1 00:00:01,839 --> 00:00:05,600 Jimmy, welcome to the Evolved Radio podcast. Thanks 2 00:00:05,600 --> 00:00:09,205 for having me on. So happy to be here. Appreciate it. Awesome. 3 00:00:09,525 --> 00:00:12,965 So this is gonna be great. I think we'll we'll just sort of jump right 4 00:00:12,965 --> 00:00:16,725 into it. I think an interesting place to start is just 5 00:00:16,725 --> 00:00:20,320 a reflection on the sort of modern version of 6 00:00:20,320 --> 00:00:24,160 LLMs and AI. I mean AI has kind of existed forever in a 7 00:00:24,160 --> 00:00:27,914 lot of ways like I remember very early on my computer 8 00:00:27,914 --> 00:00:31,755 days the, the SoundBlaster sound card that I got came with like 9 00:00:31,755 --> 00:00:35,440 the Parrot AI or the and like a 10 00:00:35,520 --> 00:00:38,480 it was like a psychologist that you could talk to and it would repeat your 11 00:00:38,480 --> 00:00:41,280 questions and ask you how you feel about that. Sort of like what I imagine 12 00:00:41,280 --> 00:00:45,055 is sort of the very first iterations of this. But, like, the very 13 00:00:45,055 --> 00:00:48,755 modern version of the LLMs that that are really revolutionizing 14 00:00:49,135 --> 00:00:52,975 everything right now have only kind of existed for about 15 00:00:52,975 --> 00:00:56,670 2 years which I find kind of mind blowing, because of all the 16 00:00:56,670 --> 00:01:00,030 things they're impacting. So, any thoughts on sort of, 17 00:01:00,030 --> 00:01:03,515 like, that this sort of first two year stage and 18 00:01:03,515 --> 00:01:06,875 and how things are sort of progressing so quickly just to open 19 00:01:06,875 --> 00:01:10,520 up? Oh, I remember the first time I used 20 00:01:10,600 --> 00:01:14,359 chat gpt, it was, like, using the Internet for the first time or high 21 00:01:14,359 --> 00:01:18,039 speed Internet for the first time or something like that. Like, it was just, 22 00:01:18,039 --> 00:01:21,215 oh, this is gonna change everything we do. 23 00:01:21,914 --> 00:01:25,755 And sitting back and watching it like magic, like, I couldn't believe that 24 00:01:25,755 --> 00:01:28,950 it was doing what it was doing, and it could talk like a person and 25 00:01:28,950 --> 00:01:31,450 it could answer things really quickly. 26 00:01:32,310 --> 00:01:36,115 And, you know, seeing that definitely opened my eyes to 27 00:01:36,115 --> 00:01:39,875 the future. And here I am, you know, less than 2 years 28 00:01:39,875 --> 00:01:43,635 later starting a company. So, like, sir, centered around the 29 00:01:43,635 --> 00:01:47,159 thesis that, like, this is the chain gonna change all areas of business, 30 00:01:47,540 --> 00:01:51,380 and people need help to do it. Mhmm. But I I mean, I can't 31 00:01:51,380 --> 00:01:54,955 believe that it's only been around this long either. Like like I, 32 00:01:55,435 --> 00:01:58,875 it's, it's funny working in AI. I've, I've been, you know, 33 00:01:58,875 --> 00:02:02,715 publicly had this AI company for about a month now and you 34 00:02:02,715 --> 00:02:06,270 go to technically implement something. And by the time you're 35 00:02:06,270 --> 00:02:09,789 done some new version of this out and you, and had you started it a 36 00:02:09,789 --> 00:02:12,825 month later, you would have done it a different way. And that's how fast things 37 00:02:12,905 --> 00:02:16,745 are moving in this space, and it's just incredible. It's it's I'm so grateful to 38 00:02:16,745 --> 00:02:20,425 be a part of it because it's so fun. Yeah. It is wild how quickly 39 00:02:20,425 --> 00:02:24,140 change and the things are changing. I did wanna ask you, like like, what 40 00:02:24,140 --> 00:02:26,780 was the sort of the moment in time? Like, do you remember sort of a 41 00:02:26,780 --> 00:02:30,460 particular point or moment or evening where you're like, you know 42 00:02:30,460 --> 00:02:33,955 what? I'm gonna start my own AI company. What what was that moment? Do you 43 00:02:33,955 --> 00:02:37,635 remember? You know, I I I don't know that 44 00:02:37,635 --> 00:02:41,140 there's a particular moment, but there is 45 00:02:41,140 --> 00:02:44,740 definitely a number of 46 00:02:44,740 --> 00:02:48,180 interactions that I could remember where I would show AI to 47 00:02:48,180 --> 00:02:50,440 people and I would see their reactions. 48 00:02:52,405 --> 00:02:56,005 So for example, I, you know, I worked in cybersecurity and I did a lot 49 00:02:56,005 --> 00:02:59,605 of, demos of using AI to hack 50 00:02:59,605 --> 00:03:02,870 people. And the laughs that I would get, the 51 00:03:03,010 --> 00:03:06,790 inquisitiveness that I would get, the questions that I would get people stopping me afterwards, 52 00:03:07,250 --> 00:03:10,995 then became a hobby and just showing my friends. I I guess one 53 00:03:10,995 --> 00:03:14,455 moment where it was like, yes, I'm definitely gonna start a company on this was 54 00:03:14,595 --> 00:03:17,495 I build a proof of concept and 55 00:03:18,820 --> 00:03:22,180 I build a proof of concept, and I had my father-in-law test it. My father-in-law 56 00:03:22,180 --> 00:03:25,240 is in his seventies, and he is the 57 00:03:25,540 --> 00:03:29,345 executive secretary for his mosque. And he was doing a fundraising campaign and 58 00:03:29,345 --> 00:03:32,705 there was a, a group of people who raised their hands saying that they would 59 00:03:32,705 --> 00:03:36,520 donate to something, but he had to, you know, send file communication to 60 00:03:36,520 --> 00:03:40,200 actually get them to mail the check or whatever it was. And I 61 00:03:40,200 --> 00:03:43,825 gave him a proof of concept of, of early hats, before it was 62 00:03:43,825 --> 00:03:47,425 hats. And, you know, I I told him, why don't you try 63 00:03:47,425 --> 00:03:50,885 using it on this? And then he called me 2 days later and said, 64 00:03:51,640 --> 00:03:55,480 Jimmy, Jimmy, Jimmy, what's what's that website that you 65 00:03:55,480 --> 00:03:58,860 gave me that your your your AI? I used it yesterday, 66 00:03:59,235 --> 00:04:02,455 and it wrote me this really great email. I I sent it out to everyone. 67 00:04:03,075 --> 00:04:06,435 Yeah. I wanna use it again. Wanna give it to to everyone else on the 68 00:04:06,435 --> 00:04:09,939 team here. And I was like, woah. This is a 69 00:04:10,019 --> 00:04:13,799 man who doesn't, like, use 2 computer screens. He uses 70 00:04:14,019 --> 00:04:14,519 2 71 00:04:23,850 --> 00:04:27,610 to use this thing in, like, a number of minutes and immediately into 72 00:04:27,610 --> 00:04:31,290 production. Like, saw the the life changing or the the way which 73 00:04:31,290 --> 00:04:35,135 is in completely changed, you know, his his nonprofit, like, 74 00:04:35,135 --> 00:04:38,895 the business. And that was, like, okay. This is gonna 75 00:04:38,895 --> 00:04:42,655 have really lasting change. And, I I think that was 76 00:04:42,655 --> 00:04:46,100 definitely a big catalyst for me to take the jump and and go full time 77 00:04:46,100 --> 00:04:49,940 at it. Okay. Very cool. Yeah. It's I mean, just quickly on 78 00:04:49,940 --> 00:04:53,634 that. Like, it is it is wild sort of the the the adoption of 79 00:04:53,634 --> 00:04:56,935 this. Like, I can't remember the stat. You may know this, but, like, 80 00:04:57,395 --> 00:05:01,130 ChatGPT's, like, acceleration to, what was it, a a 81 00:05:01,130 --> 00:05:04,570 1000000 users or something like that is sort of this benchmark that they've used for, 82 00:05:04,570 --> 00:05:08,065 like, adoption. You know, Facebook, Instagram were all these sort of 83 00:05:08,065 --> 00:05:11,365 meteoric rises and then absolutely eclipsed 84 00:05:11,425 --> 00:05:15,140 by chat gpt adoption. Do you remember that stat? Yeah. 85 00:05:15,140 --> 00:05:18,420 It was, I think it was a 1000000 users in 2 months or something like 86 00:05:18,420 --> 00:05:21,720 that, and then a 100000000 users, soon afterwards. 87 00:05:21,780 --> 00:05:25,264 So they I mean, they did it in, like, like, 88 00:05:25,264 --> 00:05:29,105 it's in order of magnitude each time faster than it happened, like, 89 00:05:29,105 --> 00:05:32,860 each time that happened. And it was like 90 00:05:32,860 --> 00:05:36,480 2 months or something where the previous was like, you know, 12 months. 91 00:05:36,700 --> 00:05:40,535 Yeah. Yeah. It's so wild. Cut. So I guess that that leads 92 00:05:40,535 --> 00:05:43,175 well to sort of the other thing I wanted to to sort of dig into 93 00:05:43,175 --> 00:05:47,015 here is like like LLMs are all over the place, right? Like, you 94 00:05:47,015 --> 00:05:50,680 know obviously everyone knows ChatGPT, Microsoft 95 00:05:50,680 --> 00:05:54,139 Copilot is now kind of, like, rising up. And now, 96 00:05:54,820 --> 00:05:58,599 you know, Google was caught a bit flat footed, largely 97 00:05:58,599 --> 00:06:01,955 I understand because of sort their concern around safety and rollout. 98 00:06:02,415 --> 00:06:06,255 So it was like these tools existed. They just hadn't made them public. Some people 99 00:06:06,255 --> 00:06:09,289 argue as to whether or not they felt it might cannibalize the search business. I 100 00:06:09,289 --> 00:06:13,069 think that's there's there's probably some validity to that. So they've got now Gemini, 101 00:06:13,449 --> 00:06:17,215 rolling out. And then there's, you know, Meta's developing their own 102 00:06:17,215 --> 00:06:20,035 internal. There's lots of even open source LLMs. 103 00:06:21,775 --> 00:06:25,540 So I understand you are sort of, like, building kinda your own model, 104 00:06:25,540 --> 00:06:29,300 and I'm I'm interested in that. Like like like, why your own model versus just 105 00:06:29,300 --> 00:06:32,085 sort of leveraging off the existing LLMs that are out there? 106 00:06:33,044 --> 00:06:36,805 Yeah. No. That's a great question. Actually, what we're doing is we're 107 00:06:36,805 --> 00:06:40,520 empowering users to tune, or customize their 108 00:06:40,520 --> 00:06:44,199 own models. So we we aren't necessarily interested 109 00:06:44,199 --> 00:06:47,180 in in creating our own model. Maybe we will in the future. 110 00:06:47,879 --> 00:06:51,544 But we aren't beholden to one model. So every 111 00:06:51,544 --> 00:06:55,384 model that you've mentioned, Gemini, llama 2, which 112 00:06:55,384 --> 00:06:57,085 is Facebook's open source model, 113 00:07:00,260 --> 00:07:03,960 GPT 3.5, GPT 4, even Claude, 114 00:07:04,340 --> 00:07:07,764 by Anthropic, we're using internally. 115 00:07:07,985 --> 00:07:11,125 And so there's there's different ways that you can customize, 116 00:07:11,664 --> 00:07:14,790 each of these models to do what you need do, 117 00:07:16,210 --> 00:07:19,910 whether that be through context or through rag or actually tuning the models. 118 00:07:20,370 --> 00:07:23,225 And we're in the business of helping small businesses, 119 00:07:24,085 --> 00:07:27,764 do that through their MSP. I see. Okay. 120 00:07:27,764 --> 00:07:30,965 So I yeah. I misunderstood this. You it's not that you built your own model. 121 00:07:30,965 --> 00:07:34,500 It's kinda like a a department store maybe 122 00:07:34,500 --> 00:07:38,120 for, for different models for different purposes. Is that a good analogy? 123 00:07:39,620 --> 00:07:41,764 Sort of. It it's like it's like, 124 00:07:44,065 --> 00:07:47,125 there's a lot of work that goes into the customization 125 00:07:47,664 --> 00:07:51,470 rollout and access to data, of actually 126 00:07:51,470 --> 00:07:54,990 using a model at scale inside your own business or 127 00:07:54,990 --> 00:07:58,670 efficiently, whether that be business process efficient 128 00:07:58,670 --> 00:08:02,445 or cost efficient or both, and we're a platform that helps 129 00:08:02,445 --> 00:08:05,585 you manage all of that end to end. Okay. 130 00:08:06,685 --> 00:08:10,430 With that, like, maybe getting into some of the technical details but like, 131 00:08:10,590 --> 00:08:14,430 if you do training on one of your models and you're doing something in sort 132 00:08:14,430 --> 00:08:17,470 of this space but you're using a different model for something else are you able 133 00:08:17,470 --> 00:08:21,245 to sort of port some of that over through through your guys' platform? 134 00:08:21,245 --> 00:08:25,085 Is that some of the benefit there? Yeah. So so you can do that in 135 00:08:25,085 --> 00:08:28,849 a couple different ways. You can you can change 136 00:08:28,849 --> 00:08:32,370 the context so you can move let's say you write a really good prompt or, 137 00:08:32,529 --> 00:08:36,265 you have an application that's built around, specific prompts that 138 00:08:36,265 --> 00:08:39,865 you you've built inside of our, UI. So so in our system, we 139 00:08:39,865 --> 00:08:43,590 basically have a a UI for prompt engineering for MSPs, and they could 140 00:08:43,590 --> 00:08:47,130 publish apps for their end users. You can switch out the underlying 141 00:08:47,190 --> 00:08:50,870 models, very easily, in in doing it that 142 00:08:50,870 --> 00:08:54,565 way. The only limitation is the context window. So 143 00:08:54,565 --> 00:08:58,265 for example, like, Claude has a 100 k, 144 00:08:58,920 --> 00:09:02,360 token context window. A token's like a syllable. So you can think of it as, 145 00:09:02,360 --> 00:09:06,120 you know, whatever, 25,000 words or something 146 00:09:06,120 --> 00:09:09,654 like a very, very long context window where, GPT 147 00:09:09,714 --> 00:09:13,555 3.5, you know, there a lot of those, was 16 148 00:09:13,555 --> 00:09:16,695 k or, 32 k 149 00:09:17,329 --> 00:09:20,769 tokens. So, you know, 1 fourth the size. So that would be one 150 00:09:20,769 --> 00:09:24,450 limitation of it. Another something that we're doing in the future 151 00:09:24,450 --> 00:09:27,634 is is, retrieval. So it's it's RAG, 152 00:09:28,894 --> 00:09:32,274 where basically the LOM can look inside of a a database, 153 00:09:33,055 --> 00:09:36,470 or some sort of dataset and then spit out factual information, 154 00:09:37,010 --> 00:09:40,770 or reference that information. We we you you 155 00:09:40,850 --> 00:09:44,675 those are interchangeable as well where you just have to have the the vector database, 156 00:09:45,055 --> 00:09:48,755 configured properly. The the instance where you can't, 157 00:09:49,535 --> 00:09:53,110 is actually tuning. So the difference between, 158 00:09:53,410 --> 00:09:57,030 training and tuning a model, basically, 159 00:09:57,650 --> 00:10:01,165 OpenAI and, you know, meta, these developers of these 160 00:10:01,165 --> 00:10:04,845 foundational large language models, they get all this training data 161 00:10:04,845 --> 00:10:08,380 in and they, you know, publish the model. And then afterwards, when you 162 00:10:08,620 --> 00:10:12,300 feed it additional information that's generally called tuning, that's 163 00:10:12,300 --> 00:10:15,605 very expensive to do and expensive to run because you're not using 164 00:10:16,165 --> 00:10:19,605 foundational model anymore that could be on a shared server or shared 165 00:10:19,605 --> 00:10:22,505 resource or something like that or even if it's read only. 166 00:10:25,960 --> 00:10:28,380 So so that gets you know, you're paying 167 00:10:29,560 --> 00:10:33,375 GPU, like, processing power to actually, tune 168 00:10:33,375 --> 00:10:36,575 it, and then you're paying for access to it as well, whether that be through 169 00:10:36,575 --> 00:10:40,399 our platform or another. So, in in those three 170 00:10:40,399 --> 00:10:44,160 scenarios, outside of tuning, you can generally switch the 171 00:10:44,160 --> 00:10:47,575 underlying, foundational model. And that's very important 172 00:10:47,575 --> 00:10:51,335 because say say you, are an 173 00:10:51,335 --> 00:10:54,715 MSP and you do tons of tons of work 174 00:10:54,960 --> 00:10:58,420 creating an amazing dataset on how to run an MSP business, 175 00:10:59,520 --> 00:11:02,960 or perhaps one of your customers is in the 176 00:11:02,960 --> 00:11:06,785 widget factory business, and they come up with a huge, a 177 00:11:06,785 --> 00:11:10,404 phenomenal data that or a set of prompts or whatever it is, 178 00:11:10,945 --> 00:11:14,360 on how to, create, you know, build widgets in the widget 179 00:11:14,360 --> 00:11:17,880 factory. When a new foundational model comes out, if you 180 00:11:17,880 --> 00:11:21,435 invested everything in tuning, you have to pay all of that 181 00:11:21,435 --> 00:11:25,274 money again to to retune the data set. So, you know, you say 182 00:11:25,274 --> 00:11:28,574 you're doing it on llama 2, then llama 3 comes out, llama 4. Whereas, 183 00:11:30,140 --> 00:11:33,740 with with rag or with, you know, cleverly using context 184 00:11:33,740 --> 00:11:37,100 windows, you you don't have to do that. So there's a 185 00:11:37,100 --> 00:11:40,345 lot of trade offs between speed, efficiency, 186 00:11:41,125 --> 00:11:44,485 cost, complexity, future proofing, and just using 187 00:11:44,485 --> 00:11:48,185 LMS internally. And I think I think there's a lot of, like, 188 00:11:56,395 --> 00:12:00,075 clear for clarification, you mentioned RAG. I'm not familiar with the the sort of 189 00:12:00,075 --> 00:12:03,515 the the terminology there. What does that refer to? I 190 00:12:03,515 --> 00:12:04,440 believe the 191 00:12:14,275 --> 00:12:18,035 So you can use GPT 4RAG, and, that is a 192 00:12:18,035 --> 00:12:21,015 version of the model that's able to look inside of a database. 193 00:12:21,610 --> 00:12:25,209 So, say you had, all of 194 00:12:25,209 --> 00:12:28,509 your SOPs stored in a database somewhere, 195 00:12:30,275 --> 00:12:33,575 gpt4RAG could go and and say 196 00:12:33,795 --> 00:12:35,895 instead you say, hey. How do I 197 00:12:37,840 --> 00:12:40,820 restart this data server or whatever? 198 00:12:41,760 --> 00:12:45,280 It could just make something up based on its foundational knowledge and 199 00:12:45,280 --> 00:12:48,925 and, you know, probably 80% of the time it would be right 200 00:12:48,925 --> 00:12:52,605 because it's pretty good. Like, g p t four is pretty good. But with 201 00:12:52,605 --> 00:12:56,190 reg, it could actually look inside of a database and and bring 202 00:12:56,190 --> 00:12:59,870 back a PDF user manual and say based on, you 203 00:12:59,870 --> 00:13:03,709 know, page 6 of, you know, this latest user 204 00:13:03,709 --> 00:13:04,105 manual, 205 00:13:13,040 --> 00:13:16,800 Yes. Depending on yes. Exactly. So so 206 00:13:16,800 --> 00:13:20,480 that's like and and you you get, 207 00:13:20,480 --> 00:13:22,220 like, say say you could get, like, 80% accuracy on complex tasks with just foundational 208 00:13:22,220 --> 00:13:23,025 models, like, they're pretty good. 209 00:13:32,250 --> 00:13:35,850 RAG, and you'd get similar results for for for tuning 210 00:13:35,930 --> 00:13:39,610 for tuned models, maybe even worse results in some cases. Some of the studies 211 00:13:39,610 --> 00:13:43,274 I'm seeing that you're actually getting better results from right because it's, you know, it 212 00:13:43,274 --> 00:13:46,714 can be more factual. So, you know, they're 213 00:13:46,714 --> 00:13:50,420 like, there's technical trade offs between them all. Okay. That's cool. 214 00:13:50,420 --> 00:13:53,220 I appreciate the background on that. It gets into some of the use case context 215 00:13:53,220 --> 00:13:56,805 that I wanna get into as well. I I do wanna call out, like, I 216 00:13:56,805 --> 00:14:00,404 have I have this conversation with anyone whenever I'm talking about AI because to 217 00:14:00,404 --> 00:14:04,024 me, it's sort of like what I find the most fascinating 218 00:14:04,165 --> 00:14:07,880 about this technology is like I kind of relate it 219 00:14:07,880 --> 00:14:11,480 to like, it's a bit of a parlor trick. Like if you really 220 00:14:11,480 --> 00:14:14,840 understand how it works like once I started digging into this and understanding the 221 00:14:14,840 --> 00:14:18,635 technology I was really blown away by by sort of 222 00:14:18,635 --> 00:14:22,394 how it actually works under the covers. So, I I know I'm not 223 00:14:22,394 --> 00:14:25,180 exactly an AI expert so I'll go through this and then you kinda correct me 224 00:14:25,180 --> 00:14:28,140 if I get anything wrong here. But, I had my brother reach out to me 225 00:14:28,140 --> 00:14:31,020 and he's like, hey. Do do you know much about sort of chat GPT and 226 00:14:31,020 --> 00:14:33,680 all this stuff coming out? And I was like, yeah. Like, it's pretty amazing. 227 00:14:34,915 --> 00:14:37,634 I can't remember how we got into this, but I I I want I I 228 00:14:37,634 --> 00:14:40,754 guess sort of started telling him, like, this is kinda how it works. Like, it's 229 00:14:40,754 --> 00:14:44,589 actually more fascinating. The way I describe this is, like, usually if you find how 230 00:14:44,589 --> 00:14:47,870 a magic trick works, it kinda takes away the magic. You're like, oh, well, okay. 231 00:14:47,870 --> 00:14:50,670 Well, it doesn't doesn't feel quite so special. Like, I don't wanna know how the 232 00:14:50,670 --> 00:14:54,265 magic trick is done. I enjoy the magic. And this is sort of this exception 233 00:14:54,565 --> 00:14:58,405 to that rule of my mind of, like, what chat gpt and LLMs do 234 00:14:58,405 --> 00:15:02,180 is amazing. Like, it's magical in a lot of ways, But it's more magical if 235 00:15:02,180 --> 00:15:05,940 you understand how it actually works in that it's not smart at 236 00:15:05,940 --> 00:15:09,620 all. It's it's purely a prediction engine and it's really, really 237 00:15:09,620 --> 00:15:13,395 good at just sort of guessing what's next in a sequence based on on 238 00:15:13,395 --> 00:15:16,835 sort of those those tokens of, like, word groups and stuff. Right? So it's like 239 00:15:16,835 --> 00:15:20,570 this new numerical valuation that it sort of figures out on 240 00:15:20,570 --> 00:15:23,769 the fly. Right? And I've I've done some prompts and stuff like that where I've 241 00:15:23,769 --> 00:15:26,524 actually sort of broken it open and it gives you the v b script window 242 00:15:26,524 --> 00:15:30,285 where it starts actually generating. Like, it's not just the typewriter text where 243 00:15:30,285 --> 00:15:34,050 it actually starts like like filling in text and replacing text and going 244 00:15:34,050 --> 00:15:37,730 all crazy as it as it sort of does this this sort of multi line 245 00:15:37,730 --> 00:15:41,305 prediction. So I find this really mind blowing if you understand like 246 00:15:41,945 --> 00:15:45,625 it doesn't understand what it's actually replying in a lot of ways, like, 247 00:15:45,625 --> 00:15:49,404 the contextual awareness. It's it's really just sort of number sequencing 248 00:15:49,545 --> 00:15:53,180 based on these groups or, like numbers assigned to 249 00:15:53,180 --> 00:15:56,620 words, which to me, like I said, is is actually kinda more 250 00:15:56,620 --> 00:16:00,380 incredible that it's actually able to do what it's do without, without being 251 00:16:00,380 --> 00:16:04,134 intelligent at all. Right? Like, do I mostly understand that 252 00:16:04,134 --> 00:16:07,435 correctly? And are you as fascinated by how that works as me? 253 00:16:08,055 --> 00:16:11,720 No. No. You you do. You are. And and so it's 254 00:16:11,720 --> 00:16:15,259 even like so chat gbt is a chat implementation of 255 00:16:15,480 --> 00:16:19,135 a, text completion. So what what 256 00:16:19,135 --> 00:16:22,815 it actually is is every time there's a 257 00:16:22,815 --> 00:16:26,175 question response, it's just a longer, set 258 00:16:26,175 --> 00:16:29,930 of text completion being set in. So there's a system 259 00:16:29,930 --> 00:16:33,530 prompt for chat gbt, and it's it's literally 260 00:16:33,530 --> 00:16:37,145 says you are chat gbt. You can do, 261 00:16:37,145 --> 00:16:40,685 you know, these different things. You should answer users. 262 00:16:41,464 --> 00:16:45,139 The question and answers follow this this format and then 263 00:16:45,139 --> 00:16:48,579 system or or its user and then it's chat 264 00:16:48,579 --> 00:16:52,420 gbt colon. Right? And then the user 265 00:16:52,420 --> 00:16:56,045 sends first response, and it says user colon. 266 00:16:56,105 --> 00:16:59,865 And then it and then it it it leaves a 267 00:16:59,865 --> 00:17:03,689 blank spot for chat gbt colon, and then jet gbt fills in that text. 268 00:17:04,069 --> 00:17:07,270 And then it just keeps sending the same thing back over and over and over 269 00:17:07,270 --> 00:17:10,905 again. So, like like, you're not like, people think of it 270 00:17:10,905 --> 00:17:14,744 as, like, oh, I'm sending off, like, something and then it's, you know, listening to 271 00:17:14,744 --> 00:17:17,865 everything I'm doing and blah blah blah and sending it back. It's really just, you 272 00:17:17,865 --> 00:17:21,649 know, one. It's like a text file that just keeps getting a little bigger where 273 00:17:21,649 --> 00:17:25,464 it's like user response user response, which I 274 00:17:25,544 --> 00:17:29,225 I mean, once I saw that I was like, oh, really? Yeah. Yeah. 275 00:17:29,225 --> 00:17:32,265 It's wild. Like like like I said, it's just it's crazy how it works under 276 00:17:32,265 --> 00:17:35,790 the covers. So, yeah, getting getting them into into the 277 00:17:35,790 --> 00:17:39,630 nerdy weeds, but, you know, we're we're a technical group that generally listens to this 278 00:17:39,630 --> 00:17:43,095 podcast, so I'm sure people will will also find this 279 00:17:43,095 --> 00:17:46,235 somewhat fascinating. We'll switch to, 280 00:17:47,255 --> 00:17:50,860 a bit on the I guess the practical use cases, Right? So, like, 281 00:17:50,860 --> 00:17:54,540 you're you're building this specifically for MSPs. And, again, 282 00:17:54,540 --> 00:17:58,220 like like, I'd love you to just sort of expand on that. Like 283 00:17:58,220 --> 00:18:01,674 like, why why a model or this platform 284 00:18:01,815 --> 00:18:05,575 for MSPs in particular? What did you envision as being possible 285 00:18:05,575 --> 00:18:09,090 with that? So it's actually for 286 00:18:09,090 --> 00:18:12,690 MSPs to get in the AI business. So it's for MSPs to bring to their 287 00:18:12,690 --> 00:18:16,130 customers and naturally use it themselves inside their own 288 00:18:16,130 --> 00:18:19,055 business. The reason for that is I've seen, 289 00:18:20,395 --> 00:18:23,535 yeah, different mega trends in the past. You look at the move to cloud, 290 00:18:24,235 --> 00:18:27,760 which in a lot of cases increase the cost per 291 00:18:27,760 --> 00:18:31,360 seat, not necessarily the total revenue, but the cost 292 00:18:31,360 --> 00:18:34,885 per managed user or managed device, about 293 00:18:34,885 --> 00:18:38,565 50% in in additional revenue, through that 294 00:18:38,565 --> 00:18:42,170 that transition. And in many cases, a move to a managed 295 00:18:42,170 --> 00:18:45,630 billing model or the introduction of recurring revenue. 296 00:18:46,490 --> 00:18:50,165 And it took a while to get there. It meant 297 00:18:50,225 --> 00:18:53,985 much because, you know, talking to someone, hey. We're gonna move 298 00:18:53,985 --> 00:18:57,549 your server your exchange server out of the closet into Office 299 00:18:57,549 --> 00:19:01,389 365. Like, it was a it wasn't the easiest conversation to have. There's a lot 300 00:19:01,389 --> 00:19:05,125 of businesses who were reluctant, this move from capital 301 00:19:05,125 --> 00:19:08,505 expenditure model to operational expenditure model. 302 00:19:09,605 --> 00:19:13,440 But, you know, MSPs were the only group of people capable of handling 303 00:19:13,440 --> 00:19:16,960 that transition for small businesses, while big enterprises, you know, 304 00:19:16,960 --> 00:19:19,940 hired large IT teams to do it internally. 305 00:19:20,845 --> 00:19:24,684 Similar thing happened in cybersecurity, but it happened a a little bit faster. 306 00:19:24,684 --> 00:19:28,284 So, many MSPs again increased their per 307 00:19:28,284 --> 00:19:31,080 seat, revenue by about 50%, 308 00:19:32,100 --> 00:19:35,460 over maybe 5, 10 years with the, 309 00:19:36,020 --> 00:19:39,645 introduction of more cyber security services. And I'm talking about 310 00:19:39,645 --> 00:19:43,325 changing from I just offer, you know, a a web 311 00:19:43,325 --> 00:19:46,865 root or, you know, Sophos or like, I just offer 1 antivirus 312 00:19:47,100 --> 00:19:50,860 as part of your managed package in addition to your RMN to, you know, 313 00:19:50,860 --> 00:19:54,539 you're getting email protection, you're getting endpoint protection, you're getting sock 314 00:19:54,539 --> 00:19:58,315 services, MFA, like, the whole suite of it, all the 315 00:19:58,315 --> 00:20:01,455 the tools and services that m MSPs have been adding. 316 00:20:02,610 --> 00:20:06,210 Cybersecurity, hard sell. Very difficult. I've been in the 317 00:20:06,210 --> 00:20:10,049 cybersecurity sales training business through, you know, working at Scout for a 318 00:20:10,049 --> 00:20:13,825 while where, you know, you can make great cybersecurity products for small 319 00:20:13,825 --> 00:20:17,665 businesses for MSPs to deliver, but you still have to help the MSPs with their 320 00:20:17,665 --> 00:20:21,429 biggest problem, which is actually convincing people that they actually need 321 00:20:21,570 --> 00:20:25,330 need the damn thing. And and AI 322 00:20:25,330 --> 00:20:28,935 is just different. So it it it happened 323 00:20:29,095 --> 00:20:32,855 cybersecurity happened faster with the movement to cloud, and I think AI is 324 00:20:32,855 --> 00:20:36,640 gonna be a similar scenario where small businesses are gonna need help. 325 00:20:36,640 --> 00:20:40,240 They're all gonna need to integrate AI into their business, whether it's 326 00:20:40,240 --> 00:20:43,885 through, you know, what we see today with interactions with large 327 00:20:43,885 --> 00:20:47,485 language models and those use cases, which, you know, you might have been asking 328 00:20:47,485 --> 00:20:50,720 about, like, writing job descriptions, doing SOPs, 329 00:20:51,120 --> 00:20:54,640 documenting things, summarizing conversations, helping with customer service 330 00:20:54,640 --> 00:20:57,860 workflows, lots of text heavy, tasks. 331 00:21:00,304 --> 00:21:03,904 But I think it's gonna happen way faster because AI is show, 332 00:21:03,904 --> 00:21:06,460 don't tell. And it's an operational 333 00:21:07,559 --> 00:21:09,500 10 xer versus 334 00:21:11,240 --> 00:21:14,885 a cost center. It's still, you know, it's still a cost, but but it's 335 00:21:14,885 --> 00:21:18,645 it's much easier. Hey. You know, you know, this thing that's been taking 336 00:21:18,645 --> 00:21:22,325 your employees, whatever, 3 days to 337 00:21:22,325 --> 00:21:26,090 do here with AI, they can do it in in 20 minutes. Like, 338 00:21:26,230 --> 00:21:29,530 who's gonna say no to that? So I think the explosion 339 00:21:29,990 --> 00:21:33,715 with MSPs is gonna happen way faster than anyone's ready for. 340 00:21:34,674 --> 00:21:38,355 That's interesting so like I guess it's sort of dual purpose like there's definitely some 341 00:21:38,355 --> 00:21:41,790 things you can do to leverage the the models 342 00:21:41,790 --> 00:21:45,630 internally like you said like finding relevant SOPs for a particular issue kind 343 00:21:45,630 --> 00:21:49,010 of you know copilot for MSP 344 00:21:49,630 --> 00:21:53,434 type data, but also, you know, how is how is that 345 00:21:53,434 --> 00:21:57,195 being leveraged and utilized, for for the client base. Right? So 346 00:21:57,195 --> 00:22:01,010 I I like, I had a conversation with someone recently on the podcast 347 00:22:01,010 --> 00:22:04,769 and and it had occurred to me that, like, I I was surprised 348 00:22:04,769 --> 00:22:07,855 I hadn't thought of this earlier of I think what you're sort of starting to 349 00:22:07,855 --> 00:22:11,695 describe here is, like, the consulting opportunity around how you actually 350 00:22:11,695 --> 00:22:14,895 roll out, implement, and leverage AI in in, 351 00:22:15,850 --> 00:22:19,210 as an MSP in your client businesses is gonna become very 352 00:22:19,210 --> 00:22:21,470 relevant very fast. Right? 353 00:22:23,049 --> 00:22:26,835 Well, I think right now, if a small business needs help writing a prompt, 354 00:22:26,835 --> 00:22:30,535 who can they go to for help? Yeah. Yeah. That's a good point. 355 00:22:31,715 --> 00:22:34,330 I mean, MSPs are gonna get those questions eventually. 356 00:22:35,110 --> 00:22:38,810 And I would bet your average level 1 technician 357 00:22:38,950 --> 00:22:42,175 is better at writing a prompt than the average CFO of a, 358 00:22:42,815 --> 00:22:46,415 of a, you know, $10,000,000 small whatever. 359 00:22:46,415 --> 00:22:49,840 $5,000,000 a year small business. Probably safe safe 360 00:22:49,840 --> 00:22:53,440 bet. Yep. But like what about like, you know, 361 00:22:53,440 --> 00:22:56,980 like you leveraging AI, I suppose, in in in their workflow, 362 00:22:57,635 --> 00:23:01,175 even some training around like Copilot, like you're rolling out 365, 363 00:23:01,475 --> 00:23:05,235 you know, hey, we're gonna add Copilot for you guys and, you know, some 364 00:23:05,235 --> 00:23:08,950 training around how to utilize that. I think those are, you know, some 365 00:23:08,950 --> 00:23:12,410 interesting use cases, very valuable to the client as well. 366 00:23:12,790 --> 00:23:16,335 What about sort of, like, more complex integrations of of 367 00:23:16,335 --> 00:23:20,174 language models in, you know, workflows, right, like interactions with, 368 00:23:20,495 --> 00:23:24,230 their clients or, you know improvements in workflows internally. 369 00:23:24,230 --> 00:23:27,909 What about some of those more complex use cases? Have you thought put put some 370 00:23:27,909 --> 00:23:31,510 thought towards what those would look like potentially? Yeah. 371 00:23:31,510 --> 00:23:35,254 Yeah. So so for our, what what 372 00:23:35,254 --> 00:23:38,294 we're building at ads and when we're releasing, I don't know, it might be out 373 00:23:38,294 --> 00:23:40,530 at the time of the release of this episode, Is, 374 00:23:42,110 --> 00:23:45,870 part of this Just quickly on that, what's what's your release date? In 375 00:23:45,870 --> 00:23:49,515 in the 1st March is our is our first product just going out. So 376 00:23:49,595 --> 00:23:53,115 Perfect. K. So by by this the time this is live, this is very 377 00:23:53,115 --> 00:23:56,315 likely to be live. So right out and check it out. There you go. There 378 00:23:56,315 --> 00:23:59,820 you go. Yeah. But what we have is an is an, 379 00:24:00,140 --> 00:24:03,740 AI app builder where you can, do all the prompt 380 00:24:03,740 --> 00:24:07,495 engineering work and build all the inputs, into a dynamic 381 00:24:07,495 --> 00:24:11,175 prompt. You know, if then then this type of thing, and then 382 00:24:11,175 --> 00:24:14,960 publish that to the end users. The end users just upload a file, press 383 00:24:15,440 --> 00:24:18,740 enter, or, you know, type in yes, no, maybe so. 384 00:24:19,840 --> 00:24:23,440 So so examples are marketing. Right? Marketing is a is an example because it's 385 00:24:23,440 --> 00:24:27,274 synonymous with all all small businesses, and small businesses generally struggle 386 00:24:27,274 --> 00:24:30,715 with marketing. They don't have enough money to pay an outside firm or they're paying 387 00:24:30,715 --> 00:24:32,735 an outside firm. They're not doing a great job. 388 00:24:34,110 --> 00:24:37,789 Say, the very least, every time you release a new 389 00:24:37,789 --> 00:24:41,549 product, you wanna post about it on social media, and you don't have 390 00:24:41,549 --> 00:24:45,235 anyone to write about it 391 00:24:45,235 --> 00:24:48,615 on social media. So you you copy and paste 392 00:24:49,155 --> 00:24:52,660 your your MSP or, someone internally builds an app, 393 00:24:53,200 --> 00:24:56,980 that generates the social media posts, and it takes a specific input, 394 00:24:57,520 --> 00:25:01,325 whether that's a PDF of the product documentation that you're releasing 395 00:25:01,325 --> 00:25:05,165 or an email update or just a summary describing it. And 396 00:25:05,165 --> 00:25:08,789 they add some additional system context about the business and who it is and the 397 00:25:08,789 --> 00:25:12,570 tone that they like to use. And then the end user, all they're doing is 398 00:25:12,710 --> 00:25:16,205 copy paste. I want this to be specialized for ins 399 00:25:16,284 --> 00:25:19,725 for Facebook, for Instagram, for LinkedIn. I want it to be this 400 00:25:19,725 --> 00:25:23,470 length and and press enter. Another use case 401 00:25:23,470 --> 00:25:27,310 might be for job descriptions. Right? You could very easily like, that's a 402 00:25:27,310 --> 00:25:31,150 one to many, type scenario, example where 403 00:25:31,150 --> 00:25:34,884 an MSP could build, a an app 404 00:25:34,884 --> 00:25:38,725 that generates, job description where you input the title of the 405 00:25:38,725 --> 00:25:42,170 job, the salary, whatever the, expected inputs 406 00:25:42,170 --> 00:25:46,010 are, and then, it it, you know, generates 407 00:25:46,010 --> 00:25:49,770 the whole job description for them, and the MSP could populate that to 408 00:25:49,770 --> 00:25:50,635 all of their different 409 00:25:57,835 --> 00:26:01,660 of aspect of these models is like like there's often people 410 00:26:01,660 --> 00:26:05,179 talk about sort of future jobs are going to be prompt engineering and like maybe 411 00:26:05,179 --> 00:26:08,735 that actually gets sort of like like, sort of born out of this and and 412 00:26:08,735 --> 00:26:12,255 sort of becomes less prevalent than it currently is. But I don't think people really 413 00:26:12,255 --> 00:26:15,475 understand how different and how much better 414 00:26:15,760 --> 00:26:19,460 information you can get from these systems if you engineer the prompt 415 00:26:19,520 --> 00:26:23,360 just right. Like using sort of like like language variables and things like 416 00:26:23,360 --> 00:26:27,205 that. So like maybe just, like, a a quick bit on this, 417 00:26:27,205 --> 00:26:30,885 like, kinda your perspective because, like, I think a lot of people understand this is 418 00:26:30,885 --> 00:26:34,530 just, like, the ChatGPT interface. I ask some questions, I ask it to build me 419 00:26:34,530 --> 00:26:38,370 a, you know, a job description and it comes back with 420 00:26:38,370 --> 00:26:42,055 some pretty good stuff like like better than what most people would write 421 00:26:42,055 --> 00:26:45,495 but the difference between that and having a really good 422 00:26:45,495 --> 00:26:48,875 prompt that is designed in a way that actually outputs 423 00:26:49,230 --> 00:26:52,769 something that is like a 100 times better than what it would just generically 424 00:26:52,990 --> 00:26:56,750 spit out. I think it's something people don't quite understand. You wanna 425 00:26:56,750 --> 00:27:00,555 expand on that? Yeah. So with 426 00:27:00,555 --> 00:27:03,595 some of the newer models you can have a very large context window like I 427 00:27:03,595 --> 00:27:07,390 said earlier. So you can provide 5 examples of 428 00:27:07,550 --> 00:27:11,230 of job descriptions, and and write ups on them. You can 429 00:27:11,230 --> 00:27:14,910 provide nuances and, you know, if it's like this then do this 430 00:27:14,910 --> 00:27:18,655 and and, really spending time to make a very 431 00:27:18,655 --> 00:27:22,335 good prompt. But the the the thing that's changing each 432 00:27:22,335 --> 00:27:25,730 time is just, you know, the title and a couple words about the 433 00:27:25,730 --> 00:27:29,490 summary of what the job is. So we're we're 434 00:27:29,490 --> 00:27:32,725 actually, releasing a a course and it should be 435 00:27:32,885 --> 00:27:36,405 released actually. As this is released, you go to our 436 00:27:36,405 --> 00:27:38,745 website and get access to it on hats.ai, 437 00:27:40,389 --> 00:27:44,149 on on how to do basic prompt engineering and and the basics around 438 00:27:44,149 --> 00:27:45,769 it and and how to do it well. 439 00:27:47,945 --> 00:27:51,565 Yeah. I mean, I think it's a skill that's extremely 440 00:27:51,625 --> 00:27:55,270 relevant. You could think about how when Google first came out, the people who could 441 00:27:55,270 --> 00:27:59,030 get Google right away and a lot of those people own MSPs now. 442 00:27:59,030 --> 00:28:02,710 Right? They could get the information they needed and other people would just type in 443 00:28:02,710 --> 00:28:06,315 the wrong thing. So it's a similar skill. 444 00:28:06,855 --> 00:28:10,055 I can relate to that actually because I used to work at an ISP way 445 00:28:10,055 --> 00:28:12,750 back in the day, and we used to run these open houses where we, like, 446 00:28:12,830 --> 00:28:16,210 sort of educate people how to use the Internet and how to use search engines. 447 00:28:16,350 --> 00:28:18,990 I used to challenge people in the room when I was doing these sessions of, 448 00:28:18,990 --> 00:28:22,585 like, name something. I can, like, I can find relevant information 449 00:28:22,645 --> 00:28:26,325 somewhere and and they're like, oh, okay. Whatever. Like, how to build a box for 450 00:28:26,325 --> 00:28:29,045 ferrets? And I'm like, alright. Here you go. Like, can they it would spit it 451 00:28:29,045 --> 00:28:32,580 up. Like, like, who's the lead for this year's f one? And I'm like, 452 00:28:32,740 --> 00:28:35,620 easy. Here you go. Right? And, like, people were kinda blown away by this because 453 00:28:35,620 --> 00:28:39,060 they couldn't find that information. They'd have to they'd really struggle. And it was sort 454 00:28:39,060 --> 00:28:42,514 of you're right. Like, that's early prompt engineering is a good Google 455 00:28:42,514 --> 00:28:46,355 search. Like, maybe it's, again, less prevalent now. But, you know, it's a I 456 00:28:46,355 --> 00:28:50,120 think an interesting analogy. I guess like the one of the other 457 00:28:50,120 --> 00:28:53,640 sort of elephants around this is is data security. Right? I think is a really 458 00:28:53,640 --> 00:28:57,275 important, aspect of this and I'm curious sort of how you guys are considering this 459 00:28:57,275 --> 00:29:00,415 in in the product that you're building because, you know, 460 00:29:01,195 --> 00:29:04,919 MSPs hold a lot of, sensitive data and they're 461 00:29:04,919 --> 00:29:07,880 you know, I guess one of the another one of those things that people don't 462 00:29:07,880 --> 00:29:11,639 quite understand about these models is if you're using just sort of like 463 00:29:11,639 --> 00:29:15,465 free version of Chappy GPT you're potentially providing information to 464 00:29:15,465 --> 00:29:18,825 be fed back into them into the the training model. Like it's 465 00:29:18,825 --> 00:29:22,640 not this is this stuff is not as sort of of private as maybe 466 00:29:22,640 --> 00:29:26,080 people assume it would be. So, like, what are your thoughts about your product? How 467 00:29:26,080 --> 00:29:28,820 you're building it for sensitivity and care around 468 00:29:29,875 --> 00:29:33,395 potential prevention potentially sensitive data, that 469 00:29:33,395 --> 00:29:37,235 MSPs will be holding and wanting to leverage in this but are sort of 470 00:29:37,235 --> 00:29:40,980 cautious around security implications. Yeah. I I 471 00:29:40,980 --> 00:29:44,820 think just AI rollout as a whole or the goal AI 472 00:29:44,820 --> 00:29:48,394 transition, the biggest problem that we're seeing is, 473 00:29:48,635 --> 00:29:52,475 data readiness. So, for example, why not just turn 474 00:29:52,475 --> 00:29:56,180 on, Copilot, right, on on 475 00:29:56,180 --> 00:30:00,020 an organization? Like, Microsoft makes you do a whole data readiness thing where 476 00:30:00,020 --> 00:30:03,805 it's like, okay, this thing's gonna have access to all your PDFs, like, that 477 00:30:03,885 --> 00:30:07,565 you have in these folders. Like, did you, you know and 478 00:30:07,565 --> 00:30:11,245 and the concern is right, like, I'm a I'm a marketing intern and I 479 00:30:11,245 --> 00:30:14,500 say, how much budget should I allocate for, 480 00:30:15,200 --> 00:30:18,720 q 2, for this program? And it says, well, the 481 00:30:18,720 --> 00:30:22,255 CMO's salary is is, you know, this much money 482 00:30:22,255 --> 00:30:25,715 and based on this, you know, spreadsheet that I found. 483 00:30:25,855 --> 00:30:29,615 So so you need to be really careful there. I think 484 00:30:29,615 --> 00:30:33,040 we set this platform up, to be a 485 00:30:33,040 --> 00:30:36,800 secure, safe, alternative to sort of everything 486 00:30:36,800 --> 00:30:40,405 out there where we're not trying to monetize your data. We're trying to put you 487 00:30:40,405 --> 00:30:43,625 in the in the AI business where you have very granular control. 488 00:30:44,405 --> 00:30:47,820 And and a big part of that and, you know, it's it's a lot more 489 00:30:47,820 --> 00:30:51,580 work on our end to make to keep the data separate from the 490 00:30:51,580 --> 00:30:55,405 models. So we you can very easily switch, models 491 00:30:55,405 --> 00:30:59,185 in the future. Right. Or as you iterate or as you build. 492 00:30:59,965 --> 00:31:03,620 Another piece of it, I think is just the 493 00:31:03,620 --> 00:31:07,400 the use cases. So people are very quick to make publicly 494 00:31:07,620 --> 00:31:11,355 facing, applications with with generative AI, and that's how 495 00:31:11,355 --> 00:31:15,034 you get problems like the GM dealership or 496 00:31:15,034 --> 00:31:18,810 the Chevy dealership that, you know, like, people 497 00:31:18,810 --> 00:31:22,410 are getting it to generate Python scripts and sell them Teslas 498 00:31:22,410 --> 00:31:25,665 and Elon Musk is screenshotting it. And then all of a sudden GM's got a 499 00:31:25,665 --> 00:31:29,505 big PR problem because some random, you know, Chevy dealership and I don't 500 00:31:29,505 --> 00:31:33,185 know where. Right. Like set up a chatbot on their public facing 501 00:31:33,185 --> 00:31:36,770 website. Like, I I I you 502 00:31:36,770 --> 00:31:40,470 need control over what users are doing, 503 00:31:40,930 --> 00:31:44,335 and you need unification especially in, like, a customer 504 00:31:44,475 --> 00:31:48,235 service environment. So that's why, you know, we build things one to 505 00:31:48,235 --> 00:31:51,730 many where you can control things at the MSP level or at the 506 00:31:51,730 --> 00:31:55,250 admin level and edit prompts, as a whole. But 507 00:31:55,250 --> 00:31:58,610 also, you know, educating users, like, this is your first 508 00:31:58,610 --> 00:32:02,174 draft. You should review this before you set it publicly. 509 00:32:03,034 --> 00:32:06,715 Start with use cases like marketing and maybe not, you know, 510 00:32:06,715 --> 00:32:10,169 entering code into your, right, production, 511 00:32:10,950 --> 00:32:14,009 Linux terminal that, you know, hasn't been tested or vetted. 512 00:32:15,830 --> 00:32:18,645 So it's, you know, it's a way to do things faster and then a way 513 00:32:18,645 --> 00:32:22,325 to do things a little better. But, like, we have one product that is 514 00:32:22,325 --> 00:32:26,085 external facing, for example. It's a it's a phone customer service agent that you 515 00:32:26,085 --> 00:32:29,620 can call and it can take notes, transfer you, create a ticket, 516 00:32:30,080 --> 00:32:33,540 you know, send email, whatever. And the amount 517 00:32:34,240 --> 00:32:37,775 of guardrails we have to put in to, you know, end the 518 00:32:37,775 --> 00:32:41,295 conversation if it starts drifting, like, it's a lot. And and 519 00:32:41,295 --> 00:32:44,355 there's, like, AI, large language model security, 520 00:32:45,590 --> 00:32:48,950 is is just beginning because these tools were used 521 00:32:48,950 --> 00:32:52,630 internally for so long. Right. So I think that there's gonna be a 522 00:32:52,790 --> 00:32:56,515 like, there's gonna be browser plugins to prevent people from 523 00:32:56,515 --> 00:33:00,195 putting social security numbers or company information at a chat gbt that 524 00:33:00,195 --> 00:33:03,670 pop up. Like, there we got a ways to go on all this. 525 00:33:04,230 --> 00:33:07,910 Yeah. No. It's a part of the I suppose part of the issue of of 526 00:33:07,910 --> 00:33:11,615 such a fast evolving field is like, you know, this stuff takes time and 527 00:33:11,615 --> 00:33:15,135 has to the the the guardrails and safety has to evolve as it 528 00:33:15,135 --> 00:33:18,655 evolves as well. Right? Yeah. I mean, it's like 529 00:33:18,975 --> 00:33:22,440 imagine that we took, you know, like like, 530 00:33:23,620 --> 00:33:27,140 it's it's it's almost like it's like, email 531 00:33:27,140 --> 00:33:30,825 was not set up for security, and here we are still using the 532 00:33:30,825 --> 00:33:34,664 same version of it. Late I mean, yeah, we've we've improved a 533 00:33:34,664 --> 00:33:38,220 little bit. Right? Still evidenced by the fact that the 534 00:33:38,220 --> 00:33:41,900 number of fishing and, you know, user training that's still required. It's an 535 00:33:41,900 --> 00:33:45,500 inherently unsafe system for sure. Yeah. Like, it 536 00:33:45,500 --> 00:33:49,235 wasn't necessarily designed with bad actors in mind. And I and and to some extent, 537 00:33:49,235 --> 00:33:52,295 I would argue that that the initial large language models, 538 00:33:52,675 --> 00:33:56,290 weren't, you know, designed with that. And and now there's talk of, 539 00:33:56,830 --> 00:34:00,545 you know, we need a, a sovereign AI, 540 00:34:00,785 --> 00:34:04,005 like like, or AI infrastructure should be, 541 00:34:04,705 --> 00:34:08,465 regulated at the national level, where you have, you know, say, 542 00:34:08,465 --> 00:34:12,250 the large language model and the, GPUs that run them that are, 543 00:34:12,250 --> 00:34:16,010 you know, dictated by the federal government and, you know, maybe the Saudi 544 00:34:16,010 --> 00:34:19,665 government has a different one. And things will probably go that 545 00:34:19,665 --> 00:34:23,364 way, and I'm not, you know, like, I'm I'm I'm in no position to 546 00:34:23,505 --> 00:34:26,910 comment. Maybe some, I don't know, like, FNAI company, but that's about it. I could 547 00:34:26,910 --> 00:34:30,750 have an opinion, I guess. Yeah. Yeah. Yeah. Yeah. I I should say I'm 548 00:34:30,750 --> 00:34:34,545 not, an expert in foreign policy and Sure. Yep. Live 549 00:34:35,245 --> 00:34:38,685 liberty, like, civil liberties and and 550 00:34:38,685 --> 00:34:42,400 copyright and all of that. But one thing I know for sure 551 00:34:42,799 --> 00:34:46,239 is the technology will 552 00:34:46,239 --> 00:34:49,920 evolve faster than the regulations. Agreed. So it'll it 553 00:34:49,920 --> 00:34:53,635 it almost doesn't matter. Yep. Like, OpenAI is 554 00:34:53,635 --> 00:34:57,315 getting sued by the New York Times, but everybody's already using them. So Right. I 555 00:34:57,315 --> 00:35:00,780 think. Yeah. I mean, like, you know, like, Meta 556 00:35:00,780 --> 00:35:04,540 and, all of the the social platforms have been around for 557 00:35:04,540 --> 00:35:07,580 a long time. They still haven't figured out how to regulate them either. Right? So, 558 00:35:07,580 --> 00:35:11,265 yeah, I don't think they're gonna be, I mean honestly maybe they're they're quicker with 559 00:35:11,265 --> 00:35:14,785 AI at least they're they're trying to work on it on some regulations around it 560 00:35:14,785 --> 00:35:18,590 but agreement on how that's actually gonna play out I think will be 561 00:35:19,630 --> 00:35:23,250 pretty pretty, messy for a little while yet. I guess 562 00:35:23,310 --> 00:35:27,135 that naturally leads to, you know, prediction time, right? 563 00:35:27,215 --> 00:35:30,655 Predictions are are terrible because, you know, if you get them wrong then people may 564 00:35:30,655 --> 00:35:34,255 remember. If you get them right, it'll see it'll seem, obvious in in 565 00:35:34,255 --> 00:35:38,040 hindsight but, I would love to sort of since you've put a lot of time 566 00:35:38,040 --> 00:35:41,480 to this and and you're passionate about the field I'm I'm curious about sort of 567 00:35:41,480 --> 00:35:45,005 how you feel AI, writ 568 00:35:45,005 --> 00:35:48,625 large will impact and change the the the MSP 569 00:35:48,685 --> 00:35:52,509 business model for, sort of in the next five to 10 years? That's 570 00:35:52,509 --> 00:35:56,030 a I think in 5 to 10 years, 571 00:35:56,030 --> 00:35:59,250 MSPs will be managing, potentially 572 00:35:59,390 --> 00:36:02,694 tuning large language models and their, 573 00:36:02,934 --> 00:36:06,555 relevant insert in infrastructure around them like vector databases, 574 00:36:07,255 --> 00:36:11,089 for the majority of their customers. And and that 575 00:36:11,089 --> 00:36:14,710 may be, like, every small business, has 576 00:36:14,770 --> 00:36:18,234 some version of a customized large language model that they 577 00:36:18,234 --> 00:36:21,835 use inside their business. And the 578 00:36:21,835 --> 00:36:25,570 MSPs are managing the infrastructure for that. I also think that there's going to 579 00:36:25,570 --> 00:36:29,410 be a movement back to private cloud and on prem for some of this 580 00:36:29,410 --> 00:36:32,390 stuff. And the most qualified 581 00:36:33,330 --> 00:36:37,085 people I know to go set up a server room or a server 582 00:36:37,085 --> 00:36:40,925 rack with a bunch of GPUs and virtualization and, 583 00:36:40,925 --> 00:36:41,700 you know, management 584 00:36:52,935 --> 00:36:56,535 ongoing maintenance of it. But, and and I 585 00:36:56,535 --> 00:37:00,079 also think that, you'll start to see super winners in 586 00:37:00,079 --> 00:37:03,920 in in in industry pop up. So I don't know if it'll 587 00:37:03,920 --> 00:37:07,595 happen in MSPs, but say one of these really big MSP platforms 588 00:37:08,215 --> 00:37:11,975 gets really smart about, organizing their data, and 589 00:37:11,975 --> 00:37:15,800 they they create, you know, this thing that everybody's been trying 590 00:37:15,800 --> 00:37:19,560 to sell. Like, the early AI people have been, you know, dreaming of of 591 00:37:19,560 --> 00:37:23,375 this AI that solves tickets for you automatically and can do your 592 00:37:23,375 --> 00:37:26,815 job. Like, say, say, say somebody figures that out and then starts 593 00:37:26,815 --> 00:37:30,175 licensing that out to all the other MSPs. And they're no longer in the MSP 594 00:37:30,175 --> 00:37:33,630 business anymore. Right. They sell all that off. They're just in the data 595 00:37:33,630 --> 00:37:37,310 business and own, own the dataset that, you know, can solve tickets. Like that 596 00:37:37,310 --> 00:37:40,644 stuff's going to happen in every industry, 597 00:37:41,105 --> 00:37:44,144 I think. Yep. Yeah. I think there's 598 00:37:44,865 --> 00:37:47,744 it'll be wild to sort of see you, like, there's there's sort of, like, some 599 00:37:47,744 --> 00:37:51,130 of these things I think are are somewhat predictable and then there's gonna be 600 00:37:51,130 --> 00:37:54,730 absolute wildcards that in, like, 3 to 5 years especially we're like, 601 00:37:54,730 --> 00:37:58,508 oh, interesting. Did not see that coming. Right? So 602 00:37:58,508 --> 00:38:00,875 it would be a fun space to watch for sure. 603 00:38:02,535 --> 00:38:06,250 Awesome. Well, this has been great, Jimmy. I appreciate your time and, best of luck 604 00:38:06,250 --> 00:38:10,010 with this. I think it'll be really fascinating to see how this evolves 605 00:38:10,010 --> 00:38:13,775 and, really, really happy to see, you in particular as one of the 606 00:38:13,775 --> 00:38:17,295 people sort of at the at the head of this spear head of the spear 607 00:38:17,295 --> 00:38:20,970 for the MSP industry in in evolving the AI workload 608 00:38:20,970 --> 00:38:24,810 for them. Thank you so much. Really appreciate it. Really 609 00:38:24,810 --> 00:38:27,470 enjoyed the conversation today. Alright. Take care.