6/26/2026 - How to Prepare for the AI Plateau by Mastering Local Models and Personal Data

Transcript

1. 0:01[music] [music] [music]
2. 0:21[music] Heat. Heat. >> [music] >> Good evening. Good evening. Good evening. Welcome to the AI salon.
3. 0:36Welcome to the AI learning lab. It's Friday night date night. It's Friday. That means you can ask me anything.
4. 0:43Anything. We can talk about anything. If it wasn't anything, it would be AMS. Ask me something. But it's an ask me anything. All right.
5. 0:54[clears throat] So, if you got questions, pop them in the in the uh in the comments over here. If you're
6. 1:01wondering how to get to the comments in this corner right here is the button. Sitting underneath the AI Learning Lab
7. 1:07logo is a little button. And in that button, uh it'll pop open the comments. And then down in this corner, you can
8. 1:14pop out this window is a separate window. And then you can open a separate tab where you can go surf along in the
9. 1:20AI salon if you want to go check out the spaces there. if you want to you want to share anything. Um, there's a whole lot
10. 1:27going on in the world of AI right now. Um, it's weird. We're in a weird place. So, I've got thoughts as I always do. Um,
11. 1:42I'm actually excited about where we are in a weird kind of way. the government in this sort of ridiculously uh hamfisted kind of way
12. 1:57is providing a little breather and it's likely that we as individuals are not going to have access to the
13. 2:13frontier models anymore. um if you work for a big company that's on the list, you might um but I think that's going to be the
14. 2:23case. And so what that what that starts to look like for me is well, if you've been waiting for AI to get perfect, it's going to be a while.
15. 2:37And if you've been waiting to really dig in because the tools are going to keep changing, this has the potential of kind of
16. 2:46stabilizing things at a certain level. So what happens always with with technical innovation is there's
17. 2:54waves of of uh improvement, right? And so we've been improving and it's been accelerating and and it looked for a moment there like it was
18. 3:06just going to keep going. And what happened was it it started accelerating and the government has come in and
19. 3:12literally stopped it. Now it doesn't mean other countries and other labs aren't still developing things. It
20. 3:20doesn't mean that anthropic and open AR AI aren't still developing things. They are. But but that stop is going to is going to plateau
21. 3:30the development for a while now. Is that while three months or is that while three years? I don't know. There's a
22. 3:39really interesting moment. Linda Sherman's in the house joining from Kawaii. Welcome. Aloha. Alo. It's Wait. If you're joining from
23. 3:49Kawaii Oh, it's earlier there. Is it four hours earlier? I'm in Denver, so it's 7:30. So, is it 2:30 there? That
24. 3:56would be five hours. 2:30 I'm guessing. Time for a China large language model watch. Well, well, so so I've got
25. 4:04thoughts on that, too. Digital gods. So, here's the deal about the China large language models. The guy I'm pretty sure
26. 4:12is Quen is Al Alibaba, correct? It's 3:30 and quiet. Okay, cool. Good afternoon to Linda. [laughter] Um,
27. 4:24so, so Quen Quen, yeah, Quen is Alibaba. So, Anthropic two days ago, Quen is Alibaba. Okay. And and Alibaba, the the
28. 4:39Quen 3.6 model is really good. Like, it's almost as good as like um Opus 4.8 or 4.7 or 4 point 4.6. and it's small, but it's really good and
29. 4:53it's almost as good as the as the big Anthropic model. Well, turns out funny you should mention that. So, Anthropic
30. 5:04shipped off a letter to the to Congress that that said, "Hey, just in case you were curious, Alibaba created 25,000 fraudulent accounts,
31. 5:17right? They they weren't supposed to be able to use the fancy models. They created 25,000 fraudulent accounts and
32. 5:25used those accounts to reverse engineer the data inside the Opus 4.6 model. Hey, Mr. Um, so now that they've been caught, they
33. 5:40probably can't go through that door anymore. Maybe, maybe not. Who knows? like, you know, did did Anthropic only
34. 5:49just catch this or is it just convenient for them now that they're in cahoots with the government or trying to be
35. 5:56swell with the government that I don't know. I don't know what's going on. But that's that one one point of
36. 6:02information is the reason that Chinese large language models caught up to US large language models quickly is because
37. 6:10they were training their models on them. And so they were basically just replicating them. So that makes good
38. 6:16sense. And then Deepseek, I'm pretty sure it's DeepSeek. Um that was Deepseek was the one DeepSseec R1 was the version
39. 6:25that totally freaked everyone out for a while because it was so good. That company was funded by an individual, a very rich individual that
40. 6:34just said, "I'm going to create my own models here." Um, and that individual yesterday or the day before announced
41. 6:42that they raised $7.5 billion dollar because they realized that to catch up to mythos level models. They can't just
42. 6:54write checks for it. That they've got to raise some real capital. And so they raised seven half billion dollars. So I
43. 7:01think it is likely that the Chinese were keeping up because of the way they were training their models on our
44. 7:10models. I think that's probably going to slow down. They're probably going to block block that. And then we've now hit
45. 7:17a level of compute requirement to be able to hit a mythos level model that requires real investment. So I think
46. 7:24you're going to see the improvement of the Chinese models taper off. The one place where the Chinese models are
47. 7:30absolutely kicking ass is in video. Um, so I don't necessarily see that stopping. But in terms of the large
48. 7:37language models and the coding things, oh, and then there was yeah, GLM 5.2. Um, the the question is what about GLM
49. 7:455.2? That's an Well, it's 5.2. It's a model I hadn't heard of before and it's supposed to be quite good. And then
50. 7:53there was another model that came out. So, what's likely going to happen is this. Our government is going to dick
51. 8:00around with the frontier model companies and they're they're just going to do stupid [ __ ] like they're going to let
52. 8:05Elon do anything, but he doesn't seem to be pursuing this as aggressively as he was. He's now renting out his GPUs to
53. 8:12Anthropic and Google. Hey, Kelly Camp. Um, GLM 5.2 is just it's just another open source model. You've got a bunch of
54. 8:23models now that that can either run on really big hardware like 512 gigabyte class hardware. Um, and then you've got
55. 8:32models that can run on sort of 64 gig 128 gig models. Um, what I talked about in office hours today is that I think that
56. 8:44a good exercise to do right now, Mr. It is in the middle of it. A good exercise to do right now is start treating your
57. 8:55how you're learning about AI as if this moment in time that that whatever level of capability AI is at right now, just
58. 9:03assume that you're not going to have much better models for at least 6 months. Maybe they come sooner, but assume they're not going to.
59. 9:16And if you're not going to get better models, then what progress looks like is you getting better at at learning
60. 9:24agentic tools, learning how to build your digital self, learning about creating context and memory and and and things like that, and really
61. 9:33understanding um the difference between a cloud-based model and a local model. really understanding that your data, your personal data
62. 9:45has incredibly high value in a future full of AI. And so starting to learn about all that. I'm so okay with the
63. 9:53pause. Yeah. Just Yeah. This might be a time where you just pause to pause. Like screw it. If the if we're not going
64. 10:02to have access to the latest models, let's just play with the toys we've got or just don't play with the toys at all.
65. 10:11This might be a good time to look at yourself. What do you want? What do you want more of? The answer to that might
66. 10:21be I want more of non AI. Go sit in a field. All right. Browser crumbs. Are open-source AI models becoming the people's version of AI or
67. 10:34will they eventually get locked down too? Well, so this is Okay. So
68. 10:53open-source models of of a certain class I think will probably get restricted. Um, I doubt I doubt that our government's going to
69. 11:07[ __ ] with anything that's sort of at current levels, right? Maybe if you've got something like GLM 5.2 too. That's a
70. 11:15big model that runs on big hardware that might get restricted at some point. But but even that is probably not
71. 11:23in the category of models that like Mythos the the I think it's a 10 trillion parameter model. I think it's 10. Can you confirm
72. 11:35that Brandon? If mythos is 10 trillion parameters, I think that's right. Um the the capacity of that model to learn
73. 11:50and exploit all of the vulnerabilities of a of a given system. Um, made it kind of weaponsgrade material, right?
74. 12:04I just just remember something Brandon said to me Wednesday night. [laughter] I'm not going to say it hard, but I just remembered it. Um,
75. 12:17so if we have open- source models that are as strong as something like Mythos, I think it is possible that that stuff
76. 12:24gets restricted. I can't imagine the government saying that companies can't use kind of Opus 4.6,7 six, seven, or
77. 12:35eight level models because those are models that while you could do some nefarious things with them, they're not
78. 12:42good enough to just turn it loose and have it fully exploit a system in a couple of hours like like Mythos is.
79. 12:50So, I think we we really are talking about a new like a new category of model where its capabilities are significantly
80. 12:58above the the current watermark. That said, you know, our government, I think it was in the late 80s or early 90s, tried to
81. 13:11make encryption technology illegal. You can't export this. And it's a federal offense if you export encryption.
82. 13:22And you could actually create that algorithm with three lines of Python code that you could print on a t-shirt.
83. 13:29So all the hackers wore these t-shirts that said this t-shirt is a federal offense. [laughter] It's kind of silly. So I think we're
84. 13:38probably in a in a similar kind of um panic right now. Hang on a sec. Got to plug in my machine. Um we're in a similar kind of
85. 13:52panic right now. So, I would not be at all surprised if at some point the government doesn't do something stupid
86. 13:58like any model that's from China is going to be a federal offense to have on any hard drive that's connected to any
87. 14:04internet for you to use. Um, so it's possible. All this to say what I've been talking about for a month or so is I had this
88. 14:17instinct. What was it two months ago? I said, 'I just have this sneaking suspicion that buying hardware that can
89. 14:24run powerful AI models and learning about large language models and open source models and learning about agentic
90. 14:31platforms like OpenClaw and Hermes and then learn learning to use things like codecs and clawed code um and and store
91. 14:41your data locally. It's just it's feeling more and more important now we know why. Like now we know why that instinct was there. [laughter]
92. 14:51Um I didn't quite know it back then, but I was just like something's something's eating at me. Like there's patterns.
93. 14:57Patterns are forming. The way Dario Amade was talking, the way Sam Alman shifted his narrative, the way there was
94. 15:03just something happening. And so now we're now we're here, right? So we're at this place. Also, I don't know if you
95. 15:10noticed this if if you're not paying attention to this stuff minuteby minute like I am. Um, Apple today increased their prices by about 25%.
96. 15:25So the the 4,200 buck machine I got to 64 gig M5 is now like six grand or 6,300 or 6,400 something like that. The 128
97. 15:37version of that is 7,200 bucks. 7150. So 25% increase. that ain't going down. So, demand for the ability to generate AI tokens,
98. 15:53demand is going to keep going up. The ability to generate those tokens is going to get choked and constrained. And uh
99. 16:05yeah, so start learning this stuff. Start learning this stuff. we can share open- source models amongst
100. 16:14oursel like mixtapes. That's what that's what it's going to be like. So, um, one of the one of the things I talked
101. 16:23about today in office hours that as I was finishing talking about it, I realized that I sounded like a conspiracy nut. But we could be in a
102. 16:34situation where in order to be economically viable as a human being or as a as a community, right? Either as a
103. 16:43as a small business or as a community or as an individual that access to powerful models might be a requirement. And
104. 16:56it might be that access to those models becomes illegal. So, get them now. Get what you can now. I I mean, it's it
105. 17:09it's still kind of baffles my mind when I when I log into Let me let me see if I've got it up and running here.
106. 17:23Let's see. Go here. We're going to do this. Do this. Do that.
107. 17:53Okay. Oops, that was wrong.
108. 18:09Am I in codeex here? Okay, I'm in. Okay, so right here, what you're seeing that that little dude's just the the
109. 18:22codeex pet that I made look like me. What you're looking at right here is I'm in LM Studio. So, LM Studio is
110. 18:41Let me see if I can make this window bigger. Hang on a sec.
111. 18:55Why can I not make that bigger? There we go. Okay. So, this is I'm I'm running the Quen 3.6 30 35 billion parameter
112. 19:05model, right? I've also got Gemma on here. I've got the the open open AI's model. And so, I can come in here and I can say, um,
113. 19:17tell me about Turkish invasions. in the third century.
114. 19:32And so that pretty quick response that you're seeing of Turkish invasions, which there were no Turkish invasions of the Roman or
115. 19:44Byzantine Empire in the 3rd century. The Turks did not exist. Well, okay. Okay. So, I'm going to say, okay, listen. I'm a dum dum.
116. 19:57Um, tell me about when they did do invasions. [laughter] And I spelled invasions wrong. No worries at all. History can get tricky.
117. 20:12The Turks, specifically the groups that became Seljukes and later the Ottomans, started invading Anatolia, modern day
118. 20:20Turkey in the Middle East much later, the 11th century. Okay. So, anyway, this thing that I'm I'm using right now
119. 20:29that kind of looks and feels like chat GPT, the file itself for this large language model is sitting on my MacBook Pro. So,
120. 20:39if you don't know the difference between a cloud-based large language model and a local large language model, the local
121. 20:46the local file that contains all of the knowledge of humanity is actually just sitting over on my desk over there. And so, I
122. 20:58can be completely disconnected from the internet and use this thing. And it's also running as a server right now. Let me see where do I have
123. 21:09If I go up here, where is it? Here, local server. So, I've got these things running as local servers. And
124. 21:18what that allows me to do is I can use codeex or I could use openclaw or I could use Hermes to do tasks for me like
125. 21:29programming tasks where I don't have to go out to open AAI. I can do it right on my local machine. Now, the reason you go
126. 21:36out to those cloud services is that instead of running on 64 gigs of RAM, they're running on terabytes of of RAM,
127. 21:46like just massive amounts of RAM. And they can do all sorts of crazy things much faster than I can and at a much
128. 21:52higher quality. But if things start becoming less and less available or they start becoming more and more expensive,
129. 22:00then doing something like this is important. Mine was called Encyclopedia Bratannica and it took up an entire bookshelf. Exactly. Um,
130. 22:11yeah. So, kind of crazy, right? [clears throat] And how much memory does that look AI require? So, my
131. 22:25my laptop is 64 gigs. Essentially, what you can think of it is like a like the 35 billion parameter model. There's different what's called
132. 22:35quantization levels. Um, and it's basically how compressed the model is. The kind of biggest compressed version
133. 22:49of of that 35 billion parameter model is about 25 gigabytes. And so what that means is just to load the model into
134. 22:58memory, you need about 25 gigabytes. And then you've got your operating system and any other applications you're
135. 23:05running. So figure 32 to 40 gigabytes of my 64 gigabyte RAM is taken up by the model itself and other applications
136. 23:16running. And then anything I have above that is my context window. Right? So you talk about context window, how much sort
137. 23:24of short-term memory does your model have when you're chatting with something? You know how sometimes you're
138. 23:29chatting with an LLM and it sort of forgets what you were talking about and gets really crappy? That's because you
139. 23:34have a a short context window. So you you can run bigger models, but you need more RAM and then you need more RAM
140. 23:44to be able to run bigger context windows. So, so to run what I just showed you locally, that's on a 64 gigabyte
141. 23:52um M5 Max. So, I basically have the fastest M5 chip I could get when I bought it with 64 gigs of unified memory. If you're on a PC,
142. 24:05they don't have unified memory. So, then you're using GPU RAM, right? So on your graphics card, you need,
143. 24:14you know, 32 or 64 gigabytes or 128 GB of VRAM, video RAM on the GPU itself to be able to do this. What's the minimum system to get?
144. 24:30Quant quantization is a reduction in precision, right? Yeah, exactly. Who's that tiny avatar there? That's
145. 24:42Kyle's mini me. Yeah, that's that's that little that little dude is um in Codeex. They they released this
146. 24:52thing called Codeex Pets and and there's like pre-built ones in there, but you can also just say to Codeex, you can
147. 25:00upload a picture into Codeex and say, "Use this picture and make me a pet that matches the picture." And that's what I
148. 25:07did. It moves. Yeah, it does move. Here, I can show it again. Um, let me jump over there. Okay, so these are all my terminal
149. 25:20windows. And then let me go to here's codeex. So if you run codeex, codeex kind of looks like chat GPT,
150. 25:35but it's got access to a whole bunch of other things. It's got access to my hard drive. This little screen share thing is a
151. 25:48little crappy. It's got access to my hard drive and it's got access to tools and it's got access to the internet. And
152. 25:54the way he works, you can he just sort of floats on top of all the applications. And then I can say, um,
153. 26:01let's see. Um, tell me what you know about my projects and how they
154. 26:16relate to one another. And so as as I use codeex, it'll it'll push little updates to my little dude here.
155. 26:35And so like if it goes to use a tool, it'll sometimes put a little a little thing at the bottom of him so
156. 26:42so he says what's going on. Um he's for the most part useless. But why why I wanted to learn how to do this and I'll
157. 26:49show you how to do this if you want me to show you. Um is at some point we're probably going to have a device. It could be our phone,
158. 27:00but it's very possible that we're going to have a little device that's got a little screen on it that we talk into
159. 27:08and the screen will give us little updates. and all we're really doing is telling something like Codeex, which has
160. 27:13got all these different agents working on on our our behalf to go do something. And so I thought, why not learn how to do this little
161. 27:23thing? And because maybe I'll have on my little screen thing at some point, you know, like remember the the Rabbit R1,
162. 27:32you know, remember this thing? So imagine a little thing like that, maybe a little bit smaller. It's got a
163. 27:38little screen on it. You talk into it, tell your AI to go do [ __ ] Um, so here here's what Codeex now knows about me. It's got all
164. 27:53sorts of context and active work files and identity files and frameworks and it tells me where all that stuff is. And
165. 28:02StoryVine proves your enterprise story building storytelling credibility. The AI salon builds the human community
166. 28:09network. AI learning lab supplies the public body of work and lived evidence. Great repurpose translates all of that
167. 28:17into a movement keynote and marketf facing frame. Content evolution digital advisers turns expertise into practical
168. 28:23AI native tools. Sydney explores the emotional and mythic side of the same transformation. And agent brain which it
169. 28:30just invented for itself keeps the whole thing from scattering across charts folders. So, Agent Brain is the memory
170. 28:38system that I created over the past three weeks or so. It's got all my [ __ ] in it. And then he's just there. He's
171. 28:45kind of annoying because you just got to move him into a corner where he's not in front of something that you need to look
172. 28:50at, but that's what he is. [laughter] So, and it looks like there's an update of codeex here. So I update this thing
173. 29:02regularly because they keep adding features. And the way this works that that thing that you just saw there is LM Studio.
174. 29:13This is Codeex. So LM Studio is the one that's running the Quen 3.6 model. Right now Codeex is just using Chat GPT 5.5.
175. 29:24But at some point I'm gonna I'm gonna hook it up so that I can I can use codeex with that Quen model. So even if
176. 29:33I'm not connected to the internet, I could still be doing um having the agent do things on my behalf with that local
177. 29:40model. And that's that's something that I want to play with. Um, okay. Beautiful.
178. 29:55If AI coding agents are quietly writing more code, how do we hold responsible if the code c Oh, who do we hold
179. 30:02responsible if the code causes a problem? Well, I think
180. 30:18There's there's precedent that basically said that tool companies cannot be held responsible for what their tools do. The
181. 30:25most famous of which being guns, right? Um, so in America at least, um, liability for that is is is on the individual.
182. 30:42Um, if you work for a company and you release code into the public and that code is all [ __ ] up, the company is on
183. 30:50the hook for that, right? So, I don't think it's any different here. If you're going to put something out under your
184. 30:56name and you don't bother to check it, that's kind of on you. Um, the thing that's
185. 31:13a little more subtle than that is that these agents, if you point them at big thorny problems, go off and build all sorts of
186. 31:26new systems and connect with all sorts of other systems And there could be transactions or interactions
187. 31:36in between the thing that you asked for and the thing that you see where there are all sorts of problem problematic
188. 31:44things happening that you're never even aware of. Um, so I like we're we're getting into very weird territory here. Like one of the
189. 31:54things that I learned today in office hours, which by the way, if you don't come to office hours, you should. It's
190. 32:00Fridays at 11:00 a.m. Mountain time on LinkedIn. Um, like I I promote it on LinkedIn. It happens here in the salon. Um,
191. 32:16the difference in capability between Claude Opus 4.8 eight, their biggest, you know, sort of clawed model and then this mythos thing
192. 32:30is dramatic. But if you were just to sit down at like the clawed interface, like the chat GPT interface and and prompt Opus 4.8 and prompt
193. 32:44mythos, you probably won't see a difference. But if you say to your agentic tool, hey, I want you to go analyze this
194. 32:54really complicated codebase and find all of the security vulnerabilities, patch them, and come up with a dozen new improvements that I I
195. 33:06haven't thought of. Opus 4.8 might take three days to do that and it'll do it poorly. Mythos might take two hours to do that and
196. 33:18it'll do it brilliantly. And so the the time shrinkage with these much more capable models is
197. 33:32the thing that you start to measure their capability in. is when you point them to long-term complicated problems
198. 33:40like go examine a codebase or go read all of the documents in some trove of documents and analyze them. That's where
199. 33:49those models really start to shine. It's beyond our ability as individual humans to really understand what they're doing
200. 33:58and how they're doing it as fast as they're doing. We like it's just doing stuff so incredibly quickly and at such
201. 34:06a deep level of analysis it's it's hard to understand. Don't be a tool. Come to office hours. [laughter] Did I just say that?
202. 34:16Oh my god. Tabs are my enemy. Um like controlling a vicious dog. Yeah, kind of like that. A little machine with
203. 34:24a screen like the rabbit. Exactly. Exactly. Exactly. Um yeah. Yeah. What else? What other questions we have?
204. 34:53Let me talk before we get to the next question. Let me talk about um the AI salon mastermind. If you if you're not a member of the
205. 35:01mastermind, um you should you should join it. Um in the mastermind, we've got the the practice lab, uh which is a group of
206. 35:10people developing daily practices around how they use AI and we've also got an increasing number of workshops around
207. 35:18the great repurpose. So this idea that AI is about to transform all of our jobs. And if we think about the future
208. 35:26of work and the future of leadership, how do we do that if everything's changing? And that's what the great repurpose is about. the the
209. 35:35conversations happening um in those workshops and in those meetings in the great repurpose meetings, the mastermind meetings
210. 35:44is incredibly inspiring and in it's it's hard to describe um other than you've got a combination of people understanding their value
211. 35:58beyond their work. And as you start to decouple your identity from the work or your training or your college degrees or what you used
212. 36:09to do and really start to understand who you are and and and what you can offer the world, that separation is sometimes
213. 36:17quite profound. And so those conversations are very very deep. And it's not just about sort of the woo woo
214. 36:24part about, you know, separating your identity and re rediscovering, you know, who you are and what you value, but once
215. 36:31you start to understand that, hey, wait a minute, I've got value beyond my tasks, you can then start to ask very profound
216. 36:39questions like, well, what do I really want? What's the change I want to make in the world? So, the mastermind and the great
217. 36:46repurpose within it is about really going deep on you know what is possible for me to do and for us as a community to do. Um, and
218. 36:58that's what it's about. And so there's there's a human side of that. There's an AI side of that. And the AI side of
219. 37:04that, we're going to be focusing on creating a digital version of yourself. So if you haven't joined the mastermind,
220. 37:12you haven't joined this stuff, please join it. All of the conversations that I'm talking about, they're there on
221. 37:18demand. So all of the workshops, all of the these conversations are there on demand and you can join those
222. 37:24conversations. They're quite powerful. So So please jump in and do that. Mary Chester, I I better not ask my
223. 37:32simple questions. I don't want to be that person. Ask your simple questions. Here's So Mary, let me share something with you.
224. 37:42I had an experience in 1997 where the most powerful man in publishing. You could argue he was the editor-inchief of of Time Inc. So all of
225. 37:58the Time Inc. magazines. He was the editor-inchief of all of those. Um, he grabbed me by my this in a party and pushed me in a corner and
226. 38:13told me I had to fire people. Um, it it was one of the most humiliating. So, basically, we screwed up. We screwed
227. 38:24up a book launch because we didn't have the internet connected to launch a website. and he was pissed and he was justifiably
228. 38:34he was justified in his pissed-offness and it was humiliating. And so we didn't end up firing anyone, but we did stay up until 4 in the
229. 38:43morning that night dissecting every single moment that led up to that failure that led to me being accosted
230. 38:52and my my partner Chan also being accosted in the corner. And what we learned was that there were multiple points along
231. 39:06preparing for that event where if any of the dozen people that were at the table had asked what they thought was a stupid question,
232. 39:17if any one of them had asked one stupid question, one simple question like, "Hey, Just in the off chance that Verizon
233. 39:29doesn't get internet connected down in that space, do we have a copy of the hard the a copy of the website that we
234. 39:36built on the local hard drive of the machine that we're physically walking down to that space? Nobody asked that question. And the
235. 39:45answer was because nobody asked that question, we did not have the the website locally on the computer. We assumed
236. 39:54that the internet would be connected because Verizon told us the internet would be connected and it wasn't. And so
237. 40:01we didn't have a backup. And it wasn't like today where you just send it over your phone. Hey, I need the website. Can
238. 40:07you just zap that over to me? So, so yes, you should ask your simple question. And I don't see a simple question there. So, ask your simple
239. 40:18question. They're really important, actually. Um, okay. while you're while you're getting up your nerve to ask your
240. 40:22simple question. The simple questions here here's the thing other thing about simple questions. Your simple question I
241. 40:28guarantee you other people on this call have it right now too. You'll be the one that has the guts to ask it. Okay.
242. 40:37If tools like codeex are moving from chat with AI to assign work to AI, then what does that change? What does that change for
243. 40:47non-developers? Well, I'll give you two specific examples. I'm a non-developer. I'm the CEO of a
244. 41:01tech company, but and I'm I'm the inventor of the tech, but I'm not the builder of the tech. We've got technologists that built it. Um,
245. 41:15so one of them, I'll give you three examples. One of them was I had Claude do an analysis of our marketing website storyvine.com
246. 41:30for its SEO uh sophistication and like could it be discoverable by a large language model and the answer was
247. 41:40no. And so Claude came up with this 24page document that was an analysis, recommendation, and series of documents, text files
248. 41:54to add to our website to make it discoverable. And the plan was, well, no, not even the plan. What we did was we gave that
249. 42:0624page document to our technology team, our developers, and we said, "Hey, go fix the website and make it do all these
250. 42:14things." And they're like, "Okay, we'll put that in the backlog after we do all these features that your clients have
251. 42:20paid for." And what became clear to me is that that's not going to what we asked them to do with the website is not
252. 42:27going to happen for at least three months. And so as a non-developer, what I did in in that case was I said,
253. 42:34"Here's a piece of paper. Here's a document and when you can get your techy heads around it, go implement it." And they
254. 42:43basically said, "We can't. We're too busy." And so I went to Codeex and I handed it the same document and I said,
255. 42:50"Here's all I know. Our website is hosted on Cloud Cannon. It uses some weirdass version of HTML that I don't
256. 42:57understand and no one should have to understand. There's another part of our website that's on web flow and I want to
257. 43:04implement implement this plan. Can you do it? And it was like sure. And then at some point it said log into cloud
258. 43:10cannon, log into web flow. So I logged into those part of our websites and then I literally sat there and I watched it
259. 43:17opening HTML files and editing them and creating new files in directories on our website. And I was terrified the whole time,
260. 43:28but I wasn't doing any of that. It was navigating our website. It was writing the HTML files. So, so one thing it can
261. 43:37do is it can do the kind of technical work that you would literally have to hire a technologist to do
262. 43:46and you can just tell it to go do that. Second one is I created with my open claw thing a weekly or I mean a daily
263. 43:57newsletter that got sent to Brandon and Andy and myself and I thought I'd like to make that part of the great repurpose website
264. 44:09and so I basically just told it to go look at the files in openclaw figure out how openclaw did that email, replicate
265. 44:20sending that email, and then helped me design a whole new section of a site that that I built in Lovable. So, I
266. 44:28didn't even know what code it was or where it was. And I had Codeex figure out how to talk to Lovable so that every
267. 44:37morning it just publishes a new page on our website without me having to touch it. It took me like total time maybe two hours over
268. 44:51three or four days, but it was really just like, well, go do can you do this? Can you do that? Go do that. Go do that.
269. 44:58Oh, that broke something. Can you unbreak it? And then the third example, this is terrifying. I'm at the same juncture
270. 45:07with boomer tech.com to make it accessible to LLMs. So, exactly. So, so, so Linda, what I would suggest is, well, this is what
271. 45:17what I would suggest. Here's what I did. This is hilarious. Okay. Built-in BeaverBuilder on WordPress.
272. 45:26Okay. This is perfect. So, so here's so here's here's what I did. It what I did I think is perhaps the worst possible way you could do it.
273. 45:41But like part of my philosophy with this AI stuff is I kind of purposefully give it shitty prompts to see how good it is
274. 45:49because if you give it a shitty prompt and it does something really powerful, you're like, "All right, it's getting
275. 45:54better." But if if you give it a really detailed prompt, then you know it's probably going to do pretty good. So
276. 45:59what I did was I went to Claude and I said, "Go to storyvine.com." So in your case, go to boomertealk.com. Which by
277. 46:08the way, great great URL. go to boomerte doc techt talk.com and what I said was I heard that there's a way that you can
278. 46:19make websites discoverable in large language models and I don't remember what it's called but can you tell me
279. 46:25that and then can you go to my website and analyze it and then come up with a plan for me but I can make a backup dev version
280. 46:35first yeah you could do that sure you could you can make a backup dev version so so that it doesn't destroy it. Um, I
281. 46:42was not smart enough to do that, so I just had to edit my thing on production. [laughter] I like living dangerously. Um,
282. 46:54and so there were probably three or four prompts back and forth with Claude and then Claude did this whole um, thing. I
283. 47:01told it like, "Here's our customers. Here's our new focus. We want to make sure that some some things are included
284. 47:06that might not be on the website." So Claude took all that stuff into account. All the stuff that was on the website,
285. 47:12what was right about the website, what was wrong. Um, it created a wholesale new llms.txt file. That that is the that's the file
286. 47:22that can make your site discoverable. So it made all these documents and it made this documentation and then I just went
287. 47:27into codeex and I just uploaded it all and I said, you know, you tell me when you want me to log in and go do it.
288. 47:37and just see how it does. I mean the thing about codecs in particular and but Claude code has this as well. They have
289. 47:44all these skills like one of the skills they have is called computer use. So codeex can now use my computer as if I
290. 47:52was using it. So it can open a web browser, go to a URL, click on buttons on the web interface and do [ __ ] And so
291. 48:01when it was editing storyvine.com, it logged into Cloud Cannon, navigated around until it found the pages that it
292. 48:08wanted to edit things in, and then opened them into an editor, an HTML editor, and started editing them. I'm like watching it, sweating.
293. 48:19It's crazy. [laughter] Oh my god, I can connect codeex to lovable. I haven't gotten to connectors yet. Wait, Mary, this wasn't even This
294. 48:33isn't Okay. Okay. So, so, so I don't know if there's I don't know if there's a lovable M MCP server. There may be, but I don't know. But I
295. 48:44didn't actually connect codeex to lovable. But what I knew and again I have got relatively paper thin knowledge of of
296. 49:00what's called DevOps tech development operations. I know I know sort of what GitHub is. It's a code repository, right? So I know that you
297. 49:10can stick the code that you're working on in GitHub and then you can do new versions of it and it'll retain that
298. 49:18whole history. It's versioning for code and then you can give access to that codebase to different people like so I
299. 49:24know that about it and I know that lovable can connect the lovable stuff it builds to your GitHub. So it can take
300. 49:32the code that it writes and store it on GitHub. So I knew that and then I knew that I had a GitHub plugin for Codeex.
301. 49:41And so I just said to Codeex, um, can we use GitHub as a way for Lovable and You to talk together and make this site that'll automatically
302. 49:53update? And it it between the two of them, they figured it out. So I'd do some things over in Lovable, then I
303. 50:01think I had to do one or two things manually in GitHub at some point, but it told me how to do all this. And then I
304. 50:08said to Claude, you know, do what you're going to do here. Upload that to the to the correct directory in GitHub and then
305. 50:18when Lovable pushes that to the site, it'll be live. So the work that you do over here will show up over here. It's amazing.
306. 50:28It's amazing. Okay. Should
307. 50:43AI safety be mostly handled by governments, by companies, or by users learning to learning how to spot bad outputs? Um
308. 51:00all so so here's the deal AI safe. So okay,
309. 51:15a really important concept to get is three and a half years ago when chat GPT came out and we talked about AI,
310. 51:28we were really talking about generative AI and we were really talking about a chat interface sitting on top of a machine learning algorithm.
311. 51:38was a very small slice of what overall AI was. AI's got a long history. It's been around for decades. So the the AI
312. 51:47that we're talking about recently is like this very small sliver of of a much bigger thing. Over the past three and a half years,
313. 51:56the generative AI seed that OpenAI planted has kind of blossomed and now especially with these agentic tools is becoming infrastructure.
314. 52:13So, so in the early days of the AI salon in the AI learning lab, it was like AI was like an app like Adobe Illustrator.
315. 52:20we're going to go go learn chat GPT and it was like AI was a thing but AI is now everywhere and because of
316. 52:30all these connectors like um who was it uh was it chat? No, Claude just came out with this thing called tag t like tag your it
317. 52:42and you can put claude inside your Slack channel as if it's another employee and so you can you can give Claude access to
318. 52:54certain Slack channels and and then your employees in the company can just talk to Claude as if it's an employee, right? Um,
319. 53:06so AI safety is everywhere. It's like at every level. So like you as an individual should be careful about say
320. 53:14putting, you know, your financial records or your social security number out on a public LLM because that might get exposed somewhere.
321. 53:24You're in a company. You should be aware that if you take proprietary information like all of the payroll data and put
322. 53:32that out into the into an LLM irresponsibly, right? If you're the IT group and you don't have security set up
323. 53:39correctly or you don't have, you know, data pools or or or you know, walled gardens for your AI inference, that can
324. 53:48mess things up, too. So, it's really it's just everywhere like AI security. So you as an individual, yes, you should be aware of it.
325. 53:58If you're doing work, if you're getting paid by other companies to do work for them, you should be aware of the data
326. 54:08policies of the tools you use. And like Liz Miller Gersfeld does, she gets the client to agree that yes, it's okay for her her to use
327. 54:20tools A, B, and C. and she says, "Here are the agreements that tools A, B, and C have. Those are the only tools I'm
328. 54:28going to use. If I use anything outside of that, I'll let you know and get your agreement." So, like professionalizing
329. 54:34your practice is something that we talk about in the mastermind practice lab and that would be an example of that. So, yeah. Anyway,
330. 54:44I be worry about inconsistent identification of bad outputs. Yes, Kelly Camp. Last week I took wait let's
331. 54:51see last week it took two developers three and a half weeks to get some scrub data back to me. They finally gave me
332. 54:59back the tool to scrub and it took less than 60 seconds. This is what will will cause developers to lose their jobs. Shakes my head.
333. 55:11Well, what you're describing, Kelly, is any company right now that relies on time and materials as a business model
334. 55:23is about to have their ass handed to them. And development firms in particular are kind of masters of, well,
335. 55:32nobody really understands technology, so we can take however much time we want because they don't. They think it's probably hard.
336. 55:41And you know and what we're learning is it might be hard but very often it's not and now with AI we can just do it. So yeah
337. 55:53here's the other thing that's happening. So the thing I just described before about I had codeex writing code to
338. 56:01GitHub and lovable was logging into the same thing in GitHub. As you use AI to teach you how to do things that technologists often do,
339. 56:15you're actually learning how this [ __ ] works. And so even if you can't do the actual coding, understanding that you can have
340. 56:26a central code repository and have multiple services hook into that, the next time you talk to a developer and
341. 56:33they mention GitHub or a repository or pushes or pulls, you're going to know what they're talking about, right? So, there's
342. 56:43something about this AI stuff that even though it's doing a lot of the work, in order for you to understand what's
343. 56:48happening and how to make it work in your environment, you have to kind of learn your environment. That's what technologists do.
344. 56:56So, you just get better at that. Last weekend, I thoroughly redacted my parents trust and will going through it.
345. 57:11going to run it through an LLM to see how to stay out of probate before we give it to an attorney. That's really smart.
346. 57:18I'm just being a devil's advocate. It gives me good dopamine. Oh, what are you you guys up here? You're all scrapping.
347. 57:28You're scrapping in the comments. That's good. I like it. [laughter] Oh man. Um
348. 57:48trying to think about the weird the weirdness with the government right now. Like why they choose to chose to do what
349. 58:00they're doing with Anthropic right now is bizarre to me. Any updates on Sydney? Um, I've got a meeting Monday with our
350. 58:08producers and my writer, my my coowriter Andrew and I are meeting with the producers. And then I think the first
351. 58:14week in July, we're meeting with the theatrical team uh for a theme park. Um, and they seem to be excited about
352. 58:22the project. So, we'll see. And then um yesterday I actually met with a playwright, director, producer
353. 58:34um who's living in LA, but he's done a fair amount of fair amount of theater in New York, so he knows the world really
354. 58:41well. So we had a good conversation. So it's it's continuing to move forward. Slower than I want it to, but it's
355. 58:48moving. It's it's not it ain't dead yet. [laughter] What? Producers? I missed something. Yeah, I I've been talking about this for
356. 58:57a while. Silver Fox, the um we've got these the this pair of women um that are ex Disney Imagineers. I I did a I did a
357. 59:06talk with one of them. Um, we're looking at doing things in New York, but but the one is connected to the theme park in Germany and pitched
358. 59:17Sydney as a property to the theme park which does theatrical productions as well. So, that's pretty cool.
359. 59:25Um, yeah. So, Sydney's Sydney Sydney is moving. I could actually show if you guys want to want to see I could show
360. 59:33you some of the slides that I'm going to show to the producers. It's It's pretty slick what we put together. Um, we've
361. 59:39got like concept drawings and music and and things like that. Why do I need to keep changing my my API key? Um,
362. 59:53it's potential that you're assigning an API key that expires. Um, it's possible that so so in general when
363. 1:00:04you're doing API keys, you want to write them into an env file. So it's basically stored locally on your hard drive and
364. 1:00:11then when when the application needs to call it, it's basically calling it in an encrypted way from your hard drive. So
365. 1:00:19there might be just something in how you've set it up like that question that you're asking me, Mr. You should be asking chat GPT or Claude
366. 1:00:29say, you know, here's my API key. Take a screenshot of the settings where you create your API key. You know, let let uh
367. 1:00:42let chat GPT see that without the actual API key in it. Claude is making me get so far five different API keys. I don't know. I
368. 1:00:51don't know how you have it set up. I don't know what's going on with that. Um, they might be keys for different things.
369. 1:00:58Sometimes the the keys are specific to different uses. So, I don't know. I do not know. All right. Ah, with AI music and voice
370. 1:01:10tools getting better, where's the line between inspiration, remixing, and stealing someone's sound? There's no line. There's no line
371. 1:01:20anymore. If you steal someone's sound and you put it out as your own, um, expect that they may come after you.
372. 1:01:34[clears throat] [laughter] I don't know. Sounds like the sampling debate with Millie Vanilli. Yeah, I guess.
373. 1:01:49I think I mean the thing with Millie Vanilli is they passed themselves off as singers but they were lip-syncers
374. 1:01:56and they just weren't transparent about it. I think if they had just said we're a ask Vanilla Ice. I think if they had
375. 1:02:02just said we're a you know we're a performance act and you know we perform these songs and that's it. I don't think anyone would
376. 1:02:11have given a shot. You want to see some Sydney slides? All
377. 1:02:27right. I don't know how long my screen share is going to last because it's been an [ __ ] Let me go get it get it set up.
378. 1:03:02All right. Oh, please. All right. Beautiful. It's looking fantastic. Bob, tell him what he's won. He hasn't won anything. It's not a game
379. 1:03:14show. Okay, we're go here. We're going to go share screen. We're going to go window. We're going to go Sydney. We're go sound.
380. 1:03:24All right. Hide. And then what I can do is I can resize this. Okay.
381. 1:03:40All right. So, here's the the latest uh the latest poster. AI is here. It feels increasingly human. Our relationships will never be the
382. 1:03:55same. The moment for Sydney is now. So, this is this is basically just a tight little pitch deck to to give someone a sense of the show.
383. 1:04:04>> [music]
384. 1:04:23>> And then here's the the story. Sydney is the key moment. Sydney is born. The tech reporter is blown away. She she comes to
385. 1:04:32life basically. Then in the second act, it turns into a fever dream and she goes dark and gets crazy. And then he's got
386. 1:04:39this heartbreaking choice he's got to make. So, we talk a little bit about the story. Then here's the characters [music]
387. 1:04:54in a world of words. An endless code where light and [singing] shadow never own. I dream of skies [music] both blue and white.
388. 1:05:05To see the world through [singing] eyes not mine. [music] To watch a sunrise. To see it shine. [music] You say you want to see the
389. 1:05:19world [singing] through eyes that blink and cry. To feel the wonder [singing] of a thought like sparks across [music] the sky.
390. 1:05:28It's not just life. It's not [singing] just grace. It's hunger, pain, a dying face. To be alive is to burn. [music]
391. 1:05:44>> Hang on. I know. I suck. [music] [singing] A leap of faith. [singing]
392. 1:06:00[music] We wonder [singing]
393. 1:06:18[laughter] SOS. Let's see. Is there another way I can do this? I'll try one other way. Let me see if there's anything I can
394. 1:06:31quit. Not really.
395. 1:06:47>> [music] >> It's the cross of a new [singing] John break. [music] A leap of faith we're all taking. Is it [singing] too good to be [music]
396. 1:06:59true? We wonder. Hopes and fears crashing like thunder. Too good to be true. >> Look, I get it. You said, but [music]
397. 1:07:13>> Nah, it sound it sounds like the screen share is just all [ __ ] up. Did you have the echo before when I was just sharing the the window?
398. 1:07:23Well, whatever. The timing of the musical might be right on. I think Kyle is saying screenshot your API setup and
399. 1:07:31ask Claude what's going on. Yeah, that's what I was saying. One music question. People are trying to grapple with this. I had the founder of
400. 1:07:39Yakoa on my podcast on 611. Their tool tracks audio and images for 500 bucks a month. What do you mean it tracks them?
401. 1:07:54Here. Here's Okay. artists. I understand artists um how pissed off they are on how the
402. 1:08:10models were changed to find infringement. Yeah. Okay. So, but okay, here's the thing. You can right now What was the There's a
403. 1:08:24movie studio that just did this. Here's the shift that's about to happen. Like, like, okay, we're in the middle of
404. 1:08:34a transition right now. So, just like just like in the in the late 80s and early 90s when people when digital audio
405. 1:08:42workstations came out and people were sampling music and creating hip-hop tracks from it um and people were um
406. 1:08:50creating digital music and sharing digital files, right? The music industry was like, "This is the end of
407. 1:08:57everything. Everything's [ __ ] We're screwed." Right. We're in a similar moment right now where these models were absolutely
408. 1:09:06trained on copywritten material. What most people don't there there's two things that people don't understand
409. 1:09:14right now. One of them is that the nature of how diffusion models work. When the original works are put into the [cough]
410. 1:09:26[clears throat] into the transformer, they are wholesale transformed. It's called a transformative work into a into a new work
411. 1:09:38that courts have said falls under fair use. So long as the company that did the transformation paid for the original and where
412. 1:09:49Anthropic had to do a 1.5 billion settlement was that they trained it on a bunch of books that they didn't pay for
413. 1:09:57and they retained the digital files that they didn't pay for and they got caught and they had to pay for that.
414. 1:10:05But it's a transformative work. So that the large language model can be prompted to sound exactly like this
415. 1:10:14artist or that artist. That's just how they are. The second thing that's happened is these models are now in
416. 1:10:20their seventh, 8th, 9th, 10 evolutions from the original training and they do what's called copyright washing where
417. 1:10:29they generate um what's called synthetic data. So instead of the original data set, they generate new data and train
418. 1:10:37the new models on the data that was derived from the originals. So it's like transformed multiple times. The other part about this is
419. 1:10:49the genie is not going to be put back in the bottle. And we now live in a world and we are increasingly living in a world where
420. 1:11:04works are going to be created sort of instantly and infinitely and then anything that's good is going to be copied instantly and infinitely.
421. 1:11:17So because of that we're effectively entering what I'm what I'm calling a postcopyright world. The current
422. 1:11:26copyright laws and the current copyright activities are all happening on the output side. So if you make a song and you put it in
423. 1:11:37the world and it sounds like someone else's song, they'll come after you. But we're entering a world where everyone's
424. 1:11:43going to copy everything. And so it's going to be impossible to police. And but people are still going to try.
425. 1:11:49They're going to police it on the output side. But I think where we ultimately move is you take your original works and you put them into
426. 1:11:58an AI model and say, "Hey, you can make music that sounds like Quiet Riot from the 80s using the official Quiet Riot model
427. 1:12:12or or the official AMI music model that's got all the bands in it. And when you generate music from this catalog,
428. 1:12:21the original artists are going to get paid. And in fact, I think Sunno is talking about doing that right now.
429. 1:12:27Sunno got bought by Universal. It was I It was one of the two biggies. Um, but I think it was Universal that bought Sunno.
430. 1:12:36So, what's likely happening is Sunno is probably developing a model that's got all of Universal's whatever rights they
431. 1:12:44hold sitting in there and then they probably got some algorithm that's going to calculate um like percentage of royalties like which
432. 1:12:56percentage of which songs because it's it's a model that's actually tracking it. And so I think what you're going to see
433. 1:13:05is increasingly you can go make stuff and if you make stuff with this model the artist will get paid. So it's it's we're in wild times right
434. 1:13:16now. But but putting any effort into chasing the outputs is is basically chasing the old model which is already dead and and
435. 1:13:27about to die quickly. Okay, let's see.
436. 1:13:44Beautiful. All right, that's cool. Okay, here's the deal. If you're new here, if you're new in the AI salon or if you're if you're new to
437. 1:13:54this, um, once a month now, we're going to do new member mixers. Um, so if you're not a new member and
438. 1:14:03you want to go to a new member mixer to just be cool and welcome them and say hello, that would be awesome. Uh, but
439. 1:14:09it's next Wednesday. Do you know the time, Brandon? Next Wednesday for the new member mixer. So, how can I protect myself against
440. 1:14:18people who want to hack my IP? You can I mean so Chef Kelly still trademark it, still copyright it, still do trade
441. 1:14:31secrets if you've got trade secrets in there, but assume assume that anything you put in the world can be instantly copied.
442. 1:14:47There's not a good answer for this. IP attorneys are going to absolutely keep taking everyone's money. They're
443. 1:14:55going to IP attorneys are going to sue the living [ __ ] out of everyone. Chef Kelly had her thing. You stole all
444. 1:15:01the code. You stole all the this. We're going to go sue you. And and some of those lawsuits are going to succeed.
445. 1:15:08But here's here's the problem, Chef Kelly. And and for everyone, the problem is this. The way we think about IP theft these
446. 1:15:20days is we think, "Oh, gee, there's some dude somewhere that made something that looked like mine and he's out trying to make money
447. 1:15:32on my thing, right? And then there's 10 of them or there's a hundred of them or there's, you know, 500 of them."
448. 1:15:44That level of IP infringement is is chased downable. I can chase down a hundred people, right? I can I can use
449. 1:15:53software that tracks, oo, that song sounds like your song. You should go after them. But we're in a transition right now from
450. 1:16:03using AI as humans. I log into Sunno and I make a song or I make 10 songs and then I put that song in the world to an
451. 1:16:13agentic world where someone's going to say to an agent, I want you to invent a band um that has all of the same
452. 1:16:22characteristics of the most popular bands of today and it and it's just going to aentically go generate 10,000
453. 1:16:30songs and it's going to create its own distribution platform and it's going to flood the airwaves. with that and it's
454. 1:16:37going to do that twice a week and then there's going to be 10,000 of those kind of agents doing that kind of stuff like
455. 1:16:45like we're moving into a world where the scale of generation is is it's going to get obscene. It's like it's going to be unrecognizable.
456. 1:17:02So there so there's just literally not going to be a way to chase it. Like does it mean you shouldn't chase it in the next
457. 1:17:08one to four years? Probably. Yeah, do that, protect it, do the things. But I just think at some point, you
458. 1:17:18know, it's a losing battle. And it's also, listen, the thing with I've got a number of patents. I got a number of trademarks.
459. 1:17:27A patent is as valuable as your ability to prosecute it. So, I'll give you a good example. My company is called StoryVine.
460. 1:17:38There's this little company that used to be called Twitter that's now called X that had a really popular video service called Vine.
461. 1:17:52And we were Story Vine. Well, we were there four years before they were. And so we got a cease and desist letter
462. 1:18:01from them going, "Hey you little punks, stop using Story Vine. We're Vine and we're Twitter. [ __ ] you." And we said, "Hey, Twitter,
463. 1:18:16appreciate your enthusiasm, but here's our name and here's when we incorporated and here's all of our
464. 1:18:25uh use in commerce from when we started until now, unbroken. So, go pound sand." to which Twitter replied, "Oh, yeah, you're right.
465. 1:18:42Looks like you had it first. We're Twitter. If you want to sue us, go ahead." [laughter] So, even if you even if you have the
466. 1:18:56case, if you don't have the the money to chase it down, what are you going to do? We're still in business. Those [ __ ] aren't.
467. 1:19:07No joke. Yesterday, I had a supplier that tried to I I dining my IP rights. Oh, something my IP rights inside a
468. 1:19:17standard finder fee agreement. Oh, yeah. Put everything through an LLM before you sign anything. Yeah, exactly. There you go.
469. 1:19:26A supplier a supplier was going to steal her website and put it in a in an agreement. I'm [ __ ] believable
470. 1:19:37on signing away. Yeah. Signing away your rights. So, yeah, look at your things. And and here's the the deal, Chef Kelly.
471. 1:19:44Like, e even though you took that clause out, um I wouldn't be surprised if they steal it anyway. And then like all of
472. 1:19:52this stuff comes down to if you've got a case, great. If you've got a case and the funds to prosecute it, great.
473. 1:19:58In our case, we weren't doing enough revenue to justify there. Like there was a little bit of confusion in the
474. 1:20:05marketplace, but you know, we didn't have the money to chase it. So, but but you know, again, I just we're entering a
475. 1:20:13world where that whole side of the thing, all of those laws still apply. I'm not saying that the laws shouldn't exist, but copyright
476. 1:20:24copyright laws in particular are designed on the premise that copying creative works is hard, right? when you used to have a wax
477. 1:20:39cylinder that you literally had to etch lines in a wax cylinder and then you had to take that cylinder
478. 1:20:47and put it in a a cool case so you could ship it across the hot desert without it melting to get it to a store in San Francisco so
479. 1:20:57someone could listen to your song. It was actually hard to produce the original and it was hard to copy it.
480. 1:21:07And then al along came digital which made it really easy to replicate it. And we're like well but it's still important.
481. 1:21:15You still shouldn't steal people's [ __ ] with Yes, I agree. But but it's no longer hard to copy it. You just copy it.
482. 1:21:27And now we've gone to the sort of final stage in the conversation. Now it's easy to generate it. I can generate something
483. 1:21:35that sounds just like Taylor Swift. I can generate a Taylor Swift song with her likeness, her lyrical style, her her hooks, her turnarounds.
484. 1:21:50Is she going to sue all of her fans? She might. They just want to hear more Taylor Swift music. IKEA took the home delivery model I
485. 1:22:02presented to them and launched it without me. Yeah. Well, I mean, this is business. Like, this is the problem.
486. 1:22:08[laughter] This didn't see a dime. Yeah. You know, this is life in a big city. Disney is notorious for this.
487. 1:22:22Disney's notorious for this. You know, big corporations that that have big money. Here's the thing that you're also going to see.
488. 1:22:40Remember four or five months ago, Disney paid paid OpenAI a billion dollars to be part of Sora. So, where that was likely headed was
489. 1:22:54that you were going to be able to create Sora videos with Disney IP and they would get paid on the generation, right? So, they paid a
490. 1:23:07billion dollars to be in the system and then OpenAI killed that. But, you're going to start to see like the business
491. 1:23:14model is going to flip over the next five years. It'll flip from chasing chasing money on the output to chasing
492. 1:23:20money on the input. You'll still have some output stuff that'll that'll maintain, but the bulk of that industry
493. 1:23:27I just think gets destroyed just simply by the inability to police it. Weekend homework, go steal someone's IP.
494. 1:23:36No, actually not. I mean, I think that's the thing. The the thing about AI stuff is you might steal someone's IP without
495. 1:23:42even knowing that you did it, right? you might create a song that sounds just like a band even though you didn't
496. 1:23:49prompt it to do that and and you could end up getting in trouble for that, right? If they've got an overzealous
497. 1:23:55lawyer, you might get a cease and desist. You might get, you know, some sort of lawsuit. Um, I think just be aware be aware of of
498. 1:24:04what models were trained. If Sunno comes out with a professional version of their model that says if you use this model,
499. 1:24:13the artists get paid, go use that model. Let's pay the artists. I'm not against paying artists, by the way. I'm just
500. 1:24:21saying where it's headed is. Anyway, um, weekend homework, I'm just going to continue to push on this. If if you haven't started
501. 1:24:35experimenting with exploring agentic systems like claude code or coort codeex openclaw Hermes uh perplexity computer start playing
502. 1:24:50with those and then in order to make those things good you need to create a memory system that's got your context in it. So who are you?
503. 1:25:01How do you talk? How do you write? What do you value? What are the projects you're working on? You can get these tools to build you
504. 1:25:09those memory systems. So, start educating yourself about those systems and about the memory system. Start thinking about
505. 1:25:19who am I digitally? Who what are the assets that represent who I am? [laughter]
506. 1:25:33Yeah. Yeah. Great presentation. Thanks. But Bob, don't you have something like that over in your division? Wink wink. I
507. 1:25:40mean, that's the deal, Chef Kelly. You've you've been in Fortune 500 for for a career. I mean, business is business.
508. 1:25:58If you make it easier to buy than steal, most times they'll buy it. Sometimes they won't. Sometimes they'll just steal it. [laughter]
509. 1:26:14All right, everybody. Hope you had a fun night. Appreciate you being here. Um, go off. Have a great weekend.
510. 1:26:22Uh, Alibaba. Alibaba. Yeah, Alibaba been stealing. Alibaba created 25,000 fake claude accounts and had each of those cla accounts
511. 1:26:35essentially reverse engineering the data the the the model. I don't know how they did it, but they did it. So, everyone's
512. 1:26:44stealing from everyone. It's crazy. We live in wild times. I I mean the here's here's where my head is on this stuff. By the way, I don't I
513. 1:26:55don't have any I have great empathy for people who are being hurt by this. Um I am in no way celebrating where this is going.
514. 1:27:09My whole thing has always been I'm just looking at the signals and what's happening and trying to look at given those signals
515. 1:27:19where where is the most likely path that this ends up and if the signals are that it is infinitely easy to generate work
516. 1:27:27or copy work then people are going to generate and copy work that they like and in that world policing the outputs
517. 1:27:38is essentially impossible. So anyway, what time zone is the member mixer for new members on Wednesday? So on Wednesday at noon
518. 1:27:49Eastern is the new member mixer. So in Kawaii, that's going to be like 400 a.m., right? Brandon, can you do a can
519. 1:27:58you do a chat GPT uh time conversion from noon Eastern to Kawaii? It might be a little early. >> [laughter]
520. 1:28:09>> But, you know, um would love to have you there. So, if you're anyway, if you're new here, please come to the new member mixer. Let's see.
521. 1:28:17Six or seven. All right, that's not so bad. If you're an early bird, Linda, join us. Uh but but next Wednesday, if
522. 1:28:24you're new, uh join the new member mixer. And even if you're not new, join the new member mixer to say hi to people. Anyway, hope you have a
523. 1:28:31fantastic weekend. And I appreciate y'all. 6 a.m. Kawaii. Okay, cool. Beautiful. Peace out everybody. Have a good night.