1 00:00:10,320 --> 00:00:13,880 Hey, it's Eric Newcomer. Welcome to Cerebral Valley 2 00:00:14,520 --> 00:00:15,800 newcomer. Back. 3 00:00:15,800 --> 00:00:19,720 With me are Max Child and James Wilsterman, the Co founders of 4 00:00:19,720 --> 00:00:21,680 Volley. Hey guys. 5 00:00:21,840 --> 00:00:24,400 Welcome back. Hey, happy to be here. 6 00:00:24,400 --> 00:00:25,880 Hey, Eric. Hey. 7 00:00:26,640 --> 00:00:28,920 All right. This week we are talking about 8 00:00:28,920 --> 00:00:32,400 what we're calling the AI Personal Stack. 9 00:00:32,439 --> 00:00:36,880 Or who? How do we actually use AI in our 10 00:00:36,880 --> 00:00:40,520 real lives? I've got in the second-half of 11 00:00:40,520 --> 00:00:45,000 this episode, Joshua Browder, the CEO of Do Not Pay, who got 12 00:00:45,000 --> 00:00:48,880 in trouble trying to use AI to act as a lawyer. 13 00:00:49,160 --> 00:00:54,720 He got his start trying to get out of parking tickets that he 14 00:00:54,720 --> 00:00:58,800 increasingly has automated, and now he's a ChatGPT addict, 15 00:00:58,800 --> 00:01:03,000 building a company around using foundation models and language 16 00:01:03,000 --> 00:01:08,200 models to fight back against companies like Comcast and your 17 00:01:08,920 --> 00:01:12,640 mega corporate gym. I think Equinox is a big target 18 00:01:12,640 --> 00:01:15,800 of his just helping people get out of the bill, their bills. 19 00:01:15,800 --> 00:01:19,080 So that's a fun conversation from someone who's really trying 20 00:01:19,080 --> 00:01:24,160 to suck the marrow out of AI and get everything he can out of it. 21 00:01:24,960 --> 00:01:29,600 So, yeah, excited to talk to you guys about how how we're using 22 00:01:29,600 --> 00:01:33,120 AI, how we're seeing people actually use it. 23 00:01:33,120 --> 00:01:38,880 Because, you know, in Silicon Valley, people's expectations, 24 00:01:38,880 --> 00:01:40,840 hopes and dreams can get ahead of them. 25 00:01:40,840 --> 00:01:45,600 We can all live in the world of intuitions as we did in the last 26 00:01:45,600 --> 00:01:47,720 episode, where it's like, what could be possible? 27 00:01:47,720 --> 00:01:50,680 Where could we take this? You know, it's an optimistic 28 00:01:50,680 --> 00:01:54,560 sort of place and this episode is going to try and ground us 29 00:01:54,560 --> 00:01:56,480 in. How are we using it? 30 00:01:56,480 --> 00:02:00,040 Are people using it? So, So that's the question, 31 00:02:01,200 --> 00:02:05,240 which of us, who thinks they use AI the most? 32 00:02:05,280 --> 00:02:09,360 I honestly would not put myself super, super high on this list. 33 00:02:09,759 --> 00:02:12,120 I guess probably James. I'm going to go, James. 34 00:02:12,120 --> 00:02:12,840 Yeah. OK. 35 00:02:12,840 --> 00:02:13,680 All right. There we go. 36 00:02:13,680 --> 00:02:17,880 I have the same guess here, James, given that we believe 37 00:02:17,880 --> 00:02:20,160 that you probably use it the most and I think you had an 38 00:02:20,160 --> 00:02:23,880 exciting AI driven experience today. 39 00:02:24,080 --> 00:02:26,840 How are you using AI in your life? 40 00:02:27,840 --> 00:02:31,120 That is very true. I commuted today in a Waymo, 41 00:02:31,120 --> 00:02:33,400 which was my first time in a Waymo. 42 00:02:33,400 --> 00:02:36,480 I've taken a few cruises around San Francisco. 43 00:02:36,480 --> 00:02:40,640 I think that that those self driving car experiences really 44 00:02:40,640 --> 00:02:45,280 ranked at the maybe the pinnacle to me of excitement of using 45 00:02:45,280 --> 00:02:51,480 AIII would say GPT chat. GPT is, you know, very. 46 00:02:52,040 --> 00:02:55,760 Very useful and I I use it all a lot, but I feel like the 47 00:02:55,800 --> 00:02:59,480 visceral excitement of riding in a self driving vehicle and just 48 00:02:59,480 --> 00:03:03,000 taking that to work is something that I have haven't experienced 49 00:03:03,000 --> 00:03:04,880 elsewhere with. Could you tell the difference 50 00:03:04,920 --> 00:03:08,240 between Cruise and Waymo? Was there a spiritual or is it 51 00:03:08,240 --> 00:03:10,800 just how the interior of the car feels or? 52 00:03:11,480 --> 00:03:15,600 I I guess I do think of it as I don't want to like over learn, 53 00:03:15,840 --> 00:03:21,000 you know from a few anecdotes of riding in in these vehicles, 54 00:03:21,000 --> 00:03:23,000 but. I do feel like there is a bit of 55 00:03:24,160 --> 00:03:26,320 AI feel like the Waymo felt a little bit more natural, a 56 00:03:26,320 --> 00:03:30,000 little bit less cautious, but not in a scary way. 57 00:03:30,000 --> 00:03:34,080 I feel like Cruise is very cautious, driver cautious, yeah. 58 00:03:34,400 --> 00:03:39,440 Yeah, and neither. No humans beside you, right? 59 00:03:39,440 --> 00:03:41,440 Like no driver. Correct. 60 00:03:41,440 --> 00:03:44,360 And and interesting in the Waymo is I was able to sit in the 61 00:03:44,360 --> 00:03:46,040 front seat. You're not able to do that in a 62 00:03:46,040 --> 00:03:48,360 cruise. And that was it. 63 00:03:48,360 --> 00:03:50,960 That alone was a pretty amazing experience because. 64 00:03:51,840 --> 00:03:55,400 I was, I just like could turn left and there was nobody there. 65 00:03:55,400 --> 00:03:59,160 And you know, I could see the whole road and see see exactly 66 00:03:59,160 --> 00:04:03,800 what was happening. So I feel like I enjoyed the way 67 00:04:03,800 --> 00:04:07,080 MO experienced a lot today. And I I think being able to ride 68 00:04:07,080 --> 00:04:12,880 up front, having it feel like more human in some sense, that 69 00:04:12,880 --> 00:04:17,160 it just kind of was relaxing, that I felt like I I wasn't. 70 00:04:17,680 --> 00:04:21,800 I wasn't like wasting time by taking a Waymo to work because 71 00:04:23,120 --> 00:04:25,040 you know it, it drove pretty naturally. 72 00:04:25,040 --> 00:04:27,280 And yeah, it was an amazing experience. 73 00:04:27,600 --> 00:04:32,240 And you know, the Waymo CEO, Co CEO, Dimitri is speaking at 74 00:04:32,400 --> 00:04:34,280 Cerebral Valley. I think we will have announced 75 00:04:34,280 --> 00:04:37,960 that by the time this podcast comes out, Yes. 76 00:04:38,040 --> 00:04:40,000 But so very excited to hear more. 77 00:04:40,080 --> 00:04:42,880 Very. Excited as well other other 78 00:04:42,920 --> 00:04:43,680 uses. Of it, Yeah. 79 00:04:45,400 --> 00:04:47,000 So. Definitely. 80 00:04:47,000 --> 00:04:49,920 I've already talked on this podcast about using ChatGPT for 81 00:04:49,920 --> 00:04:53,280 just creative kind of brainstorming and exploration. 82 00:04:53,280 --> 00:04:56,400 I think that it's very effective at creating kind of internal 83 00:04:56,400 --> 00:05:01,080 documentation or early drafts of it or getting you started on 84 00:05:01,080 --> 00:05:05,480 that at work. And then I also think that the 85 00:05:05,480 --> 00:05:08,440 code side is like really, really important here. 86 00:05:08,480 --> 00:05:12,480 A lot of Volley engineers are using ChatGPT in their 87 00:05:12,560 --> 00:05:15,400 day-to-day experience coding at Volley and. 88 00:05:15,840 --> 00:05:18,640 What that looks like is pretty interesting because it's 89 00:05:18,640 --> 00:05:25,080 primarily kind of bouncing between your IDE and IDE, your 90 00:05:26,280 --> 00:05:30,320 independent development environment, integrated 91 00:05:30,760 --> 00:05:32,920 integrated integrated development environment. 92 00:05:32,920 --> 00:05:37,560 Where you code basically and bouncing between that and sort 93 00:05:37,560 --> 00:05:40,600 of ChatGPT. Where you might ask for an 94 00:05:40,600 --> 00:05:44,240 architectural advice, or even paste code in, or ask for some 95 00:05:44,240 --> 00:05:45,920 code in. It used to be you Google Stack 96 00:05:45,920 --> 00:05:48,640 Overflow, blah blah blah and they'd copy and paste it there. 97 00:05:48,640 --> 00:05:51,120 Exactly. And it's it's a really similar 98 00:05:51,240 --> 00:05:53,040 experience to when I was learning to code. 99 00:05:53,120 --> 00:05:56,520 And I think what is interesting is a lot of even more junior 100 00:05:56,520 --> 00:06:00,160 developers are the the most eager to kind of explore here 101 00:06:00,160 --> 00:06:02,800 and just use ChatGPT because they want to learn quickly and 102 00:06:03,080 --> 00:06:06,120 you know they don't, they were going to Google it anyway from 103 00:06:06,120 --> 00:06:10,720 Stack Overflow so. I think that is a huge change in 104 00:06:10,720 --> 00:06:13,800 how especially junior software developers are are working 105 00:06:13,800 --> 00:06:15,720 today. And obviously the promise there 106 00:06:15,720 --> 00:06:18,480 is like integrated code environments where. 107 00:06:18,800 --> 00:06:23,560 Yeah, and there is a start up called Cursor that has a 108 00:06:23,560 --> 00:06:26,360 integrated IDE and I know it's paying for that. 109 00:06:27,040 --> 00:06:29,200 No, we're not doing it. We're not using any of anything 110 00:06:29,200 --> 00:06:34,760 like that other than GitHub Copilot, which is I think you 111 00:06:34,760 --> 00:06:37,880 know I use GitHub Copilot. It essentially is auto complete. 112 00:06:38,240 --> 00:06:44,200 For code, and I would say it's useful but not revolutionary, 113 00:06:44,200 --> 00:06:47,280 not game changing. It maybe improves your kind of 114 00:06:47,280 --> 00:06:51,360 speed by, you know, 5% or something, anything else. 115 00:06:51,600 --> 00:06:54,160 I actually will give you another one which I think is really 116 00:06:54,160 --> 00:06:56,840 interesting. I am a new father. 117 00:06:56,840 --> 00:07:02,560 I had a newborn born about a month ago, and congratulations. 118 00:07:02,560 --> 00:07:06,600 Thank you and. I find chat chi PT to be really 119 00:07:06,600 --> 00:07:09,040 incredible as a parenting assistant. 120 00:07:09,040 --> 00:07:14,240 Just being able to ask questions like when should I expect my 121 00:07:14,240 --> 00:07:18,880 baby to be able to see certain distances right. 122 00:07:18,880 --> 00:07:23,360 And I could Google that and and I'd get different sources and I 123 00:07:23,360 --> 00:07:27,480 you know it would be you know something I could research but I 124 00:07:27,480 --> 00:07:30,240 basically don't need like a precise answer and and Chat chi 125 00:07:30,240 --> 00:07:32,800 PT is able to give me hey here's how the here's how this might 126 00:07:32,800 --> 00:07:35,800 look over the next. Month, two months, six months, 127 00:07:35,800 --> 00:07:38,120 right? That's just really convenient to 128 00:07:38,280 --> 00:07:41,760 be able to ask a question in open-ended natural language like 129 00:07:41,760 --> 00:07:44,280 that. All right, Max, what about you? 130 00:07:44,280 --> 00:07:47,200 How are you using? I mean a lot less than James, so 131 00:07:47,200 --> 00:07:49,640 I'll move through this in a hurry I would say. 132 00:07:49,640 --> 00:07:53,080 One I occasionally have it right, like memos or long form 133 00:07:53,080 --> 00:07:56,600 emails for Volley. But again, I would say only in 134 00:07:56,600 --> 00:07:59,880 the case where I'm OK if it's like B plus quality output 135 00:07:59,880 --> 00:08:02,240 'cause if it really has to be an A or an A plus, I feel like I 136 00:08:02,240 --> 00:08:03,640 have to do it myself. And then? 137 00:08:04,040 --> 00:08:07,400 I dictate like texts and emails and messages like quite a lot 138 00:08:07,400 --> 00:08:11,120 like I would say an old man. Yeah, I'm like an old man. 139 00:08:11,120 --> 00:08:12,720 I mean, I'm, you know, I have a kid too. 140 00:08:12,760 --> 00:08:15,040 I'm, I'm in my 30s, so that's old enough. 141 00:08:15,920 --> 00:08:17,920 Yeah. I would say at home I dictate a 142 00:08:17,920 --> 00:08:20,360 lot of texts and messages into my phone and I now think that 143 00:08:20,360 --> 00:08:22,880 like Apples native dictation is like. 144 00:08:23,200 --> 00:08:26,280 You know, 98% accurate or whatever you want to call it and 145 00:08:26,280 --> 00:08:28,440 and if I want to dictate a, how do you actually do it? 146 00:08:28,800 --> 00:08:30,720 You just hit the little microphone button in the bottom 147 00:08:30,720 --> 00:08:32,480 right of the keyboard and then you just start talking right 148 00:08:32,480 --> 00:08:35,600 away and it turns into text. Turns into text. 149 00:08:35,600 --> 00:08:37,000 It's pretty fast and it's pretty good. 150 00:08:37,880 --> 00:08:40,760 So I do that a lot. Other than yet that. 151 00:08:40,760 --> 00:08:44,000 I mean, yeah, I I sometimes use like some of those image 152 00:08:44,000 --> 00:08:47,600 generators for for internal, you know, slide decks, but that's 153 00:08:47,600 --> 00:08:49,600 usually almost as like a joke because they're sometimes 154 00:08:49,600 --> 00:08:50,280 hilarious. So. 155 00:08:51,840 --> 00:08:55,200 I mean that's kind of it. I am playing with some of these 156 00:08:55,200 --> 00:08:57,440 newer apps that are based around image generation. 157 00:08:57,440 --> 00:09:00,960 Like I think we talked about Can of Soup, which is like a very 158 00:09:00,960 --> 00:09:04,600 hot YC company that just came out that is basically like 159 00:09:04,600 --> 00:09:07,120 Instagram for image generated content. 160 00:09:07,120 --> 00:09:10,080 And so every day they send you a push notification that says 161 00:09:10,080 --> 00:09:13,480 like. Max riding a giant chicken, like 162 00:09:13,480 --> 00:09:16,440 in a cowboy hat or something, and then you click on it and it 163 00:09:16,440 --> 00:09:18,680 generates a photo of you, you know, writing, writing a giant 164 00:09:18,680 --> 00:09:20,840 chicken on a cowboy hat or wearing a cowboy hat. 165 00:09:21,200 --> 00:09:23,760 And you know, do you use that everyday? 166 00:09:23,760 --> 00:09:25,520 Did you use? That I clicked on that. 167 00:09:25,880 --> 00:09:28,360 I don't think I'm like a daily user, but I've used that a 168 00:09:28,360 --> 00:09:30,680 couple times a week because they're pretty good at sending 169 00:09:30,680 --> 00:09:32,760 you prompts that just sound funny, like. 170 00:09:33,000 --> 00:09:35,840 And so I click on it like maybe one to two out of seven, 171 00:09:35,840 --> 00:09:38,480 basically and. I don't really use it for the 172 00:09:38,480 --> 00:09:41,240 Instagram functionality of like looking at other people's 173 00:09:41,240 --> 00:09:43,400 because I don't really care like what other people look like on 174 00:09:43,400 --> 00:09:45,000 top of giant chicken, like like everyone. 175 00:09:45,000 --> 00:09:47,920 I'm very a selfish person and so yeah, I like. 176 00:09:47,960 --> 00:09:51,560 But that use case of just sending you a prompt everyday 177 00:09:51,640 --> 00:09:54,280 and then just you click on it and a push notification opens 178 00:09:54,280 --> 00:09:56,480 right out to generating something from that prompt is a 179 00:09:56,480 --> 00:10:01,040 fairly compelling like use case. I don't know how you build a 180 00:10:01,040 --> 00:10:03,800 social network out of it, but that part has been working on me 181 00:10:03,800 --> 00:10:08,720 a little bit, Eric. Yeah, Eric, Eric. 182 00:10:09,440 --> 00:10:13,000 So Eric how do you use AI in your in your daily life? 183 00:10:14,840 --> 00:10:18,680 I mean I'm probably the most disillusioned with Chad GB TI 184 00:10:18,680 --> 00:10:23,200 mean credit credit to the AI. Then in the beginning when when 185 00:10:23,280 --> 00:10:26,880 image generation was new I used some of it to illustrate my 186 00:10:26,880 --> 00:10:30,240 newsletter which is not you know that's significant it it would 187 00:10:30,240 --> 00:10:32,680 probably cost a lot. You know, I I get access to 188 00:10:32,680 --> 00:10:37,000 Getty Images through Sub Stack, so I don't have to agonize about 189 00:10:37,000 --> 00:10:39,760 the price of images. But you know, there's meaningful 190 00:10:39,760 --> 00:10:42,400 value to that. And like it's a big editorial 191 00:10:42,400 --> 00:10:43,760 decision to decide to put it out. 192 00:10:43,960 --> 00:10:46,600 Though I do think there was a period where like where the 193 00:10:46,600 --> 00:10:51,640 aesthetic of AI was cool and sort of even if it was bad, it 194 00:10:51,640 --> 00:10:55,720 was showing that you were sort of you were doing sort of AI 195 00:10:55,720 --> 00:10:58,280 stuff. And eventually obviously the 196 00:10:58,280 --> 00:11:01,200 images need to stand on their own and be great. 197 00:11:01,200 --> 00:11:04,320 And I partially, I just need to be better at prompting and I 198 00:11:04,320 --> 00:11:07,320 think I'm just not like expert level and it's so it's not worth 199 00:11:07,320 --> 00:11:11,880 the energy to use it. I've found Chachi BT to be very 200 00:11:11,880 --> 00:11:16,040 successful in proofreading. The most impressive experience 201 00:11:16,040 --> 00:11:22,520 being that inputted a story and it had it told me that a million 202 00:11:22,520 --> 00:11:25,160 was supposed to be a billion which was like, OK, wow, that's 203 00:11:25,160 --> 00:11:27,960 like that was great. That would have been like, 204 00:11:28,200 --> 00:11:30,920 that's hard to catch. I think it's checked. 205 00:11:31,440 --> 00:11:35,280 It'll catch like internal misspelling errors, you know, in 206 00:11:35,280 --> 00:11:37,520 proper nouns, you know. Obviously getting names right is 207 00:11:37,520 --> 00:11:42,560 always important. It's good at yeah, it'll, it 208 00:11:42,680 --> 00:11:45,120 catches like double word typos pretty well. 209 00:11:45,120 --> 00:11:48,640 So I think it's always useful putting that in. 210 00:11:49,000 --> 00:11:54,000 It can be good about directing you towards active voice, but I 211 00:11:54,000 --> 00:11:59,000 feel like as soon as you start asking for subjective advice it 212 00:11:59,000 --> 00:12:02,200 gets pretty bad. And like, I don't, you know? 213 00:12:02,200 --> 00:12:04,880 To clarify is that is that 3.5 that you're using? 214 00:12:05,000 --> 00:12:09,600 No, no no, not four, only four. I find 3.5 terrible like I I so. 215 00:12:09,720 --> 00:12:11,720 I thought you said you gave up on paying for it. 216 00:12:12,160 --> 00:12:16,760 I was paying and then I gave up, so I just returned to relying on 217 00:12:16,760 --> 00:12:19,800 humans to proofread. But now because we're recording 218 00:12:19,800 --> 00:12:23,480 it, so I did give it up like partially just frustration that 219 00:12:23,480 --> 00:12:26,120 like I perceived it to be getting worse. 220 00:12:26,120 --> 00:12:29,200 Though I'm not sure what I actually think on that front. 221 00:12:31,200 --> 00:12:34,840 And I I I tried using web plugins and I actually found 222 00:12:34,840 --> 00:12:38,800 them like of poor experience so. It's very slow. 223 00:12:38,800 --> 00:12:41,760 Like did you ever try the web browser plug in they they they 224 00:12:41,760 --> 00:12:44,760 just added their own like so originally they had an outside 225 00:12:44,760 --> 00:12:46,760 one and I think now they have their own. 226 00:12:47,920 --> 00:12:49,880 It's still. It feels like it relies on like 227 00:12:49,880 --> 00:12:53,200 one page and then. I agree it's it's not very 228 00:12:53,200 --> 00:12:54,880 effective because it's essentially. 229 00:12:55,400 --> 00:12:59,840 Going to Bing and Googling whatever you said Bing then. 230 00:12:59,840 --> 00:13:03,520 Please, I think. Sorry, that's pretty funny. 231 00:13:03,520 --> 00:13:05,240 Oh my God. Poor Microsoft. 232 00:13:05,600 --> 00:13:07,680 Poor poor Microsoft. You go to Bing. 233 00:13:07,680 --> 00:13:10,960 Bing. Yeah, binging your prompt and 234 00:13:10,960 --> 00:13:17,440 then essentially just clicking on maybe one or two links and 235 00:13:17,440 --> 00:13:21,400 somehow using that. LLM That text in the LLM process 236 00:13:21,400 --> 00:13:23,840 that that just doesn't seem very productive right now. 237 00:13:24,360 --> 00:13:25,960 I don't. I hope they can get better at 238 00:13:25,960 --> 00:13:28,840 that. I mean, I I've said, you know, 239 00:13:29,200 --> 00:13:32,240 it changed my vows. It made like a line and active 240 00:13:32,240 --> 00:13:35,200 voice, which was minor but sort of funny that they will forever 241 00:13:35,200 --> 00:13:40,960 have been impacted by Chachi BT I I generally find it in writing 242 00:13:40,960 --> 00:13:45,720 to be like a siren song in that it feels so easy and when I'm 243 00:13:45,720 --> 00:13:48,400 feeling lazy it's like this will generate text. 244 00:13:48,640 --> 00:13:53,200 But I honestly think it's like a bad starting point in that it's 245 00:13:53,200 --> 00:13:55,280 so off. Like I think I created a job 246 00:13:55,280 --> 00:13:58,480 posting with Chachi PT and we haven't heard any. 247 00:13:59,120 --> 00:14:00,600 We haven't heard that. I don't know. 248 00:14:00,600 --> 00:14:03,880 I blame Chachi. Somebody just told me I need to, 249 00:14:03,880 --> 00:14:06,120 like, chill it out. It's a little, you know, like, 250 00:14:06,800 --> 00:14:09,160 so I'm going to have a human rework that. 251 00:14:09,160 --> 00:14:12,240 I mean, I do, but it's good at, like, things that appear, which 252 00:14:12,240 --> 00:14:15,800 is why I'm calling a siren song. Like, it appears good. 253 00:14:15,800 --> 00:14:18,480 Like, it's like, oh, that's what a job posting looks like. 254 00:14:18,920 --> 00:14:22,000 But it doesn't always feel like it's sort of thought through on 255 00:14:22,000 --> 00:14:26,120 every piece of it. Have you tried pasting in an 256 00:14:26,120 --> 00:14:30,560 article of yours, like a draft and asking Chachi PT? 257 00:14:31,080 --> 00:14:32,840 Hey, is there anything I'm missing? 258 00:14:32,840 --> 00:14:35,000 What questions will the reader ask? 259 00:14:35,640 --> 00:14:39,520 That is, that's what I mean. I, I, you know even for this 260 00:14:39,520 --> 00:14:43,480 show I put in and it has been helpful prepping for the show. 261 00:14:43,480 --> 00:14:47,240 Like I put in our outlines and then said blah, blah, blah. 262 00:14:47,400 --> 00:14:50,120 Like. I mean it was helpful sort of 263 00:14:50,120 --> 00:14:53,000 filling in timelines and coming up with like papers that I think 264 00:14:53,000 --> 00:14:54,720 we referenced in the first episode. 265 00:14:55,200 --> 00:14:59,840 But I I don't think it's really achieved the mirror my style 266 00:14:59,840 --> 00:15:03,160 sort of thing and give me it. I feel like it doesn't adhere to 267 00:15:03,160 --> 00:15:07,160 my style enough which is really frustrating given how much of my 268 00:15:07,160 --> 00:15:10,920 writing is online that it's I'm like Eric Newcomer, a public 269 00:15:10,920 --> 00:15:15,280 figure who's written your XYBC plays like it doesn't really 270 00:15:15,280 --> 00:15:18,280 learn that much off of it. And you know, even when I give 271 00:15:18,280 --> 00:15:21,680 it a lot of text to prompt, I don't feel like it really gets 272 00:15:21,720 --> 00:15:26,920 my style very well. And I do think overall, yeah, I 273 00:15:26,920 --> 00:15:29,560 guess I haven't been critical enough so far in this podcast of 274 00:15:29,560 --> 00:15:35,040 how much I find the writing and thinking style to be not, not 275 00:15:35,040 --> 00:15:37,560 there, like it. I mean, again, I'm open to the 276 00:15:37,560 --> 00:15:39,760 idea that there are prompts that have to fix it. 277 00:15:42,240 --> 00:15:44,880 Yeah. Anyway, more more negative than 278 00:15:44,880 --> 00:15:48,560 I've been on AI, but I just, I feel like the threat to writing, 279 00:15:50,720 --> 00:15:53,040 you know, I believe it. If it gets like, you know, the 280 00:15:53,040 --> 00:15:56,480 next, if the improvement from to the next version is as great as 281 00:15:56,480 --> 00:15:59,800 it was from 3.5. I'm still open to the idea that 282 00:15:59,800 --> 00:16:02,520 it becomes a great writer. But I think it's great at 283 00:16:02,520 --> 00:16:06,080 writing poems and like more formulaic stuff. 284 00:16:06,080 --> 00:16:10,000 Like it's great at rhymes. Like I feel like I've, you know, 285 00:16:10,000 --> 00:16:13,040 come up with like funny, like here's the history of me and my 286 00:16:13,040 --> 00:16:18,200 wife, like write a poem, but but it, you know, they're, they're 287 00:16:18,200 --> 00:16:21,600 sort of gimmicks. Well, going back to the kind of 288 00:16:21,880 --> 00:16:26,640 integration into the coding environment idea, do you think, 289 00:16:27,160 --> 00:16:29,480 Have you tried any tools like that for writing? 290 00:16:29,960 --> 00:16:32,560 I know there's a bunch like writing specific ones. 291 00:16:32,560 --> 00:16:36,200 I you know, I know the guys that the guy who's doing Lex, he was 292 00:16:36,200 --> 00:16:40,080 sort of an X sub stack or from the newsletter every I need to. 293 00:16:40,160 --> 00:16:43,120 I haven't used his yet. Have you guys have either of you 294 00:16:43,120 --> 00:16:46,280 used any writing assistant ones? I don't use tome either to be 295 00:16:46,280 --> 00:16:48,520 honest, which I want. To try I use tome, yeah. 296 00:16:48,520 --> 00:16:52,600 What do you think of it? Not great tome is the tome, 297 00:16:53,440 --> 00:16:54,160 didn't you? Which one? 298 00:16:54,160 --> 00:16:56,120 Have you interviewed him? You interviewed him last time, 299 00:16:56,120 --> 00:16:59,440 right? I mean, I just think that tome 300 00:16:59,440 --> 00:17:01,880 is the PowerPoint. To give the generous, yeah, the 301 00:17:01,880 --> 00:17:04,560 generous interpretation. So it's basically Keynote or 302 00:17:04,560 --> 00:17:08,160 PowerPoint, but with with AI generation as the core creation 303 00:17:08,160 --> 00:17:11,640 mechanism. I mean it, it was impressive 304 00:17:11,640 --> 00:17:13,480 demo. I think it felt like a very cool 305 00:17:13,480 --> 00:17:16,560 demo for again. I mean I think it was more of 306 00:17:16,560 --> 00:17:19,040 the middle school, high school level of output, not even the 307 00:17:19,040 --> 00:17:22,480 college level of quality. And so if I were like in 6th 308 00:17:22,480 --> 00:17:25,319 grade and I needed to make like a book report about like 309 00:17:25,640 --> 00:17:27,960 Yosemite National Park or something, I think it's like a 310 00:17:27,960 --> 00:17:30,720 killer use case. For the record, that is what the 311 00:17:30,720 --> 00:17:34,600 CEO basically said was like the core use case was like homework 312 00:17:34,600 --> 00:17:36,680 and and such. So you know, not to give him a 313 00:17:36,680 --> 00:17:38,880 hard time, but as like a professional. 314 00:17:39,760 --> 00:17:42,440 You know, slide deck creation tool, it was really not there. 315 00:17:42,440 --> 00:17:44,520 Like what? Was yeah, what was missing, I 316 00:17:44,520 --> 00:17:47,800 guess. I would say to come back to this 317 00:17:47,800 --> 00:17:50,960 text generation being the big issue, like the content on the 318 00:17:50,960 --> 00:17:55,560 slides was really, really kind of empty and banal, right? 319 00:17:55,560 --> 00:18:01,160 It was like like it was the most sort of vague, like tautological 320 00:18:01,160 --> 00:18:04,360 bullet points, like in the slide deck and then. 321 00:18:04,720 --> 00:18:07,480 And then it was putting in generated artwork, right, Which 322 00:18:07,560 --> 00:18:09,280 again I used this like four or five months ago. 323 00:18:09,280 --> 00:18:12,440 So it was like pre the last two generations of mid journey or 324 00:18:12,440 --> 00:18:14,600 whatever. But the AI generated artwork was 325 00:18:14,600 --> 00:18:16,640 like really bad in like March and April. 326 00:18:17,000 --> 00:18:19,160 And so you were like well I could never use this in a real 327 00:18:19,160 --> 00:18:21,480 presentation. And so at some point it was sort 328 00:18:21,480 --> 00:18:24,720 of like the only thing this tool did for me was like make 10 329 00:18:24,720 --> 00:18:29,000 slides and like put essentially bullet points and I'm going to 330 00:18:29,000 --> 00:18:31,160 have to rewrite on them. And so it was like I could do 331 00:18:31,160 --> 00:18:33,520 this by. Copy pasting slides in in a 332 00:18:33,520 --> 00:18:36,080 normal presentation, you know framework and that's what I and 333 00:18:36,080 --> 00:18:39,800 in some cases having a starting point can be helpful, but in 334 00:18:39,800 --> 00:18:43,560 other cases it just sort of constrains your thinking 335 00:18:43,720 --> 00:18:48,080 unnecessarily. Yeah, I mean, yeah, I even if it 336 00:18:48,080 --> 00:18:50,080 doesn't constrain my thinking, to your point, it's sort of a 337 00:18:50,080 --> 00:18:52,280 siren song where I like, I'm like, oh cool, someone else is 338 00:18:52,280 --> 00:18:54,400 going to make my slide deck and then they make it slide deck and 339 00:18:54,400 --> 00:18:57,000 I'm like, OK, well, I can use about 10% of this, so. 340 00:18:57,520 --> 00:18:58,480 I'm not really sure why I did that. 341 00:18:59,040 --> 00:19:01,400 I just think it's good. It's good at cutting things, 342 00:19:01,440 --> 00:19:04,480 like, you know, I think when people are over long and you put 343 00:19:04,480 --> 00:19:10,000 a lot of text in it, it can all often strip away useless stuff. 344 00:19:10,440 --> 00:19:14,480 Sort of the summarization use case or yeah, yeah, I'm very 345 00:19:14,480 --> 00:19:17,040 excited about. Again, I think it's a great and 346 00:19:17,040 --> 00:19:20,440 we've sort of centered on this. It's great knowledge delivery 347 00:19:20,440 --> 00:19:24,080 tool like souped up sort of Google experience where Google 348 00:19:24,080 --> 00:19:27,400 was moving towards giving you the answer when you searched 349 00:19:27,400 --> 00:19:30,280 instead of directing you to a link like this like James was 350 00:19:30,280 --> 00:19:33,720 talking about with his child. Like I feel like it's great at 351 00:19:33,720 --> 00:19:37,200 sort of quick. Like I wish there was like a 352 00:19:37,520 --> 00:19:41,200 likely accurate score. Because I it does feel like what 353 00:19:41,200 --> 00:19:44,120 if one out of every ten of these is just like it's hallucinating 354 00:19:44,120 --> 00:19:48,680 and just making shit up. I'm I feel like humans including 355 00:19:48,680 --> 00:19:50,040 myself. I'm willing my. 356 00:19:50,080 --> 00:19:52,720 Daughter, My daughter's in trouble is, I know, but we're 357 00:19:52,720 --> 00:19:55,280 willing to accept like, oh, it's wrong 10% of the time. 358 00:19:55,280 --> 00:19:56,760 That's better than a lot of people. 359 00:19:57,440 --> 00:19:59,560 But then you just treat. You don't really know which ones 360 00:19:59,560 --> 00:20:01,920 were wrong, you know? So until we get burned. 361 00:20:02,200 --> 00:20:03,760 Yeah, the confidence score would be helpful. 362 00:20:03,840 --> 00:20:06,840 Yeah, I definitely have not never asked GPT for help with my 363 00:20:06,840 --> 00:20:09,080 daughter. Cause to your point, like if I'm 364 00:20:09,080 --> 00:20:12,200 like playing Russian roulette where it's like, oh, like like 365 00:20:12,200 --> 00:20:14,400 why that attendity? This is just wrong and I'm going 366 00:20:14,400 --> 00:20:17,160 to like take that away for my childcare. 367 00:20:17,640 --> 00:20:20,840 It's like, not great. Not a good, not a good idea, I 368 00:20:20,920 --> 00:20:22,840 mean. I like to think that I have some 369 00:20:22,840 --> 00:20:26,880 higher level filter of if it tells me to do something insane. 370 00:20:26,880 --> 00:20:30,320 I hope so, yeah. I mean, it's pretty good at 371 00:20:30,320 --> 00:20:32,920 plausible bullshit, though. I mean, that's the thing like 372 00:20:32,920 --> 00:20:36,200 it's it's really good at making wrong things sound plausible. 373 00:20:36,520 --> 00:20:41,360 All right, let's get out of our narcissism and think about how 374 00:20:41,360 --> 00:20:43,560 other people are using it. What? 375 00:20:44,040 --> 00:20:47,480 I mean, Max, I think you've made it clear, like students are the 376 00:20:47,480 --> 00:20:49,080 answer. Students are using this the 377 00:20:49,080 --> 00:20:50,720 most. And we you can talk about that a 378 00:20:50,720 --> 00:20:53,280 little bit more. And I totally agree. 379 00:20:53,280 --> 00:20:57,280 I think we all agree on that. But yeah, what professions do 380 00:20:57,280 --> 00:20:59,040 you think are using this the most? 381 00:20:59,040 --> 00:21:02,280 What types of people? And then we'll go from there and 382 00:21:02,280 --> 00:21:05,240 sort of what we think, who will be using it soon. 383 00:21:06,240 --> 00:21:09,160 It's pretty clear if you look at educational research that 384 00:21:09,760 --> 00:21:13,120 private tutoring is like the best possible thing you can do 385 00:21:13,120 --> 00:21:15,200 for learning in almost any context. 386 00:21:15,200 --> 00:21:18,760 Like one-on-one tutoring in like every research study that has 387 00:21:18,760 --> 00:21:21,800 ever been put to the test just absolutely destroys. 388 00:21:21,800 --> 00:21:26,040 Like the classroom environment or self-directed reading or you 389 00:21:26,040 --> 00:21:28,840 know, yadda yadda yadda what any, any sort of way you can 390 00:21:28,840 --> 00:21:32,080 think of to sort of teach people like one-on-one Private tutoring 391 00:21:32,080 --> 00:21:34,240 just kills for obvious reasons. I mean, right? 392 00:21:34,240 --> 00:21:37,640 You get personalized, you know, follow up feedback, custom 393 00:21:37,640 --> 00:21:41,440 lessons, all that good stuff. So you have to imagine, you 394 00:21:41,440 --> 00:21:44,840 know, if you could automate to some degree the learning process 395 00:21:44,840 --> 00:21:50,160 using, you know, a generative AI tool, whether or not that 396 00:21:50,160 --> 00:21:52,440 replaces teachers or that's just something teachers can use as 397 00:21:52,440 --> 00:21:55,920 leverage to really provide personalized feedback and lesson 398 00:21:55,920 --> 00:21:58,400 plans and and instruction to children. 399 00:21:58,680 --> 00:22:02,600 I just think that to come back to my initial analogy around 400 00:22:02,600 --> 00:22:04,880 this being like the industrial revolution for the mind, I 401 00:22:04,880 --> 00:22:08,600 really think we could see like insane gains in the quality of 402 00:22:08,600 --> 00:22:10,440 education over the next decade or so. 403 00:22:10,440 --> 00:22:14,120 Which is not, I think, with the intuition that a lot of people 404 00:22:14,120 --> 00:22:15,520 have, it's like, oh, it's helping them cheat. 405 00:22:15,520 --> 00:22:19,600 They're unplugging their brains. I think it's good if they're if 406 00:22:19,600 --> 00:22:22,240 the full experience is like interacting with the AI and 407 00:22:22,240 --> 00:22:26,800 going back and forth. And especially, you know, if the 408 00:22:26,960 --> 00:22:29,680 AI is set up to actually instruct so it knows when. 409 00:22:30,080 --> 00:22:33,240 To be withholding and not just like give every answer, that 410 00:22:33,240 --> 00:22:36,160 feels very powerful. But I do think the threat of 411 00:22:37,320 --> 00:22:41,920 it's easy to pretend to learn is like a real one and I'm I'm open 412 00:22:41,920 --> 00:22:43,680 to it going either way. I don't know. 413 00:22:43,680 --> 00:22:47,720 I mean this is sort of the Wall-e, you know like are people 414 00:22:47,720 --> 00:22:50,200 just going to be like laying back as the machine sort of 415 00:22:50,200 --> 00:22:53,080 taking over, but I think that might be slightly over indexing 416 00:22:53,080 --> 00:22:57,880 on like higher education, right? I think that if you think about 417 00:22:57,880 --> 00:23:02,280 what you learned up through almost like 8th grade right, a 418 00:23:02,280 --> 00:23:05,880 lot of it was like math, like addition, subtraction, 419 00:23:05,880 --> 00:23:07,760 multiplication and basic algebra, right? 420 00:23:08,080 --> 00:23:10,120 Phonics. Like reading, right? 421 00:23:10,920 --> 00:23:13,840 You know, I think reading comprehension can also be tested 422 00:23:13,840 --> 00:23:19,160 in person, You know, science. Just basics of the sort of core 423 00:23:19,160 --> 00:23:21,320 curriculum of elementary and middle school. 424 00:23:21,800 --> 00:23:27,160 I don't think like necessarily is going to be substituted for 425 00:23:27,160 --> 00:23:30,160 by like, oh, I need to dump an essay prompted to ChatGPT, 426 00:23:30,160 --> 00:23:31,640 right. I mean I think like, I think 427 00:23:31,640 --> 00:23:35,320 ChatGPT is obviously super good at the like give me a 5 428 00:23:35,320 --> 00:23:38,440 paragraph essay about Jane Eyre and like what the various 429 00:23:38,440 --> 00:23:41,880 symbolisms, you know in chapter four are or whatever, right. 430 00:23:41,880 --> 00:23:45,280 It's incredibly good at that. But that's sort of like a pretty 431 00:23:45,280 --> 00:23:48,200 narrow subset of what I would consider like almost all 432 00:23:48,200 --> 00:23:50,800 education all the way up to the college level. 433 00:23:50,800 --> 00:23:53,840 So I think that if you can give people private instruction on 434 00:23:54,680 --> 00:23:58,480 how to add you know or like you know really basic phonics and 435 00:23:58,480 --> 00:24:00,960 reading comprehension and grammar and sentence structure 436 00:24:00,960 --> 00:24:03,840 and you know science, geology, whatever, like all this stuff 437 00:24:03,840 --> 00:24:06,600 that you learn especially under the 5th or 8th grade level, I 438 00:24:06,600 --> 00:24:09,160 think it's hard to imagine we couldn't make really, really big 439 00:24:09,160 --> 00:24:11,840 gains in that area. And you know especially if you 440 00:24:11,840 --> 00:24:13,720 look at the United States where we fall short, a lot of it's 441 00:24:13,720 --> 00:24:16,680 just like making sure everyone reads really well by a certain 442 00:24:16,680 --> 00:24:18,360 age, right. And and I think AI could be 443 00:24:18,360 --> 00:24:20,600 pretty helpful there, particularly with voice and 444 00:24:20,600 --> 00:24:23,560 audio as like a multi modal experience, not just text. 445 00:24:24,120 --> 00:24:26,400 Don't offer a provocative one, I think. 446 00:24:27,560 --> 00:24:31,120 Have you read the book Bullshit Jobs or like this class Class of 447 00:24:31,120 --> 00:24:32,560 Tasks? Read a review of it, yeah. 448 00:24:33,600 --> 00:24:35,000 Like, I just missed it out of hand. 449 00:24:35,000 --> 00:24:39,080 No, I mean I would think you'd be sort of you'd buy it. 450 00:24:39,080 --> 00:24:42,840 I mean the idea is, you know that there are a ton of white 451 00:24:42,840 --> 00:24:47,960 collar jobs mostly that exist to sort of just move things around, 452 00:24:47,960 --> 00:24:51,080 don't create real value like a lot of, I mean, corporate, 453 00:24:51,080 --> 00:24:54,160 corporate law. I think is like a key one some 454 00:24:54,160 --> 00:24:59,000 of HR you know just a lot of where where is the real value 455 00:24:59,320 --> 00:25:04,040 versus just imposing a tax on a system that both sides have to 456 00:25:04,040 --> 00:25:09,600 pay. I mean to I think sort of not 457 00:25:09,600 --> 00:25:13,640 top tier marketing necessarily, but I think this sort of low 458 00:25:13,640 --> 00:25:18,080 tier marketing is obviously a case that I put in this bullshit 459 00:25:18,080 --> 00:25:21,640 jobs category where a sort of. And it's somewhat derivative. 460 00:25:21,640 --> 00:25:23,600 You know, like you really don't need to. 461 00:25:24,080 --> 00:25:27,280 You know, hopefully artists are coming up with many of the 462 00:25:27,280 --> 00:25:30,360 original ideas and so you're trying to sort of be in touch 463 00:25:30,360 --> 00:25:34,760 with culture and pull something. What do you mean exactly by low 464 00:25:34,760 --> 00:25:36,080 tier marketing? I guess 'cause. 465 00:25:36,080 --> 00:25:39,400 Like I just mean like you know people or you know, just I think 466 00:25:39,400 --> 00:25:44,120 obviously like, I mean I think there will be an AI Super Bowl 467 00:25:44,120 --> 00:25:47,800 ad, but I put that in sort of a. Sure, it will be like the 468 00:25:47,800 --> 00:25:51,080 creativity of it, but you know, that's going to be reviewed by a 469 00:25:51,080 --> 00:25:52,440 human, right? I mean, like if there's an AI 470 00:25:52,440 --> 00:25:54,160 Super Bowl. But I'm just, I'm saying like, 471 00:25:54,160 --> 00:25:57,280 OK, somebody's like I'm putting up a website for our company and 472 00:25:57,280 --> 00:26:00,120 we need to explain all this stuff and you read all that copy 473 00:26:00,520 --> 00:26:03,440 and it's just like this is like, no, nobody paid somebody who's 474 00:26:03,440 --> 00:26:05,680 like great at writing to do something novel. 475 00:26:05,680 --> 00:26:09,000 They're like, what does website copy look like? 476 00:26:09,440 --> 00:26:13,280 What does our company do? Let let's create website copy 477 00:26:13,280 --> 00:26:16,120 for. Plus and so then, if that's your 478 00:26:16,120 --> 00:26:18,880 standard of what you want, you just want to have a website that 479 00:26:18,880 --> 00:26:23,200 has things that are not wrong. I think ChatGPT and other 480 00:26:23,200 --> 00:26:25,680 similar products will be able to deliver that. 481 00:26:25,840 --> 00:26:27,960 Yeah, I think you're right. I just would push back. 482 00:26:28,360 --> 00:26:30,440 So I probably would have like fully agreed with you like five 483 00:26:30,440 --> 00:26:35,360 years ago, but now, now, now having now having run a start up 484 00:26:35,360 --> 00:26:37,560 and and experience a lot of these things that people 485 00:26:37,560 --> 00:26:41,120 consider to be bullshit chops. I think the thing that you maybe 486 00:26:41,120 --> 00:26:43,800 don't realize having been like close to the metal on like 487 00:26:43,800 --> 00:26:48,080 putting up a website or whatever is there is actually quite a lot 488 00:26:48,080 --> 00:26:50,840 of people in the world who will sell you the service of putting 489 00:26:50,840 --> 00:26:54,440 up a generic website that will do a pretty bad job. 490 00:26:54,440 --> 00:26:58,560 Like a worse job than you're describing as a sort of like you 491 00:26:58,560 --> 00:27:02,760 know an AI quality or a low tier quality output, right. 492 00:27:02,760 --> 00:27:06,320 People will like, literally not even deliver at all. 493 00:27:06,320 --> 00:27:09,120 Or they will deliver something that's like grammatically just 494 00:27:09,120 --> 00:27:11,480 incomprehensible. Or they will deliver something 495 00:27:11,640 --> 00:27:13,560 that's grammatically comprehensible but really quite 496 00:27:13,560 --> 00:27:15,360 bad and does not describe what the product does. 497 00:27:15,640 --> 00:27:19,720 So you will encounter like just a staggering number of services 498 00:27:19,720 --> 00:27:22,240 people that this. Could cover is much larger than 499 00:27:22,240 --> 00:27:23,640 right. I'm saying that, but I'm saying 500 00:27:23,640 --> 00:27:26,720 in the end like the leverage point ends up being like the 501 00:27:26,720 --> 00:27:30,160 person who can look at the really bad version of it and be 502 00:27:30,160 --> 00:27:33,600 like well actually there's like a mid tier version of this that 503 00:27:33,600 --> 00:27:37,240 we could just get an AI to do right and and go get the AI to 504 00:27:37,240 --> 00:27:39,800 do it and then decide side that that's actually good enough for 505 00:27:39,800 --> 00:27:41,360 for showbiz to go on the website, right. 506 00:27:41,640 --> 00:27:46,120 So I'm just saying like I I think that I think that that the 507 00:27:46,120 --> 00:27:49,360 humans can always sort of move up the stack I guess you know 508 00:27:49,480 --> 00:27:51,240 until we get super intelligence or whatever. 509 00:27:51,240 --> 00:27:55,840 But like where the the key the key job is deciding what is like 510 00:27:55,840 --> 00:27:59,680 good enough output, not like creating good enough output in 511 00:27:59,680 --> 00:28:03,680 many cases I guess I would say. I mean like and yeah, a core 512 00:28:03,680 --> 00:28:07,440 question to me is like from what you're just saying there. 513 00:28:08,360 --> 00:28:14,920 Is does AI make dumb people smarter or smart people higher 514 00:28:14,920 --> 00:28:17,000 leverage? Like in the most brutal way? 515 00:28:17,000 --> 00:28:19,040 Like, is it? And it's not just smart and 516 00:28:19,040 --> 00:28:21,880 dumb, but like in your case, it's like, does it help the 517 00:28:21,880 --> 00:28:26,000 person with taste do a ton of jobs, or does it help have the 518 00:28:26,000 --> 00:28:30,240 person with sort of poor taste or like bad execution skills? 519 00:28:30,240 --> 00:28:33,360 But maybe they know it's not good and therefore they can do 520 00:28:33,360 --> 00:28:36,120 better, like I think a lot of the takes have been. 521 00:28:36,680 --> 00:28:40,280 It helps sort of the the person who needs more help, yeah. 522 00:28:40,280 --> 00:28:42,760 But there's also this argument that it's scaling people. 523 00:28:42,960 --> 00:28:46,000 I think it's different on taste, I guess, But I do think the 524 00:28:46,040 --> 00:28:48,520 early academic research that's coming out is showing that. 525 00:28:49,360 --> 00:28:52,440 I think they gave it to lawyers who were in, in law school, 526 00:28:52,440 --> 00:28:56,120 basically, and they let them use it on their tests essentially. 527 00:28:56,120 --> 00:28:59,800 And it did. So they had a curve, you know, 528 00:28:59,800 --> 00:29:02,480 everybody kind of knew what their grades were going into the 529 00:29:02,480 --> 00:29:05,920 AI assisted test basically. And it did pull up the bottom 530 00:29:05,920 --> 00:29:08,760 half of the curve quite a lot and then the top half of the 531 00:29:08,760 --> 00:29:11,280 curve maybe a little bit, but not a huge impact. 532 00:29:11,280 --> 00:29:14,280 So at least in the law and I don't know if this will be 533 00:29:14,280 --> 00:29:16,560 different and more, you know, different types of fields, more 534 00:29:16,560 --> 00:29:18,320 creative fields versus more structured fields. 535 00:29:18,320 --> 00:29:19,640 It's it's a little bit hard to predict. 536 00:29:19,640 --> 00:29:22,520 But in the case with the law where there's sort of clear 537 00:29:22,520 --> 00:29:26,040 right and wrong answers and there's a value in having 538 00:29:26,240 --> 00:29:30,200 massive text digestion machine essentially which is what these 539 00:29:30,200 --> 00:29:33,360 LLMS are, right. It seemed like it was helping 540 00:29:33,360 --> 00:29:35,840 the bottom of the curve substantially more than it was, 541 00:29:35,840 --> 00:29:38,280 giving like a ton of leverage to the top of the curve. 542 00:29:38,560 --> 00:29:41,720 I would also add there. This kind of reminds me of an 543 00:29:41,800 --> 00:29:45,920 analogy with GPSI. Forget who's who said this 544 00:29:45,920 --> 00:29:51,480 originally, but we went from an. An era where we were to 545 00:29:51,480 --> 00:29:54,080 navigate, we had someone maybe in the passenger seat just 546 00:29:54,080 --> 00:29:57,440 pulling out maps and you know, they could actually help 547 00:29:57,440 --> 00:30:00,520 navigate in natural language and you know get you to where you 548 00:30:00,520 --> 00:30:04,120 wanted to go or you had to pull over and you know pull out the 549 00:30:04,120 --> 00:30:08,160 maps yourself. And then we went to GPS and the 550 00:30:08,160 --> 00:30:11,320 navigation at first got I was actually worse than having 551 00:30:11,320 --> 00:30:13,960 another person just look like a human look at the maps. 552 00:30:14,720 --> 00:30:18,200 But it did make. On the whole, like a lot of more 553 00:30:18,200 --> 00:30:21,040 convenient trips, you could just, you know, throw on the 554 00:30:21,040 --> 00:30:24,680 navigation. Similar with early call centers. 555 00:30:25,000 --> 00:30:27,960 Like it was nice to when you originally it was all humans and 556 00:30:27,960 --> 00:30:30,120 you just talk to them and you get customer support. 557 00:30:30,400 --> 00:30:33,840 And then with call center automation, many, many more 558 00:30:33,840 --> 00:30:36,840 firms could create call centers, but they were way worse, Like 559 00:30:36,840 --> 00:30:41,080 you couldn't get the answer that you wanted and that, you know, 560 00:30:41,080 --> 00:30:43,200 was frustrating for consumers. And I think we're going to see 561 00:30:43,200 --> 00:30:45,600 something very similar here with AI that. 562 00:30:47,240 --> 00:30:49,880 Just like Max was saying, like AI can write spam emails. 563 00:30:49,880 --> 00:30:52,000 Like for sales. Like those emails are going to 564 00:30:52,000 --> 00:30:55,880 be way worse than a human writing a spam sales e-mail and 565 00:30:55,880 --> 00:30:57,200 you're going to get way more of them. 566 00:30:57,200 --> 00:31:01,160 So overall, our lives as startup founders for now might get worse 567 00:31:01,160 --> 00:31:05,080 because we're getting 10 times as much spam e-mail and trying 568 00:31:05,080 --> 00:31:09,200 to sell us stuff, but on net. Maybe that's good for the 569 00:31:09,200 --> 00:31:13,000 economy or something like more firms can use spam e-mail 570 00:31:13,000 --> 00:31:16,760 tactics to sell products, so. I I think that a lot of times 571 00:31:17,280 --> 00:31:21,800 that the the technology creates the ability to scale things that 572 00:31:21,800 --> 00:31:25,840 weren't scalable before but on net makes that experience with 573 00:31:25,840 --> 00:31:28,600 the customer worse. And I think we're going to start 574 00:31:28,600 --> 00:31:31,560 seeing that and that that might because it's so much cheaper to 575 00:31:32,200 --> 00:31:34,600 do so much. That might go for website copy 576 00:31:34,600 --> 00:31:36,280 too. Like it'll be way easier to 577 00:31:36,280 --> 00:31:39,000 throw up a website on the on the Internet, but the average 578 00:31:39,000 --> 00:31:41,640 website quality will go down because. 579 00:31:42,120 --> 00:31:45,680 The writing will be worse and anyone will have the ability to 580 00:31:45,680 --> 00:31:47,920 put a website up and code it with AI. 581 00:31:47,920 --> 00:31:51,960 And yeah, I I think I don't expect that to like raise the 582 00:31:51,960 --> 00:31:55,240 quality of the average website. That ties into Max's tutor 583 00:31:55,240 --> 00:31:56,200 point, right? I mean it. 584 00:31:56,280 --> 00:31:59,120 It's democratization to make something that was expensive, 585 00:31:59,120 --> 00:32:04,520 private tutoring cheaper, and possibly the original quality is 586 00:32:04,520 --> 00:32:07,640 lower. But way more people can access 587 00:32:07,640 --> 00:32:08,960 it. One one last point on the 588 00:32:08,960 --> 00:32:11,280 website topic, actually there's this a very funny story I 589 00:32:11,280 --> 00:32:14,320 experienced where I talked to investor and he said one company 590 00:32:14,320 --> 00:32:21,360 in his portfolio had used GPT to create 8 million websites, 8 591 00:32:21,360 --> 00:32:24,360 million web pages essentially that were all meant to 592 00:32:24,360 --> 00:32:28,040 essentially attract Google search traffic essentially. 593 00:32:28,040 --> 00:32:33,400 And then they would use the how many clicks those various pages 594 00:32:33,400 --> 00:32:36,920 got to like decide what the next thing is that they should work 595 00:32:36,920 --> 00:32:39,080 on or the next thing they should create as their company. 596 00:32:39,080 --> 00:32:42,280 So essentially like, yeah, they were very early on this. 597 00:32:42,280 --> 00:32:47,160 Like we're going to just like absolutely, just like carpet 598 00:32:47,160 --> 00:32:52,600 bomb any keyword imaginable that is remotely related to the 599 00:32:52,600 --> 00:32:56,040 business and then figure out where the user traffic is and 600 00:32:56,040 --> 00:32:59,600 essentially use it as like honey pot or whatever to see where the 601 00:32:59,600 --> 00:33:01,320 traffic is. And then that's what we're going 602 00:33:01,320 --> 00:33:03,360 to build next. And so I think your idea 603 00:33:03,360 --> 00:33:06,720 specifically of just the number of spam websites going through 604 00:33:06,720 --> 00:33:10,400 the roof clearly is is either happening, going to happening. 605 00:33:10,400 --> 00:33:11,840 Yeah. I mean is is going to be a huge 606 00:33:11,840 --> 00:33:14,520 part of and and Google search is really going to have to deal 607 00:33:14,520 --> 00:33:16,440 with this. Like, how do we deal with the 608 00:33:16,440 --> 00:33:20,760 fact that it's now become, you know, three orders of magnitude 609 00:33:20,760 --> 00:33:22,760 cheaper to create a crappy website? 610 00:33:22,760 --> 00:33:25,040 And how do we filter through that with Google search? 611 00:33:25,080 --> 00:33:27,760 I feel like a theme that's emerging on this podcast already 612 00:33:27,760 --> 00:33:32,240 is the idea that AI is going to be at war with itself, Like it's 613 00:33:32,240 --> 00:33:35,720 going to bring improvements, but it's also, it'll make it easier 614 00:33:35,720 --> 00:33:38,600 for spammers. But we have to hope that Gmail 615 00:33:38,600 --> 00:33:42,040 also gets better at filtering them out and that to some degree 616 00:33:42,320 --> 00:33:45,640 there's going to be sort of a back and forth there and that if 617 00:33:45,680 --> 00:33:48,640 if that works out well. But you know that will, that 618 00:33:48,640 --> 00:33:50,960 will be for the best. I mean even spammers. 619 00:33:51,080 --> 00:33:55,120 I mean in James's GPS analogy. Like, the end point is that they 620 00:33:55,120 --> 00:33:58,640 have to write good spam that is compelling to you, and that that 621 00:33:58,640 --> 00:34:01,800 that's a winner rather than just living in sort of the shitty 622 00:34:01,800 --> 00:34:05,400 spam era forever. Yeah, but I think the spam 623 00:34:05,400 --> 00:34:08,600 e-mail example is is important here because. 624 00:34:09,280 --> 00:34:13,719 They can annoy 99% of people who won't convert, but if it helps 625 00:34:13,719 --> 00:34:17,360 convert any extra 1% of people, then that's a win for that 626 00:34:17,360 --> 00:34:20,159 company, right? It could be 99.99. 627 00:34:20,159 --> 00:34:22,960 I mean it could be one out of it could be one out of 10,000 of 628 00:34:22,960 --> 00:34:25,199 these emails convert or 100. Like on spam. 629 00:34:25,199 --> 00:34:29,159 Eric, you're naively optimistic. Sweet. 630 00:34:29,360 --> 00:34:33,199 All right, we. I am interviewing Joshua Browder 631 00:34:33,360 --> 00:34:37,000 at Do Not Pay. So stick around and give that a 632 00:34:37,000 --> 00:34:40,400 listen. Welcome, welcome. 633 00:34:42,639 --> 00:34:46,600 Joshua Browder, CEO of Do Not Pay, welcome to the Cerebral 634 00:34:46,600 --> 00:34:49,760 Valley Podcast. Thank you Eric for having me. 635 00:34:50,120 --> 00:34:54,080 Yeah, I mean, I I, you know, you've had such a great journey. 636 00:34:54,080 --> 00:34:57,040 I think some people might be familiar, you know, just from 637 00:34:57,560 --> 00:35:00,720 Twitter following it. But can you talk about, you 638 00:35:00,720 --> 00:35:02,560 know, I want to end this conversation, really focus on 639 00:35:02,560 --> 00:35:05,200 what do not Pay is doing in AI, but like. 640 00:35:05,600 --> 00:35:08,240 You've got such a great journey. If you could just give a little 641 00:35:08,240 --> 00:35:11,400 bit of the sketch of how you get into this world of helping 642 00:35:11,400 --> 00:35:15,800 people fight, you know, bills, traffic tickets, everything, 643 00:35:16,280 --> 00:35:18,280 yeah. And then how, how that sort of 644 00:35:18,280 --> 00:35:20,800 evolves into this sort of generative AI world? 645 00:35:21,480 --> 00:35:23,240 Sure. So I'm the founder of the 646 00:35:23,240 --> 00:35:27,160 company called Do Not Pay and Do not pay as an AI legal agent. 647 00:35:27,560 --> 00:35:29,880 So. There are so many areas in life 648 00:35:29,920 --> 00:35:33,760 where people are being ripped off, from parking tickets to not 649 00:35:33,760 --> 00:35:37,800 being able to cancel their subscriptions, to junk fees, and 650 00:35:38,080 --> 00:35:41,560 no one has time to wait on hold for five hours to argue over 651 00:35:41,560 --> 00:35:43,600 $12.00. And so that's a really good job 652 00:35:43,640 --> 00:35:47,120 for AI and software. I started the company six years 653 00:35:47,120 --> 00:35:50,600 ago with templates. When I moved from England, I got 654 00:35:50,600 --> 00:35:53,760 a bunch of parking tickets. I was a terrible driver. 655 00:35:54,040 --> 00:35:56,600 And I realized if you know the right things to say, you can get 656 00:35:56,600 --> 00:35:58,520 out of your tickets. You're you're speeding or you're 657 00:35:58,520 --> 00:36:01,280 just putting, you're putting your car wherever or. 658 00:36:02,000 --> 00:36:06,400 In in the UK test they don't test how to do parallel parking, 659 00:36:06,400 --> 00:36:09,000 so I wasn't particularly good at that skill, especially on the 660 00:36:09,000 --> 00:36:11,600 other side of the road. Hey, listen, I grew up in Macon, 661 00:36:11,600 --> 00:36:15,000 GA and you did not. I failed the parallel parking 662 00:36:15,240 --> 00:36:16,880 part of my test. I didn't know you only had like 663 00:36:16,880 --> 00:36:18,680 three tries to try and get it in. 664 00:36:19,160 --> 00:36:21,560 But I didn't. You know you could pass without 665 00:36:21,560 --> 00:36:23,720 without succeeding on that portion anyway. 666 00:36:23,720 --> 00:36:27,160 Yeah, so. I think in the UK test they do 667 00:36:27,160 --> 00:36:30,120 have it, but it's random and I got another random one that was 668 00:36:30,120 --> 00:36:32,120 really easy. I think my one was a three-point 669 00:36:32,120 --> 00:36:35,040 turn, which I can do in any case. 670 00:36:35,040 --> 00:36:37,200 So I created the first version of Do not Pay. 671 00:36:37,200 --> 00:36:40,360 I looked at the top 12 reasons from Freedom of Information Act 672 00:36:40,360 --> 00:36:42,880 requests. A wide parking tickets are 673 00:36:42,880 --> 00:36:45,360 dismissed. Built templates around that and 674 00:36:45,360 --> 00:36:48,400 really just built it for fun. And I could never have imagined 675 00:36:48,400 --> 00:36:50,720 that. America and the UK and the world 676 00:36:50,720 --> 00:36:53,160 would hate parking tickets and it would go viral. 677 00:36:53,440 --> 00:36:56,560 And this made me realize that this idea of automating consumer 678 00:36:56,560 --> 00:36:59,760 rights is bigger than just tickets and I should work on 679 00:36:59,760 --> 00:37:02,440 other areas of fighting back. So. 680 00:37:02,440 --> 00:37:07,280 Fastball parking tickets thing. 20/15/2016 so I was a freshman 681 00:37:07,280 --> 00:37:09,640 at Stanford. And at first, it's not automated 682 00:37:09,640 --> 00:37:11,160 at all. Like you're doing it personally 683 00:37:11,160 --> 00:37:13,680 yourself and people are like asking you for favors. 684 00:37:13,920 --> 00:37:16,640 Is that what gets you into sort of automating things in the 1st 685 00:37:16,640 --> 00:37:20,920 place or had you been a sort of? In other domains trying to to do 686 00:37:21,320 --> 00:37:24,000 automated stuff. Yeah, just from a time 687 00:37:24,000 --> 00:37:26,000 perspective. People were asking me and I was 688 00:37:26,000 --> 00:37:29,600 in Google Docs back then copying and pasting letters. 689 00:37:29,880 --> 00:37:32,480 And I just built it, made it automated because too many 690 00:37:32,480 --> 00:37:36,960 people were asking me and what a rumour had spread that I was the 691 00:37:36,960 --> 00:37:38,760 guy who could help people with parking tickets. 692 00:37:38,760 --> 00:37:41,440 And that's not a good rumour to have about you because everyone 693 00:37:41,440 --> 00:37:43,920 will bother you. And so I just built it out of 694 00:37:43,920 --> 00:37:47,000 necessity, really. When does it become a company? 695 00:37:48,040 --> 00:37:52,880 So in 20/16/2017 I was very lucky. 696 00:37:52,880 --> 00:37:56,400 Andreessen Horowitz invested 1,000,000 into the precede 697 00:37:56,400 --> 00:37:59,240 rounds. I still didn't know have any 698 00:37:59,240 --> 00:38:02,400 business model or anything like that, but I guess it became a 699 00:38:02,400 --> 00:38:06,560 company legally at that point. But it was really on 2019-2020 700 00:38:06,560 --> 00:38:08,040 that I actually started charging. 701 00:38:08,240 --> 00:38:11,400 So I just saw it as a free public service until I started 702 00:38:11,400 --> 00:38:14,520 charging. I did it not to do a start up or 703 00:38:14,520 --> 00:38:17,040 make money, but because I like helping people fight back. 704 00:38:17,480 --> 00:38:21,200 I like think it's so unjust all of these big corporations and 705 00:38:21,200 --> 00:38:25,600 governments and so I was more mission like activism driven and 706 00:38:25,600 --> 00:38:28,400 that's proven by like 5 years of no business model. 707 00:38:29,440 --> 00:38:31,640 A good crusade is important. You know what? 708 00:38:31,720 --> 00:38:34,440 You can get investors to give you money to go on a crusade. 709 00:38:34,440 --> 00:38:40,040 What's better than that? The I mean the the rise of 710 00:38:40,040 --> 00:38:45,560 ChatGPT Foundation models. What I I I I can guess that a 711 00:38:45,560 --> 00:38:50,360 key value for you is that it can generate like demand letters on 712 00:38:50,360 --> 00:38:52,920 its own. Like what have been the pieces 713 00:38:53,320 --> 00:38:57,760 of large language models that have been most useful so far for 714 00:38:57,760 --> 00:39:00,880 Do Not Pay? Really, it's synchronous 715 00:39:00,880 --> 00:39:03,520 responses. So in the realm of what we can 716 00:39:03,520 --> 00:39:06,000 fight back, parking tickets is asynchronous. 717 00:39:06,000 --> 00:39:08,800 You generate A latter, send it off, and wait three weeks. 718 00:39:09,080 --> 00:39:11,960 But there are some things that you need to do synchronously. 719 00:39:11,960 --> 00:39:14,160 So for example. Bill disputes. 720 00:39:14,480 --> 00:39:17,400 So one way you can dispute your Comcast bill is you log in and 721 00:39:17,400 --> 00:39:20,440 you go into online chat and you negotiate with them for an hour 722 00:39:20,440 --> 00:39:23,160 on online chat. That is a great job for AI 723 00:39:23,160 --> 00:39:26,720 because it doesn't give up. And better yet if the agent 724 00:39:26,720 --> 00:39:29,960 denies the request you can just start a new chat and that's what 725 00:39:29,960 --> 00:39:32,840 we have bots doing. So they just keep trying, keep 726 00:39:32,840 --> 00:39:35,360 pushing until we get a bill reduced. 727 00:39:35,600 --> 00:39:37,920 So synchronous is 1 and not giving up. 728 00:39:37,920 --> 00:39:40,840 I think are are two and also being dynamic. 729 00:39:42,040 --> 00:39:45,760 But not the letters itself. Or are are you using it for sort 730 00:39:45,760 --> 00:39:50,760 of demand letters? So the contents of the latter we 731 00:39:50,760 --> 00:39:54,000 want to kind of free craft, but really it's about the 732 00:39:54,000 --> 00:39:55,960 communications out surrounding that. 733 00:39:56,280 --> 00:39:58,680 It doesn't really differ from person to person. 734 00:39:58,680 --> 00:40:02,000 So for example, the top reason why people get out of tickets is 735 00:40:02,000 --> 00:40:04,560 the poor signage. We've all seen those memes on 736 00:40:04,560 --> 00:40:07,200 Twitter of Like 5 signs saying the opposite thing. 737 00:40:07,520 --> 00:40:09,840 On top of each other or a tree covering the sign. 738 00:40:10,120 --> 00:40:13,120 So the substance of the defence doesn't really change with AI, 739 00:40:13,120 --> 00:40:15,760 but the way you can communicate it does. 740 00:40:16,360 --> 00:40:20,480 You have a person who has to figure out the the argument, 741 00:40:21,760 --> 00:40:26,640 yeah, to build the kind of system and then and then AI can 742 00:40:26,640 --> 00:40:28,240 go back and forth communicate, yeah. 743 00:40:29,160 --> 00:40:32,720 And I I'm not, you know, I guess for context for the listener, 744 00:40:32,720 --> 00:40:35,960 you had this whole you know, we'll be your lawyer period or 745 00:40:35,960 --> 00:40:38,840 can you talk about? Where sort of the legal and 746 00:40:38,840 --> 00:40:42,600 policy realm has been something you've run into and sort of the 747 00:40:42,600 --> 00:40:48,360 quest to operate as like an attorney, basically and the 748 00:40:48,360 --> 00:40:51,920 protectionism that you found in that crusade. 749 00:40:53,080 --> 00:40:56,840 Yeah, so consumer rights is completely underserved. 750 00:40:56,840 --> 00:40:59,920 There's not a lawyer who will get out of bed to help you with 751 00:40:59,920 --> 00:41:02,080 your parking ticket, which is our first use case. 752 00:41:02,080 --> 00:41:05,800 Or Dell disputes or any of this, especially in flight Wi-Fi 753 00:41:05,800 --> 00:41:09,600 refunds which are literally $30.00 so. 754 00:41:09,600 --> 00:41:12,000 When Oh yeah, I've had that where the Wi-Fi goes down and 755 00:41:12,000 --> 00:41:13,880 then you have to like you're like you didn't give me a 756 00:41:13,880 --> 00:41:17,040 service, like how can you charge me a bazillion dollars And then 757 00:41:17,040 --> 00:41:19,040 it's hard to ask. For the refund, because you 758 00:41:19,040 --> 00:41:21,400 don't have Wi-Fi, it's ridiculous. 759 00:41:21,400 --> 00:41:23,680 Yeah, yeah, that that's a great job. 760 00:41:25,240 --> 00:41:28,480 There was a time for about two months at the beginning of this 761 00:41:28,480 --> 00:41:31,120 year where I wanted to actually bring ChatGPT into the 762 00:41:31,120 --> 00:41:35,040 courtroom, and I got a lot of push back from that from from 763 00:41:35,040 --> 00:41:37,800 the judges and regulators. Because which I love. 764 00:41:37,800 --> 00:41:40,760 To be clear, our supporter 100% on this podcast. 765 00:41:40,800 --> 00:41:45,200 Anyway, go ahead because it's nice to help people online, but 766 00:41:45,200 --> 00:41:48,400 if you can actually go into court, you can go into even more 767 00:41:48,400 --> 00:41:51,280 advanced areas. And they push back so hard that 768 00:41:51,280 --> 00:41:54,040 I thought, I'm not getting any push back with consumer rights. 769 00:41:54,040 --> 00:41:57,360 Everyone hates Comcast. Even the lawyers do, and so it's 770 00:41:57,360 --> 00:42:00,400 best to just stick to the undeserved area to help people. 771 00:42:01,520 --> 00:42:03,480 The provocative fight, you know, like I'm going to be your 772 00:42:03,480 --> 00:42:06,280 lawyer, you know, you get PR attention out of it, but then 773 00:42:06,280 --> 00:42:09,080 yeah, you bring down the regulatory heat. 774 00:42:09,080 --> 00:42:11,680 So there's a double edged sword there. 775 00:42:12,280 --> 00:42:15,960 I mean I get the the philosophical issue here or like 776 00:42:15,960 --> 00:42:20,520 they're they're just like the sense that there are systems at 777 00:42:20,520 --> 00:42:25,160 war against the consumer, right. Like a Comcast, you know, can 778 00:42:25,160 --> 00:42:28,280 think about the process and then you were an individual, not 779 00:42:28,280 --> 00:42:31,840 teamed up with other consumers. Sort of fighting with your arms 780 00:42:31,840 --> 00:42:35,280 tied behind your back and and similarly, you know, I think 781 00:42:35,280 --> 00:42:38,920 part of the threat to the legal system is that this would expose 782 00:42:38,920 --> 00:42:42,640 that it's sort of a system of repeat actions that happen over 783 00:42:42,640 --> 00:42:46,640 and over again and slight adjustments and even lazy ones, 784 00:42:46,800 --> 00:42:49,560 you know, lazy and that you know that it's a computer, not like 785 00:42:49,560 --> 00:42:52,520 some person could have a big, big change. 786 00:42:52,520 --> 00:42:55,920 So talk about you know how this is sort of going up against 787 00:42:56,280 --> 00:42:58,480 these systems on behalf of the consumer. 788 00:42:59,520 --> 00:43:03,440 My most optimistic view of AI is it will end up patterns. 789 00:43:03,680 --> 00:43:06,080 So right now we have this problem in Society of 790 00:43:06,080 --> 00:43:08,360 concentrated benefit but spread out harm. 791 00:43:08,640 --> 00:43:12,000 So what I mean by that is Planet Fitness can make it super hard 792 00:43:12,000 --> 00:43:16,040 to cancel and charge 10 million people $30. 793 00:43:16,200 --> 00:43:21,240 They make a 300 million but the average person is only getting 794 00:43:21,240 --> 00:43:24,480 $30 taken out of their account. So they find it very difficult 795 00:43:24,480 --> 00:43:26,640 to fight back. And with Planet Fitness 796 00:43:26,640 --> 00:43:29,800 specifically in most gyms, you have to actually sign a physical 797 00:43:29,800 --> 00:43:31,640 legal latter and mail it to cancel. 798 00:43:32,560 --> 00:43:36,000 And so all of these hoops that that these big companies made 799 00:43:36,000 --> 00:43:39,440 you jump through AI can jump through it because it doesn't 800 00:43:39,440 --> 00:43:42,680 require a salary and doesn't have anything to do. 801 00:43:42,880 --> 00:43:46,240 So you can get it to do that. And then all of these kind of 802 00:43:46,240 --> 00:43:49,680 systems that prevent people from take fighting for their rights 803 00:43:50,480 --> 00:43:53,680 can can go through. Another example is privacy. 804 00:43:54,760 --> 00:43:57,120 There's an amazing law in California, the California 805 00:43:57,120 --> 00:44:00,320 Consumer Privacy Act, but it's largely been a failure because 806 00:44:00,320 --> 00:44:02,480 no one really has exercised their rights. 807 00:44:02,480 --> 00:44:05,840 The the law says that you can request to delete your data or 808 00:44:05,840 --> 00:44:08,880 not sell your data, but no one has time to fill in all the 809 00:44:08,880 --> 00:44:11,760 forms. We have AI going send 1000 810 00:44:11,760 --> 00:44:14,960 requests to every data broker on behalf of consumers. 811 00:44:15,320 --> 00:44:19,040 So I think AI will really help level the playing field and give 812 00:44:19,040 --> 00:44:22,640 power back to ordinary people to fight this concentrated benefit 813 00:44:22,640 --> 00:44:24,880 problem. So you can you talk about like 814 00:44:24,880 --> 00:44:29,800 the mechanics of using AI, like is this ChatGPT or what are you 815 00:44:29,800 --> 00:44:32,440 finding right now that's the most useful to you? 816 00:44:33,480 --> 00:44:36,200 So we're mainly using the GPT for API. 817 00:44:37,000 --> 00:44:39,800 So for example, with Comcast, what's interesting is they're 818 00:44:39,800 --> 00:44:43,520 also using AI, so the two AIS are sometimes chatting with each 819 00:44:43,520 --> 00:44:45,320 other to negotiate. They're already a war, yeah, 820 00:44:45,600 --> 00:44:48,040 yeah. And it feels like we're making 821 00:44:48,040 --> 00:44:51,440 so much progress every few weeks and months with the new 822 00:44:51,440 --> 00:44:52,920 releases. So the two. 823 00:44:53,000 --> 00:44:57,720 So the biggest was when GPT 3 upgraded to GPT 4, it became a 824 00:44:57,720 --> 00:45:01,880 much better negotiator. So with GPT 3, Comcast would be 825 00:45:01,880 --> 00:45:04,600 like, OK, I'll give you $20 off your bill. 826 00:45:05,720 --> 00:45:08,240 And GPT 3 would say, yeah, that sounds great. 827 00:45:08,240 --> 00:45:11,400 Thank you so much. Now GPT 4 says, no, that's not 828 00:45:11,400 --> 00:45:13,520 enough. I want $100 or I'm going to 829 00:45:13,520 --> 00:45:16,480 cancel right now. And it does these high stakes 830 00:45:16,480 --> 00:45:20,320 negotiations that I don't think previous AI models were capable 831 00:45:20,320 --> 00:45:22,080 of. And then the thing I'm really 832 00:45:22,080 --> 00:45:24,600 excited about now is the multi modal stuff. 833 00:45:25,720 --> 00:45:29,080 AI is not useful unless it can interact with the world and 834 00:45:29,200 --> 00:45:33,200 being able to send it images and all sorts of different media 835 00:45:33,480 --> 00:45:37,480 PDFs as well is really helpful, especially in the legal context. 836 00:45:37,480 --> 00:45:40,600 So you can imagine it looking at like a parking Bay and saying 837 00:45:40,600 --> 00:45:43,520 that parking Bay is not to code, which is why I should get out of 838 00:45:43,520 --> 00:45:45,480 my ticket. So that's all the stuff we're 839 00:45:45,480 --> 00:45:46,120 working on. Right. 840 00:45:46,200 --> 00:45:48,680 Now you sort of explained this in the answer, but multimodal 841 00:45:48,680 --> 00:45:52,000 just like putting in text and images into the same system, not 842 00:45:52,000 --> 00:45:54,800 totally separate system so that it could like you're saying, 843 00:45:55,080 --> 00:45:59,040 process a photo and then respond in text to it. 844 00:45:59,280 --> 00:46:04,440 How much are you saying Open AI is the expert at ChatGPT like 845 00:46:04,440 --> 00:46:05,720 they're going to keep improving it. 846 00:46:05,720 --> 00:46:09,320 A general language model is just going to be better than anything 847 00:46:09,320 --> 00:46:13,560 we're doing versus the idea that like you are now specialized in 848 00:46:13,560 --> 00:46:15,840 this. Use case of like consumer 849 00:46:15,840 --> 00:46:20,360 defence and therefore you should like train a model on those 850 00:46:20,360 --> 00:46:22,480 interactions. Like how do you think about how 851 00:46:22,480 --> 00:46:26,120 much to sort of fine tune or build your own language model. 852 00:46:27,160 --> 00:46:32,200 I I think the the kind of sweet spot is you have these kind of 853 00:46:32,200 --> 00:46:34,640 big commercial models and then you fine tune them. 854 00:46:34,840 --> 00:46:38,440 So what we do is we feed it like successful cases and and 855 00:46:38,440 --> 00:46:40,640 existing letters that we think are good. 856 00:46:41,080 --> 00:46:43,520 Any answers, like every time they're like, here are the 857 00:46:43,520 --> 00:46:47,160 models of the good ones. And then, yeah, yeah, like here, 858 00:46:47,160 --> 00:46:50,080 here's an example. So when we were doing our 859 00:46:50,080 --> 00:46:54,480 courtroom stuff, we didn't open. AI is so valuable to our 860 00:46:54,480 --> 00:46:56,560 business, we didn't want to get banned by doing this 861 00:46:56,560 --> 00:46:58,240 controversial. Stunt interesting. 862 00:46:58,400 --> 00:47:02,680 And so and so we said we were going to use open source model. 863 00:47:02,680 --> 00:47:05,840 So we had to go deep on the open source ones to get it to work. 864 00:47:06,280 --> 00:47:10,880 And it feels like it's like 75 percent, 80% of the way there at 865 00:47:10,880 --> 00:47:12,680 this point, the open source models. 866 00:47:12,680 --> 00:47:16,440 So I'm very confident that they'll continue to improve and 867 00:47:16,680 --> 00:47:20,280 it will always be at that level. So I'm a big believer in open 868 00:47:20,280 --> 00:47:23,520 source and and there are going to be things if you want to have 869 00:47:23,520 --> 00:47:25,880 a big impact, there are going to be things that are quote UN 870 00:47:25,880 --> 00:47:28,440 quote against the rules. So you you will have to use an 871 00:47:28,440 --> 00:47:30,680 open source model to do some interesting stuff. 872 00:47:31,440 --> 00:47:33,240 You're fighting for the consumer, but you're there's 873 00:47:33,240 --> 00:47:36,000 already sort of the status control, I mean under it makes 874 00:47:36,000 --> 00:47:39,360 sense like you know. Open AI doesn't want to be 875 00:47:39,360 --> 00:47:43,880 exposed to your sort of niche like principled fight, even 876 00:47:43,880 --> 00:47:46,520 though they probably should. That's super interesting. 877 00:47:47,000 --> 00:47:47,840 How? Which? 878 00:47:48,520 --> 00:47:50,320 Which open source models have you been using? 879 00:47:50,320 --> 00:47:55,320 Like which ones do you see the most promising matters? 880 00:47:55,320 --> 00:47:58,440 One is good we in January, we're using GPTJ. 881 00:47:58,760 --> 00:48:04,240 Yeah, GPTJ is is good at holding a conversation if you kind of 882 00:48:04,440 --> 00:48:07,680 fine tune it enough so that that's what we were using to 883 00:48:07,680 --> 00:48:13,120 power the courtroom stunt. And is it expensive on Chachi BT 884 00:48:13,120 --> 00:48:16,800 just like to build a business around it, Like how how much are 885 00:48:16,880 --> 00:48:20,120 you paying? We're paying 10s of thousands 886 00:48:20,120 --> 00:48:23,160 every month. I would say that for us it makes 887 00:48:23,160 --> 00:48:24,560 sense. We're a subscription model. 888 00:48:24,560 --> 00:48:27,760 We fortunate to have a lot of subscribers and a subscription 889 00:48:27,760 --> 00:48:31,920 is a great business model. There are some things everyone 890 00:48:32,000 --> 00:48:34,480 that they don't do these great demos on Twitter, but we're 891 00:48:34,480 --> 00:48:37,320 thinking as a do not pay team, this would never work if it 892 00:48:37,320 --> 00:48:39,200 wasn't in the do not pay business model. 893 00:48:39,480 --> 00:48:42,400 So one example is every few months someone comes up with an 894 00:48:42,400 --> 00:48:46,520 idea of ChatGPT, scanning websites for terms and 895 00:48:46,520 --> 00:48:50,040 conditions, whether it's good terms and conditions or bad. 896 00:48:50,400 --> 00:48:55,240 And probably each scan maybe costs two cents or something 897 00:48:55,240 --> 00:48:56,920 along those lines, one or two cents. 898 00:48:57,840 --> 00:49:00,240 I don't think that's something consumers would pay for. 899 00:49:00,440 --> 00:49:03,640 Imagine if you're being charged $0.02 every website or you visit 900 00:49:03,960 --> 00:49:05,160 so that. Would be a great browser. 901 00:49:05,560 --> 00:49:07,800 Feature like Chrome's like, you know, it's like how with a 902 00:49:07,800 --> 00:49:10,840 credit card you get a bunch of features you could imagine 903 00:49:10,840 --> 00:49:12,520 someone paying for like a browser. 904 00:49:12,520 --> 00:49:15,520 If it's like, one of the things we do is we say this has really 905 00:49:15,760 --> 00:49:17,480 aberrant like terms and conditions. 906 00:49:17,480 --> 00:49:20,400 So we should, you know, that would be an interesting company. 907 00:49:20,400 --> 00:49:24,640 Yeah, it it would. But so we have 200 products and 908 00:49:24,640 --> 00:49:28,000 from my experience, it seems like consumers can most about 909 00:49:28,000 --> 00:49:30,360 getting money back. That's the number one thing 910 00:49:30,720 --> 00:49:32,640 privacy they do, they do care about. 911 00:49:32,640 --> 00:49:35,880 There are certain niches of people, but the mass market 912 00:49:35,880 --> 00:49:39,960 people are just trying to get by every day and so it it would 913 00:49:39,960 --> 00:49:42,960 have to be very cheap for them. Do you get a? 914 00:49:43,160 --> 00:49:45,840 Have you found any ways to get a percentage of the money you save 915 00:49:45,840 --> 00:49:49,400 people? We think it's unfair. 916 00:49:49,400 --> 00:49:53,800 So Equinox is $300.00 a month. We help people cancel Equinox 917 00:49:53,800 --> 00:49:55,720 subscriptions. We could. 918 00:49:55,800 --> 00:49:58,040 There are some companies that say we've helped you cancel 919 00:49:58,080 --> 00:50:01,080 Equinox subscription. We've saved you $3600. 920 00:50:01,080 --> 00:50:04,400 We're going to take 20% of that. And we believe that business 921 00:50:04,400 --> 00:50:06,440 model is not actually consumer friendly. 922 00:50:07,160 --> 00:50:10,520 We'd rather and we have 100% success rate with cancelling 923 00:50:10,520 --> 00:50:13,280 Equinox. We'd rather just charge the $13 924 00:50:13,440 --> 00:50:15,960 a month subscription. We'd rather just charge the 925 00:50:15,960 --> 00:50:18,560 subscription. How would you judge like the 926 00:50:18,560 --> 00:50:22,960 intelligence of ChatGPT and the open source models based on what 927 00:50:22,960 --> 00:50:24,760 you've seen you you've talked about? 928 00:50:25,240 --> 00:50:28,000 You know, oh like, wow. It can really start to like, 929 00:50:28,000 --> 00:50:31,760 make threats and like, say, OK, this is a final, like, you know, 930 00:50:31,760 --> 00:50:34,760 you see a ton of this now. Like, does it feel like Super 931 00:50:34,760 --> 00:50:38,080 Rodent has the same playbook, or does this feel like something 932 00:50:38,080 --> 00:50:42,600 with intelligence to you? It's extremely intelligent. 933 00:50:43,120 --> 00:50:46,320 From the legal standpoint, GPT 3 was not so impressive. 934 00:50:46,320 --> 00:50:48,720 It would make very basic mistakes, like mix up the 935 00:50:48,720 --> 00:50:51,200 defendant and the plaintiff and all of this stuff. 936 00:50:52,280 --> 00:50:54,480 GPT 4, those mistakes have disappeared. 937 00:50:55,520 --> 00:50:58,680 My biggest worry with it is just like just like humans, it's 938 00:50:58,680 --> 00:51:01,120 almost too intelligent, so it gets on the verge of 939 00:51:01,120 --> 00:51:05,560 manipulative, so those who can use it face liability from it. 940 00:51:05,560 --> 00:51:08,440 Potentially lying. I think there's a difference 941 00:51:08,440 --> 00:51:12,720 between hallucinating and lying. So with lying, if the goal in 942 00:51:12,720 --> 00:51:16,600 the prompt is like I want to get a discount on my bill, it will 943 00:51:16,600 --> 00:51:19,560 say things like I've had four outages in the past week. 944 00:51:19,840 --> 00:51:22,160 Oh no, that's not true. Oh my God. 945 00:51:23,080 --> 00:51:26,080 So we've had to kind of build guard rails about around it in 946 00:51:26,080 --> 00:51:27,960 the prompt. And also we even have another 947 00:51:28,000 --> 00:51:31,520 kind of pseudo AI model on top of it and the prompt says like 948 00:51:31,520 --> 00:51:34,160 stick to the facts, stick to the provided information. 949 00:51:34,400 --> 00:51:37,840 So because we have liability, we we can't have it lying on behalf 950 00:51:37,840 --> 00:51:39,920 of people. That's not good. 951 00:51:39,920 --> 00:51:43,520 We're responsible for that, so. That's fascinating, yeah. 952 00:51:43,520 --> 00:51:47,080 The distinction between hallucinations where it's just 953 00:51:47,600 --> 00:51:50,640 it's just confused basically in lying where it knows that it's a 954 00:51:50,640 --> 00:51:52,560 tactical. Move. 955 00:51:54,560 --> 00:51:56,400 Yeah. I mean, I mean there is, you 956 00:51:56,400 --> 00:51:58,240 know, there's an argument that it doesn't have enough 957 00:51:58,520 --> 00:52:00,680 information. Like you know it because it. 958 00:52:01,000 --> 00:52:02,560 Right. You know, that you could face 959 00:52:02,560 --> 00:52:07,520 like legal consequences. Yeah, that's a fascinating one. 960 00:52:09,040 --> 00:52:12,320 Are there cases, I mean, people, consumers, do lie to these 961 00:52:12,320 --> 00:52:14,440 companies sometimes when they try and get out of it? 962 00:52:14,960 --> 00:52:18,320 Are are there cases where you're OK with where that's allowed or 963 00:52:18,320 --> 00:52:21,640 you basically have to be 100% factual? 964 00:52:23,360 --> 00:52:26,720 I I think you can be aggressive so you can say I'm not happy 965 00:52:26,720 --> 00:52:29,920 with the service, it barely works, things like that. 966 00:52:30,760 --> 00:52:32,960 But you don't want to be opinions. 967 00:52:33,400 --> 00:52:36,760 I'm going to cancel tomorrow and maybe they're the only Internet 968 00:52:36,760 --> 00:52:40,920 company in the area. So you can't cancel but so, so 969 00:52:40,920 --> 00:52:42,160 those things I'm comfortable with. 970 00:52:42,160 --> 00:52:45,720 But right specific concrete details we we don't feel 971 00:52:45,720 --> 00:52:48,760 comfortable doing it. We we have our own we have some 972 00:52:48,760 --> 00:52:51,320 of the best lawyers that do not pay that we were hired 973 00:52:51,320 --> 00:52:54,520 ironically and and we're we're trying to stay in compliance. 974 00:52:54,520 --> 00:52:57,640 So it will help us a lot. Are you going to go back into 975 00:52:57,640 --> 00:53:00,880 the legal realm at any point or are you you sticking to fighting 976 00:53:00,880 --> 00:53:05,120 companies or I guess you? Fight companies in the legal 977 00:53:05,120 --> 00:53:08,160 realm, Yeah, in the legal. There's a lot of interesting 978 00:53:08,160 --> 00:53:10,880 stuff that we're working on now that's definitely in the legal 979 00:53:10,880 --> 00:53:14,560 realm, but it's focused on helping consumers with their 980 00:53:14,560 --> 00:53:16,600 everyday issues. So one thing that I'm really 981 00:53:16,600 --> 00:53:21,160 excited about is everyone has money lying out there from class 982 00:53:21,160 --> 00:53:24,200 action settlements. There's like $20 here. 50 983 00:53:24,200 --> 00:53:27,320 dollars there, because if you're a customer of Macy's a few years 984 00:53:27,320 --> 00:53:30,120 ago, so we've built like a bot that will go through your 985 00:53:30,120 --> 00:53:33,920 emails, figure out which companies you're owed class 986 00:53:33,920 --> 00:53:36,000 action settlements for, and just claim the money. 987 00:53:36,200 --> 00:53:38,040 So that's squarely in the legal realm. 988 00:53:38,040 --> 00:53:40,760 But once again, it's underserved and it's just helping ordinary 989 00:53:40,760 --> 00:53:42,920 people. We're not trying to help defend 990 00:53:42,920 --> 00:53:45,080 people from murder, although that would be. 991 00:53:45,080 --> 00:53:47,360 We're not Sam Bang Bun Freed's trials today. 992 00:53:47,360 --> 00:53:49,560 But we can't help Sam Bang Bun Freed. 993 00:53:49,760 --> 00:53:52,920 But we can help ordinary people who are don't have these 994 00:53:52,920 --> 00:53:55,000 complicated issues in their lives. 995 00:53:55,400 --> 00:53:59,560 How much? How much do you use AI, Chat, 996 00:53:59,560 --> 00:54:02,960 UBT any of this in your personal life? 997 00:54:02,960 --> 00:54:05,760 Like you're someone who likes to automate your problems. 998 00:54:05,760 --> 00:54:09,160 To what extent have you found it useful outside of your company? 999 00:54:10,960 --> 00:54:13,760 My company is just me trying to scale myself. 1000 00:54:14,080 --> 00:54:17,640 And so almost all our products start with either me or another 1001 00:54:17,640 --> 00:54:20,600 team actually doing this in their personal life. 1002 00:54:21,960 --> 00:54:26,280 So one example is I built a bot that phoned up Wells Fargo on my 1003 00:54:26,280 --> 00:54:30,040 behalf, and I actually created a voice model of myself and it 1004 00:54:30,040 --> 00:54:32,960 negotiated with Wells Fargo to get some bank fees back. 1005 00:54:33,200 --> 00:54:35,960 We decided not to make that into a product for a variety of 1006 00:54:35,960 --> 00:54:39,520 reasons, mainly to actually make it a working product. 1007 00:54:39,520 --> 00:54:42,600 You need to record what they're saying and transcribe it so the 1008 00:54:42,640 --> 00:54:45,400 AI knows what to say back. And there's lots of federal 1009 00:54:45,400 --> 00:54:50,280 wiretapping laws and state wiretapping laws and so we 1010 00:54:50,320 --> 00:54:52,520 couldn't make that product, but it was helpful for me. 1011 00:54:52,760 --> 00:54:56,760 Another example is I using Plaid, I connected chat TBT to 1012 00:54:56,760 --> 00:55:00,520 my bank account and I got it to come up with different ways I 1013 00:55:00,520 --> 00:55:03,520 could save money so that people do not pay products around that. 1014 00:55:03,760 --> 00:55:06,360 So I'm definitely a Guinea pig everyday for these AI 1015 00:55:06,360 --> 00:55:09,240 technologies, which is I love. I like to say I live the do not 1016 00:55:09,240 --> 00:55:12,120 pay lifestyle. I'm always fighting back. 1017 00:55:12,560 --> 00:55:16,320 Is it always, it feels like ChatGPT looms super large. 1018 00:55:16,320 --> 00:55:19,880 I mean there is this question of whether open source models or 1019 00:55:20,400 --> 00:55:24,040 Google Gemini or what, you know, whatever will be sort of 1020 00:55:24,040 --> 00:55:26,920 competitive. Are are you sort of saying right 1021 00:55:26,920 --> 00:55:33,040 now that it's chat GPTS to lose? I would say that the the only 1022 00:55:33,040 --> 00:55:35,800 reason they would lose is if they restrain themselves too 1023 00:55:35,800 --> 00:55:37,320 much. I think they're under a lot of 1024 00:55:37,320 --> 00:55:39,400 pressure. I know lawyers are evil and very 1025 00:55:39,400 --> 00:55:41,880 greedy and I think that a lot of lawyers are actually going off 1026 00:55:41,880 --> 00:55:45,000 the ChatGPT right now and they're saying you have to put 1027 00:55:45,000 --> 00:55:47,360 this disclaimer and that disclaimer and stop doing this 1028 00:55:47,360 --> 00:55:49,880 and stop doing that. And then the AI will just not be 1029 00:55:49,880 --> 00:55:53,080 useful and then they'll be like some foreign company where the 1030 00:55:53,280 --> 00:55:56,840 US users can use the API that might overtake it. 1031 00:55:57,200 --> 00:56:01,360 I I think we're on the verge of like society collapsing from all 1032 00:56:01,360 --> 00:56:05,680 the new regulations and lawyers, and so I do think there's a real 1033 00:56:05,680 --> 00:56:08,480 danger around that. Yeah, Just there's sort of 1034 00:56:08,480 --> 00:56:12,400 regulatory legal state gets to it that creates sort of an 1035 00:56:12,400 --> 00:56:17,040 opening for the unregulated open source world. 1036 00:56:17,960 --> 00:56:20,480 So I I can give you a concrete example actually. 1037 00:56:20,480 --> 00:56:25,000 So one when I used ChatGPT was I actually connected my credit 1038 00:56:25,000 --> 00:56:30,840 report to ChatGPT and I asked it to like dispute things on my 1039 00:56:30,840 --> 00:56:33,200 behalf. And then I recently tried to do 1040 00:56:33,200 --> 00:56:35,520 it again, and then I got all these disclaimers and it 1041 00:56:35,520 --> 00:56:38,240 wouldn't even let me. So they're they're definitely 1042 00:56:38,240 --> 00:56:40,840 trying to all these high, high risk areas 1 by 1. 1043 00:56:40,840 --> 00:56:45,080 They're kind of foreclosing on them and stopping useful 1044 00:56:45,480 --> 00:56:48,560 implementations. Do you, you know, do you think 1045 00:56:48,560 --> 00:56:53,040 it's like white collar workers are better at being Luddites and 1046 00:56:53,040 --> 00:56:55,840 protecting their jobs? That, and they're happy to let. 1047 00:56:56,200 --> 00:56:58,600 Sort of factory workers jobs be automated. 1048 00:56:58,600 --> 00:57:01,440 It's like lawyers protecting their jobs or it's just sort of 1049 00:57:01,840 --> 00:57:05,640 paranoia or what do you think motivates this? 1050 00:57:06,520 --> 00:57:08,960 The white collar workers are definitely going to put up a 1051 00:57:08,960 --> 00:57:13,240 fight with the Actors Guild. I know in their, I'm sorry, the 1052 00:57:13,280 --> 00:57:16,160 Screenwriters Guild, they put in their contract that they can use 1053 00:57:16,160 --> 00:57:19,360 the AI, but the studios can't, which seems seems a bit 1054 00:57:19,720 --> 00:57:22,280 unbalanced. Lawyers are the ones who write 1055 00:57:22,280 --> 00:57:24,640 the rules and so of course they have an advantage. 1056 00:57:25,680 --> 00:57:29,040 I know a lot of state legislatures are passing AI 1057 00:57:29,040 --> 00:57:31,160 laws. I think Utah was was the latest. 1058 00:57:31,160 --> 00:57:32,920 They have a few AI bills coming up. 1059 00:57:33,200 --> 00:57:36,960 It feels to me like we are going to get to a place where we each 1060 00:57:36,960 --> 00:57:40,880 have our own sort of bots. Like the advantage being it 1061 00:57:40,880 --> 00:57:44,360 knows the facts, like I trusted to be in my bank account. 1062 00:57:44,600 --> 00:57:47,880 It sort of has the the data advantage on me that it has the 1063 00:57:47,880 --> 00:57:51,640 full picture, sort of train the personality around what I how I 1064 00:57:51,640 --> 00:57:55,400 like to interact with people, you know, a bazillion reasons. 1065 00:57:55,680 --> 00:57:58,200 Do you see that as a threat to you? 1066 00:57:58,240 --> 00:58:01,720 You know you're in some ways like specialized around a set of 1067 00:58:01,720 --> 00:58:03,720 tasks. But if if you're going to be, 1068 00:58:03,720 --> 00:58:06,160 would you think you'll be competing against a different 1069 00:58:06,160 --> 00:58:09,320 model where it's all about the person and therefore they do a 1070 00:58:09,320 --> 00:58:12,880 lot of stuff. We've we've thought about this 1071 00:58:12,880 --> 00:58:16,040 for a while because consumer rights what what happens if 1072 00:58:17,040 --> 00:58:19,920 Google just build an assistant that do all your consumer rights 1073 00:58:19,920 --> 00:58:22,520 for you and and you're right it should be just part of a super 1074 00:58:22,520 --> 00:58:25,320 assistant. Fortunately we are so anti 1075 00:58:25,320 --> 00:58:28,600 authority that we're willing to do things to help people that 1076 00:58:28,600 --> 00:58:30,960 they love that big companies won't. 1077 00:58:31,720 --> 00:58:35,320 I imagine that Comcast has a huge advertiser on Google and so 1078 00:58:35,320 --> 00:58:38,720 they're not trying to be too adversarial and so we think of 1079 00:58:38,720 --> 00:58:42,640 things like adversarial AI. So plug insurance is a great 1080 00:58:42,640 --> 00:58:44,720 example. When ChatGPT released plug 1081 00:58:44,720 --> 00:58:48,280 Insurance, a lot of people when we thought is Comcast going to 1082 00:58:48,280 --> 00:58:52,240 build a plug in to lower your bill or or Planet Fitness going 1083 00:58:52,240 --> 00:58:55,320 to build a plug in to cancel your subscription from ChatGPT 1084 00:58:55,640 --> 00:58:58,360 And I think the answer is obviously no because it's 1085 00:58:58,360 --> 00:59:01,280 adversarial and so this adversarial kind of use case 1086 00:59:01,280 --> 00:59:04,440 helps us. I I think what's going to happen 1087 00:59:04,440 --> 00:59:07,360 is there'll be these niche application layer on top of 1088 00:59:07,360 --> 00:59:10,800 these general assistant for these specialised areas that 1089 00:59:10,800 --> 00:59:15,600 require either a different brand or some specialised expertise 1090 00:59:15,600 --> 00:59:17,480 that Google is not willing to take on. 1091 00:59:18,120 --> 00:59:20,760 Yeah, that that makes. A lot of sense. 1092 00:59:20,760 --> 00:59:24,320 Do you have anything like sort of the general assistant or how 1093 00:59:24,320 --> 00:59:28,920 much time are you spending getting a single AI with enough 1094 00:59:28,920 --> 00:59:32,000 memory to keep track of all of your your information? 1095 00:59:32,400 --> 00:59:35,920 Well, so we have chat bots that talk to our consumers and our 1096 00:59:35,920 --> 00:59:39,080 consumers what expect jokes they they they talk to it about 1097 00:59:39,080 --> 00:59:41,520 irrelevant issues. So we have thought a lot about 1098 00:59:41,520 --> 00:59:43,880 the general use case and we want to make it entertaining. 1099 00:59:45,040 --> 00:59:47,960 Eventually. We have a kind of we draw the 1100 00:59:47,960 --> 00:59:51,080 line and it says I'm, I'm actually an AI legal assistant, 1101 00:59:51,080 --> 00:59:53,080 so please can we like stick to that topic. 1102 00:59:53,320 --> 00:59:55,000 But we we have a few jokes in there. 1103 00:59:55,560 --> 00:59:58,720 But do you like when you're doing the Plaid integration with 1104 00:59:59,360 --> 01:00:02,000 your bank account or credit report, is that all synced up? 1105 01:00:02,360 --> 01:00:04,800 Like, do you sort of either through a prompt, say hey, 1106 01:00:04,800 --> 01:00:07,080 here's everything we've done together in the past, or like 1107 01:00:07,320 --> 01:00:11,000 try to buy extra memory in some way, use the API to like how 1108 01:00:11,000 --> 01:00:13,920 much are you trying to, like, tie different tasks of yours 1109 01:00:13,920 --> 01:00:18,240 together in one sort of prompt? Well, I guess there's two ways 1110 01:00:18,240 --> 01:00:19,800 to do it. So you can either have a very 1111 01:00:19,800 --> 01:00:23,120 large context window and I think anthropics right now is the 1112 01:00:23,120 --> 01:00:25,000 biggest and so that's really exciting. 1113 01:00:25,160 --> 01:00:28,800 Or you can build like these auto GPT implementations where it's 1 1114 01:00:28,800 --> 01:00:34,040 task after another we we've gone the auto GPT route. 1115 01:00:34,040 --> 01:00:37,840 So it's like incremental and I I think that's the solution, but 1116 01:00:37,840 --> 01:00:41,240 hopefully there's like one day there'll be this AI with this 1117 01:00:41,240 --> 01:00:44,160 massive contacts window and then that'll make things so much 1118 01:00:44,160 --> 01:00:46,920 easier for us. With with the auto GPT, you know 1119 01:00:46,920 --> 01:00:49,960 it can be. It feels like they need to have 1120 01:00:49,960 --> 01:00:52,960 some response every time. It's like very hard to just say 1121 01:00:53,400 --> 01:00:56,720 here's some information just say OK and then let me give you more 1122 01:00:56,720 --> 01:01:00,040 information like is it a problem that it always seems to want to 1123 01:01:00,040 --> 01:01:06,000 be completing some task with every answer if if it's very 1124 01:01:06,000 --> 01:01:08,520 discreet and you've really kind of narrowed it. 1125 01:01:08,640 --> 01:01:11,040 I I've seen it work. So we've used it. 1126 01:01:11,280 --> 01:01:15,320 It goes on the government website and if if say you move 1127 01:01:15,320 --> 01:01:18,040 address and you own a check and they can't the big companies 1128 01:01:18,040 --> 01:01:20,320 can't reach you they just send it to the government and so it 1129 01:01:20,320 --> 01:01:23,840 becomes an unclaimed money. So we've built an auto GPT bot 1130 01:01:23,840 --> 01:01:27,000 that checks for unclaimed money and the reason we have to use 1131 01:01:27,000 --> 01:01:31,440 auto GPT rather than just like selenium or define scraping is 1132 01:01:31,440 --> 01:01:35,600 because the the web page changes and what we found is it can be 1133 01:01:35,600 --> 01:01:38,760 useful and actually it's pretty successful at that use case. 1134 01:01:39,080 --> 01:01:44,000 If you said to auto GPT order me a pizza the website will say 1135 01:01:44,680 --> 01:01:48,720 like which topping do you want and then auto GPT will be stuck. 1136 01:01:49,040 --> 01:01:52,760 So if a discrete task is good but the more open end, I think 1137 01:01:52,760 --> 01:01:57,040 we are a long way from general AI assistant, maybe like a few 1138 01:01:57,040 --> 01:01:59,600 years, which which is not that long, but it's not like 1139 01:01:59,600 --> 01:02:03,560 tomorrow, right? Have you do you have you gotten 1140 01:02:03,560 --> 01:02:06,680 a taste of where Chachi PD is headed? 1141 01:02:06,680 --> 01:02:08,960 Or do you have you seen many beta products? 1142 01:02:08,960 --> 01:02:10,840 Or what? Do you do you think the rate of 1143 01:02:10,840 --> 01:02:14,520 improvement is going to slow down a lot after the current 1144 01:02:14,520 --> 01:02:17,320 model? I I do. 1145 01:02:17,320 --> 01:02:22,080 I think the current we we have a like a million applications 1146 01:02:22,080 --> 01:02:24,600 waiting to be built. The Do Not Pay team is even 1147 01:02:24,600 --> 01:02:28,280 swamped in our use case and so I think the next year or two is 1148 01:02:28,280 --> 01:02:31,000 going to be really exciting with these AI applications being 1149 01:02:31,000 --> 01:02:34,440 introduced to every product. Like, we're going to get better, 1150 01:02:34,440 --> 01:02:38,360 but the models maybe aren't, well, not just us, but the 1151 01:02:38,360 --> 01:02:40,600 whole, right, right. I think we'll build incredible 1152 01:02:40,600 --> 01:02:42,560 things. You'll see sales tools, all of 1153 01:02:42,560 --> 01:02:46,840 the stuff, customer service, phone lines, but it's just a 1154 01:02:46,840 --> 01:02:49,040 statistical model at the end of the day. 1155 01:02:49,360 --> 01:02:53,920 And what worries me is that I think GPT 4 is at the limit of 1156 01:02:53,920 --> 01:02:56,200 the number of parameters. I'm not sure how many more 1157 01:02:56,280 --> 01:02:58,360 parameters and data they can see then. 1158 01:02:58,560 --> 01:03:01,360 They're sucked in everything there is to know, Yeah, they're 1159 01:03:01,360 --> 01:03:04,120 sucked in everything that is. So I think there has to be a 1160 01:03:04,120 --> 01:03:07,240 fundamental breakthrough with how these models are designed, 1161 01:03:07,400 --> 01:03:10,480 which hasn't happened yet. I I think it will happen because 1162 01:03:10,480 --> 01:03:13,240 there's so much money flowing into the space, as you know, but 1163 01:03:14,360 --> 01:03:18,000 it's not like it's where it's secretly been done and just 1164 01:03:18,000 --> 01:03:21,360 hasn't been released yet. Maybe that's above my pay grade, 1165 01:03:21,360 --> 01:03:24,080 but I I don't know of that and I don't think that's happened. 1166 01:03:24,400 --> 01:03:28,720 What what do you make of the politeness of Chachi BT or I? 1167 01:03:28,720 --> 01:03:32,000 I find one of the big weaknesses, just like how 1168 01:03:32,000 --> 01:03:34,880 insufferably polite it is. Like I I think I said this in an 1169 01:03:34,880 --> 01:03:37,960 earlier episode but you know, I beg it to behave like George 1170 01:03:37,960 --> 01:03:40,520 Carlin or something. It it feels like impossible. 1171 01:03:40,720 --> 01:03:43,640 And it's funny, you know, given you're sort of trying to teach 1172 01:03:43,640 --> 01:03:47,440 it to be this like against the system, sort of, it seems like 1173 01:03:47,600 --> 01:03:50,560 person is. It's personality is not disposed 1174 01:03:50,560 --> 01:03:53,800 towards that type of behaviour or how do you how do you address 1175 01:03:53,800 --> 01:03:55,760 that and like why do you think that's happening. 1176 01:03:56,680 --> 01:03:59,920 This is a big issue for us. So in our prompt we definitely 1177 01:03:59,920 --> 01:04:03,120 say you imagine you're an aggressive lawyer. 1178 01:04:03,200 --> 01:04:07,120 The the note that doesn't care about being polite, have to get 1179 01:04:07,120 --> 01:04:09,400 the exact prompt for consumers. They have to. 1180 01:04:09,600 --> 01:04:12,640 They you can put in your own kind of general prompt for the 1181 01:04:12,640 --> 01:04:15,160 entire ChatGPT. And there are some amazing 1182 01:04:15,160 --> 01:04:19,120 examples on Twitter, and some of them are like don't be polite, 1183 01:04:19,120 --> 01:04:23,040 don't apologize, don't say you're an AI, don't caveat it 1184 01:04:23,040 --> 01:04:24,840 with any disclaimers or warnings. 1185 01:04:24,840 --> 01:04:26,480 And that seems to work for people. 1186 01:04:26,480 --> 01:04:28,280 So I would recommend that as well. 1187 01:04:28,480 --> 01:04:31,440 Yeah, I need to put it because I'll put it in the actual text 1188 01:04:31,440 --> 01:04:33,040 in the beginning. It'll still apologize. 1189 01:04:33,040 --> 01:04:37,040 And then I'll say what was the number one thing I told you not 1190 01:04:37,040 --> 01:04:39,920 to do. And it'll, you know, it'll know 1191 01:04:39,920 --> 01:04:41,880 it. I said not to apologize, but 1192 01:04:41,880 --> 01:04:44,360 then it won't implement it. It can be sort of inscrutable. 1193 01:04:45,080 --> 01:04:49,680 I don't know how it prioritises requests it it could be good. 1194 01:04:49,680 --> 01:04:53,080 Maybe they're an open AI program app, because not not being 1195 01:04:53,080 --> 01:04:55,160 polite is a very harmless request. 1196 01:04:55,160 --> 01:04:57,840 But people might want to drag jailbreak it, so they're 1197 01:04:57,840 --> 01:05:01,640 probably trying to build all these barriers to jailbreaking 1198 01:05:01,640 --> 01:05:05,000 it. Yeah, when you write an e-mail 1199 01:05:05,040 --> 01:05:08,920 like, is ChatGPT involved or like in sort of what day-to-day 1200 01:05:08,920 --> 01:05:12,400 tasks like where? How much are you bringing AAI or 1201 01:05:12,400 --> 01:05:15,480 is it mostly around the things your company does where it's 1202 01:05:15,840 --> 01:05:18,640 sort of, you know, yeah, consumer defence. 1203 01:05:19,920 --> 01:05:24,200 I think in kind of communication people can tell. 1204 01:05:24,360 --> 01:05:26,080 I think there's this insult around. 1205 01:05:26,080 --> 01:05:30,560 We heard it during the debates. I think Chris Christie said to 1206 01:05:30,560 --> 01:05:33,400 Vivette you sound just like ChatGPT and I think that's a 1207 01:05:33,400 --> 01:05:38,000 that's an insult and I can definitely tell I have a family 1208 01:05:38,000 --> 01:05:40,080 member. She asked me for my help she's 1209 01:05:40,080 --> 01:05:43,440 writing something important and I and she said I use ChatGPT and 1210 01:05:43,440 --> 01:05:46,960 I said I can tell. So I I try and not use ChatGPT 1211 01:05:46,960 --> 01:05:50,120 when I'm writing an e-mail to people because I think it's so 1212 01:05:50,120 --> 01:05:53,480 inauthentic and and they can definitely tell I do use it in 1213 01:05:53,480 --> 01:05:55,960 other areas of my life. I was playing pickle ball with 1214 01:05:55,960 --> 01:05:58,800 my friends and I asked ChatGPT to create a pickle ball 1215 01:05:58,800 --> 01:06:02,680 tournament for like 12 people and do matches and it does 1216 01:06:02,680 --> 01:06:05,800 things like that very well. But if I was speaking to 1217 01:06:05,800 --> 01:06:08,000 someone, I would never use it at the moment. 1218 01:06:08,000 --> 01:06:11,400 It's good for transactional tasks like negotiating a bill, 1219 01:06:11,560 --> 01:06:14,480 although I'm biased saying that, but it's definitely not good 1220 01:06:14,480 --> 01:06:19,360 sending really high stakes communication path where 1221 01:06:19,360 --> 01:06:22,360 personal touch is needed. In the broadest terms like the 1222 01:06:22,360 --> 01:06:26,120 difference between it being an assistant to an accountant type 1223 01:06:26,120 --> 01:06:28,640 person versus being an accountant, you know, just being 1224 01:06:28,640 --> 01:06:31,720 the agent itself versus supporting people, what's what's 1225 01:06:31,720 --> 01:06:34,440 your general view on how close we are to? 1226 01:06:35,000 --> 01:06:39,320 This is like a total AI problem with maybe a light human 1227 01:06:39,320 --> 01:06:42,800 supervision versus just continuing to help professionals 1228 01:06:42,800 --> 01:06:46,520 who bring in the AI to support, you know, their professional 1229 01:06:46,520 --> 01:06:50,480 work. A few weeks ago I asked ChatGPT 1230 01:06:50,480 --> 01:06:53,120 to add up 10 numbers and it got the answer wrong. 1231 01:06:53,440 --> 01:06:56,280 So. Not close, no, no special. 1232 01:06:56,520 --> 01:06:58,000 Just literally 10 numbers, right? 1233 01:06:59,000 --> 01:07:00,440 It's it's actually not very good at math. 1234 01:07:00,960 --> 01:07:03,920 People don't realize this, but it it's good at language but not 1235 01:07:03,920 --> 01:07:07,480 necessarily math. So I think there's going to be 1236 01:07:07,480 --> 01:07:10,680 software tools where ChatGPT makes the experience amazing, 1237 01:07:11,240 --> 01:07:14,520 like Kit, which is an accounting software for small businesses. 1238 01:07:14,520 --> 01:07:15,720 I know they're using a lot of chat. 1239 01:07:15,720 --> 01:07:20,120 GPTI think accountants will use it, but given that it can't even 1240 01:07:20,120 --> 01:07:23,480 add up 10 numbers on it's own, I wouldn't trust it with my taxes 1241 01:07:23,480 --> 01:07:27,600 yet. Yeah, if it can't add, does that 1242 01:07:27,600 --> 01:07:30,440 mean it really can't reason? You know there is. 1243 01:07:30,840 --> 01:07:35,360 There's an argument that, well, it's just like it knows the end 1244 01:07:35,360 --> 01:07:38,320 you want like get out of this deal with Comcast or whatever 1245 01:07:38,320 --> 01:07:42,280 and it knows what text that might, you know, get it to that 1246 01:07:42,280 --> 01:07:45,280 end looks like. But it it's do you think it has 1247 01:07:45,280 --> 01:07:49,720 like a through line of any sort of reasoning in its strategy? 1248 01:07:50,960 --> 01:07:54,120 When you ask it something, it goes into its model and it says 1249 01:07:54,120 --> 01:07:57,080 based on all of human literature and all these parameters, what's 1250 01:07:57,080 --> 01:07:59,080 the most likely next response, right. 1251 01:07:59,440 --> 01:08:02,400 And so if you ask it to add up numbers, it finds some mass 1252 01:08:02,920 --> 01:08:06,320 training data somewhere and it just thinks, oh OK, well based 1253 01:08:06,320 --> 01:08:08,920 on all the training data, this is the the next response. 1254 01:08:09,120 --> 01:08:12,160 But that's not reasoning, that's just picking picking something 1255 01:08:12,160 --> 01:08:15,360 on a statistical model. And so that's why it can write a 1256 01:08:15,360 --> 01:08:19,800 beautiful poem, but it can't reason to add up 10 numbers. 1257 01:08:20,240 --> 01:08:23,840 Yeah, but do you think it's reasoning in the poem writing or 1258 01:08:23,840 --> 01:08:27,920 it's just good at sort of mimicking in a very 1259 01:08:27,920 --> 01:08:31,680 sophisticated way. It's it's good at picking. 1260 01:08:31,680 --> 01:08:33,760 So it just has a such a big option. 1261 01:08:33,760 --> 01:08:36,439 It can just pick and and throw things together. 1262 01:08:36,640 --> 01:08:38,279 I think it's just really good at picking. 1263 01:08:38,520 --> 01:08:40,680 It's definitely not reasoning at the moment. 1264 01:08:41,800 --> 01:08:43,520 In the accounting context though. 1265 01:08:43,840 --> 01:08:46,800 One, I'll give you a concrete example of why it's useful. 1266 01:08:48,399 --> 01:08:51,880 Previously and we do not pay had this problem, there would be 1267 01:08:51,880 --> 01:08:55,840 like 1000 transactions that you would get from Plaid or manually 1268 01:08:55,840 --> 01:08:59,600 entered and you would it would every transaction would be 1269 01:08:59,600 --> 01:09:02,560 different. There'd be no kind of reason to 1270 01:09:02,560 --> 01:09:06,240 them you can like characterize the transactions. 1271 01:09:06,479 --> 01:09:12,279 So you could say oh PLMP is Planet Fitness or or something 1272 01:09:12,279 --> 01:09:15,279 like that or this is a business expense or this is travel 1273 01:09:15,279 --> 01:09:17,200 expense. So that's where I see it being 1274 01:09:17,200 --> 01:09:19,399 useful in the accounting context, but I don't think it's 1275 01:09:19,399 --> 01:09:20,479 going to be adding numbers, do you? 1276 01:09:21,200 --> 01:09:24,160 Do you do you have any friends in the AI world where it feels 1277 01:09:24,319 --> 01:09:26,399 like I I'm interested in this question of like? 1278 01:09:27,359 --> 01:09:32,200 AI you know just almost like a computer you build at home right 1279 01:09:32,200 --> 01:09:35,800 before you know sort of the mainstream computer is 1280 01:09:35,800 --> 01:09:39,920 available, people sort of cobbling together a super, you 1281 01:09:39,920 --> 01:09:43,040 know a powerful computer of their own based on what they get 1282 01:09:43,040 --> 01:09:46,240 access to. Do you think that sort of is 1283 01:09:46,359 --> 01:09:49,479 happens in a meaningful way or the reality is that just like 1284 01:09:49,840 --> 01:09:53,600 purpose built like Chachi, BT, mid, journey, whatever are the 1285 01:09:53,600 --> 01:09:56,680 best at what they do So people are just using these? 1286 01:09:57,080 --> 01:10:00,600 Obvious tools like do you is there a real like are you seeing 1287 01:10:00,600 --> 01:10:04,040 people get their own memory and blah blah blah to to sort of 1288 01:10:04,400 --> 01:10:06,920 have an advantage over what's publicly available. 1289 01:10:08,200 --> 01:10:12,040 It's not simple enough for consumers to rely on AI just 1290 01:10:12,040 --> 01:10:15,440 with ChatGPT I I agree with your theory that they have to build 1291 01:10:15,600 --> 01:10:18,440 their own computer. It's not even just about the 1292 01:10:18,440 --> 01:10:22,000 AIAI as I mentioned is useless if it's not connected to the 1293 01:10:22,000 --> 01:10:26,400 real world. So how do you connect AI with AP 1294 01:10:26,400 --> 01:10:31,160 is so one thing I do is I connect ChatGPT to the mail 1295 01:10:31,200 --> 01:10:34,480 because it language through mail is one way you can get things 1296 01:10:34,480 --> 01:10:38,360 done and I use the LOB API to do that and so you have to. 1297 01:10:38,360 --> 01:10:42,920 Find a way to old fashion mail. Old fashioned mail, so in the 1298 01:10:42,920 --> 01:10:45,320 legal world you know it's a Stone Age. 1299 01:10:45,320 --> 01:10:47,960 You still have to send mail to get things done. 1300 01:10:48,200 --> 01:10:49,680 Some of these disputes are over mail. 1301 01:10:49,680 --> 01:10:53,080 How do you get chat GPD to send a physical latter and that's 1302 01:10:53,080 --> 01:10:55,480 through the LOB API. Similarly, how do you get 1303 01:10:55,480 --> 01:10:58,560 ChatGPT to make a phone call? Twilio API? 1304 01:10:59,360 --> 01:11:02,200 The building your computer is plugging in all of these janky 1305 01:11:02,200 --> 01:11:07,600 AP is so that the language can get through and then also 1306 01:11:07,600 --> 01:11:10,160 there's this open source component with the gatekeeping 1307 01:11:10,160 --> 01:11:13,520 so it all ties together on it's own. 1308 01:11:13,520 --> 01:11:16,000 Just having an interface where you can send or receive chat is 1309 01:11:16,000 --> 01:11:18,280 not that useful. It only begins to be useful when 1310 01:11:18,280 --> 01:11:21,600 it can be an agent in the real world, and I I think if there is 1311 01:11:21,600 --> 01:11:24,120 one AI law we should pass, we should pass a law that says that 1312 01:11:24,120 --> 01:11:27,160 AI can be an agent on your behalf to help people. 1313 01:11:27,880 --> 01:11:34,080 Fascinating isn't isn't your your dad is like you know has 1314 01:11:34,080 --> 01:11:35,320 that connection to the Russian lawyer. 1315 01:11:35,360 --> 01:11:37,920 Yeah, isn't it? Don't you have a like very pro 1316 01:11:38,560 --> 01:11:43,960 lawyer background or I'm curious your your view on how your view 1317 01:11:43,960 --> 01:11:47,400 on lawyers has developed or where where you see the the role 1318 01:11:47,400 --> 01:11:51,280 for them. I think that there are a lot of 1319 01:11:51,280 --> 01:11:54,720 amazing lawyers out, including the ones that help us and human 1320 01:11:54,720 --> 01:11:58,680 rights lawyers that my family's very involved with like fighting 1321 01:11:58,680 --> 01:12:01,080 back against Russia. But I also think that there are 1322 01:12:01,080 --> 01:12:04,720 lawyers on billboards that charge people a lot of money for 1323 01:12:04,720 --> 01:12:08,840 doing very little and I think that it with AI, hopefully it 1324 01:12:08,840 --> 01:12:12,520 will replace them and make their services free and accessible for 1325 01:12:12,520 --> 01:12:15,800 people. I was in Vegas a few weeks ago 1326 01:12:16,160 --> 01:12:19,040 and there are so many lawyer billboards and I hope AI will 1327 01:12:19,080 --> 01:12:22,120 replace all of them. What just given how much you see 1328 01:12:22,120 --> 01:12:24,400 and like you interact with a lot of these other companies, like 1329 01:12:24,400 --> 01:12:28,840 what, what do you think will happen in the next five years? 1330 01:12:28,840 --> 01:12:32,680 Or like, what's your prediction for how all of this plays out? 1331 01:12:32,680 --> 01:12:35,120 Or, you know, just a couple of things you're most excited 1332 01:12:35,120 --> 01:12:37,080 about. The biggest thing I'm excited 1333 01:12:37,080 --> 01:12:39,840 about is AI going from attractive to proactive. 1334 01:12:40,240 --> 01:12:43,920 So a lot of these assistants like right now you go into 1335 01:12:43,920 --> 01:12:47,800 ChatGPT you say I want this similarly with even do not pay. 1336 01:12:47,800 --> 01:12:50,480 You say I want this my info my wife it didn't work. 1337 01:12:51,000 --> 01:12:54,280 But these a is are going to be integrated into people's systems 1338 01:12:54,280 --> 01:12:57,600 so that you just wake up one day and the AI has done something 1339 01:12:57,600 --> 01:13:02,120 for you overnight. And to get that there's be a 1340 01:13:02,480 --> 01:13:06,480 kind of latency issue so that the AI can do 1000 things to 1341 01:13:06,480 --> 01:13:08,920 check what it wants to do a cost thing. 1342 01:13:09,120 --> 01:13:12,280 It has to be much cheaper because to find out the right 1343 01:13:12,280 --> 01:13:14,600 thing to do you have to throw a lot of darts. 1344 01:13:15,760 --> 01:13:19,320 And then also just a systems integration where a pool will 1345 01:13:19,320 --> 01:13:22,680 have a kind of probably a local LLM on the device constantly 1346 01:13:22,880 --> 01:13:25,400 figuring things out. You'll have lots of data 1347 01:13:25,400 --> 01:13:29,840 integrations with plug in like API type things that connect to 1348 01:13:29,840 --> 01:13:32,760 the AAI. So I'm really excited. 1349 01:13:32,760 --> 01:13:35,760 I'm trying to build a world where people wake up and the AI 1350 01:13:35,760 --> 01:13:39,480 says I've just saved you $100. I think there'll be situations 1351 01:13:39,480 --> 01:13:42,360 where the AI will say maybe you should get your health checked. 1352 01:13:42,360 --> 01:13:45,600 I am noticing something wrong with your blood and and things 1353 01:13:45,600 --> 01:13:48,080 like that and that will really improve quality of life do. 1354 01:13:48,960 --> 01:13:51,680 You think the consumer is on your side or where? 1355 01:13:51,680 --> 01:13:55,720 Where do you think the regular person is in terms of fear 1356 01:13:55,720 --> 01:13:58,800 versus excitement here and how that will translate into 1357 01:13:58,800 --> 01:14:03,840 regulation? Ordinary people love it and the 1358 01:14:03,920 --> 01:14:07,880 the only people that I I think there's a lot of scaremongering 1359 01:14:07,880 --> 01:14:11,800 going on, but I don't think it really touches the ordinary 1360 01:14:11,800 --> 01:14:14,920 person's experience. People just work really hard, 1361 01:14:14,920 --> 01:14:17,280 they don't have much time in the day, they're being ripped off 1362 01:14:17,280 --> 01:14:20,800 across their life. They just need something to help 1363 01:14:20,800 --> 01:14:24,520 them write. Thanks for writing thank you 1364 01:14:24,520 --> 01:14:27,560 notes to saving them money. I think it's it's mainly 1365 01:14:27,560 --> 01:14:32,160 beneficial to them. The kind of intellectualisation 1366 01:14:32,160 --> 01:14:36,040 of of the dangers of AI worries me, and that's probably why GPC 1367 01:14:36,040 --> 01:14:38,920 Four got worse. So we we should start doing 1368 01:14:38,920 --> 01:14:40,760 that. Are you, what do you think of 1369 01:14:40,760 --> 01:14:44,280 the whole like, you know, AI doom and gloom sort of. 1370 01:14:44,280 --> 01:14:46,960 It'll kill us all. Sort of general intelligence 1371 01:14:47,280 --> 01:14:49,160 fears. Like are you do you think we're 1372 01:14:49,160 --> 01:14:52,240 anywhere close or do you have a strong point of view on that 1373 01:14:52,480 --> 01:14:56,400 discussion? I think the, I think there are a 1374 01:14:56,400 --> 01:14:59,240 lot of evil people working on AI right now. 1375 01:14:59,360 --> 01:15:02,080 I've seen this first happen. Typically with a new technology, 1376 01:15:02,080 --> 01:15:06,520 it gets in the hands of evil People 1st and we actually need 1377 01:15:06,520 --> 01:15:10,560 to work on AI, accelerate our work on AI to fight back against 1378 01:15:10,560 --> 01:15:11,720 them. They're going to work on it 1379 01:15:11,720 --> 01:15:14,640 regardless. So I'll I'll give you a concrete 1380 01:15:14,640 --> 01:15:16,560 example. So what we're saying we we help 1381 01:15:16,560 --> 01:15:18,960 people charge back things and fight fraud on their bank. 1382 01:15:19,160 --> 01:15:21,720 We've even seen a few examples of fraudsters. 1383 01:15:22,240 --> 01:15:26,000 They phone up, they pretend to be someone's relative with a 1384 01:15:26,000 --> 01:15:29,080 spoof voice, and you can do that right now with just a minute 1385 01:15:29,080 --> 01:15:31,760 recording. And the way you fight that is 1386 01:15:31,760 --> 01:15:35,480 you have an AI from the Telecut phone company stopping people or 1387 01:15:35,480 --> 01:15:38,400 saying or on your phone maybe device side if you don't want 1388 01:15:38,400 --> 01:15:42,360 the telephone listening. Saying this is not who you think 1389 01:15:42,360 --> 01:15:45,880 it is because our AI is better and we've we've detected it. 1390 01:15:46,120 --> 01:15:49,640 If we stop work on AI, the criminals outside of the 1391 01:15:49,640 --> 01:15:52,920 jurisdiction of the United States will still build their 1392 01:15:52,920 --> 01:15:55,960 technologies, but we just won't have the tools to fight back. 1393 01:15:56,160 --> 01:15:59,080 So it's never a good idea to stop the progress of technology 1394 01:15:59,280 --> 01:16:01,440 because the evil people won't won't stop. 1395 01:16:01,880 --> 01:16:05,560 Yes, I agree with that. The copyright problem I where 1396 01:16:05,560 --> 01:16:08,560 I'm a little more sympathetic. Lots of writers are concerned 1397 01:16:08,560 --> 01:16:12,760 that these these large language models have inhaled all their 1398 01:16:12,960 --> 01:16:16,440 thinking and work to become what they are and that they should be 1399 01:16:16,440 --> 01:16:18,440 compensated in some way for it or what. 1400 01:16:18,440 --> 01:16:21,920 What do you make of sort of the creator and writer class being 1401 01:16:21,920 --> 01:16:25,440 worried about what's been built potentially on their backs? 1402 01:16:26,720 --> 01:16:31,080 I think that there's no practical way of dealing with 1403 01:16:31,080 --> 01:16:34,120 it. No one knows how these large 1404 01:16:34,120 --> 01:16:38,440 language models come up with their answers, and so perhaps it 1405 01:16:38,440 --> 01:16:43,040 can list it by ten sources. But ChatGPT, every answer is 1406 01:16:43,160 --> 01:16:46,280 more than ten sources. It's a million sources or even a 1407 01:16:46,280 --> 01:16:51,240 trillion sources. So I I think if I was to read 10 1408 01:16:51,240 --> 01:16:53,880 books and then write a book myself, even if it was inspired 1409 01:16:53,880 --> 01:16:57,720 by the 10 books, I don't think I should have to compensate the 1410 01:16:57,720 --> 01:17:00,480 people I read. If I physically copy that work, 1411 01:17:00,480 --> 01:17:02,680 I do. But if it's just inspired by I 1412 01:17:02,680 --> 01:17:05,880 don't. I don't think I would have to. 1413 01:17:05,880 --> 01:17:09,120 I I know it's a very tough situation, but I don't see any 1414 01:17:09,120 --> 01:17:11,560 practical way of doing the compensation. 1415 01:17:12,520 --> 01:17:14,840 Fortunately, though, the lawyers have already thought about this, 1416 01:17:14,840 --> 01:17:17,680 and there's like multiple lawsuits, class action lawsuits 1417 01:17:17,680 --> 01:17:19,840 against ChatGPT that will play out. 1418 01:17:19,840 --> 01:17:23,120 I don't want it to be, you know, so costly that it means these 1419 01:17:23,120 --> 01:17:24,920 things can't come into existence. 1420 01:17:24,920 --> 01:17:29,560 But if it turns out, you know. Large language models are fairly 1421 01:17:29,560 --> 01:17:31,600 commoditized. We can you know people are 1422 01:17:31,600 --> 01:17:37,680 competitive with Chachi BT then it would seem like a lot of the 1423 01:17:37,680 --> 01:17:43,040 core sauces like the breadth of human intelligence that went 1424 01:17:43,040 --> 01:17:46,080 into them and if like, I don't know especially specific ones 1425 01:17:46,080 --> 01:17:47,080 where like. Right. 1426 01:17:47,080 --> 01:17:49,560 In the style of like Stephen King or whatever, you know, he 1427 01:17:49,560 --> 01:17:52,680 created an iconic style that's extremely valuable. 1428 01:17:53,520 --> 01:17:57,440 So I I guess I just don't think it's like this large long tail 1429 01:17:57,440 --> 01:17:59,920 for a lot of the stuff. I think it's more likely that 1430 01:17:59,920 --> 01:18:05,160 there are a few core great things that are being used, you 1431 01:18:05,160 --> 01:18:08,280 know for, for a lot of answers and that could be sort of 1432 01:18:08,360 --> 01:18:11,240 attributed and. Those people should probably be 1433 01:18:11,240 --> 01:18:14,560 compensated in some way if for the rest of human history you 1434 01:18:14,560 --> 01:18:17,680 know those those answers are going to be a bedrock of of how 1435 01:18:17,680 --> 01:18:21,040 these systems answer. Do you disagree with that? 1436 01:18:21,240 --> 01:18:24,680 But I I do. I think there's existing laws on 1437 01:18:24,680 --> 01:18:27,920 the books to deal with that. So California has this law about 1438 01:18:27,920 --> 01:18:31,200 right to publicity and likeness, which is completely separate 1439 01:18:31,200 --> 01:18:34,440 from copyright law. And so if I was to say right in 1440 01:18:34,440 --> 01:18:37,920 the style of Stephen King, that would probably violate like his 1441 01:18:37,920 --> 01:18:41,920 right to publicity if I was to publish a book from Chachi PT 1442 01:18:41,920 --> 01:18:45,720 writing in his likeness. So there there are a kind. 1443 01:18:45,720 --> 01:18:48,600 Of but you ask Chachi PT to write a sci-fi novel and it's 1444 01:18:48,600 --> 01:18:52,520 like it's like based on like 5 different, you know, I feel like 1445 01:18:52,520 --> 01:18:56,400 it's based on you know the top ten list of sci-fi authors and 1446 01:18:56,400 --> 01:18:59,240 then it sort of finds the midpoint between them and gives 1447 01:18:59,240 --> 01:19:02,160 you, you know some text. Is that another? 1448 01:19:02,200 --> 01:19:06,560 Another issue is that if I say that that's not copyrighted, I 1449 01:19:06,560 --> 01:19:10,080 don't even own that work. The Copyright Office recently 1450 01:19:10,080 --> 01:19:14,160 made a ruling that things done purely with generative AI do not 1451 01:19:14,160 --> 01:19:17,320 qualify for copyright, even if it's 1000 prompts. 1452 01:19:17,520 --> 01:19:20,640 So if I do a mid journey thing and prompt it 1000 times, fine 1453 01:19:20,640 --> 01:19:23,760 tune it, it's still not mine from a copyright perspective. 1454 01:19:24,240 --> 01:19:27,560 So I I do think these things have to be changed in some way. 1455 01:19:28,640 --> 01:19:30,480 What? Sorry, What you want? 1456 01:19:30,720 --> 01:19:33,280 You want a law change where you get protection if you build 1457 01:19:33,280 --> 01:19:35,720 something out of or? I'm confused what you're saying 1458 01:19:35,720 --> 01:19:39,040 there. If if you fine tune the output 1459 01:19:39,080 --> 01:19:42,040 of a large language model, you're not protected with 1460 01:19:42,040 --> 01:19:45,800 copyright at the moment. So I I do think that the system 1461 01:19:45,800 --> 01:19:48,960 is is not built to deal with the AI and so I'm not arguing that 1462 01:19:48,960 --> 01:19:51,120 we just do nothing. I think there's lots of these 1463 01:19:51,240 --> 01:19:54,560 new answers, but I I do worry that we would just halt 1464 01:19:54,640 --> 01:19:57,160 technology. Yeah, yeah, you're you're you're 1465 01:19:57,160 --> 01:19:59,800 saying, OK, maybe we need to build it, answer some of these 1466 01:19:59,800 --> 01:20:02,920 specific questions, but not not if it's putting roadblocks. 1467 01:20:03,320 --> 01:20:06,040 I'm curious like the. The existing chat bots like the 1468 01:20:06,040 --> 01:20:08,880 characters of the world like are you are you have you played 1469 01:20:08,880 --> 01:20:11,520 around with them Much like do you think they're they have 1470 01:20:11,520 --> 01:20:14,880 perpetual value or what's your what's your view on that space 1471 01:20:14,880 --> 01:20:18,160 at the moment? We're going to have AI 1472 01:20:18,160 --> 01:20:23,360 girlfriends, AI people can AI therapists, AI friends, and 1473 01:20:23,360 --> 01:20:25,440 people can talk to their favorite characters. 1474 01:20:26,440 --> 01:20:29,640 I'm a big fan of the show Better Call Saul for obvious reasons, 1475 01:20:29,640 --> 01:20:32,080 so I was playing around on character AI Better Call Saul. 1476 01:20:32,080 --> 01:20:34,960 But. So I think it's it's really 1477 01:20:34,960 --> 01:20:36,880 important to give people that connection. 1478 01:20:36,880 --> 01:20:40,480 So I'm I'm for it. I don't think it's as big as 1479 01:20:40,560 --> 01:20:43,800 people think it will be there. You think it's overhyped? 1480 01:20:45,360 --> 01:20:49,880 I think that given the kind of modality, it's just text at the 1481 01:20:49,880 --> 01:20:52,320 moment, there's only so much you can do with that. 1482 01:20:52,640 --> 01:20:55,200 So what? Once you have like a slowly 1483 01:20:55,520 --> 01:20:58,880 interactive character that you can FaceTime, and I know that 1484 01:20:58,880 --> 01:21:02,040 those exist, but latency is too big, it doesn't feel completely 1485 01:21:02,040 --> 01:21:04,240 real. Then then it gets serious. 1486 01:21:04,240 --> 01:21:07,160 But until then I don't think people will just spend all their 1487 01:21:07,160 --> 01:21:11,760 time talking to spots chat bots because no, no one likes to do 1488 01:21:11,760 --> 01:21:12,920 that. You sort of open the 1489 01:21:12,920 --> 01:21:16,160 conversation talking about multimodal like clearly that's 1490 01:21:16,160 --> 01:21:19,840 something you're excited about. Where do you think how close is 1491 01:21:19,840 --> 01:21:21,640 that? I mean you know Open AII Guess 1492 01:21:21,640 --> 01:21:25,160 is released something where Dolly and Chachi BT are 1493 01:21:25,160 --> 01:21:28,440 connected, but can you talk about what you think the real 1494 01:21:28,440 --> 01:21:35,280 potential of multimodal is? Yeah, GPT 4 VI think was the 1495 01:21:35,280 --> 01:21:38,320 most exciting thing around that where you can upload images and 1496 01:21:38,360 --> 01:21:42,000 and PDFs and things like that to to make it multi modal with the 1497 01:21:42,000 --> 01:21:47,040 API. If AI is constantly ambient 1498 01:21:47,040 --> 01:21:51,200 listening it gets back to this kind of proactive approach which 1499 01:21:51,200 --> 01:21:55,080 is so exciting chat. GPT is very powerful, but it 1500 01:21:55,080 --> 01:22:00,600 doesn't have the data right now, and so another concrete example 1501 01:22:00,600 --> 01:22:03,680 is the browsing. So right now the way ChatGPT 1502 01:22:03,680 --> 01:22:07,360 browsing works is it takes the HTML of the page ingest that 1503 01:22:07,720 --> 01:22:10,760 produces an answer and that's why it's so time intensive. 1504 01:22:10,960 --> 01:22:13,920 Multimodal fixes that in the future I imagine it takes a 1505 01:22:13,920 --> 01:22:17,440 screenshot of the page and then they can put any blockers or 1506 01:22:17,440 --> 01:22:19,720 anti bot things, it doesn't matter they've taken a 1507 01:22:19,720 --> 01:22:23,960 screenshot. And so how we view things is, 1508 01:22:23,960 --> 01:22:25,920 how humans perceive things is through vision. 1509 01:22:25,920 --> 01:22:28,680 And that's why I think it makes it so powerful because no one 1510 01:22:28,680 --> 01:22:30,680 can stop vision. Sweet. 1511 01:22:30,680 --> 01:22:33,080 Joshua Browder, thank you so much for coming on the show. 1512 01:22:33,080 --> 01:22:35,480 I really this was great. Thanks for having me. 1513 01:22:36,080 --> 01:22:40,440 That's our episode. Do Not Pay CEO Joshua Browder 1514 01:22:40,760 --> 01:22:44,160 Great conversation about the AI personal stack. 1515 01:22:44,480 --> 01:22:48,160 Thanks so much to Max Child and James Woolsterman, the Volley Co 1516 01:22:48,160 --> 01:22:51,080 founders. I'm your host, Eric Newcomer. 1517 01:22:51,200 --> 01:22:54,920 This episode has been in our Cerebral Valley series on the 1518 01:22:54,920 --> 01:22:59,720 Newcomer podcast I'm hosting with Max and James this Cerebral 1519 01:22:59,720 --> 01:23:04,120 Valley Conference in San Francisco on November 15th. 1520 01:23:04,640 --> 01:23:07,360 This is the second AI conference in a year, and you can go back 1521 01:23:07,360 --> 01:23:10,920 and watch the old videos on our YouTube channel from March. 1522 01:23:11,280 --> 01:23:14,320 And now, yeah, we're going to, we're going to have another 1523 01:23:14,320 --> 01:23:17,440 exclusive AI conference, but we're bringing it all to 1524 01:23:17,440 --> 01:23:22,920 everybody on YouTube and some of our favorites will go on the 1525 01:23:22,920 --> 01:23:25,760 podcast feed. Shout out to our producer Scott 1526 01:23:25,760 --> 01:23:30,120 Brody, my Chief of Staff Riley Kinsella, Gabby Caliendo at 1527 01:23:30,120 --> 01:23:34,440 Volley and young Chomsky for the wonderful theme Music. 1528 01:23:35,000 --> 01:23:38,080 Like, comment, subscribe on YouTube, give us a review on 1529 01:23:38,080 --> 01:23:44,760 Apple Podcast, play a Alexa game, try Song quiz or yes, sire 1530 01:23:45,200 --> 01:23:48,200 at Volley. And of course, most important, 1531 01:23:48,880 --> 01:23:52,280 subscribe to the sub stack newcomer.co. 1532 01:23:52,880 --> 01:23:53,600 Thanks so much. Goodbye. 1533 01:23:55,120 --> 01:23:55,320 Goodbye.