah, API for. So when I started the when I started 898 01:15:02.439 --> 01:15:04.199 working on the book, I was like, oh, I'm probably going to 899 01:15:04.279 --> 01:15:10.319 find patterns that are very similar to the Gang of Four or you know, 900 01:15:10.720 --> 01:15:14.680 enterprise integration patterns and things like that. And there ended up not really being 901 01:15:14.720 --> 01:15:17.479 that many, if anything. I mean there's some that kind of allude to 902 01:15:18.319 --> 01:15:21.880 older patterns and API facade was the one that was kind of like very very 903 01:15:21.920 --> 01:15:30.119 similar conceptually to the integration pattern, but at the same time it's completely different. 904 01:15:30.600 --> 01:15:34.479 It's essentially earlier in the episode, I mentioned that, like, you 905 01:15:34.479 --> 01:15:38.880 know, when you're talking to an Olympia assistant, it might have like twenty 906 01:15:39.000 --> 01:15:44.960 twenty five different functions and can call. You can't, at least with the 907 01:15:44.960 --> 01:15:47.000 current state of the art technology. You can't give it like hundreds of functions 908 01:15:47.000 --> 01:15:50.279 to call. Basically you're using up too much context, it's too expensive, 909 01:15:50.680 --> 01:15:54.560 they'll get confused, et cetera, et cetera. So, for instance, 910 01:15:54.560 --> 01:16:00.760 how do we do integration with GitHub or Gmail? You know APIs that have 911 01:16:00.880 --> 01:16:05.159 surface area with like dozens or hundreds of endpoints. You use what's called an 912 01:16:05.159 --> 01:16:11.680 API facade. So there's basically one function, which is the Gmail helper function 913 01:16:12.399 --> 01:16:15.119 that's exposed to your main assistant. So anything it wants to do with Gmail, 914 01:16:15.359 --> 01:16:18.840 it can just send a request to the Gmail helper function. Gmail Helper 915 01:16:18.880 --> 01:16:26.199 function, in turn, is another AI that has all the different operations interesting, 916 01:16:27.680 --> 01:16:30.439 so you literally go to the you know, if the user asks our 917 01:16:30.520 --> 01:16:36.359 assistant, hey, check if I have any urgent messages from Charles, the 918 01:16:36.399 --> 01:16:43.159 assistant will call the Gmail helper function and say do I have any urgent email 919 01:16:43.199 --> 01:16:47.000 messages from Charles? So it's proxying it, right, And then that AI 920 01:16:47.920 --> 01:16:51.880 has the ability to scan your inbox to search your inbox, right, and 921 01:16:51.920 --> 01:16:56.399 that AI has all the full documentation and the prompt directors and things that you 922 01:16:56.439 --> 01:16:59.880 need to in order to be able to use that Agat same thing with githubs, 923 01:17:00.039 --> 01:17:02.399 we have like a get ub integration so you can actually look it, 924 01:17:02.479 --> 01:17:08.439 open prs and do operations and things like that. So that that's like twenty 925 01:17:08.520 --> 01:17:13.640 thirty functions itself. So you do it with a helper. So that's what 926 01:17:13.680 --> 01:17:15.720 I in the book. That's what I mean by patterns, right, so 927 01:17:15.760 --> 01:17:20.600 identified that this is a pattern that used more than once, you know, 928 01:17:20.720 --> 01:17:25.920 and should be reusable by other people. So I generalize it, describe it 929 01:17:26.399 --> 01:17:33.359 documented in the book. That is really interesting. I'm just imagining I don't 930 01:17:33.399 --> 01:17:39.880 know, a podcast assistant. It's you know, hey, Obi said something 931 01:17:39.880 --> 01:17:44.720 he didn't want to say at about fifteen minutes in and it sounded something like 932 01:17:44.760 --> 01:17:46.960 this, can you go take it out? And then it right, it 933 01:17:47.000 --> 01:17:51.159 goes to the audio editor assistant, and the audio editor assistant knows how to 934 01:17:51.199 --> 01:17:56.520 open up the file and scan and understand. Okay, this is where Obe 935 01:17:56.520 --> 01:18:01.760 said the thing, and then right, that's the that's the premise of the 936 01:18:01.880 --> 01:18:06.439 multitude of workers. So can I call the multitude of workers because like in 937 01:18:06.479 --> 01:18:12.720 your example, you'd have a worker who was the podcast editor worker so that 938 01:18:12.880 --> 01:18:15.479 you know, has a specific set of tools, and it just said start 939 01:18:15.520 --> 01:18:20.199 waiting for a request for that, which is why it's reminiscent of micro services, 940 01:18:20.520 --> 01:18:23.680 you know, and it's actually I think you think about it in a 941 01:18:23.680 --> 01:18:28.840 similar way. Yeah. Cool, Well, I want to be mindful of 942 01:18:28.880 --> 01:18:32.199 time, and we scheduled this to end in like three minutes. We're going 943 01:18:32.279 --> 01:18:36.840 to go over a little bit. But this has been really fascinating. I 944 01:18:36.920 --> 01:18:41.520 just went and picked up your book. I didn't know I could have it 945 01:18:41.600 --> 01:18:44.600 yet, so as soon as as soon as I looked for and I was 946 01:18:44.640 --> 01:18:48.159 like, I was like, oh, I'm buying that sucker. Yes it 947 01:18:48.560 --> 01:18:54.880 is mine. It is on my computer. So yeah, so I'm gonna 948 01:18:54.920 --> 01:18:58.600 go ahead and move us picks before we do that, though, Obe, 949 01:18:58.720 --> 01:19:02.039 if people want to connect with you on this stuff, I see you all 950 01:19:02.079 --> 01:19:06.720 the time on the Ruby AI Builders discord, and I'll get a link in 951 01:19:06.760 --> 01:19:13.039 there for that, But are there other good places to find you? I'm 952 01:19:13.079 --> 01:19:16.399 just you know, usual Twitter. I'll be Fernandez a Gmail if you want 953 01:19:16.399 --> 01:19:21.680 to email me. The products is Olympia that chat. We got a little 954 01:19:21.680 --> 01:19:29.319 community of users there. Yeah, that's pretty much it. I mean some 955 01:19:29.439 --> 01:19:30.520 of my other channel, like if you find me on Instagram and stuff, 956 01:19:30.520 --> 01:19:32.840 it's all going to be my music career stuff. So I mean, if 957 01:19:32.880 --> 01:19:35.840 you're interested, if you like electronic music, look me up on Spotify. 958 01:19:35.960 --> 01:19:44.439 I'll be Fernandez. I just recently made a spreadsheet of all the music I've 959 01:19:44.439 --> 01:19:49.680 ever made since I started, and I'm over one hundred and thirty original compositions, 960 01:19:49.880 --> 01:19:54.760 so about half of them never been released. But but yeah, that's 961 01:19:55.439 --> 01:20:00.399 come a long way with that as well, So awesome. All right, 962 01:20:00.560 --> 01:20:06.039 Valentino, what are your picks? A lot along the subject of today, 963 01:20:06.399 --> 01:20:12.159 I'm just gonna do one pick. I saw this talk from doctor Jim Fan 964 01:20:12.399 --> 01:20:17.279 who's part of Nvidia's new Gear group, which is generalist embodied agent research. 965 01:20:18.079 --> 01:20:25.199 It's fascinating stuff. But he gave this talk at the last NVIDIA conference on 966 01:20:25.640 --> 01:20:31.079 the new foundation agents that they're building around robots and teaching them to learn using 967 01:20:31.159 --> 01:20:36.640 virtual dojos. So there's like a very manual process involved in training a robot 968 01:20:36.680 --> 01:20:41.119 to do tasks where they have to do a lot of trial and of their 969 01:20:41.159 --> 01:20:45.880 own to autocorrect themselves. And so they've basically found a way to create virtual 970 01:20:45.960 --> 01:20:53.239 environments using like some previous products they did with Minecraft, to like teach skills 971 01:20:53.319 --> 01:20:58.560 to robots in virtual environments so that they can do the repetitive process of trial 972 01:20:58.600 --> 01:21:05.079 and error virtually and super fascinating stuff. And he gives a great talk on 973 01:21:05.159 --> 01:21:15.760 how these foundation agents work, and I recommend you check it out. Nice. 974 01:21:16.439 --> 01:21:18.760 I'm gonna throw in a few picks on my own. So my friends 975 01:21:18.800 --> 01:21:25.760 came over last night. We were playing board games, and yeah, I 976 01:21:25.800 --> 01:21:31.239 think people are pretty used to me picking the board games, and we played 977 01:21:31.279 --> 01:21:34.640 Lost Ruins of Arnac, So I'm gonna pick Lost Ruins of our nec. 978 01:21:34.760 --> 01:21:41.359 I just pulled up the wrong thing. So essentially, you have like a 979 01:21:41.399 --> 01:21:45.880 technology track, you have exploration, you can do, you can do deck 980 01:21:45.920 --> 01:21:54.520 building. All of those things play into your score at the end. And 981 01:21:54.640 --> 01:22:01.520 so as it's a little bit more involved than kind of your your regular game 982 01:22:01.800 --> 01:22:08.359 board game. Geek waits it at two point ninety six, but it does 983 01:22:08.439 --> 01:22:12.000 say that you can play it with the players that are ten ten plus. 984 01:22:12.960 --> 01:22:15.880 The community says ten plus, the game says twelve plus. That's probably about 985 01:22:15.920 --> 01:22:19.039 right. There were three of us playing last night. We had a great 986 01:22:19.079 --> 01:22:24.800 time. So anyway, it came out in twenty twenty. There is an 987 01:22:24.840 --> 01:22:29.600 expansion, and I haven't played the expansion in a while, so I'm not 988 01:22:29.640 --> 01:22:35.000 gonna shout that out. But anyway, Lost Ruins of Arnac. It's a 989 01:22:35.039 --> 01:22:40.039 super fun game. It's just a little more involved than sort of your Settlers 990 01:22:40.039 --> 01:22:46.039 of Katan or whatever that you know comes in about it two. So another 991 01:22:46.079 --> 01:22:51.159 pick I have. So I've been working my way through Ayoush who's not here 992 01:22:51.199 --> 01:22:58.079 today, but through his book The Rails and hot Wire Codex, and I 993 01:22:58.279 --> 01:23:06.239 actually got a working version of my I'm basically rewriting top end depths because they're 994 01:23:06.279 --> 01:23:11.399 just things I want different, vastly different. Right. You learn all your 995 01:23:11.479 --> 01:23:15.119 lessons and then it's like, okay, well I could refactor all this, 996 01:23:15.319 --> 01:23:17.279 but that would be more work than just, you know, writing it new. 997 01:23:17.319 --> 01:23:19.880 And the other thing is is I can add in things that I didn't 998 01:23:19.920 --> 01:23:26.439 have before from the start, like internationalization and accessibility and things like that that 999 01:23:26.479 --> 01:23:30.079 I didn't do last time because adding that in after the fact I miss stuff 1000 01:23:30.159 --> 01:23:35.720 everywhere. So this way I can just be complete on putting it together. 1001 01:23:35.800 --> 01:23:41.520 But yeah, so he has his chapters on Turbo Native and so yeah, 1002 01:23:41.560 --> 01:23:45.199 so I have an iOS version of the app running. I have most of 1003 01:23:45.239 --> 01:23:48.960 an Android version running. I just got it running so that it brings up 1004 01:23:49.039 --> 01:23:57.239 the website in the in the native view, but it doesn't have the navigation 1005 01:23:57.319 --> 01:24:00.159 or anything else on it yet. Turns out that's a little bit of work 1006 01:24:00.920 --> 01:24:04.520 to get that to go, but it's very cool, So I'm very happy 1007 01:24:04.520 --> 01:24:10.159 with that. I'm going to pick the Rails and hot Wire Codex by a 1008 01:24:10.319 --> 01:24:17.119 us Natia and then yeah, there was something else I was going to pick, 1009 01:24:17.159 --> 01:24:19.520 and I just can't think of what it is at the moment. Oh, 1010 01:24:19.640 --> 01:24:26.319 I've been watching the Euros. I'm a soccer nut, and so yeah, 1011 01:24:26.359 --> 01:24:28.920 I've been pretty happy with the way things are going so far. So 1012 01:24:29.760 --> 01:24:31.800 my team if they're in and they weren't in the World Cup, and it 1013 01:24:31.840 --> 01:24:36.840 broke my heart. Is Italy just because I lived there for two years, 1014 01:24:38.640 --> 01:24:42.159 actually went to a couple of soccer soccer matches while I was there, and 1015 01:24:42.239 --> 01:24:45.399 I just I just love watching those guys play. If it's not Italy, 1016 01:24:45.479 --> 01:24:51.119 then it's usually England, France, US and honorable mention is Netherlands. And 1017 01:24:53.399 --> 01:24:58.840 so all those teams are playing in the EUROS because it's European Championship and except 1018 01:24:58.880 --> 01:25:02.479 for US, and all of them when their matches. So I'm I'm enjoying 1019 01:25:02.520 --> 01:25:08.560 so far watching them win. I think the next game on the schedule that 1020 01:25:08.680 --> 01:25:11.680 I'm going to be watching though, is France versus Netherlands, and so of 1021 01:25:11.680 --> 01:25:14.520 course I have to it for France, so we'll see how that goes. 1022 01:25:14.560 --> 01:25:17.720 But yeah, it's it's been awesome. So if you enjoy soccer, let 1023 01:25:17.720 --> 01:25:23.680 me know you might see me tweet about it some. But yeah, those 1024 01:25:23.680 --> 01:25:29.600 are my picks. Uh pick, yeah, do it. I've been playing 1025 01:25:29.600 --> 01:25:34.760 with this thing called udio. It's like audio but without the a uh huh 1026 01:25:34.880 --> 01:25:39.920 except i'm you. I'm so. I think generating a song is a you 1027 01:25:39.960 --> 01:25:43.960 know, like a full song, especially if use ai lyrics is a just 1028 01:25:44.000 --> 01:25:49.319 a novelty. I don't know that this will have any sort of transfer transformational 1029 01:25:49.319 --> 01:25:55.680 impact on anything. However, it's really fun for doing things like impressing your 1030 01:25:55.720 --> 01:26:00.000 barber, which I did here in Mexico City a week ago. Uh, 1031 01:26:00.159 --> 01:26:01.880 because we were he asked me about AI, and I said, did you 1032 01:26:01.880 --> 01:26:05.439 know I can make music? And then I made a like a like a 1033 01:26:05.520 --> 01:26:13.840 bonda, you know, kind of a thing about Marcos, the barber who 1034 01:26:13.880 --> 01:26:18.319 had and and he was so florid that he was laughing his ass off and 1035 01:26:18.520 --> 01:26:25.520 it literally interrupted all the other barbers to show them in the shop. So 1036 01:26:26.079 --> 01:26:30.880 it has good novelty value, you know, making operas about your coworkers and 1037 01:26:30.880 --> 01:26:34.439 stuff. But what I what I'm using for as as a semi professional producer, 1038 01:26:35.319 --> 01:26:42.920 is to generate vocals and to get ideas based on my own music. 1039 01:26:42.960 --> 01:26:48.000 Because a week ago or so they added the ability to upload bits of music. 1040 01:26:49.119 --> 01:26:54.560 So now now that turns it into a bit of a professional tool, 1041 01:26:54.960 --> 01:26:58.359 because, for instance, I can take an instrumental that I've made, I 1042 01:26:58.439 --> 01:27:00.840 make mostly instrumentals, and I can upload it and then I can put in 1043 01:27:00.880 --> 01:27:04.399 my lyrics and it will sing my lyrics, and I can iterate on that 1044 01:27:04.439 --> 01:27:08.600 in the same way that you would iterate an image with my Journey or Dolly 1045 01:27:09.640 --> 01:27:14.000 until you get something usable, and then without getting into the workflow too much, 1046 01:27:14.000 --> 01:27:16.600 you can extract the vocals and actually integrate them into your music. So 1047 01:27:18.720 --> 01:27:23.399 for any of you out there listening who like me, uh you know, 1048 01:27:24.000 --> 01:27:26.800 love the idea of writing lyrics or singing, but you can't sing worth it, 1049 01:27:26.880 --> 01:27:31.640 dam then this is a really awesome tool to use for that. And 1050 01:27:31.680 --> 01:27:41.079 it's a it's a lot of fun awesome that's my big cool all right, 1051 01:27:41.079 --> 01:27:48.279 Well, thanks for coming Obie, this was likewise thanks, yeah, looking 1052 01:27:48.319 --> 01:27:51.760 forward to reading the rest of the book for sure, right on. Yeah, 1053 01:27:51.840 --> 01:28:00.319 let me know what you think. And uh oh yeah that's it for 1054 01:28:00.960 --> 01:28:02.279 no, it's all good. We'll just wrap it up though, and until 1055 01:28:02.319 --> 01:28:06.039 next time, folks, Max out m