8/7/2026 - Agentic AI, Local Models, and Building a Personal AI That Works for You

Transcript

1. 0:03[music] [music] [music]
2. 0:20[music] [music]
3. 0:34Good evening. Welcome to the AI learning lab here in the AI salon. Um, happy Friday night. It's Friday night date night. I'm
4. 0:43putting away my little, what do they call these? Gimbals. Little gimbal. Putting away my gimbal. Is it a gimbal?
5. 0:48I think it's a gimbal. Wasn't gimbals a shopping store when I was a kid. I think it was [laughter] a department store.
6. 0:57You going down to gimbals today? That was the accent I grew up with. You going to change your oral in the car? You going to them all?
7. 1:09Yeah, I grew up in Pennsylvania. All right, producer Brandon is back. All right, so here's the deal. Welcome to
8. 1:17the AI salon. Welcome to the AI learning lab. Um, for the next 90 minutes, uh, we're going to talk about all things AI,
9. 1:25like literally. If you have any questions, um, pop them in the in the chat over here to the side. If you're
10. 1:32like, I don't see a chat. How do I get to the chat? The chat, there's a little button down in this corner that will pop
11. 1:38the chat out. So, you click that, pop the chat out. I think if you're on mobile, the chat's below this stream, I
12. 1:44think, but I don't know cuz I could. It just doesn't matter. It's not important to you. [laughter] I'm here but can't see what I'm texting.
13. 1:54Or actually, she's like, I'm here but can't see what I'm yexting. because she can't see it, she writes words like yexing. [laughter]
14. 2:09Um, okay. If you're new to the AI salon, welcome. So, a couple of things about it. Um, it's a remarkable community and it's
15. 2:20more remarkable when you participate. So, it doesn't cost you anything to participate. Well, it cost you your time.
16. 2:28The benefit of interacting in a community like the AI Salon is you get to meet people and you get to build
17. 2:35trusted relationships with people that you resonate with. So down the side of the um the the interface of the AI
18. 2:42salon, there's all those different tabs, right? All those different what we call spaces. All those spaces are different
19. 2:50conversations going on with different kinds of people. So there's community chat and a community feed. Um there's
20. 2:56there's business spaces and and life hacks and um ethical AI. There's all sorts of cool things down the side. So,
21. 3:04go check all that stuff out. Toward the top of that list is an events tab and it'll show you all the events and all
22. 3:11the workshops and the free education. It's all in there. Um, the AI salon is full of, smart, nice,
23. 3:25humble, generous people that haven't forgotten what it feels like to just be starting with this. So, if you're new to this AI stuff and
24. 3:35you're like, "Oh, these people know so much more than I do." Guess what? We all feel that way still because AI is
25. 3:43moving. so fast like no matter how much you think you have it figured out about three months later you're like oh I'm
26. 3:49clueless again it's just part of the state of the state of being in this world so um so with that welcome welcome
27. 3:57welcome um tonight is an ask me anything night so an AMA so just feel free to pop questions in there I see a couple
28. 4:05popping up right now so feel free to put them on screen Brandon whenever you got your thing and I've got a couple of
29. 4:11couple of news items that I talked about in office hours today that I think are worth that I think are worth putting up
30. 4:17there. Um, let's see. All right. For non-technical workers, what does agentic AI actually mean in practice?
31. 4:27Fewer meetings, better workflows, or just another tool to manage? Yes. [laughter] What aentic AI means.
32. 4:40So, we went through a phase about a year ago where everyone was talking about AI agents and they weren't AI agents. They
33. 4:46were just either automations or they were just like chat bots like a custom GPT. Those weren't agents. The the thing that makes something
34. 4:59agentic is that it it operates independent of you prompting it. So, if you go to chatgpt.com
35. 5:13and you just see the little prompt hole, the prompt box, and you type in your prompt, it will give you an answer. And
36. 5:20it's this back and forth conversation between you as the human and this large language model, this this compaction of all of human knowledge
37. 5:32into this thing that answers you really quickly. And it's this back and forth. That's that's called a chat interface.
38. 5:40the agentic interface like in in chat GPT there's two tabs at the top one says chat and one says work and if you've got
39. 5:47a paid subscription you can do this work and what happens with work is you put in a prompt just like you put in chat the
40. 5:55difference with aentic stuff is it will go off and do things on your behalf. It will go search the web. It will take
41. 6:02what it finds from that search and come back and write that into a word document and then put that word document into a
42. 6:09Google drive. It it has tool use. So you can give it access to your Google drive or you can give it access to to
43. 6:15different tools like Figma or or I don't know all sorts of things. Travel sites if you want to do the like the classic
44. 6:23example is you know it can go research a trip for you and then book all the tickets. Um, in theory, all of that means that you can just
45. 6:38literally Hey, Quinn. Uh, I'm going to need you to go research uh tickets to Porta Viarda
46. 6:55uh for September 15th. Got it. Let me check that. >> And then she goes off and does that. Um, in an agentic system, she can do
47. 7:09multiple things in a row and use different tools and spin up sub aents that go do other things for you. So in
48. 7:16theory, what an agentic workflow is, you can just say, "Hey, you have a goal of for this amount of money, go book, you
49. 7:27know, a vacation for my family and create the itineraries and send everyone a personalized itinerary and take into
50. 7:35account all of our preferences and all that sort of stuff." In reality, the tools just aren't there yet.
51. 7:42And so you can get to about 80 or 90% of what you want to do and then it just still sucks. [laughter] And I mean,
52. 7:58so especially if you're new to AI, if what you're saying is why would I put myself through that? If the tools don't
53. 8:08work yet, why are you people here? Why are you doing this? That is a great question. And that has to do there's a there's a
54. 8:16book called Crossing the Chasm. And and the chasm is this kind of dead zone between the early days of technology in
55. 8:27when the masses adopt it. Right? So with the worldwide web in the early days, you had a bunch of geeks that were like, "What's this HTML
56. 8:38thing? this looks pretty interesting. And they would figure out how to make web pages and they would put web pages
57. 8:44on these things called internet servers with their dialup modems. And then there would be pages with hyperlinks on them
58. 8:50and other people would click on them and go, "Oh, that's kind of cool." And it was very similar to way where AI is
59. 8:57right now where it was kind of cool, but it was like, "Yeah, it's a little it's not quite there." And it took a lot of years
60. 9:05for those early web pages to develop into into something that was useful enough that normal people could look at
61. 9:14it and go, "Oh, that's cool. I'll use that. I could use one of those website thingies for my business." [cough and clears throat]
62. 9:23The the chasm is between what are called early adopters and the early majority. So, the early majority are willing to
63. 9:30try stuff, but only if they understand what the value is. Early adopters are like, "Let's tinker. Let's go play." I'm
64. 9:38one of those. I love tinkering. I love playing. I love playing with the tool and seeing where it's going to be in five years and
65. 9:48imagining, oh crap, what's the world going to be like if this thing that's crappy now gets not crappy?
66. 9:55The early majority is like, as soon as it can take hours out of my day, count me in. You know, I can still do good
67. 10:04work and, you know, be at the grocery store talking into my phone. All right, cool. But it has to be like consistent
68. 10:14and good and valuable, right? There's value to be had. [laughter] 100% still bad. Apple, the early majority. You know,
69. 10:28it's funny. I I gave [ __ ] to Apple for for a number of years. Um because they just missed the boat on this AI stuff
70. 10:36and they didn't invest in it. It pissed me off because if they decide to do it, it's going to be really good. And I
71. 10:42thought I thought they could absolutely be a player in being a frontier model, being one of the leaders in this new
72. 10:51thing called intelligence. And they they they have avoided it. They're basically just letting, you know, letting Anthropic and Open AAI
73. 11:06and Google and Elon Musk just have a mud fight and they're just sitting back in their spaceship shaped building going,
74. 11:14"Yeah, let them fight. Let them fight." You know what we should do? Let's Let's not make the car. Let's not make all
75. 11:21these crazy ass devices. What we should do is we should make computers that are really good at running AI. So when these
76. 11:30Yahoos out here having the mud fight get some really good technology that actually works that our consumers would
77. 11:38be able to use, then we'll just adopt it and and they can run it on our machines better than any other machines.
78. 11:46And so they're going to sell more computers and then when they decide to roll out actual AI, like right now they have that
79. 11:54Apple intelligence [ __ ] that's just it's built on shitty models and they're just sort of placating the masses. Yeah,
80. 12:00you want AI, here's some AI. Um, but when they really come out, they're going to come out with badass computers
81. 12:11that have badass models models on them with badass consumer interfaces that are going to be good enough
82. 12:19that early majority people are going to go, "Oh, I'll take one of those. It does that. Does it do that?" Because I've
83. 12:26used Siri and Siri sucks, right? for for a decade. For a decade, Siri has sucked. Sucked and then it's it's not going to. So
84. 12:41anyway, that's what's going on. All right. On 7:31, you spoke about about setting up a separate computer for
85. 12:49using codecs to let access local files. Yes, I'm about to do that for having codeex help with a couple of WordPress
86. 12:57sites. Perfect. Aside from pro protecting my hard drive on my main computer, anything else I should watch
87. 13:04out for on the dedicated machine with 16 gigabytes of RAM? Okay. So there's there's a couple of things,
88. 13:14Linda, that not necessarily things to watch out for, but it's just important to understand that um to really understand the distinction
89. 13:29between using cloud inference or cloud compute and local compute on a 16 gigabyte machine. As of last week, it probably wasn't
90. 13:41powerful enough to run a decent model, but as of this week, it is. Brent Brent Peterson on office hours today. Were
91. 13:48Were I Were you at office hours today? I can't remember. Um, there's a new model out from Liquid. That's a a 2.6
92. 13:57billion parameter model. It's like liquid 2 2.5 like version 2.5 2.4 or 2.6 six billion parameter model anyway um that you can
93. 14:14run locally and and basically what that means is you could if you wanted to like organize files on your computer or
94. 14:23something like that you wouldn't have to send instructions out to chat GPT or Claude um to do all that all that stuff um
95. 14:35I think the most important thing to do Linda is just you So look up, have Grock or ChatgBT go do research on how
96. 14:46to set up a brain for an agentic system like codeex where you've got your user file that talks about who you are and
97. 14:54you've got your agent file agent.md that talks about how you want that codeex agent to act on your behalf. um
98. 15:03those two files and then memory is files that talk about here's the projects that I'm working on and here's the the stuff
99. 15:10that um I'm really excited about. Mel is going to be giving a uh an LOL a learn out loud. I don't know if it's scheduled
100. 15:19yet, but but look for it um on how she she's set up these three files and how she uses them. You don't have to be
101. 15:26really fancy with it. You can get really good value out of that. I would say like just start really simply. Start with understanding what
102. 15:35your user file is, what the agent file is, and and a basic memory system. These can literally start out as folders with
103. 15:41with text files in the markdown files, and Codeex will know how to set them up. Just do the research to tell Codeex how
104. 15:49to set them up. Um, and then you can have it interview you, and you just answer the questions it asks you. You
105. 15:55can even do that on, you know, speech to text or, you know, the the advanced voice stuff and then it'll take your
106. 16:02conversation and turn it into all those files. That's probably the biggest thing. And and just initially just use chat GPT's,
107. 16:10you know, cloud-based inference and you can start to switch between expensive models and cheaper models and things
108. 16:17like that. You can experiment, but I would just experiment and then do things like, hey, you know, you're going to have to do things like
109. 16:25enable the tools. You're going to have to enable um access to the web. You're going to have to enable I would assume
110. 16:32that there's a WordPress MCP or if not, you might have to log into your WordPress site and give um Codeex access
111. 16:40to it. You might have to give Codeex um computer use. So, it's a it'ss ability to use your computer and then do things
112. 16:48like um hey, I want to go edit the the title on this web page and just watch it go do its thing, log into WordPress,
113. 16:59make a file, tell you that it's going to, you know, update your website, and then it's probably going to ask you
114. 17:06permission. Is this okay? And then say yes, and then just pray it didn't screw up your website. Um, I did that about a
115. 17:13month ago and it was completely nerve-wracking and then it did it fine. It was really good. So, here you go. All right. Let's see.
116. 17:24Maybe Apple is just going to Oh, capitalize on the hardware. They might, but they I mean, the thing with Apple is they've always their their
117. 17:35hardware is always amazing, but they always have the software layer layer on top of it that makes it useful. So, I
118. 17:41think it's going to have to be both for them. But again, they're just waiting until we're pretty close. Like once what
119. 17:48in two months is is my guess, we're going to really start hitting this cycle where the the AI starts improving
120. 17:55itself. And at that point, there's going to you're going to have very small models that can run on 16 gigabyte
121. 18:01machines that are incredibly good. And then it'll just get better and better and better and better. Um
122. 18:08Mel working on it. Mel is working on her LOL. Hey Mel, here's the thing. You are more of an expert than you realize and
123. 18:17you're you're way more prepared than you think you are. I know you're like, listen, God love you. Like, I know you
124. 18:24want to do something good and you want it to be great and that's awesome, but even if you just took what you what you
125. 18:30showed us, the little preview that you showed us is like banging. Like people, it will change people's lives. It's
126. 18:37really good. It's really really good. Um, Frostbitten Skills, plugins, MCPS all have their goods and bads. Yes,
127. 18:46thank you, Kyle. Yeah, you're welcome, Linda. Um, yeah, I was at the office hours today. Yeah, so you heard you
128. 18:52heard Brent talking about the that small model. I played around with it a little bit. I ran it on my this this computer,
129. 18:59my M1, um, which is stupid because this has got all my files on it. So, running a running an agentic harness on a
130. 19:08machine that's got all your files on it is what would be called not recommended. Uh, but I just wanted to see if it would
131. 19:16run and it ran and it was like a decent amount of token speed. Um, this is a 16 gigabyte machine here. It's an M1. Um,
132. 19:23okay, one more question. Um, wait, hang on. Linda Sherman has one. Let me let me see this one. One more question. Codeex won't already know
133. 19:34about me from chat GPT chats. It doesn't connect. It does not connect. Um well, I don't know. Maybe I don't know. Okay, so let's
134. 19:50go back to boxes. Okay, this this is your computer. This is your your 16 gigabyte dedicated codeex computer.
135. 20:06This is Chat GPT living up in the big server cloud. The the data centers that no one wants while they type into chat GPT and
136. 20:17watch Netflix. That's what this is. Okay. Um, all of your history of your chats in chat GBT is here. It's up in
137. 20:28the cloud. It's in It's in the data centers. Everyone's doesn't want in their neighborhood. They They want this,
138. 20:36but they don't want that. [laughter] This has got actually this in here. See all these little filing cabinets? This is
139. 20:49your computer. So what codeex does is it uses the agentic tools. It goes to chat GPT. It goes to the cloud for the
140. 20:57compute for the AI inference. You give it your you give it a prompt. It goes and does the inference out there
141. 21:05and then it it brings the answers back and writes it into files on your local computer. So, so what happens with
142. 21:11codeex is all your memory stuff, that user file and the agent file and all your project status and progress and and
143. 21:19updating your website on a daily basis, all that stuff that you're doing is now stored locally on this computer. Like
144. 21:25one of the things that you're going to want to do at some point is you're going to you're going to want to have your
145. 21:30codeex agent say to um uh to to these files, we're going to now go to GitHub and we're going to back you guys up up
146. 21:40on GitHub. So So in case you your computer gets wiped out, GitHub will still be there. You'll have a backup up in the cloud. Um,
147. 21:50I would imagine that there is going to be some sort of um, memory exchange between all the crap you have in the cloud and what Codeex is
148. 22:01doing. But right now, it's really sketchy. Um, and it's not clear to me and it's it's not clear in the interface
149. 22:09what is a cloud-based file and and context interaction memory [ __ ] and what is a local one. And you also have three
150. 22:18different products. Well, no, you have you you have more than three. You've got chat GPT on the web that you you
151. 22:26interact with through your browser. Then you have the chat GPT app. Within the chat GPT app, there's chat GPT chat,
152. 22:35chat GPT work, and codeex. And then you have the phone which you can connect to codecs or chatgpt work on your computer if it's
153. 22:49running. It's so it's a very complicated ecosystem of capabilities that ostensibly kind of looks like this.
154. 22:58If you're running the [ __ ] in a browser, all of your files and all of your memories are living on OpenAI's servers.
155. 23:06If you're running the chat GPT app, it's saving [ __ ] to your local computer and it can interact with your local computer
156. 23:13and it can write files to it and it can delete files from it, which is why to make a dedicated computer for this.
157. 23:19Right now, these are kind of two distinct worlds with incredibly mushy interface where it's impossible to know
158. 23:26what's what without really really thinking about it and architecting it. So if you really want to understand
159. 23:33this, it's probably worth even like diagramming out what happens in chat GPT in the cloud, what happens here on my local computer,
160. 23:42what's the difference between codeex and work and chat. It I don't know. They're they're kind of mostly the same. Um it's bizarre. On on
161. 23:54the Claude side, you've got Claude code and claude co-work which are essentially the same thing. It's just Cloud Code has
162. 24:01a few more bells and whistles for geeks like pull requests and you know GitHub native GitHub integration and IDE
163. 24:09connections and things like that. There's all sorts of fancy [ __ ] that coders use that you don't need to use for co-work but the agentic
164. 24:19system is essentially the same system. So it's it's a mess. Okay. When people talk about upskilling for AI, what what should act actually
165. 24:31what should someone actually practice this month? Prompting reviewing AI work, building workflows, or understanding risk.
166. 24:42That question sounds like it was written by AI Laundry Zeppelin. Um, [laughter] but it's a good question.
167. 24:58All right. Here's the thing about upskilling for AI. What does it say? What should I practice this month? Prompting, reviewing AI
168. 25:09work, building workflows, or understanding risk. None of those things. Here's the thing about AI.
169. 25:24It's weird. It does things that you don't think your computer can do. It does things that you don't think your computer should be able
170. 25:34to do. But if you just treat AI like it's going to make me more efficient at writing emails or more efficient at writing
171. 25:45marketing copy because that's what I know how to do and I'm going to upskill so I use AI to do the thing I already
172. 25:52know how to do a little bit faster, a little more efficiently. You're not actually learning what AI is. You're just learning how to take
173. 26:07you know, an entire factory and you're just using like the vending machine in the corner and you're like, "Oh, cool. I if I put
174. 26:17in a dollar, I can get out a a a Snickers." There's a whole [ __ ] factory that you have access to, and you're in the corner
175. 26:28getting Snickers out of the vending machine if you if you use AI like this. here's what I know how to do and I'm
176. 26:34going to use AI to make that more efficient. I don't like to use the term skill although there there's some skill
177. 26:45involved but I think skills are transitory. Skills are are related to tasks. I've got some skills. I can make a
178. 26:53PowerPoint presentation. I can make a word thing. I I got skills. And so upskilling sort of implies I just need
179. 26:59to learn new tasks. AI is going to do the tasks. But if you really want to level yourself up, there's two things that have to happen.
180. 27:14One is you need to not be so attached to your job, your current job, because all of our current jobs are about to change and get weird.
181. 27:25And so to the extent that you're tied to I'm a project manager and I do project management like this and I'll always be
182. 27:32a project manager. Well, at some point guy's just going to do it, do the tasks. Well, who but who do you have to be to
183. 27:41be a project manager? Oh, well, I'm organized and I'm thoughtful and I like to make sure things get done. Okay,
184. 27:46those things are valuable. Like how you show up is valuable, right? So, so the first thing to do is kind of separate your
185. 27:57identity of tasks equal value to I am valuable because I care about things getting done right and I'm passionate
186. 28:07about excellence and whatever are the things that make you a good project manager. Those become more important than the actual tasks.
187. 28:17And that sounds easy in theory, but if you've been doing those tasks for 30 [ __ ] years, what's your value now? If they're
188. 28:29valueless, the tasks a because the AI does them. That's some work to do. So part of this is a, you know,
189. 28:38a brain a brain thing, a being thing. Who am I? So that's part of it. And then the other part is understand
190. 28:49that the only way to really understand how AI might be relevant to you and what you care about is to just play.
191. 28:59Literally play. Go just go play with the tools. Play with tools that you think you should have no right doing. I'm
192. 29:06really a horrible musician. Go to Sunno and make music. You will be an excellent musician. [laughter] guitar players will disagree with you,
193. 29:16but you'll be able to make music that is actually ideas out of your head that you turn into a song in a way that you never
194. 29:24thought possible. If you're really good at music, go have it make a spreadsheet with very fancy formulas and pivot
195. 29:34tables on it. Have it go do some data analysis on on the local economic data for your little town. and blow your [ __ ] mind. Just
196. 29:44literally go play. So if you if you're trying to get into this, the the first step is just to understand that AI makes things possible
197. 29:57that you could have never imagined like never imagined. Like there's a dude, I think he was in France, not a scientist.
198. 30:12This is like six months ago. His dog got cancer, bad cancer, like stage 4 nasty. We're going to have to put him down. And this guy
199. 30:28used chat GPT to figure out how to do RNA sequencing. and he took a sample of his dog's tumor and pulled some [ __ ] DNA out of it
200. 30:41and figured out how to do some [ __ ] with crisper and he made his dog an mRNA vaccine for his specific tumor and shot it in him
201. 30:54and cured his dog of the cancer. That doesn't make sense. It doesn't make sense that we live in a world that
202. 31:01someone could just do that. But someone just did that. Scientists had chat GPT
203. 31:15design DNA sequences for new kinds of viruses, bacteria phages, which can't hurt humans, but they can hurt bacteria.
204. 31:28And they came they they they theoretically designed I think 256 new novel bacteria phages phages that had specific
205. 31:37properties and they narrowed it down to 16 and they made new bacteria phages and Brent Peterson told us that today in
206. 31:51office hours that one of them is so novel it's got such unique characteristics as a life form that it should actually be classified as
207. 32:02a new species. That shouldn't be possible. But that happened this week. So in an environment like that,
208. 32:16using AI to make your email writing more efficient is kind of laughably limited. And if all the jobs are going to change anyway, what if
209. 32:29you played enough with AI to to to like internalize, oh, holy [ __ ] I could actually kind of do anything I want.
210. 32:41I don't have to go back to school to learn X. I don't have to put in a bunch of years over there to learn Y. I can
211. 32:48just have one of these agentic platforms be really good at project management and I can suck at it and it will do the
212. 32:55project management for me. Yes. All of that. If we live in a world where your options are you can do anything.
213. 33:09Now there's only one question. It's very simple. It's not easy. It's in fact it's profound. But if you want to get good at AI,
214. 33:25there's one question. Uno, that is the most important question. What do I want?
215. 33:39If you don't have a good answer to that, then what you're going to do is you're just going to spin your wheels and
216. 33:45you're going to go play with these toys and you're just going to do like masturbatory AI crap. It'll be fun, you know,
217. 34:03but what do you want? What do you want more of? If you could do anything, what would you choose to do and why? Oh, well, I value
218. 34:13this. I value people. I value pets. I value fun. I I love making things. I love making things excellent. I love
219. 34:21making things crappy. Remember uh what's her name? Simone, the inventor chick who she would make all these crazy
220. 34:30like a like her glasses were dirty and she wanted them washed. So she'd make a hat that had like a hand that came down
221. 34:37with a bottle and it would squirt, you know, Windex on her glasses and then the hand would wipe them off. She
222. 34:44purposefully made crappy machines cuz she thought it was fun and interesting and it it was art and it was performance and it was science,
223. 34:58you know. And if you want the answer, like what you should do this weekend is you should
224. 35:12go buy my 49.99 prompting 101 course on structured prompt management for maximizing chat GPT.
225. 35:26You're missing the boat. That prompt course isn't going to give you the skills you think it is. What you're going to learn
226. 35:38in that course is like, "Oh, I need to ask for what I want." Yes, that's what it's going to teach you. Now, it's going to teach you a very
227. 35:49specific way to ask for that. But here's a funny thing. Some of the newer, more powerful, more powerful models actually get worse if
228. 35:59you give them highly structured prompts. What? Two years ago, highly structured prompts were the [ __ ] man.
229. 36:10I've got a prompt library of all of the best prompts in the history of prompting. $49.95 99 1494750.
230. 36:26By the way, I don't know who I'm making fun of when I when I do that character. It makes me laugh.
231. 36:40The thing to the thing to learn is who you are and what you want. and then just learn enough of this AI stuff to understand that
232. 36:52there's lots and lots of capabilities. And then how you end up filtering what you actually learn is back to you. Once
233. 37:00you understand what you want, what you want to do, then you can go, wait a minute, how do I use this AI stuff to pull this off?
234. 37:09Oh, well, I'm going to need something to help me do planning. Okay, which of these things is good at planning? And
235. 37:15I'm going to need some creative concepts. Okay, which of these things is good at creative concepts? And then you
236. 37:20will start looking for specific AI solutions that you learned about when you were playing and you'll be able to stitch together
237. 37:29some string of three or four tools that when you put them together to to try to do what you want, we'll blow your mind.
238. 37:37You'll be like, "Holy [ __ ] that's that's not supposed to be possible. How is that possible?" Like, that's where you want to get,
239. 37:49right? And you don't get there from a [ __ ] prompting class. All right. What other questions you got? Did that answer the question?
240. 38:01Is anyone still here? [laughter] This is more like a meltdown Monday. [snorts] Okay. Yeah, let's do some news items. Okay. Couple of things.
241. 38:12Um, if you're new to the AI salon or if you've been a member of the AI salon for a while and you haven't joined the mastermind,
242. 38:25you need to go check out the mastermind. So, in the top in the top of the spaces in the in the AI salon, there's there's
243. 38:32a a a space there that will tell you about what the mastermind is. The mastermind is a subscription area of the
244. 38:39AI salon that has remarkable people in it, remarkable courses in it, remarkable content in it, and we're doing a lot of
245. 38:49things in there that are very, very in innovative. Um the mastermind practice lab is a place where people are um
246. 38:57creating a daily practice and learning how to use AI in this more mindful way in this more um in this way that is attached to to kind
247. 39:09of who you are. Um learning something new is always challenging. So one of the things that that I personally use the practice lab
248. 39:20for is a place to just process that running something like the AI learning lab I have put myself in a position of
249. 39:33um knowing more than other people like you know at least you know on the surface and one of the ways that I use
250. 39:40the practice lab is to just make it okay to not know and make it okay to feel crappy about not knowing and just sitting in the ick.
251. 39:51Right. So, for for me personally, I hate not knowing. Um, and this is a place where I can do that. You know, some of the other things we're
252. 40:03doing in the in the uh mastermind um are the great repurpose and we're doing um workshops and courses on leadership and
253. 40:13mastery and um decoupling your identity from from your work and reclaiming your value, all the things that are part of
254. 40:22the great repurpose. Um a lot of that work is being developed in the mastermind. Um, and so people are literally changing their lives. So if
255. 40:33you haven't checked that out, go check that out. Okay, I want to talk about some news now. So commercial over.
256. 40:39Just take a few minutes. It's not bad. And listen, to be clear, the the cost of the mastermind is is ridiculously low. It's it's a cost, but
257. 40:54for what you get from it, it is ridiculously low. So, I would encourage you to go check it out if you haven't. Okay. Couple of things,
258. 41:03news items that I think are worthy of you paying attention to. One of the things that's been weirdly clear in the past three months
259. 41:16is that anthropic and open AI are really battling for superiority when it comes to these these frontier models
260. 41:25and everyone else is sort of, you know, second or third tier beneath them. For about a year, Google and Gemini were
261. 41:36were right up there with them. And in the past three months or so, maybe more, maybe six, in terms of like that frontier model,
262. 41:46Gemini has clearly dropped to a second tier. Now, Gemini does have some really incredible tools like the the Gemini Notebook. It
263. 42:00used to be called Notebook LM. It's now called Gemini Notebook. that remains a really strong application, but that's a
264. 42:07specific application that does very specific things and it doesn't need the most powerful model. So there's some
265. 42:14things that are that Google is still doing quite well. Their video models are really quite good. You know, notebook is quite good,
266. 42:23but like their agentic framework and their frontier models are just they're not great and no one's using them. So
267. 42:32yesterday, I think it was yesterday, Demisabus, [clears throat] who's the guy that founded Deep Mind and the brains behind
268. 42:41Alphafold, uh the the protein folding thing that he won the Nobel Prize for in chemistry. Um he is stepping down as CEO
269. 42:50of of Google DeepMind. So Google DeepMind essentially turned into Google AI and Demis was the CEO of that.
270. 43:00he is no longer the CEO of that. Um, while they're looking for a CEO, I I just saw a news item earlier that
271. 43:07apparently Sergey Brin, one of the founders of Google, is actually going to head that up until they find someone to
272. 43:14run it. So, Demis is out. there's there's kind of an implicit acknowledgment that Demis' approach wasn't working and how it was different
273. 43:29than OpenAI and Anthropic is those two companies have really leaned into the large language model as the model. Demis
274. 43:38was leaning into what what's called a world model which is based on video and you know physics and you know all sorts
275. 43:47of things that are beyond language and Google has all of the YouTube data you know 25 how many 20 21 years of
276. 44:00YouTube data right 21 years of videos they've got all that so they can they can make a world model and and the the
277. 44:08theory is that a model that understands the world is going to be a way more powerful than a model that just
278. 44:14understands words. I think long-term that's true. Short term it's cost them their lead. So there's big changes going
279. 44:21on there. Um I'm still bullish on world models. I don't know what it takes to make them, but it just makes sense to me
280. 44:28that a model that understands how the world works is going to be more powerful than [laughter] a model that understands
281. 44:34how language works. But where we are right now is the language is the thing and everyone's focusing on coding and
282. 44:41that's language. So um so that's an important an important shift and something to pay attention to. If they get this right,
283. 44:52Google's got the money and the smarts and the data to quickly be able to catch up, but they've lost their lead. Um so that's that. um
284. 45:07the government yesterday or maybe it was Wednesday uh came out with a little statement that said, "Oh, by the way, if you're a
285. 45:19Frontier Model Company, looking at you anthropic, looking at you Sam Alman, and your cute little open AI company, then
286. 45:26you're going to need to go on ahead and come on in on Saturday and let us, the government, review your
287. 45:33models. before you can send them live to the to the public. Okay. Okay. And then part of the [cough] rejiggering and manipulation that those
288. 45:43two companies were trying to do was they were trying to get the government to make open weight models, the open- source models
289. 45:53illegal [cough] because most of them are from China and they've been distilled from American models. So they said it's IP theft and
290. 46:02you should make them illegal. What the government said is OpenAI and Anthropic have to run their models by
291. 46:09the government before they can launch them and open weight models are exempt. So the open weight models are going to
292. 46:16keep coming. The fact that they just made open weight models exempt from that scrutiny um likely means that China is if they
293. 46:25weren't already accelerated, they're probably going to accelerate model development. and then anyone in the US
294. 46:29that's doing openw weight models um it's it's going to encourage um innovation here as well. So I think we'll start to
295. 46:37see lots and lots and lots of new models that are as good or potentially better than these frontier models because
296. 46:46they're going to be purposefully slowed down because people in the government have to review them for their safety.
297. 46:54How is anyone in the government going to know what the hell is safe or not safe in these models? The scientists that
298. 47:01built them don't understand what's safe and not safe in them. Anyway, [laughter] now if you're all freaked out about,
299. 47:15wait a minute, why is that a good thing? Can't people do bad [ __ ] with the openweight models? Yeah, they can. and
300. 47:23the white hat hackers. So, this is this is the way it's always been since probably the 60s or 50s. The new technology comes out, some
301. 47:32people try to do bad [ __ ] with it, and then there's the other people that try to stop the people doing bad [ __ ] with
302. 47:38good [ __ ] That's not going to stop. If the government locked down the openweight models, what would happen is
303. 47:46the hackers would do the bad [ __ ] and the the white hat people wouldn't have the best models to prevent those attacks.
304. 47:54So this is ultimately a safer route even though on the surface it feels more dangerous. Anyway, all right. Beautiful, beautiful,
305. 48:04beautiful, beautiful. What else was happening that was worth uh talking about? Oh, Teraphab. Let's go. Let's go
306. 48:10look at Terraab, shall we? [cough] This is crazy. Oh, wait. You know what I want to do? I
307. 48:26think I'm going to do it a different way. Hold, please.
308. 48:42By the way, Bionic is LM Studios new um agentic harness um where you can run that liquid model that that uh Brent Peterson was talking
309. 48:54about today on your local machine. Um, all right. Let's go to the X and then let's go to what did I want to show you? Oh, Terra
310. 49:10Fab. Terra Fab. So, this is Elon Musk's new pet project is called Terra Fab. Here, let me We'll do this as a
311. 49:51Notice all the Tesla semis. Look at that thing.
312. 50:18I don't know if the audio or if that was just a video overlay, but I'm not really sure what that was. >> Were was there no audio?
313. 50:27>> We could not hear the audio. >> God damn it. Really? >> Hang on. It took me a minute with you being quiet to figure out that there was
314. 50:39supposed to be audio because it was a very cool visual, but um >> Well, it it's really it's really just >> Can you hear it now?
315. 50:48>> You hear it now? >> Yes. >> Oh, that was it was user error. >> Anyway, it's it's all it's just it's all
316. 50:58just music. There's no narration to it. But let me let me show you something. scale of this thing here. It's a 100
317. 51:05million square feet. Um, this little thing right here is the Pentagon, 6.6 million square feet. Um, that big giant
318. 51:15Apple thing that we talked about is this teeny little thing right here, 2.82 million square feet. The Gigafactory,
319. 51:21the Tesla Gigafactory, which was this biggest manufacturing thing in history, is this little teeny thing here to the
320. 51:27left. So, so Terraab, this is a chip manufacturing plant, right? So, this is right now if you if you want high-end AI
321. 51:38chips, um they they only come from one country, from one company. Um, and so this has been revealed. I don't know how long. Let's see.
322. 51:52There's never been a building this large. It will be the most valuable building by far. He's he's calling it like basically like a city. Um
323. 52:03I don't know what the timing of it is. It's incredible, but look at the scale of trees. Yeah, this is the apple the apple ring which
324. 52:14is absolutely massive and it's just tiny in comparison. 100 million square feet. Anyway, um yeah, there's like 50 pentagons inside
325. 52:27it or something like that. Yeah, [cough] it's crazy. Um okay, what else was I going to share? I think that was it. Oh, uh just just
326. 52:39talking about this. So, um, the bottleneck the bottleneck right now for us to be able to do more and better
327. 52:50AI development is compute. It's just it is. And um one of the things that Musk said when he was talking about this was, "Hey, if if
328. 53:05TSMC, the company that makes all the chips that supplies the world, um if you think, you know, this is this is about
329. 53:12competing with you," he basically said, "We will buy as many chips as you can manufacture for the foreseeable future."
330. 53:20like like this what he's saying is that this thing at full capacity can't keep up with with the demand that's coming.
331. 53:28And so he's basically, you know, trying to bring this level of chip manufacturing to the United States,
332. 53:34which would be huge. Um but but we're this is we're in a we're we're in a we ain't in Kansas anymore. We're the the
333. 53:42scale of what we're about to enter is is really kind of kind of bonkers. All right, let's see. Mel, why did you say Gemini is second tier
334. 53:55now? I haven't explored it a lot yet, but I wanted to know about Gemini Google Workspace benefits versus our GPT or
335. 54:02claude. And then it seems like goo Google is incorporating a lot of their apps into Gemini plus power of Google
336. 54:09could be really powerful. Yes, I agree with that. Also, it seems like Gemini is better at knowing real-time button
337. 54:16configuration for when you're setting things up. Would love to know your thoughts. Okay, so there's there's a couple of different games.
338. 54:32There's a couple of different games being played right now. Mel, one of Google's advantages is also one
339. 54:41of its weaknesses, which is they have so many products and they have such complicated products that incorporating Gemini
340. 54:52well across all of them is essentially impossible. like it is still the case when a new feature comes out from Gemini,
341. 55:02I can use it on my individual Google account, but I can't use it out on Workspace. So, I can use the new stuff on my free
342. 55:10account, but I can't use it on my paid account because it's new and they haven't figured out all of the
343. 55:16permissions and, you know, how they're going to charge for it and all that [ __ ] right? So, so, so in theory, yes. Right. And if you
344. 55:27look at a product like notebook, um, like like Google notebook or the fact that you can use Gemini across Google
345. 55:35Drive and Gmail and all that stuff seamlessly, there are huge advantages there. The place they're falling behind
346. 55:43is the actual frontier model itself. So the the core intelligence capability that is going to be the thing that that
347. 55:54businesses transform themselves on. They have failed to launch Gemini 3.5 Pro. So they've been talking about it for a
348. 56:03while. It slipped for a month, then it slipped for another month. Now they're just talking about GPT or Gemini 4 is
349. 56:09going to be awesome. And and they're launching um what are they called? Their their little models are called not light. They're called
350. 56:22I don't know the the the teeny models. Not not Gemma. It's it's the it's it's the size of them. They're they're the uh
351. 56:31I forget Flash. Yeah, the Flash models. So So Gemini has been putting out these flash models. Hey, we've got a new
352. 56:38Gemini 3.5 Flash, which is our little small version of what the Pro version would have been had it launched, right?
353. 56:46So they're saying these small models can run on your local computer and they're really good and they're probably really
354. 56:50good but where they've lost the lead is in that big frontier model the core intelligence of of the big big models like like
355. 57:00mythos and fable for um anthropic and GPT 5.6 and GPT6 for open AI. Google doesn't have that right now. Um,
356. 57:15uh, SpaceX AI doesn't have that right now. Meta certainly doesn't have that right now, right? So, you've got you've
357. 57:22got three really big companies that are playing in the AI space. Google should absolutely be up there with these guys,
358. 57:30but something happened with the 3.5 training. I don't know if it was they were focusing on world models and not
359. 57:36LLMs and they just depprioritized innovating there. I like I don't know. I I I have no idea. But that shakeup, that's what that
360. 57:46shakeup's about. Um what I would do, Mel, is look really specifically at your use cases and at the actual
361. 58:01integrations and how those things are integrated into the products. Because you can take a lesser model and if it's well integrated into
362. 58:12applications. You don't need the Frontier model, right? Notebook LM is incredible and it's not using the latest models.
363. 58:23It's using their cheap models and it's using nano banana to make images and it's using Omni to make videos and you
364. 58:29know their music generation [ __ ] to make music and then it's it's using some sort of you know probably proprietary
365. 58:38rag you know retrieval augmented generation technology to take all of your data and allow you to search against them, right? Like that's an
366. 58:45incredible product, but it's but it's not one of these leading things that's going to be solving physics and solving math and you
367. 58:53know the thing where all the the big ass innovation is going to happen and then other products will trickle out of that
368. 58:59innovation. That's where Google should be leading and they're not right now. So So it doesn't it I'm not saying don't do
369. 59:06Google. I'm just saying in that game they're they're they're losing. They've lost the lead. Um, and I I would say at
370. 59:14this point they've lost being considered in in that class. There's only two right now. I would love to have them back up
371. 59:20there. I would love to have SpaceX AI up there. It at some point you would think with all their [ __ ] money meta should
372. 59:27be back up there, but they just they they really screwed it up on talent acquisition. Okay, [clears throat] good question. Thank you.
373. 59:42If this was your real first week paying attention to AI news, what's the one story you actually need to understand? H
374. 1:00:10I think it's the um I think it's the the the government exempting openw weight models from
375. 1:00:25from scrutiny that that basically they're letting openweight models be accessible by Americans. Um like on the surface there's nothing
376. 1:00:38sexy about that. But if they had done the opposite, if they had restricted access to those and you know green lit you know open AI and
377. 1:00:48anthropic making it easier for them them to get their closed models out there that would have locked something down. Um
378. 1:00:59because so and here's why I think it's the most important story. It will become very clear three years from now
379. 1:01:14that those people who have taken the time right now to understand what it means to have local data, data sovereignty, and to have their own
380. 1:01:24brain, the the LOL that Mel's going to teach us all how to do. Um, and she's doing it in this really simple way that, you know, once you set
381. 1:01:33it up, you can just use it with any old AI. You don't need a fancy setup. You don't need harnesses. like you can do
382. 1:01:39all that other [ __ ] but what she's teaching is like a really simplified version of this thing. Understanding that your ability to have
383. 1:01:52to have files that represent who you are, what you care about and value, and what your projects are, and be able to give that
384. 1:02:03to some AI tool and have the AI go work on your behalf. Those people are going to be incredibly powerful people three years from now.
385. 1:02:14And there'll be all sorts of tools to manage it and things like that. But like that understanding h happening right now
386. 1:02:22like within the AI salon like a lot of the conversations right now are circling around this idea of data sovereignty and
387. 1:02:29local models versus cloud models and wait who really owns my data and how do I manage this data and what actually is
388. 1:02:35this data because when you hear data like for when I hear data I just think like data science and I it just feels
389. 1:02:42big and nebulous and no we're talking about very specific speific files here. A file that describes who you are. User.md. You're the user.
390. 1:02:57There's a text file that says Kyle is an artist innovator that likes this and he runs the AI salon and he does
391. 1:03:05this AI learning lab thing and he's got this company Story Vine and he's written a musical called Sydney and
392. 1:03:12and this is how he thinks and this is what he likes and this is how he writes and this is what he doesn't like.
393. 1:03:18He doesn't like bombastic bloiators bloiating about [ __ ] while he bloiates about bloiators. That file
394. 1:03:31tells any AI here's who this guy is. And then there's another one called agent MD that tells the AI how it's supposed to
395. 1:03:41act on my behalf. And then there's this other file that's like, here's the [ __ ] I'm doing right now that I care about
396. 1:03:47that I want you to pay attention to for me. So those three really specific files is what Mel's going to teach her LOL on.
397. 1:03:56And and those are the, you know, those are kind of the core files that all of these agentic platforms, these harnesses they call them,
398. 1:04:06are using. And the people that are figuring out how to do that and understanding that that data is incredibly valuable
399. 1:04:18combined with openweight models that the nerds out there, they're going to take these really big models that you
400. 1:04:25can't run on your local laptop and they're going to distill them down and quantize them and do whatever the hell
401. 1:04:31they do to make them small enough to run on our little MacBook Pros or our Mac minis. And then we can have essentially all of
402. 1:04:42human knowledge in some, you know, slightly reduced form sitting on our little local computer with our files of our stuff,
403. 1:04:56working for us 247. So that's the one I think is most important. Um, I don't know. [laughter]
404. 1:05:17Mel, is your LOL tomorrow? I don't think it is. I can't see it in I can't see Mighty Networks while I'm watching this.
405. 1:05:22I don't think it is. Mary, I don't think she's scheduled it yet. Um, all right. What other questions we got? Got about 20 minutes left.
406. 1:05:44Isn't that terra fab thing insane? I saw a cool meme today. Let me see if I can go find this meme. Um, future versus reality.
407. 1:06:03Tesla
408. 1:06:26Yeah, I can't find it. There was this cool thing that had like um screenshots from movies of the future and then
409. 1:06:33screenshots for what Tesla has actually designed and is building and it's pretty cool. But like we're we're pretty Elon
410. 1:06:42is pretty much building shitty seen in sci-fi movies. [laughter] It's kind of trippy to have a uh you
411. 1:06:51know a trillionaire actually doing this. Okay. Are AI agents getting genu genuinely useful or did this week prove
412. 1:06:58we still do not know how to keep them inside the lines? No. I mean they still suck. Todd, happy Friday night. Checking in
413. 1:07:12from the Tahi Bluegrass Festival. Oh, that's so cool. [laughter] Everyone's watching the uh the Fiddler
414. 1:07:19and you're listening to the AI Learning Lab. It's Friday night date night and we're out smoking a Jay in the Bluegrass
415. 1:07:27Festival. Beautiful. What's happening? What's happening, Todd? Um, grievy gy gravy. Dallas is testing the Tesla cabs here.
416. 1:07:41Interesting. The robo taxis. Um, Denver just got we just got the Whimo um the weirdass looking vans with the with the
417. 1:07:52siren on top, the LAR siren. Um, but they're all like I don't know. They're they're all being driven by human beings
418. 1:08:00right now. So I maybe that's a testing phase that they're doing local testing for a given city that they pay drivers
419. 1:08:07for some amount of time. Um speaking of speaking of uh here's an unscrupulous uh uh thing that I that I heard that from Corey Doctro was on
420. 1:08:22Joe Rogan or or John Stewart I can't remember which one. I think it was John Stewart and he was talking about how
421. 1:08:42so when you're an Uber driver, you say, "Okay, I'm ready to I'm ready to drive someone somewhere." And the app presents you with a with a
422. 1:08:52job. And the job might be go to the airport, right? And they have things like surge pricing and and you know, if you're a rider and you
423. 1:09:04ride during surge pricing, you get charged $60 instead of $40. And it used to be that what the drivers got paid was
424. 1:09:12just some percentage of what Uber was getting. And if it was surge time, then the drivers would take surge time. And
425. 1:09:19then what Uber started doing was I don't know if it was randomly, but they would they would put an offer to a driver that, you know,
426. 1:09:33you're going to drive this person from here to the airport and instead of paying them $45, they would say, you
427. 1:09:39know, we'll pay you $35 for this. So, they would purposefully lowball them. Even though they were charging 45, they
428. 1:09:48would lowball the driver and say, "We'll pay you 35 for this." And then drivers who are desperate for fairs take that.
429. 1:09:55And then that established a new bottom for that driver that the next time they did it, they would lowball them again. What? We we
430. 1:10:05have another ride to the airport. Now this time it's $30. And if the driver takes the the shitty offer, they
431. 1:10:11basically just keep driving the prices down. So it's like the [ __ ] level of evilness. I mean, sure, Uber had to get
432. 1:10:22profitable, but it's just like, you know, you know, the drivers still have to pay for their own gas and they have
433. 1:10:27to pay for their maintenance and now they're, you know, they're taking jobs where they're not getting surge pricing.
434. 1:10:33They're just getting the base rate and then they're getting [ __ ] on the base rate. It's just amazing. Just amazing.
435. 1:10:40Lots of Jay's around here. That's hil hilarious. Yeah. And every everything's going to go to driverless anyway. So,
436. 1:10:46yeah. Oh, you have Whimo, too. Interesting. All right. Any other questions? Any other questions? Thoughts?
437. 1:10:56Thoughts? I got to drink water. I'm a little dry.
438. 1:11:20Who's got a question? If AI coding agents are already showing up quietly across open-source projects,
439. 1:11:30if AI coding agents are already showing up quietly, should repos start labeling what was written by humans, co-pilots,
440. 1:11:38or agents? No. That's a stupid question. I don't even understand that question. I mean, in general, no. Labeling [ __ ] AI can Okay,
441. 1:11:52I'll You want me to rant? Turn the question off because that's a shitty question. But you want me to rant? I'll rant.
442. 1:12:02Whether or not something was generated with AI is as irrelevant as if an article was written with Microsoft Word. It doesn't [ __ ] matter.
443. 1:12:17What matters is that a human being generates something and attaches their name to it or a company generates
444. 1:12:26something and attaches their name to it. And if what they generate and put in the world is a piece of horseshit,
445. 1:12:36then that company or that person should be held accountable for this steaming pile of poo that they've put in the world.
446. 1:12:47You can make crappy content without AI. You can make brilliant content with AI. If it's brilliant, it's brilliant.
447. 1:13:02It doesn't matter what tool was used in the chain of craft to create a thing. But if you are a creator of stuff,
448. 1:13:15company or individual, and the stuff you create is garbage, then [ __ ] you. But I don't care if you did the garbage
449. 1:13:26with AI or not AI. And I don't care if you did the brilliant thing with AI or not AI. Who gives a [ __ ] It's good or it's not. [screaming]
450. 1:13:40I'm blowing my vocal cords out. And we got to label this AI with the deep fake people. This one was made by AI and there
451. 1:13:54therefore you can judge it as less than because it's AI because someone just pushed the button and out came the thing.
452. 1:14:02You can use AI like that. Push a button, out comes the [ __ ] You know what that's called? Lazy. If you're a lazy piece of [ __ ]
453. 1:14:18that makes shitty work, AI lets you make more shitty work faster, I suppose. But the the generator is independent
454. 1:14:34of the quality of the work. I don't don't we don't need to label it. Everything will be AI in three years.
455. 1:14:49Even the [ __ ] pottery that you buy at the [ __ ] farmers market from the artisan potter, half of that is going to be designed by AI,
456. 1:15:01squirted onto a [laughter] platter with a 3D nozzle and baked in a kil. >> [laughter] >> You're going to be drinking from the
457. 1:15:16coffee mug going, "Hey, I wonder if this was designed by AI. I don't I shouldn't be drinking from this. I only want pottery with
458. 1:15:26fingerprints in it." Okay, then go find that. But if it's a nice mug and you can drink [ __ ] out of it and does it matter?
459. 1:15:42I mean, quite frankly, the potter that figured out how to use a 3D printer to make pottery with some sort of vision of how to
460. 1:15:52design it in a new and unique and interesting way is an artisan in a different kind of way. You can judge it. You can say Corey
461. 1:16:04Sandler pottery throwing her her mud on the thing and doing it with her hands is superior to someone that figured out how
462. 1:16:11to do it with a machine. But in the end, if you like one or the other, isn't it independent of how it was made? Now, if you want to support
463. 1:16:22Corey, then do that. But then the thing isn't just a cup, right? Then the thing is a Corey Sandler cup.
464. 1:16:36I have an infographic about Terraab size in the salon chat. Okay, I have to admit that that's an insightful rant, Kyle.
465. 1:16:44Thank you, Linda. I listen I understand I understand the instinct of wanting to label stuff AI. AI is dangerous. AI is this. AI is going to
466. 1:16:59lie. AI is crap. AI is stolen from us. Humans are better. Humans are better. Humans are better. Humans are better. If we label it, it's bad.
467. 1:17:11Unless it's not. >> [laughter] >> Where's Corey when we need her? [laughter] Cammy, let's see. Um, I'm pretty blown
468. 1:17:25away reflecting on the tools a few years ago versus today. There's still a lot of noise, a bit of jank, but man, the tools
469. 1:17:32are getting so good. I'm playing with GPT sites feature right now, and it's giving GHL AI Studio a run for its
470. 1:17:39money. I You know what's funny, Cammy? And welcome. I haven't seen you in a while. I haven't even um I haven't even
471. 1:17:47played with sites yet. Like like there's major features in chat GPT I haven't even played with yet. I've been playing
472. 1:17:53with other stuff. Um it's cool to hear that they're good. Um wild.
473. 1:18:07Uh I would like to have a human in the loop somewhere. AI slop that is not checked by anyone is annoying.
474. 1:18:14So, here's the thing, Linda. I I don't disagree with you. [clears throat and cough] And and in the end, I really do think
475. 1:18:23whoever put their name on it is responsible for what they put out there, right? If you [clears throat and cough] if you steal something
476. 1:18:33and, you know, lean on someone's trademark, there are trademark laws for that. Like, you can be sued for that,
477. 1:18:39right? So um so I think in the end people have to put their name on it. They have to be accountable. I do I do think that there is a
478. 1:18:57I do think that there is a possible future where there can be content created at scale that is actually really good
479. 1:19:08and and and it will be and there will be people that try to play that game. But we're already at the point where 47%
480. 1:19:17of the content on LinkedIn right now is AI generated. So even if it's generated from a human being um [clears throat]
481. 1:19:24and ostensibly paid attention to, you know, we're we're very very quickly going to saturate and we're going to get
482. 1:19:30to a point that 98% of content is going to be fully AI generated and a lot of it's going to be really good. The problem is
483. 1:19:42while the content scales the number of hours of human attention that can pay attention to it does not scale.
484. 1:19:56So if if this so if this is human attention all of human attention is this much and this is how much content if this is
485. 1:20:08human attention for a year we're going to quickly get to a point where here's the amount of content generated every day by AI and humans
486. 1:20:17right and we're probably there already but it's going to be like ridiculous like this is a year's worth of attention
487. 1:20:23of all humans if they watched eight hours a day times 8 billion people. This is how much attention they have and
488. 1:20:32here's every day how much content is produced. Like that's where we're headed. So, [clears throat] so even if we get to a place where those
489. 1:20:43content farms generate good content, the idea of mass media is essentially dead. And what it goes back to is community.
490. 1:21:00Being in a community like the AI salon having a circle of somewhere between 10, 50, 100, 150 Dunbar's number
491. 1:21:12of you know the amount of human beings a single human being can maintain some number of relationships with is about 150.
492. 1:21:24So if you think all of us have you know somewhere less than 150 people in our trusted circle those are going to
493. 1:21:34be where the valuable interactions happen and those are going to be where where the valuable content is exchanged
494. 1:21:42and makes a difference for people. Now, occasionally some piece of content or whatever, an app, whatever, you know, in one of these
495. 1:21:51circles may escape and strike a chord and you have a Billy Isish that just, you know, is so different that everyone
496. 1:21:59pays attention to that. I think you'll still have those moments, but but they're in a sea like it'll be much more
497. 1:22:06like playing lotto where, you know, your chances of winning that are just astronomically small. Like that's where we're headed.
498. 1:22:16[clears throat] Why can't I even see my post before before I oosted it to catch my mistakes? [laughter] Exactly, Vicki.
499. 1:22:33Point made. [laughter] No. Is that true? Corey and Corey and Vicky Baptiste are partnering on 3D pottery. Okay, this is so cool. THIS IS
500. 1:22:48WHAT I'M TALKING ABOUT. This is what I'm talking about. Vicki passionate about making [ __ ] and technology and doing cool new things and
501. 1:22:59has a whole 3D printing business. Corey Sandler really good with AI but like is a potter and an artist. Those two
502. 1:23:07collaborating. That's going to be so [ __ ] cool. This is what I'm talking about. We're entering a new world.
503. 1:23:20A new world. New things are possible. Those kind of collaborations are going to create something that Vicki couldn't
504. 1:23:28have done on her own. Corey couldn't have done on her own. Together, they come together. They do this crazy ass
505. 1:23:32thing. It's going to make some kind of new thing possible. Now, is it going to resonate and take off? I DON'T KNOW. WHO KNOWS? THIS IS
506. 1:23:39WHY YOU MAKE [ __ ] and put it in the world. Sometimes you make brilliant [ __ ] and no one cares. Sometimes you make mediocre
507. 1:23:47[ __ ] and everyone cares. I don't know. Hopefully they make brilliant [ __ ] and everyone cares. That's really cool. Oh,
508. 1:23:54it's not true. Oh, damn it. It should be. You should reach out to Corey Vicki. I think that would be a great idea.
509. 1:24:03Ah, I got all excited there. Damn it. Damn it, Vicki. [laughter] Gifts and guitar, Yamaha drum, Sennheiser, microphone. H.
510. 1:24:14All right, let's see. Should we have a co-op for the 15minute pre-release of international affairs announcements? I
511. 1:24:26don't even know what that means. I love Billy. Yeah, I know. I mean, Billy Isish is a really interesting one, right? Like
512. 1:24:32it's like it's like what is it like what was the unique combination of time in history level of talent,
513. 1:24:47how she she was raised, the connections that she had. Like what was that rare combination of just kind of an infinite number of small
514. 1:24:59improbable things that added up to she's just like bang Billy isish my nozzles won't accommodate her clay [laughter]
515. 1:25:15we're crazy but not that crazy. Um weekend homework. All right let's do some weekend homework here. Let me think about this.
516. 1:25:26If you haven't played with codeex or clawed code with the with the agentic harnesses, um those are the two I'd start with
517. 1:25:35because they're just they're just there. You can just use them. You download the app and you kind of prompt it like you
518. 1:25:41prompt chat GBT and just watch what it does and then you're like, wait, why doesn't it know who I am? Because you
519. 1:25:45didn't tell it. You got to figure out these, you know, brain files. So, if you haven't played with that, I'd say that
520. 1:25:52would be something good to play with. Um,
521. 1:26:11if you're super nerdy and really want to geek out, downloading Buzz, Buzz.xyz is probably something to nerd out on. If
522. 1:26:19you're not, it's way too early. The chat is doing you dirty. I hate this character. [laughter]
523. 1:26:34You guys are hilarious. Um, let me see if there's one other thing that could be fun for weekend homework.
524. 1:26:42So, play with a harness if you haven't. If you want to geek out and just learn something new, Buzz is is worth your
525. 1:26:48time. Go go go make an agent. Like if you've got if you've got a local large language model running on a machine,
526. 1:26:58go into Buzz and make yourself an agent that uses that local compute. That could be something interesting. And see if you
527. 1:27:06can share it with someone. That could be fun. And then here's here's another one. Go to notebook.
528. 1:27:21Gemini notebook. There's a bunch of new features in Gemini Notebook. Um I don't know what the URL for Gemini Notebook is. Hang on a sec.
529. 1:27:31[clears throat] I think it's still notebooklm.com. Gemini notebook. Let's see what they Notebooklm.g Google still blog.g Google make use of
530. 1:27:49all right notebooklm.google. I guess they'll change it at some point. Go upload a bunch of documents
531. 1:28:04into into Notebook LM and then just play with the the outputs. The outputs are all new now. There's all sorts of stuff there.
532. 1:28:13Oh, I hate chat. I can't see. I can't see [ __ ] Yeah, I know. Are you Vicki? Are you on Are you on a mobile device?
533. 1:28:23I was typing everything backward. It was It was typing everything backwards for me on Wednesday. Oh, use voice use voice
534. 1:28:31texting. Okay. All right. That's kind of a cool idea. Chatter chatter chat not character chat not character [laughter]
535. 1:28:44comments have been noted. Um we are talking to we are getting comments back to Streamyard who is working on this in
536. 1:28:53real time you know as we're building it. So um I don't know if they'll listen to us but we are getting comments to them.
537. 1:29:02But I miss what he's saying Danielle. Oh, yeah. When you're when you're voice texting. Yeah, that's a good point. All
538. 1:29:07right. Anyway, listen. Um, appreciate y'all being here. I am going to go. Have a beautiful, beautiful weekend. Um, and
539. 1:29:14uh and let's see what else. I think that's it. Have a good weekend. I'll see you Monday and we will uh we'll do it
540. 1:29:21all over again. All right. Peace out. Have a good one, everybody.