6/15/2026 - Why the Claude Fable Shutdown Signals a Major Shift in Geopolitical AI Control

Transcript

1. 0:05[music] [music] Heat. Heat. [music]
2. 0:26>> [music] >> Uh oh. What did I just do? There we go. Hello. Hello. Hello. Welcome to the AI Learning Lab. It is Monday night. It is
3. 0:43June 15th in the year of 2026. Um it's Monday night. We still have no Fable or Blair's in the house. Yo, what's
4. 0:56happening? Sharon Crawford came for the loop talk. Staying for the entropic drama. [laughter] Oh, Dario, Dario, Dario. Isn't it amazing that
5. 1:10in December of 2025, Anthropic launches Opus 4.6, six. The tech community falls in love with it. People start fleeing Open AI
6. 1:25because they hadn't really released anything notable in a while. Then Open Claw comes out. People are still excited about Anthropic.
7. 1:38Then Anthropic launches Opus 4.7 and it's a dog. and then quickly launched Opus 4.8 and it's still kind of a dog.
8. 1:49Careful what you wish wish for, Nario. Daario, we are back indeed. And we're back. Thank Kelly Camp. So, welcome everybody.
9. 2:00Welcome to the AI learning lab. If you want to chat, if you want to see the chat and be able to chat because
10. 2:06chatting is where it's at, you go right there and that's the link to pop out the chat. If you want to be able to like
11. 2:15surf the AI salon while you're watching this, you can do that, too. But you have to have two tabs. And in the lower
12. 2:22corner right here, you click on that and it'll pop this out as a window. And then in a separate tab, you can surf the
13. 2:29salon. So, you can have the live stream here. That's on desktop. If you're on mobile, you can kind of either chat or
14. 2:36watch. It's not great. Um, we we are giving feedback to Streamyard and uh and Mighty Networks about making that
15. 2:45experience suck less, but we'll see. We'll see how they do. It takes me three sessions to get Claude to refresh a
16. 2:52project these days. 2024 me is impressed. 2026 me is impatient. Yeah. Yeah. So, if you don't know, um, on Friday night,
17. 3:08I'm like getting ready to ask questions and answer questions and and Dr. Jay, Claire Jacobs, was like, "Uh, Kyle,
18. 3:20is uh is it true that Fable's gonna go away?" >> [laughter] >> And I was like, "Claire, have you been smoking the wacky weed again?"
19. 3:32And uh turns out Claire was uh correct. [laughter] So um I wish I had a clear answer for you. There's been some investigative
20. 3:44reporting, but it all seems to be contradictory. I'll tell you the the I'll tell you the sort of sequence in the order that I heard them was
21. 3:58well based on how the so what happened was the fable model launches which is the the safe version of the mythos model
22. 4:08which which they announced two months before and they said it's really dangerous and we're going to give it to our special friends
23. 4:16to to harden some of the systems. And then they launched Fable and and you know, it comes out to kind of mixed
24. 4:23reviews. Some people love it, some people hate it, but it's clearly more powerful. Like I did I did a couple of
25. 4:28things with it. It was it's it's just clearly on a different level for whatever reason. If they're if they're
26. 4:34just putting more thinking behind it or whatever it is, it's some amount 10 trillion parameters, whatever it is, big
27. 4:41big ass model. So it's going well. So they they slap it with export controls at 5:20 on Friday night, last Friday
28. 4:51night. And the export controls basically say that any foreign national, anyone who's not an American citizen, um can't use it, including
29. 5:05the foreign nationals that work at Anthropic that made it. So, so there's no way for there's no way for Anthropic to know if someone using
30. 5:18it is a US citizen or not because that's just you can't do that. So, they had to shut the thing down. So, the government
31. 5:24knew they were going to do that. Um, the first little tidbit of news I heard was that it had been jailbroken. And that was
32. 5:39actually kind of reported in the days before. people were like that someone had jailbroken it. But every time one of
33. 5:45these models comes out, the hacker community immediately goes in and tries to jailbreak it. Like jailbreaking one
34. 5:52of these models is not a new thing. Like every time they come out, you see posts about so and so jailbroke it and what
35. 5:59and whatever. They're off doing their thing. It's the that's the hacker community. That's what they're going to
36. 6:03do. So that so none of that seemed important. But then, you know, they get hit with this export thing. So then I heard
37. 6:19that Amazon, who is one of Anthropic's biggest investors, that their their um secure their cyber team are actually the ones that
38. 6:32jailbroke it. and Amazon reported anthropic to the government which doesn't track for me. Maybe that happened. I don't know.
39. 6:42That's what was sort of reported. Yeah, this the system prompt has been leaked as well which which they always
40. 6:50say. Here's the thing about all those things about [laughter] we now live in a world of AI. So news, you can't you couldn't trust news
41. 7:02before. You really can't trust it now. Um so so that was that was the first thing. Then then I heard no what actually happened was
42. 7:18there was this jailbreak whether it was from reported by Amazon or not but there was some sort of jailbreak and I guess
43. 7:27Anthropic had been in touch with the White House over something and they said hey patch patch that thing fix it so
44. 7:35it's not jailbroken and Daario told him to go pound sand and that and that the shutting it down was this vindictive
45. 7:44move which given the administration that doesn't surprise me. That's first thing I said on Friday was you know this
46. 7:52doesn't smell like a security problem this smells like a political problem. Um so that was sort of that and then you
47. 8:02know along with that everyone's calling for Daario. Dario's got to go. You know, it's bad. He's the worst CEO ever. How
48. 8:11could he blow this lead? Meanwhile, Sam Alman and um Elon Musk. Not a peep. Not a peep. Their competitor is driving
49. 8:23his fancy truck into a brick wall over and over and over again. Just let him let him keep doing it. So then today, I
50. 8:30heard that no, no, no, it wasn't that. It was actually that this thing got leaked to a Chinese, you know, to to
51. 8:39Chinese operatives in their equivalent of the CIA or some [ __ ] like that. So, I have no [ __ ] clue. I don't
52. 8:47know. Like, it's definitely political. Um, it is definitely
53. 9:02this absolutely shifts AI from here's a technology that we should all learn because it's going to change things to a geopolitical
54. 9:19um issue. Now, I think it always was a geopolitical issue, but but I think I think anthropic, however they whether they wanted it or
55. 9:30not, you know, there there's also speculation about this is exactly what Daario wanted. This isn't what Daario wanted. Um,
56. 9:39but this is clearly now like we're now in the territory where these these powerful AI models are, you know, are a geopolitical issue. Um,
57. 9:56and so what's that mean for us? I don't know. I don't know. um to a great degree. I don't think it matters for most people right now.
58. 10:10Um there's a guy on on X named uh I think his name's Marco. He's a he's a designer dude and he's been he was using
59. 10:20Fable to make SNES games. He had he had games that he had that he had written and he converted them into Super
60. 10:27Nintendo games and he'd done like two or three of them [snorts] and he was on his third one or his fourth one and Fable went away and
61. 10:36he just kept working and he and he said, you know, if you've got the right prompts and if you're using these
62. 10:43agentic tools, you don't need something the level of Fable. Um Fable was very expensive. Um, I think people who are just messing
63. 10:52around with AI when when Fable went away, they're like, "Oh, well, you know, we can't do anything good now." That's
64. 10:59not the case. The current AI models are fine. They can code well. If you pair them with something like OpenAI's codeex
65. 11:07or you pair them with something like Claude Code, Claude Co-work, um, you know, they're good. They're good.
66. 11:16They're fine. Same with Gemini with anti-gravity, things like that. Buffering for me. Um I'd love to be on
67. 11:24the fly on the wall at Anthropic for that con conversation. This is why we come to you for the news. Well, so to be
68. 11:31clear, I like I don't know. I am I am I am chat TMZ. I'm totally chat TMZ without the investigative reporting. [laughter]
69. 11:42It's me just looking at the looking at the noise. I I just look at the noise and I'm like, where where is the noise
70. 11:49centering? And right now it's just noise. Right now it's just noise. Um we will eventually find out what happened
71. 11:56here. Um I would not be at all surprised if Daario gets ousted. Um here's the reason why. Okay, SpaceX just went public.
72. 12:13The majority of their revenue came from XAI, not from their models, but from renting out their compute centers. So they went public.
73. 12:27So Anthropic and OpenAI both have filed S1s. The government shutting down Anthropic's biggest model makes it very very dangerous from an IPO
74. 12:44perspective for both of them, right? Like the risk the risk of [laughter] like them not being able to monetize what they've created goes up
75. 12:56dramatically if the government can just go [ __ ] it, we'll take it. we'll either shut it down or we'll take it from here.
76. 13:04Um, so I think that's probably the reason we haven't heard from Sam Alman and and probably frankly why we haven't
77. 13:10heard from Musk. Like they just went public. So he's not going to talk about AI. He's going to talk about rockets. We're going
78. 13:16to make more rockets. Rockets big. Rockets good. Rockets go up. Rockets land. We catch them on sticks,
79. 13:25right? He he doesn't want to flush his currency that he just made down the toilet. Um so you know the Open AI team can't be
80. 13:39very happy about this and frankly no one should be very happy about this. You know just about the time AI is getting good
81. 13:48you know this happens. Um, yeah. So, I don't know. I don't know.
82. 14:04There's a couple of things that this opens the door for. This opens the door for um Chinese open source models that are
83. 14:14really powerful. So, expect to see those more aggressively dumped um as as they as they, you know, start building them.
84. 14:23Um because they want to sew chaos and the government isn't going to be able to control this for for super long.
85. 14:32They might make it illegal, but remember back in the late 80s, early 90s, they made they made cryptography illegal and
86. 14:40then you could put three lines of pearl on a t-shirt and, you know, export it. [laughter] Um, just so you know, um, tonight for
87. 14:51this week, tonight is it. Um, I'm traveling. I leave tomorrow afternoon going to Connecticut um on Wednesday and
88. 15:00then New York and New Jersey on Thursday and I'll be back on Friday. But it means Wednesday and Friday are both screwed up
89. 15:07for this for this time slot. Um if anything happens that's worth going live for during the week and I've got a
90. 15:16slot, I may just go live. So just pay attention on Tik Tok. I probably won't go live on StreamYard or or in here. I
91. 15:22don't know. I'll figure it out. um if if it's worth if it's worth doing. But right now, I don't know. It's whatever.
92. 15:30Like, they're just playing [ __ ] games. And I don't think we're going to see I don't think we're going to hear a peep from any of Google,
93. 15:40Open AI, XAI. If Meta ends up doing something, I don't think we'll hear anything from them. You might hear
94. 15:47something from Apple because they're sort of outside of the fray of this. Um, but I think until this blows over and we
95. 15:56sort of figure out what's going on here, like what were the motives, um, we'll see. The reason I think that Daario might go
96. 16:05down is that the markets the markets do not want the markets that have invested trillions of dollars into these AI platforms
97. 16:20do not want the ability to take them public to not exist. And so I think there's probably got to be a fall guy for this anthropic
98. 16:31debacle. And it I like it has to fall on Daria. Um Danielle, I would love a Tik Tok live. Maybe I'll do that. At least
99. 16:40you're honest about it. Oh, about about not knowing what the hell I'm doing with the news. Yeah. Um got some new folks. Can you show them
100. 16:49where the chat is again? Yes. So, welcome everybody. Um, this is the AI learning lab live as you know and we're
101. 16:56here in the AI salon. If you want to pop the chat out, if you want to be able to chat like the other people are chatting
102. 17:01and I'm looking at their chats in this upper corner right here, click on that. That'll pop out the chat. Now, there's a
103. 17:08cool thing you can do. You can have one tab open where you can surf the AI salon and go to the different spaces and do AI
104. 17:15salon stuff. You can also then pop out this live stream in a window in a separate tab. So if you create two tabs
105. 17:24and one of them has this in it right down in this corner, right? Yeah. On mobile comments work, picture and
106. 17:37picture doesn't. So down in this corner is a button for picture and picture. So if you're on desktop, click that.
107. 17:42That'll pop this window out as a stream. And then in a separate tab, you can have the thing going. Um,
108. 17:50digital gods, the only ones. Oh, wait. Mary Carol, great job. Wait, [laughter] Nate. Okay, Mary Carol, great job today
109. 17:58in the business group, Kelly. Looking forward to hearing more and super excited. Joe Joe came too. Beautiful. Um,
110. 18:05digital gods. The only one doing any big evil with AI is governments and state sponsored actors. Yeah. Yeah. Exactly.
111. 18:12Exactly. Yeah. Th this is all this is all gamesmanship right now and distraction and what whatever it's like
112. 18:20it feels very politically motivated. Um it one of the things that I talked about year one in this channel
113. 18:35was that what's very clear to me is that AI is going to threaten all of the centers of power. And when centers of power get
114. 18:50threatened, they do weird [ __ ] And so I think the next 10 years, it's just going to be weird [ __ ] after
115. 18:58weird [ __ ] after we weird [ __ ] So the weird [ __ ] is starting. So I don't know. I don't know how this plays out. I think
116. 19:05we probably get some version of Fable back in the next two weeks. It's probably not this week, but it's probably two weeks.
117. 19:16I wouldn't be surprised if Daario goes and then there'll be some sort of announcement that makes it clear that that
118. 19:30it's okay to go public. You can if you're an open AI or if you're one of these companies and you want to take it
119. 19:36public, it's okay. We're not gonna [ __ ] with it. this thing with Daario was a one-off thing. He's been slapped. He's
120. 19:42been fired. You know, some sort of thing like everybody's cool. Everybody calm down. Let's let's get the markets back
121. 19:50up and to the right. Up and to the right. Let's go. So, I think we'll see that. And then the minute you have that kind of signal,
122. 19:59then, you know, we'll get we'll get the the latest from Open AI and the latest from whoever. Um, Lori Blair.
123. 20:13All right. All I do is I tap the video and the box appears in the upper right. Let's chat. Picture and pick. Does work
124. 20:20on mobile iOS, but you have to go to the StreamYard link at the way bottom of the page. Looks like a compass. And then upper right
125. 20:29dotted line. Okay, everybody. Everybody's supporting one another with how to use this in different places.
126. 20:36Okay. Um, beautiful Tom Perez. Knowledge is power. Power corrupts. Absolute power corrupts. Absolutely. Yeah. So, so I
127. 20:45think I think that this is just the beginning of we will start to see weird [ __ ] either from the government or from big
128. 20:54multinational corporations. I wouldn't be surprised at all if we saw something um where big medical breakthroughs start
129. 21:05happening and all of a sudden that becomes too dangerous. Boweapons we this it's danger. We can't be curing diseases. We can't be using
130. 21:16we can't be using this AI stuff to cure the diseases because there's all this risk over here and really the pharma companies should
131. 21:26be the ones you know managing the uh really powerful healthc care AI. I I think you're going to see lots and lots of those kind of
132. 21:41like power power retention strategies. Must not lose power. Must not lose power. Um, again, a lot of this is way above my pay
133. 21:55grade, but it's just like this. It's I think it's fairly common sense, right? If the people that have all the
134. 22:01trillions want to keep their trillions and this AI risks those trillions, [laughter] [ __ ] gonna happen. All right. Okay. Um,
135. 22:15so that's that that's that's Fable. Um, so until that thing plays out, I wouldn't expect [ __ ] from Open AI. Okay. So, what I promised
136. 22:31last week was that I would talk about loops. Um,
137. 22:51so if you haven't actually it's really interesting this anthropic thing. Here's a fascinating thing. So
138. 23:08where this idea of loops is coming from is from these agentic platforms like claude code claude co-work openai codeex open claw Hermes
139. 23:19these agentic platforms openclaw and hermes are open-source versions of these claud code and co-work are from anthropic
140. 23:29and codeex is from open aai I've been up in the air about open AI or or claude code claude co-work all of a sudden [laughter] I'm like hm
141. 23:48my my instinct was codeex so so codeex is the thing I have installed codeex is the thing I've been learning and teaching myself on um slowly
142. 24:01I got a little shell shocked with Adam about how bad Adam was and how bad I was at making Atom. And I thought I'll just
143. 24:09I'll wait for the technology to get a little better. And so I've been slowly learning Codeex. My instinct has been Codeex is the one is
144. 24:18the one to go with, but Claude Code and Claude Co-work everyone talks about and they're the ones that developed MCPs and they're the
145. 24:26ones that develop skills. So Anthropics got some serious engineering going on. And now they [ __ ] up. [laughter]
146. 24:38Now the government has effectively shut them down. So the question then becomes, is anthropic a liability? Well, I mean,
147. 24:48I guess this is what if you're an investor, I guess this is what investors do, right? Did did whatever they did to
148. 24:55to maneuver to get to this position with the government, did that just [ __ ] up their IPO possibilities? Anyway, okay.
149. 25:14Trying to think where to start. So, if you start with something like chat GPT or claude, just regular chat bots.
150. 25:25when when the regular chat bots started they were disconnected from the internet and they had limited um historical knowledge they were only
151. 25:36so recent right and it was usually like the cut off time was like a year or two before present their knowledge was cut off so
152. 25:45like chat GPT3.5 the early version of chat GPT4 you could you could look up anything so long as it you know wasn't newer than a
153. 25:56year ago or two years ago. And then at some point in 2023 when when chat GPT40 came out and then probably a
154. 26:06little bit in 2024 when when 4 came out the the omni model. Um I think it was in 2023 when they connected chat GPT to the
155. 26:18internet. So, not only could you search the model which had a cutoff date, but you could also now search the internet
156. 26:25which extended the cutoff date and this is going somewhere. So then chat GPT
157. 26:43allowed us to make custom GPTs and these were these standalone we could we could create our own system prompt put it in a
158. 26:49custom GPT and you had like your own personalized version of chat GPT and a lot of people in the AI salon built a
159. 26:56lot of those. A lot of people I think Kelly Camp's still making a lot of those. I think Vicky Baptist is still
160. 27:01making a lot of those. Custom GPTs are still very powerful, right? They're these standalone applications that have
161. 27:08all the power of AI, but then they have the sort of specific knowledge that you put in your custom prompt, your your system prompt for them.
162. 27:19Um, I think it was in 2025, Anthropic, it might have even been late 24, but it's probably early 25. Anthropic comes up with this thing
163. 27:30called MCP, model context protocol. In fact, let me see when it came out. >> Um, when did MCP come out from Anthropic?
164. 27:46November 2024. Okay. So, it's been out for a while. So, they came out with model context protocol and model context protocol MCP. Okay.
165. 28:01In computer programming, there's a thing called an API, an application programming interface that if I've got a website like um
166. 28:17Soundcloud, right, and I've got a website and people can come to my website and they can click on songs to
167. 28:23play, right? through my interface, I can create a thing called an API, an app application programming interface
168. 28:33that allows developers to use the services of my website without having to use the front end. So, they could build
169. 28:40their own music player that's effectively hitting the play button on Soundcloud, but it happens on on their
170. 28:46server, right? It's a it's a programming connection between the two. Those APIs were fairly complicated and you had
171. 28:55to be fairly complic you had to be fairly sophisticated to be able to use them and understand you had to read the
172. 29:01API documentation and you had to understand what was possible. So in late 2024, Anthropic invents this
173. 29:10thing called the model context protocol, which basically is kind of like an API where the the developer of a website can
174. 29:19say, "Hey, I'm going to take all these functions of my website and I'm going to make them available to a large language
175. 29:24model. I'm going to describe them and then I'm going to describe if you want to use them, how you use them, and we'll
176. 29:31there we'll have some connectivity." But they were they were much less complicated. They don't require API keys. They don't require as much
177. 29:41programming. You can just basically connect Chat GPT or Claude to one of these MCP servers and then it can do it. I actually connected
178. 29:52ChatGpt today to runway the video producing thing. And now in chat GBT I can say go make me a movie of this and
179. 30:03it'll just you know I type slashrunway and it'll go render stuff on runway and bring it back into chat GBT. So that happened
180. 30:16and so what ended up happening is that chat GBT went from being this kind of standalone application where you would
181. 30:26ask it for something to something where you could ask it for something but then you could ask it also to go use a tool
182. 30:33or you could ask it to go search the internet, right? But it was still very much us being the interface kind of
183. 30:41explicitly telling it go over there, go over there, go do that thing, bring me back the answer and then I as the human
184. 30:48will look at the answer and say is this good or not. So around the time that all this is happening, the engineers within all of the frontier
185. 31:02model companies including I assume Meta and XAI and and Anthropic and OpenAI. They started building their own developer tools. Um
186. 31:14there's a tool called cursor there. There you know clawed code came out relatively quickly. Codeex came out. So
187. 31:21all of these initial tools were designed for software developers to be able to use large language models in their core
188. 31:33development environments. So, so one of the things that MCPs allowed was for things like cloud code to connect with other services
189. 31:43like GitHub and you know all of the you know hosting facilities and code repositories all that sort of stuff
190. 31:55and then they also connected those programming tools to the terminal they call it CLI command line interface.
191. 32:06So on your computer, if you're on a Mac, it looks all pretty, but sitting underneath it is a BSD Linux server. And
192. 32:14how you interface with Linux is through the command line. And so programmers love the command line because they've
193. 32:20got direct direct access to all of the functionality of the machine. They can just call whatever the things are. They
194. 32:27can write files. They can change file names. They can do batch things. They can do automations. they can just do do
195. 32:34it all through this command line. So all of these programming tools had access to the computer they were the the engineer
196. 32:42was on via the command line. And then what happened was some of those geeks were like, "Hey, wait a minute. This access to the
197. 32:54computer and access to all this programming stuff can also be used for things like doing [ __ ] that isn't programming.
198. 33:04And so all of these tools are now starting to evolve to be able to through a visual front end. So Claude co-work if
199. 33:14you download the Claude application on Windows or Mac. You've got Claude Code in there is one tab and then you have
200. 33:22Claude Co-work which is another tab. They're both doing effectively the same thing. They're doing some use of tools.
201. 33:29They're accessing the internet and then they're writing files to your local computer. But the cloud co-work is much more for
202. 33:39like knowledge work. Like if you want to automate a marketing campaign or if you want to, you know, automate some media
203. 33:46buying or some research, you can use the same tools that developers were using to do all that. So the things that make these tools what
204. 33:59they are is they're connected to a large language model just like chat GPT was. They're connected to the internet just
205. 34:06like chat GPT was. Um they're connected uh to tools like using like MCPs being able to use websites. Um, and then the new thing is they're
206. 34:19connected to your local computer and you could sort of connect Chat GPT to your local computer by connecting it to Google Drive through MCP.
207. 34:30So you're starting to have chat GPT is turning more into an agentic platform and these programming tools are turning more into
208. 34:42kind of a consumer version of an agentic programming platform and they're sort of coming together. Where I think this ultimately goes is
209. 34:55that every one of these frontier model companies, the thing that we will interact with is essentially what Elon Musk calls the
210. 35:05super app or the everything app. And this this is again this is something we've talked about in this channel for
211. 35:11years that the everything app is coming and the everything app can just do everything. And so the rumor that what's coming from
212. 35:19OpenAI is that they're going to combine their web browser um chat GBT codeex um and computer use. So it's ability to do
213. 35:33all the local stuff on the computer into a single app. Okay. So,
214. 35:49so what's that mean for all of us? And what's that got to do with loops? What the [ __ ] are loops, Kyle?
215. 35:56Is anybody still with me? [laughter] Commercial first. Okay, we'll get back to loops after the commercial. So, um,
216. 36:08welcome everybody. If you're new here, welcome, welcome, welcome. Um, I'm loving having Claude interact with my
217. 36:14Google Drive, but it makes me run out of tokens all the time. Yeah, we're I'm we're gonna get to that. And we're going
218. 36:20to get to why you're running out of tokens. Thanks for the recap. You're welcome, Herb. Um, here's the deal.
219. 36:28One of the ways that we support the AI Salon and support this channel um is is some of it's through sponsorship, but
220. 36:37some of it's through um revenue. The mastermind is a subscription area of the salon and it's getting incredibly
221. 36:45robust. So, if you're not a part of the mastermind, go check it out. If you scan the code or just go to the uh there's a
222. 36:52mastermind section where you can read about what it what it's all about. So, if you haven't joined yet, I would
223. 36:57consider joining. Um, if you're a part of the mastermind, we've got the AI salon mastermind practice, um, which is
224. 37:06a which is a framework for doing a daily practice for yourself around how you use AI. Um, we've got the great repurpose,
225. 37:15um, which is about the future of work, creating the future of work. Today we just kicked off uh a series on leadership and um language and
226. 37:26agreements and and how to be more powerful in the world. Um all of these skills are going to become increasingly
227. 37:34important as AI gets more and more powerful. What the what the mastermind is really about is about reclaiming your
228. 37:40agency in a world where it feels like everything is out of your control. The whole point of the mastermind is people
229. 37:47intentionally figuring out, you know, who are they, what do they value, who do they care about, and what do they want
230. 37:52to do in the world, and then how do we use AI to do that? So, it's a very thoughtful group. It's a very professional group. So, if you're
231. 37:58thinking about leveling up your AI, um consider joining joining the mastermind. If you missed Dakota Coons's um
232. 38:07leadership series today, um I could only be there for the first half hour of it today because I had a client thing. Um,
233. 38:13but I'm going to go watch the second half of it. If you're in the mastermind, go watch that session. It's a really
234. 38:18good session. Um, and uh, so go watch that session and then join that series. There's just lots and lots of good stuff
235. 38:25going on in in the uh, in the mastermind. Okay, cool. Okay, back to loops. So, so we've now got these two sort of flavors of
236. 38:40what look and feel kind of like the same thing. Like if you look at the interface of chat GPT and then you look at the
237. 38:47interface of codeex, they kind of look the same and you can kind of do the same [ __ ] right? You can put in a prompt and it'll
238. 38:57give you an answer. You can hook up MCP servers. You can connect chatgbt to other websites like Canva and and uh
239. 39:06runway like I did today. You can connect codecs to those things. You can use chat GPT to write files to Google
240. 39:16Drive. You can use codecs to write things to Google Drive. Here's the primary difference. The primary difference is that with chat GPT,
241. 39:31you personally are the orchestrator. You're the one that says, "I'm going to tell it what I want. It's going to give me something back. I'm
242. 39:46gonna tell it if it's good or not. It's I'm right." And so, and if you're pretty sophisticated, you can give it lots of context.
243. 39:57You can go to Notebook LM and get a bunch of context and put it in your chat GPT and max it out and do really sophisticated
244. 40:06things, but it's it's ultimately you choosing to spend time, your time with this thing. Now, as the large language models that sit
245. 40:20behind chat GPT have gotten more sophisticated, if you give them a complicated prompt, they can work for sometimes they'll work
246. 40:28for 10 or 15 or 20 minutes because they're out researching things. You turned on deep research, right? And then
247. 40:36you're having it go out and research and you're having it write you something and write you a report and that can take some time.
248. 40:43It's still ultimately you doing the driving. So that's prompting these agentic platforms like Codeex and Claude Co-work.
249. 40:56You're actually interfacing with an agent that will do the prompting for you. So you're basically just telling it, I want
250. 41:05you to go do some things for me. And you can tell it, you can string those things together. And in fact, what
251. 41:13you can now do, they've they've just added new features to both claude code and codeex called slashgoal goal where you can give it a goal.
252. 41:27Like the goal might be I want to start a business about selling artisal snacks. Go. And that coordinating agent, if you've
253. 41:40got your codeex set up right, will just go and say, "Oh, I need to do some research. In order for me to do
254. 41:48this, I need some research. So, I'm going to spin up a sub agent that goes and does research. And then I'm going to
255. 41:56spin up an agent that understands small business dynamics. I'm going to create a small business specialist, and they're
256. 42:03going to go research what's the best business model, and what's the best business structure. Oh, and then I
257. 42:07probably need some legal advice. So, let me spin up a little legal version of an agent, right? And so all of these things
258. 42:15are kind of out there just performing independently on your behalf. The loops, as we talked about loops, is
259. 42:31you could have it do the same thing over and over again. Like for example, one of the loops I created with Adam
260. 42:39with my my crappy agent, OpenClaw agent, Adam is on a daily basis when he works [laughter] on a daily basis. I think his daily
261. 42:51crown jug job turned turned into a weekly crown job for some I don't know how that happened because because I suck. Um but
262. 43:00on a on a daily basis he understands what the great repurpose is and he goes out and he and he finds news articles within the past week
263. 43:14related to the big themes of the great repurpose and he finds the five most relevant articles and then he does a
264. 43:23synopsis and kind of the three bullet points of those things and it happens every day. So that loop of, you know, go
265. 43:31out, read a bunch of articles, find the five most relevant, write this structured email that's structured the
266. 43:37way we talk about, and then send that email to the recipients I told you to, that loop can just be done over and over and over again.
267. 43:47And so when these guys are talking about shifting from prompting to loops, the big concept to get is that these
268. 43:58things can just do [ __ ] for you. You don't have to be there waiting for the answer to come back. You can design the loop.
269. 44:09You can you can do some back and forth. You could even do some back and forth with chat GPT to kind of design what you want it to do.
270. 44:16And then you can go over to these agentic platforms and say, "Hey, we're going to do this, then we're going to do
271. 44:20that, then we're going to do that, then we're going to do that." And then I want you to do that on this regular schedule.
272. 44:27And those things are designed in such a way that they can just pick that [ __ ] up and just go. Yeah. By the way, the little scroll down
273. 44:36there, I'm not going to be here Wednesday and Friday. So, this is it for this week. Um, if anything if anything
274. 44:42big and interesting happens this week with Anthropic or OpenAI or really anyone, I might do a spontaneous live
275. 44:50probably on TikTok, but I won't be here. But where is the human in the loop? Great question. So, one of the things that Anthropic
276. 45:01talked about in an article that that they published shortly before they got shut down [laughter] stupid idiots. Um, is that
277. 45:17we're very much on the verge of what's called recursive self-improvement where the AI is actually good enough
278. 45:31to design the next AI. So, so it figures out what it's good at, what it's what it's bad at, and then it can choose
279. 45:39what are what are the areas of research that it wants to go work on for the next version, right? So, and and then if that
280. 45:48gets better, then that version will design the next version. And that's the recursive self-improvement. We're not
281. 45:53there yet. In that article, what Anthropic talked about was the tactical execution of the research. AI is way
282. 46:08better than humans. The coding, the the analysis, the the the strategy, just all the all the stuff that was the doing.
283. 46:22The AI is already better than the humans. where the humans still have an edge is in is in the the the taste
284. 46:32for like what areas of research to pursue like the instinct like AI is not good at instinct. AI is not good at huh I have a hunch right
285. 46:43that human huh I have a hunch it said the AI is getting better like the gap is shrinking but humans still have the edge
286. 46:52there I would argue that for a lot of work humans will retain the edge there and that and that our job the the human
287. 47:03in the loop is once once you understand that these systems can loop and loop and loop and loop and do all these things
288. 47:10for you. The real question becomes Herb. What do you want to do? Our job is to go oh well what do I want to do?
289. 47:25Well, what can I do? Well, these tools are getting PhD level smart at everything and within the next two
290. 47:37years, you know, you'll be able to have 50,000 geniuses waiting for your command. What would you like me to do? That question
291. 47:51is our job and I think it remains our job. Our job is to go, huh, you know what would be cool this. Oh, I can
292. 48:00design some loops that do that, right? Or I can sit over here and chat GPT and just, you know, ad hoc figure out the
293. 48:08piece parts of it. Then once I have my head around this thing, I can go do that. The other thing you can do with these systems
294. 48:15is you can say, "Hey, here's a manual process that I do. Can you design some loops for me?" and it will go design the loops for you. You
295. 48:25don't even have to learn what loops are. You just need to know that they can do them. But the human in the loop part is
296. 48:32what are you going to do? What does success look like? What would not only make you happy if it were built, but would get you excited
297. 48:41that it were built and got you excited to scream at the top of a roof, here's what I made. Here's what I told my thing to make. Right?
298. 48:54When we say AI is better than the human, how do we know who makes that determinations? Who's checking for
299. 49:00errors? Well, the the who's checking for errors part. So, like we're kind of in the third tier of of the way these systems work. The
300. 49:11first tier was to next token prediction. Spew out an answer. The hallucinations were horrible. The next one was spew out
301. 49:20an answer. tell it if it got it right and it would sort of, you know, repair it. Um, the third area that we're in
302. 49:28right now is it spews out an answer, but before it shows it to you, they're they're now creating so some of these
303. 49:35systems now are building agents to check the work. So, it produces the work. It uses some some um um like reinforcement improvement
304. 49:48like like reasoning, right? It added reasoning. It used to just spew out [ __ ] then it added reasoning. So, it's
305. 49:54thinking through something as it's writing it. Now, it thinks through it, produces it, and then another agent
306. 50:01actually checks it. And and people that are that are decently sophisticated with prompting, they they instinctively
307. 50:10discovered that anyway, right? They would produce something with a reasoning model over here. They would then take
308. 50:15that from chat GPT over to Claude and say, "Check this for me." Then they'd take the response of that over to
309. 50:22Perplexity and say, "Research this for me." And then both of those would have recommendations that they'd bring back
310. 50:28to chat GPT and say, "Go fix this." I saw a cool Tik Tok today where a guy said, um, you have chat GPT produce what it's
311. 50:38going to produce and then you say, um, what are the five most important things you missed? Um, and so you could you can get these
312. 50:47things to actually analyze their own work. So there's there's some work heard that you can You can um
313. 51:02you can just say did it get the math problem right? Yes or no, right? Did it did it prove the theorem? Yes or no. And
314. 51:10then you know was it correct? We can verify that that was correct. Right? Did it make a program that worked? Yes. Did
315. 51:17it make a program that worked that is human readable? Maybe or maybe not. Did it work that a program that worked was
316. 51:24efficient was human readable? Maybe, maybe not. So all of the different kinds of knowledge work are going to have
317. 51:30different different kinds of benchmarks to know if the machine is better than the humans. The the pure qualitative piece of taste
318. 51:42like is that one better than that one? Huh? I like that one better. And then someone else says, "Well, I like that
319. 51:49one better." Right? The chocolate and vanilla ice cream thing. That's where humans, I think, will always play a role, right?
320. 51:58Humans are the ones that are going to say, "I've got all these capabilities. I'm going to use them to express myself
321. 52:04in this way." And you might look at that and go, "I would never do that. I would express it this way." Well, then go do
322. 52:12that. Right? And then now we have two different things. And some people will love my thing. I squirted it out. And
323. 52:17some will like yours you squirt it out. All right. Our job is to tell them they don't have taste. [laughter] Oh my god.
324. 52:37I I saved our town from a center. A center? What kind of center? Because I was able to bombard select them before
325. 52:44they changed Oh, data center. They changed the law. Yeah. What does the success look like? I think that's up to us as well. Is that
326. 52:57is that like the thinking models in chat GPT? Well, all of listen all all large language models at this point have reasoning in them.
327. 53:07Um what do they call it? They call it um something time compute. Not in time compute, not real time compute. starts with a D.
328. 53:26Um, basically they just make them think more. They make them do more cycles, they get better. Um, but but so all of
329. 53:34the models right now have reasoning, but this this ability to to have it just go and do things for you is pretty crazy.
330. 53:42Runtime compute, that's what it is. Um, thank you, Brandon. Okay. So, so that's loops. The other thing that's kind of
331. 53:54mind-blowing about these models, and this is this is a thing I'll show you. So, I'm going to I'm going to flip over and and demo some
332. 54:03some stuff here. What are we doing timewise?
333. 54:21Oh, I know what I need to do. All right.
334. 54:38It was a DHA intake center, not data center. Yeah. And listen, I mean, there you go. That's because you've got the capacity to use a
335. 54:54this is so so what Loriy's talking about right here is she's got something she cares about. um she wants to take action. Historically,
336. 55:09you know, she would have to spend massive amounts of time either researching or figuring out how to write
337. 55:16letters or, you know, doing whatever it is depending on how involved she is. Now, with AI, she can kind of amplify
338. 55:21her intent, right? And so, these tools become an amplifier of anything we want to do. Yeah, there you go. small town
339. 55:30was was busy trying to be sneaky. Um, she found out about it and she said, "Oh, let's let's let's use AI to bring this out in the open.
340. 55:41Let's make it noisy." Right? And so and so that stopped. Um, so so that kind of becomes our job. I want to show you something
341. 55:53that I'm not going to show you all of it because some of it's proprietary, but I want to show you kind of what happened. Let's see. I'm
342. 56:04not going to kick that off just yet. Okay. So, I talked about this a little bit on
343. 56:18um on office hours last week, but I didn't really show it. So, I've created this thing called um I'm part of this group
344. 56:30called Content Evolution. And a couple years back, we we created digital twins, digital advisors, and they were digital twins of the
345. 56:41members of content evolution. So for every person that we wanted to make a digital twin for, we did an interview with them, a video
346. 56:52interview where we took the transcript of that interview. It was a structured interview designed to sort of capture
347. 56:58all of who the person is and we took the transcript of that interview and that became the the context, the data set for a custom GPT.
348. 57:10So, we had a we had a prompting. We had a prompt for the custom GPT. Here's how to use that interview. Here's how
349. 57:16[clears throat] to behave. Here's how to talk. Um, hang on. Um,
350. 57:31and then we made we made a bunch of them. And then we made a a big GPT that had them all sitting inside there. So,
351. 57:40so I'm now in this is this is Claude or this is OpenAI's codeex. So, you can see in the in the little menu here in the top. Um, this is
352. 57:50>> you would be able to see if your screen was sharing. >> Oh, that's fantastic. I'm so good at this. Hang on.
353. 57:58That wasn't my machine not sharing it. That was me just No, wait. I thought I did. Maybe I didn't. Oh, I think I just
354. 58:06spit out the uh No, I was sharing. It's We're gonna We're gonna have this issue again. Okay, let me I'll do this so I can
355. 58:15I can see. Oh, you know what I'll do? I'm going to move this over here. And I think I'm going to share in a
356. 58:25different way. Hang on a sec. I think this might be more successful. Okay.
357. 58:44Okay. So, what you're looking at here is the prompt that I wrote. So, I'm I'm in codeex. So, this is a downloaded application
358. 58:53that you can download the chat GPT application. This is the codeex application. So again, think chat GPT as
359. 59:02you're the one doing the interaction and codeex is the agents are doing the interaction and you're guiding the
360. 59:08agents and the agents are doing the work. And so what I wanted to do was I wanted to create a um
361. 59:25I wanted to see how quickly Codeex could take the work that we did creating all these custom GPTs and turn it into a dynamic system
362. 59:39um that would allow me to easily put together custom combinations of these advisors for different clients. Okay.
363. 59:51And so so I said here's all of the interviews that we did. Here's the prompt that that was used in the custom
364. 59:58GPT. And then I talked about Oh, this is another thing I didn't talk about. I talked about MCPS. The other thing that um
365. 1:00:06that Anthropic invented, I think in the past year, year and a half was something called skills. And skills is a markdown.
366. 1:00:15It's a text file that's kind of just structured text that is literally just English instructions for how to do
367. 1:00:23something. And so you can create your own skills or you can use other people's skills. But there might be a skill like
368. 1:00:29here's how to write song lyrics or here's how to write like Shakespeare or here's how to create a certain kind of
369. 1:00:37do a certain kind of math, right? There's all sorts of skills you can do. So you can have these tools create their
370. 1:00:45own skills. What's cool about skills is that because they're just text files, you can literally just save that text
371. 1:00:52file to your hard drive and now all of a sudden your agents have whatever skills you created. So, what I wanted to do was
372. 1:01:01say, how quickly could this thing turn this sort of blob of a data set into some very discreet skills? So, I'm just
373. 1:01:08going to kind of read some of this prompt. Um, I'd like to have a series I'd like you to have you create a series
374. 1:01:15of skills that allow me to bring these digital advisors to life as skills that you have access to. There should be an
375. 1:01:23overall skill to create a new digital advisor. Um, there should be individual skills for each person that captures
376. 1:01:30their individuality based on their interview transcript that also includes relevant metadata about each person.
377. 1:01:38There should be a skill that lets me present a problem and it will select a subset of digital advisors of five to
378. 1:01:45seven content evolution advisor. So I just I just literally kind of brain dumped here's all the different kinds of
379. 1:01:52things I'd like it to be able to do and then I told it to go make it and it it said it took five minutes. And so
380. 1:02:05in five minutes and 25 seconds it replied, I've built and installed the complete content evolution digital
381. 1:02:12advisor suite. 28 individual advisor skills. Right? So one for each person, four shared skills, create a content
382. 1:02:22evolution digital advisor, assemble a panel, write an article, or find CE advisors, which is something that I
383. 1:02:30asked it to do. Um Mark Pransky was uh one of our members of Content Evolution Collab. He passed away last year. Um so
384. 1:02:38it respectfully labeled him as a legacy adviser postumous. Um it said it's missing calendar and contact preference data. It said if it
385. 1:02:48was missing contact and calendar preference data, it marked it as unavailable. And then it said all 32 skills passed validation. So, it took
386. 1:02:58five minutes to to effectively um do what I asked it to do. And then I I I I had it I you know did some follow-ups and I made it more
387. 1:03:08sophisticated. Um but what I want to show you is it just stopped sharing. At least I can see it now.
388. 1:03:34Okay, here's pages. Um, let me get rid of that. Go to 300% so you can see it. Ah, it just killed it again.
389. 1:03:51This is maddening. Maddening. And I just restarted my machine. I don't know what's going on. Okay. So, one of the things that I asked
390. 1:04:02it to do was to write documentation for how to use the skills because I'm just learning how to use codecs. Like I really don't know
391. 1:04:16what I'm doing. Like again, I have it kind of conceptually in my head, but unless you're using this stuff kind of
392. 1:04:24day in and day out, you know, I'm I'm still running my company, so I don't have I don't have the time I need to to
393. 1:04:31really dig into this. So So I wanted it to teach me how to use the skills I just asked it to make for me, right? And so
394. 1:04:41here's what it did. And it like it's it's amazing, right? It it wrote all of this. it formatted it, you know, in this
395. 1:04:48nice way. So, here's the skills that you call. So, the dollar sign CE um create a digital advisor. Here's
396. 1:04:58assemble an advisor panel. So, so it tells you what the skill is and what what the purpose is. This will select
397. 1:05:05five to seven relevant advisors, con convene their modeled perspective, and synthesize their recommendations.
398. 1:05:12Um, then here's the individual people skills, right? So, there's the Kyle Shannon advisor. Um, you know, here's all the different
399. 1:05:23people within, you know, within content evolution and what the skills are. You know, here's how the skills work. Um,
400. 1:05:31here's how to use them. You can consult a single advisor. You can convene a pa panel by using the, you know, assemble
401. 1:05:39the advisor panel thing. You can use it to write an article. Here's people to um find people to reach out to. Here's how
402. 1:05:46you work with collab. And then here's interesting use cases. Strategic red teams, executive decision support, um
403. 1:05:54trust and attribution, and then you know there's the source where everything came from. So, this was I did all of this work within
404. 1:06:09um I don't know maybe a 45 minute period while I was on another call that I didn't have to pay attention to. I would
405. 1:06:17occasionally be called on so I could do that. Let's see. Can people co-author an article from the panel? Yes.
406. 1:06:27Damn, this app is per Snick. I think that's good. [laughter] But here, let's go. Let's go play with it. Let's Let's see. Um,
407. 1:06:39so if I do, um, let me see. Do a new chat.
408. 1:06:53Oh, wait. Okay, there we go. Okay. So, this is write an advisor article. So, here's what we're going to
409. 1:07:08do. I'm going to go um slash paste. Is that how I do it?
410. 1:07:24No, I think I just go hang on. I just go um use there's the skill. Okay. Write the advisor article. Here we go.
411. 1:07:40Okay. So, use that skill. So, so there it shows up as a skill. See how it's showing up as this blue thing? Use write
412. 1:07:46the article to write a LinkedIn
413. 1:08:00article about the latest reporting on the Fable Five.
414. 1:08:15Um, government takeover. I want
415. 1:08:29you to use and then I'm going to go get mine, my skill.
416. 1:08:52Okay. I want you to use CE advisor Kyle Shannon skill um to be the primary author of the article.
417. 1:09:10And I want you to include three quotes.
418. 1:09:43Let's see. Three quotes um from the Oh, come on. Could someone use when Claude Fable comes back, could someone use it to
419. 1:09:58please fix OSX and screen sharing? I think this is a memory thing. Um, I want you to include three quotes um from the most
420. 1:10:13relevant most relevant CE pro digital advisors.
421. 1:10:34um as well as two diverging POVs um
422. 1:10:50from the CE universe to keep Kyle honest. Okay, so
423. 1:11:06So now I'm going to do this. So it says thinking and this this is my little my little pet, the codeex [clears throat] pet.
424. 1:11:14And this sort of what he does is sort of shows you what's going on, what the model's doing. So it says writing the fable five. I'll
425. 1:11:26use the CE article skill as the editorial workflow and Kyle Shannon's advisor skill is the primary authoral
426. 1:11:34lens. First, I'll verify what Fable 5 government takeover refers to in current reporting. Then, I'll put together the
427. 1:11:43advisor roster and the Kyle context of the piece. Right? So what this is kind of doing is I basically took the the data from our custom GPTs
428. 1:11:57and this is sort of effectively turned it into this dynamic suite of skills that I can explicitly tell it to draw on. But I could turn
429. 1:12:08this into a loop, right? I could say every day I want you to find the most interesting issue on AI that day. I want
430. 1:12:18you to, you know, write this from using the Kyle Shannon point of view or just from some random CE point of view,
431. 1:12:25right? We could we could, you know, say I I want to write an article from each of these nine people. So just cycle
432. 1:12:33through those nine every, you know, every nine days, we'll just repeat the this the loop and then I want you to create
433. 1:12:44quotes from these people, things like that. So this is like one teeny tiny little thing that's possible. And Herb,
434. 1:12:52this gets back to your I don't know if you're still here, but this gets back to the thing that you were talking about where um
435. 1:13:04you know h [laughter] how do you choose what to do here? Well, you've got to you've got to figure out well well what are you personally
436. 1:13:15what's something that you care about? Because the downside of a system like this is that you just have it spewing out [ __ ]
437. 1:13:26and it it means nothing to you. And it's just like these systems have already like last week we we already crossed the
438. 1:13:35threshold where there's more content being produced every day by bots, by agents than by humans. We just crossed
439. 1:13:43that threshold a week ago. So that's going to get worse and worse and worse, right? That's going to it's going to
440. 1:13:48absolutely kill the internet. But what's going to survive, what's going to rise above the noise is content and things that are produced
441. 1:13:59that people have trust in. And you don't we we will cease to have trust in the outputs. we will increasingly have trust in the
442. 1:14:14creators of those outputs. And so if someone gets trusted for creating things where their apps are just slightly better, the things that
443. 1:14:24they write have really interesting takes on them, all that sort of stuff, then you know those are the going to be the
444. 1:14:32things that rise above the noise. So it comes down to each of us individually figuring out what we give a [ __ ] about
445. 1:14:39so that when these s we design these systems to go do stuff for us. The thing that comes out the other end represents
446. 1:14:46what we wanted it to do in the first place. And then we could talk about it. Hey, I'm doing this cool thing and once
447. 1:14:52a day it creates an app that does XYZ and here's why I give a [ __ ] about that. I would take those nine articles, put
448. 1:15:00them in Notebook LM and create a podcast. So that's exactly and and you can probably I haven't even tried this
449. 1:15:07but I would assume you could just make a skill that has this thing take those articles put them in notebook LM and turn that
450. 1:15:17into a podcast and you could have a content evolution collab right so let's see it's done I wrote the LinkedIn so this will be
451. 1:15:27fascinating we'll go look at this article we're coming up on time so this is good we'll go look at this article
452. 1:15:34It's supposed to write it like me, which that'll be fascinating to see if it did. Um, let me share my screen again because it
453. 1:15:41just crapped out. This is cool, right? And again, like I'm just barely scratching the surface here, right?
454. 1:15:56Um, but but so here's what's important. God damn that thing. Um, here's what's important. The genesis of this channel
455. 1:16:10was all of us paying attention to all of the tools, trying to keep up with all of the tools, trying to learn all of the tools.
456. 1:16:21That ship has long since sailed. And we know we we know we've given up on that. We're now in an era where a single tool
457. 1:16:31like codeex is so complicated and requires so much understanding of so many of the piece parts of all the
458. 1:16:39things that we've done for the past three years that we actually do have an advantage that we at least can
459. 1:16:45understand what are the piece parts that this thing can do something with. Um, and the time we're investing now is wait
460. 1:16:57a minute, I need to understand how to even set this thing up so that it can do what I want it to do. Like one of the
461. 1:17:03cool things that happened is all those skills that it described are now just text files sitting on my computer that I
462. 1:17:11can turn into a zip file, which I had it do, and write an installation instruction. And so I can now give that
463. 1:17:18zip file to any members of content evolution to to take these skills and put it in their claude code or their claude
464. 1:17:29co-work or their codeex or their open claw or their Hermes. They can take the work that I did here and they can just
465. 1:17:36ingest it and and have the same same thing. Um, being able to even understand what these things are doing for the next two years
466. 1:17:53is going to be a massive advantage for you. Massive like in in the workplace, right? If if if you like it if like we have
467. 1:18:03proven that just being curious about AI, if you talk about it is valuable in the workplace. understanding these systems
468. 1:18:10and understanding the the difference between prompting and looping and sort of massive context generating and and
469. 1:18:19centralized data management. All of the stuff that it's required to use this tool. Um those are skills that are going
470. 1:18:26to be incredibly valuable. Okay, let me try this one more [ __ ] time. It's driving me crazy at this point. Okay, I treated CD
471. 1:18:38universe as the CE advisor universe. That's good. It picked up my typo and fixed it. I used modeled Kyle Shannon as
472. 1:18:45the primary invoice. I included three modeled advisor pull quotes and two counterweight POVs from Richard Carson
473. 1:18:53and Richard Boyd to keep the piece honest. I also verified the factual spine against current resources
474. 1:19:00including anthropics. So, let's go look at the article. Uh, it's in Markdown. So, I'm going to say um I've asked you before
475. 1:19:12to turn anything you want a human to read to be in doc x format,
476. 1:19:27not md. Please fix and remember this. So let's see and remember
477. 1:19:42this. Um so AMD file is a markdown file and I could read them. They're they're easily readable but they just don't look
478. 1:19:51pretty. So I'm just telling it make me a a Word document. You're right. I should have caught that preference and
479. 1:19:57delivered the readable artifact as docx. I'll convert this draft now and add the preference to durable Kyle context. So
480. 1:20:05future humanfacing deliverables default to Word. Okay, one last concern. Several agents have gone rogue and caused massive damage.
481. 1:20:15Yes. Is there a possibility of a you loop making that error and then keep repeating it until you destroyed your computer? Yes. So
482. 1:20:29there's all sorts of ways that you can protect against it. One of the reasons that people are buying standalone Mac
483. 1:20:36minis. So So this um this codeex that I'm running is running on a dedicated computer that's got its own data um and
484. 1:20:48its own everything. And so it is it's kind of sandboxed on that computer. Um, so the computer that I'm working on,
485. 1:20:58it does not have access to that. It's it's got access to its own computer. With these things, I would start very
486. 1:21:05simply like I'm doing this very slowly. One of the cool things about the world we live in right now. Oops. Office
487. 1:21:13unexpectedly quit. Oh, s office unexpectedly quit. reopen.
488. 1:21:27Huh. That's weird. All right. Looks like it's still doing that. Um, like I would start slow and start like
489. 1:21:39this little thing that I did here is a very like the the the worst thing this could do is it it starts running those
490. 1:21:46loops and it just writes more and more and more articles. Like I'm not doing anything here like you know organize my
491. 1:21:53files and delete the the oldest ones, right? And then it just starts deleting the oldest ones until it deletes them
492. 1:22:00all, right? Like so just just be careful like start slowly with these things. You know if if you're just getting your head
493. 1:22:09around chat GPT you can just start to do things like connect different applications to chat GPT and start using
494. 1:22:16that if you want to understand what these agentic platforms are capable of. Use chat GPT to teach you right or use
495. 1:22:24claude to teach you what these things are how to use them where to start. When I started with Adam, I I said, you know,
496. 1:22:32what is the the safest way for me to start? What's sort of the least damage it can do? Okay, opening pages. Let's go see our article.
497. 1:23:07Okay. The Fable 5 shutdown is not just an AI safety story. It's a trust story drafted in a modeled Kyle Shannon voice
498. 1:23:15using content evolution digital advisor. On June 9th, Anthropic launched Claude Fable 5 and Claude Mythos 5. It's new
499. 1:23:23mythos class models blah blah blah. Okay, that is the clean factual version. That line right there is very AI and is not
500. 1:23:32something I would do. But that's okay. It's AI writing still sucks to be clear. The less clean version. Okay, let's see.
501. 1:23:41People are calling it a government takeover. Let's see the question. Wait, so why do we trust? So, one of the
502. 1:23:47things that I will likely do is I will go through and I will tell Codeex why I think what it wrote sucks and to fix it and to add that to my
503. 1:24:01little thing so that it gets better at my writing. Okay. Um, let's see. This is why the fable five moment feels like a threshold. Okay.
504. 1:24:133CE advisor lenses modeled by Dr. David Bray. Trust is the willingness to be vulnerable to an actor you cannot directly control.
505. 1:24:25In AI governance, both the company and the government have to demonstrate benevolence, competence, and integrity
506. 1:24:33or the public will trust neither. David's lens is helpful. Cindy [ __ ] I don't know if Cindy's on tonight. Do not
507. 1:24:42predict one future from this event. Build several. In one, this becomes a template for responsible intervention.
508. 1:24:49In another, this becomes a precedent for chaotic chaotic model politics. Your job is to act now so that better futures
509. 1:24:57become more likely. Cindy's threat casting lens. So, like it it understands who Cindy is, you know, what her
510. 1:25:04background is. Here's Rick Lroy. He's a PR guy. In a crisis, the information vacuum becomes the story. If the facts
511. 1:25:13are sensitive, um, say what can be said, what happened, what is known, what is unknown, who's reviewing it. So, anyway, um,
512. 1:25:22there's some counterpoints. So, here's our article. So, we've got an article. Now, is it good? No. Is it is it
513. 1:25:28something I would publish? Not even close. Um, but does it prove that I could take some blob of data, turn it
514. 1:25:39into some replicable thing that I can put into these loops? Like once I get this thing refined, right? Then,
515. 1:25:49you know, then we're off to the races. And then there's probably, I don't know, a hundred thousand different little things like this you
516. 1:25:58could do. Which ones are you going to do? That comes back to what do you care about? All right. So, that's that.
517. 1:26:13Let me bring this over here. Um, what are your thoughts? So you have to download all different apps on that computer, different accounts.
518. 1:26:25Um, no, no, because most of the apps that Codeex is using are web- based apps. But if I want like if I wanted it to be able to um
519. 1:26:43because it's got computer use if I wanted it to be able to open to create an image in chat GPT open it in Photoshop
520. 1:26:56into a template with a watermark on it. Export that out into a specific drive on my computer. then I would need to have
521. 1:27:03Photoshop licensed and running on that computer. But if I if I just had an MCP connection to Photoshop online,
522. 1:27:14I could do the same thing without having to have the app on a computer. So most of the apps that that that thing
523. 1:27:22most of the tools it's using it's doing it's doing out in the world. The skills are things that you can build you can
524. 1:27:29build locally and they can be any combination of any of the other things, right? So you can have a skill that says
525. 1:27:36go do some research, come back here, turn that research into a research brief that you then take to Perplexity, have
526. 1:27:45Perplexity do its thing, write a report, make a website, take the URL of that website, you know, and and the image
527. 1:27:53prompt from the article that it wrote, come back here, make an image, add that image into a blog post, into
528. 1:28:01the site that we built, like all that sort of [ __ ] So that chain of events could become the loop, right? You need to start thinking
529. 1:28:09about what's the loop. You can also say to codeex, what are all the repeatable things that you know we've done a bunch of stuff
530. 1:28:16here. Come up with a series of repeatable things and when you think those things should trigger and what should trigger them.
531. 1:28:24Okay, on deck new LOL this Saturday. Okay. So again, this week, this is it for AI learning lab. Um I am I am
532. 1:28:35traveling this week. So Wednesday I'm not here. Friday I'm not here. If anything big happens this week, I'll go live on TikTok. Um
533. 1:28:45and I'll coordinate with producer Brandon. We'll we'll figure out a way to communicate if something big goes down.
534. 1:28:50He's he and I are generally um back and forth on big things happening. So, um, so that's that. This coming
535. 1:29:00Saturday is an LOL, a learn out loud with no Emmy. Do you know what she's what she's teaching? I didn't I didn't
536. 1:29:07catch that, Brandon. Um, and then we've got um we've got the AI salon mastermind practice lab is Thursday at 10 a.m.
537. 1:29:17Mountain time uh with Sid Harrow and myself. Um, and the and the the people that are in the mastermind practice lab
538. 1:29:25who are great. So, if you're in the if you're in the mastermind, uh, join the practice lab if you haven't been in a while. No,
539. 1:29:36more vibe coding. So, Brandon just finished a fiveweek lovable course this Saturday. No is teaching some more vibe
540. 1:29:43coding. So, NMI has created uh a native Swift app and is selling it on the Apple App Store. She's not a programmer.
541. 1:29:55She's got an app on the App Store. She's going to teach you how to do that [ __ ] Very thorough as always. Lots of food
542. 1:30:02for thought. Um always loops back to what what do I want to do it this so Herb. Yes. Like all of this [ __ ]
543. 1:30:12[clears throat] all of this [ __ ] for me. And I I said I was going to talk about disclosure, but we're out of time.
544. 1:30:21Disclosure day. It all comes back to who are we? Who are we? Who are you? Who am I? What do I care about? Who do I care about? What
545. 1:30:37difference do I want to make in the world? That's always been the question. It's always been the question. But like
546. 1:30:48life and work and just the [ __ ] the complexity of this planet buried that. And I think what AI is doing is it's
547. 1:31:00going, look, I'm going to handle all the crap. I'm gonna handle all the [ __ ] that's gotten in your way and I'm gonna allow you to just do
548. 1:31:11everything, literally everything you could have ever imagined you ever wanted to do. I'll do all that. You have one job.
549. 1:31:21Tell me what that is. Tell me what you want to do. And it's a hard [ __ ] question. So, thank you, Claire. I appreciate that. So
550. 1:31:34my homework for you this week is just continue to explore. If if you if you have not started a daily practice around
551. 1:31:41how you use AI, now would probably be a good time. Um I would encourage you to think, go sit on a bench and think about who am I
552. 1:31:50and what do I want? Who do I care about? Have a safe and productive trip, my man. I appreciate that, Silver Fox, very very much. Um
553. 1:32:01yeah. Yeah. Go sit on a bench. Go sit by a fun. I I started fly fishing again. Look, a Colorado fishing license. [laughter] Um I'm official,
554. 1:32:20you know, and I'm going out there so I don't have to think about anything like I want this gone so I can go, huh, what
555. 1:32:28do I actually want? What do I care about? All right, beautiful. Thank you all so so much. I will uh see you next
556. 1:32:38week. Um I'll be there for should am I going to be there for office hours? I should be there for office hours on
557. 1:32:44Friday. I think I scheduled my flight so I can be there for office hours on Friday. So I'll see you Thursday at the
558. 1:32:50practice lab. I'll see you Friday at office hours. And then you're on your own for Wednesday and Thursday night.
559. 1:32:56You got a deal. Go get your own date. It can't be your date. Got to get your own date. Peace. Thanks everybody.