5/29/2026 - Vibe Coding vs. Regular Coding: How Anyone Can Build Apps with AI

Transcript

1. 0:05[music] [music]
2. 0:22[music] [music] >> Good evening. Good evening. Friday night
3. 0:36date night. Welcome everybody. Welcome to the AI learning lab here in the AI salon. My name's Kyle Shannon and uh
4. 0:46we are off. It is uh if you got a date tonight, awesome. If you don't have a date tonight, I'm your date. We'll have an evening of it. Um
5. 0:54Sharon has joined the chat. Fantastic. Uh if you're looking for the chat, so if you're watching here and you're looking
6. 1:00for the chat, the chat is sitting underneath this thing. And if you want to pop this window out, one of the cool
7. 1:06things you can do is if you open up two tabs in the AI salon and one of them is this, you can pop this window out right down here.
8. 1:15Pop it out. And then in the other tab, you can look at stuff in the AI salon. You can you can be multitasking. Um we are
9. 1:25neurodivergent friendly around here. So, I don't expect you to even pay attention to anything I'm saying. I expect you I
10. 1:32hope you are multitasking. I hope you're trying to figure out I don't know, teach yourself quantum theory and I I hope you're
11. 1:41inventing zero-point energy as we speak while I'm ramble on. >> [laughter] >> Danielle's in the house. Danielle's son graduated high school
12. 1:51today. Congratulations. That's a big deal, big milestone. Congratulations. And as I recall, Danielle, your son is
13. 2:00not vehemently against AI. Is that correct? Maybe not into it, but but but willing to at least play, right?
14. 2:08Um my guys my guys are 26 and they're not they just don't care. They're not vehemently against it. They just don't care. Um
15. 2:21if you are Now, wait. Hey Brandon, I have a question for you. Is the is the interface of the AI Salon the the left-hand navigation? Is that
16. 2:32over here for you or is it over here? A or B? My right, so over here. Okay. Okay, so if you're in the AI Salon now if you if
17. 2:47you click out to other spaces, you'll lose this unless you do the two tab thing. But anyway, you want to check out
18. 2:53the different spaces in the AI Salon. If you're new here, uh toward the top is the events tab, so you can see all the
19. 2:59stuff going on in the salon. There's lots of different spaces um of people doing different things. There's a
20. 3:04community feed. There's a community chat. There's there's this AI Learning Lab and a regulars channel.
21. 3:11Tonight he's 17 and he uses AI all the time. Terrific. Fantastic. We have one young person that hasn't been destroyed by our our educational
22. 3:21institutions uh about AI. Um the uh the the the graduating ceremonies are getting more and more painful as the people
23. 3:32giving the the commencement addresses are trying to navigate talking about AI or just avoiding it altogether. >> [laughter] >> It's crazy.
24. 3:44Um anyway, if you have questions about AI, pop Does Sleep Town is multitasking? It absolutely he Um some people take sleeping pills,
25. 3:53other people watch the AI learning lab. I'll put you right down. It's I've got a droney voice. I just go on and on. Some
26. 4:01people are just like, does he ever stop? No. No, that's the whole thing. The whole thing This is it. This is it. You get the fantastic
27. 4:11backdrop that looks like an actual home office, but it's not. This is all CGI. This is all AI generated. My hair I'm bald.
28. 4:21This is all AI generated. I'm wearing like a wife beater t-shirt right now, but it looks like I'm wearing some some some some threads.
29. 4:34I have 20/20 vision. Why do I put these on? TO LOOK HIP. >> [laughter] >> IF PEOPLE ask you what you do Friday
30. 4:48night, you could tell them I watched an AI avatar. At least going to sleep to >> [laughter] >> Kyle as a natural remedy. It's It's
31. 4:57semi-natural. >> [laughter] >> I'm I am I am fueled by Kool-Aid. Did you ever Did you all see the the Tom
32. 5:05Segura and Bert Kreischer Kreischer episode of of Two Bears, One Cave? Where Tom Segura discovers that Bert's drinking 64 oz of
33. 5:16Kool-Aid in the morning? >> [laughter] >> It's Look up Bert Kreischer Tom Segura Kool-Aid if you want to see a If you
34. 5:24want to see two two grown men just lose it. Uh, and you can't not laugh with them. It's absolutely hilarious. Ah, [sighs] okay.
35. 5:35Let's get the show on the road. So, if you have any questions, pop them in the comments uh to the right. Um, Brandon's got some
36. 5:41that have come in throughout the week, so he'll pop those up. Um, the The idea here is, you know, I'll
37. 5:47try to unpack what I can on office hours today. Um with a lot of people sort of I asked people to talk about what they do outside of
38. 5:58sitting at a computer. Um a couple of people are like, "I'm just sitting at the computer." Um there are a fair amount of people that had
39. 6:05actually really nicely balanced lives. >> [laughter] >> The weirdos. And they and they had them before AI and
40. 6:14then during AI and then now. They're they're they're just balanced. You know there's something else going on
41. 6:21there. There's something There's something deeper in that river that we probably don't want to know anything about. And
42. 6:27then there are other people that are kind of newly discovering unplugging from AI um as as a as a way to, you know, keep
43. 6:35the batteries charged. Mostly balanced life. I'm not quite there yet, but getting closer. That's great. That's beautiful. >> [sighs] >> Um
44. 6:47while we wait for some questions to come in, I um next Tuesday is AI Salon presents, so mark that in your calendar.
45. 6:57Uh or or even better, go to the events tab up here and RSVP for the event. We've got Marta um I'm going to butcher her last name.
46. 7:09Cichocka. Marta Cichocka. And she's going to be talking about the capacity gap. And um she just got back from leading a
47. 7:20vision quest in Moab, Utah and um doing some remarkable things here in the states. Um her plans changed, so she thought she
48. 7:31was going to be in the states when she was doing AI Salon presents, but she's actually back in Poland now.
49. 7:37So, Tuesday night when when it's 5:00 for me, it's going to be like 1:00 in the morning for her and she's still doing it. So So my uh
50. 7:46my deep appreciation for her, but it should be a really powerful evening. So So please join. Um my 10-year-old son uses AI, too. I
51. 7:56think I encourage my boys to use it all the time if they have questions or want to want to learn about something. That's cool. That's really
52. 8:04cool. I laser cut some designs after work today to get my charge. Very cool. Love that. Nothing like a good laser cutter. Beautiful.
53. 8:16All right. Ah, okay. Cool. Um Can you pop that question into the into
54. 8:30one of the uh into the pop-ups there, Brandon? Yeah, that's a good one. Sammy is my daddy. Okay. What's the
55. 8:40actual difference between vibe coding and regular coding with AI? So vibe coding is a term that was coined
56. 8:48by Andrej Karpathy, who's one of the co-founders of OpenAI. He just accepted a job at Anthropic, probably
57. 8:57millions of dollars in the making to get him over there. Um he's a really smart guy. He's the one that coined the term vibe coding. And
58. 9:07what what he was talking about was things like lovable and uh cloud code and cursor. I think it was cursor he might have been using when he
59. 9:18coined that term. Um OpenAI has Codex, Perplexity, uh or love there's lovable, um and and what these platforms do is you can just talk
60. 9:32in words, "Hey, I want a program that does X." And so vibe coding is just the idea that you're never really looking at the code,
61. 9:40right? You're just you're just assuming that it's doing good enough, and it makes a thing, and then you're like, "Hey, that thing
62. 9:48doesn't work." And then it goes, "Oh, what a brilliant thing for you to point out for me. Let me go fix that." And
63. 9:54then it'll go try to fix that, and then it'll break something else, and you're like, "You broke that other thing." It's like, "Ah,
64. 10:01how wonderful of you to comment on my shortcomings as a coding platform. Let me go try to fix that." And it'll fix
65. 10:07that, and and so, vibe coding is literally just vibing along, and just not worrying about the code. And Karpathy just talked about this
66. 10:17moment where he went from it was kind of an ego detachment thing that he had, where he went from this ego
67. 10:26attached thing, like, "I've got to check the code because I'm a coder." to realizing that it doesn't really matter.
68. 10:33And he And the The amazing thing about coders that vibe code is that if it goes off the rails, they can actually go in
69. 10:40and look at what's going on, and potentially fix things, or, you know, guide it with a little more, um, with a little more precision.
70. 10:50That's not necessarily an advantage, right? Sometimes just asking a simple thing, like, "Give me a game that stuff drops from the sky."
71. 11:00is all the instruction that it needs, and if you know too much, then that can mess it up. Regular coding using AI,
72. 11:08it Think of that more like auto complete, right? Like, I've got an idea for a program. I know that I want to
73. 11:14create, you know, this particular component that's going to make some API calls, and do that. So, I'm going to use
74. 11:20something like ChatGPT, or Claude, or or whatever, Grok. Um, I'm going to I'm going to tell it I need some code that
75. 11:28does this, and I maybe I paste in some other code, and I say I need an API call that calls this thing. And and then I'm
76. 11:35going to take that chunk of code and I'm going to copy and paste it into my development environment and right. So, it's it's
77. 11:44Think of rather than auto completing lines of code, you're using AI to generate blocks of code, but then you're still manually kind of
78. 11:51taping those chunks of code together. That's not vibe coding. That's just more like AI assisted coding. Um Where we're headed,
79. 12:07it gets even trippier because vibe coding right now, if if you use something like lovable, you log into it, you say you want it to
80. 12:19do something, and then it shows you something, and then you give it feedback, and there's this back and forth.
81. 12:25But even lovable just added these things called sub agents, where you can give it a goal, and you might be interacting
82. 12:32with whatever it's producing here, but it might generate five different sub agents that are in the background off doing their own thing.
83. 12:40And then it's got an orchestrator agent that is going to coordinate all that stuff. And where we're moving is you might just
84. 12:48give something a goal, and it will produce code for you. Um At the Google I/O conference, Google demoed a version of search. This is wild.
85. 13:03Google demoed a version of search where it recognize it recognizes patterns in your search. So, the example they gave was a guy like
86. 13:14he was planning a trip for his kids or whatever, right? You know, to a water park. And he had done a couple of searches on
87. 13:21hotels and water park and activities and whatever the the stuff, right? And at some point it says, "Hey, it look
88. 13:28it looks like you're you're planning a trip. Do you want me to create a trip planner for you?" And then the guy said, "Yes." And it
89. 13:36just started writing code. So, that's a case where the context of what someone was doing in search became the prompt. They you know,
90. 13:48a person who's just searching for stuff wouldn't even necessarily know that they're prompting an AI tool. They're
91. 13:54just searching, and then the tool says, "Hey, would you like me to make you a trip planner based on the stuff you're
92. 14:00looking at?" And someone goes, "Okay." And then and then just code starts happening and then I'm like, "What's that?" And then pling
93. 14:10there's a trip planner. And you can click on stuff and it'll be automatically connected to your accounts
94. 14:16that are connected to your Google [ __ ] "You want me to turn that into a spreadsheet for you?" Sure, okay. And then off it goes.
95. 14:25So, we're we're we're moving into a place like I think vibe coding we're going to look back on lovingly as you
96. 14:31know, vibe coding was kind of like hand coding was 3 years ago. And then we're just going to be living in this world where you just talk to
97. 14:38[ __ ] and it goes off and does stuff. So, we'll see. All right, got some more questions here. Let's see. There's been a lot of
98. 14:48a lot of doomer news regarding data center water use and specifically water system damage. Thoughts on having this
99. 14:55conversation with people who are concerned? Um Yes. >> [laughter] >> Um
100. 15:19I don't even I don't even know where to begin with it. The amount The amount The demand for for compute
101. 15:37is about to go off the charts. And it's largely based on what I just talked about where if you if human beings are using tools
102. 15:48like lovable or chat GPT or Claude, where it's a human being giving a prompt, getting an answer, going back
103. 15:55and forth, they're going to use a certain amount of tokens, right? Of of compute. A certain amount of cycles on the chips up in the cloud.
104. 16:06As people start shifting over to these agentic platforms that that actually operate 24/7 and and go decide what they're going to
105. 16:15do on their own, you might have someone who goes from you know, I don't know, 10 million tokens a month to half a billion token
106. 16:23tokens a month. Without them as a human being, they're not doing anything really all that different. They're just saying like I
107. 16:30want you to go do stuff, not really aware that this thing's about to go do lots of different things. And and those
108. 16:36tools within corporations are going to dramatically spike um compute demand in those, you know, in those companies.
109. 16:48So, we're all we're already seeing the frontier model companies running out of compute. So, so at the beginning of the year, Claude
110. 16:58had put out Claude 4.6 and Claude 4.6 was awesome. And I know I haven't really answered the water question yet. I'm I'm circling
111. 17:05back to it. I just want to give you some context. Claude 4.6 was awesome. A bunch of customers from OpenAI came over to
112. 17:13Claude and they said, "This is the best. This is the bee's knees. And then Anthropic launched Claude 4.7 and they
113. 17:21made it dumber. They they turned down its ability to think think long. Um and they did something in that model
114. 17:27that made it not as great. And and basically the Claude Opus product just went to [ __ ] and people started really complaining about it.
115. 17:37And it it got bad enough that they actually cut a deal with SpaceX to rent or lease their Colossus supercomputer cuz SpaceX has already
116. 17:48built Colossus 2. So they had this Colossus 1 com- compute center data center that Anthropic now uses to meet the demand. And so Anthropic's
117. 18:03get getting good again. If the demand goes even 5x what it is right now, that won't be enough compute. And then this is
118. 18:14going to happen across all the major frontier model companies, right? So [clears throat] as a as a country if we want to keep up
119. 18:23with what what the rest of the world is doing with AI, we need more compute centers. We need more data centers.
120. 18:32The speed with which they're trying to put these things in the ground, they're they are not designing them for
121. 18:39non-ecological impact. So they're so they're designing them essentially crappily. >> [laughter] >> And and just trying to
122. 18:48you know, rush them rush them into the system. So so that part's bad. And I think I think holding those companies accountable to
123. 18:58not just, you know, be like 1970s super polluters is actually a valuable thing. My fear is that the the dialogue about
124. 19:09data centers right now isn't it doesn't do our water good and let's let's have a conversation about how we meet this compute demand
125. 19:18um and do it in a less destructive way. It's just like [ __ ] it. I want no data centers in my in my community. Related
126. 19:25question on screen. Can we figure out how to build data centers in the ocean? So that's that's So this is where I was going.
127. 19:31So there's a couple of things. One is data centers can be built more responsibly with things like closed-loop systems
128. 19:39where you you put all the water into a system. First, it's a closed-loop system and then you're not sort of pulling in cold
129. 19:46water at the top and putting it back into the environment on the other end, which is what I think a lot of the
130. 19:52current ones do or they do it down in the ground and back up and it's [ __ ] up groundwater. Um but if they did if they do a closed-loop
131. 19:59system that might be less. There's two other options that are potentially very viable. I met with um with an entrepreneur the last time I
132. 20:11went to Washington D.C. One of the entrepreneurs I met with with was this woman who's building ocean-based data centers
133. 20:18that she's building on the on the ocean floor and they're basically using the the cold water of the ocean for the cooling um and
134. 20:28and then I don't know what they're doing for the power. I think they're they're putting them in oceans that are near power centers
135. 20:35near near the shore. Um so I think that's how they're doing it. And then I I saw another ocean-based
136. 20:41one that's doing them on the surface and they're basically doing big floating data centers where again they're using
137. 20:48the ocean water for cooling and they're using wave motion for energy generation. So those are two things that are that
138. 20:57are actively being funded right now and so I think that that's a very much a viable option. The third option and the one that I
139. 21:06think ultimately becomes where most of the compute goes is into space. And so Elon Musk is big onto this.
140. 21:14There's a There's a a company They might be called Planetary. I forget what they're called. Um where they've designed data centers that
141. 21:22basically sit in orbit. So So imagine a you know, a bunch of H100 chips, you know, supercomputers sitting on a rack
142. 21:32floating in space with big giant solar sails that are capturing the energy from the sun. It's 24/7. Right? So you don't need batteries
143. 21:42because you have the sun 24/7. And then space is cold. You have to have special manifolds for for dissipating things in
144. 21:49space, but they figured that out. They know how to do it. And so Elon Musk with Starlink has proven you
145. 22:00can get lots of things into space, and space is huge, right? And so So I think that what starts to happen is
146. 22:06a lot of the compute will go to space. And so the data center problem remains a problem. There are alternatives. There are ways
147. 22:18to do land-based data centers more responsibly. Um and there's urgency. So >> [laughter] >> So So I you know, I think what you're
148. 22:30going to have is just a series of This is probably the largest um infrastructure investment by the private sector in history is my guess.
149. 22:42And so you're going to have people like you've got uh Kevin O'Leary has bought 40,000 acres in Where is it? Idaho or I don't know,
150. 22:51Wyoming, somewhere up there. People are pissed off about it, but he's like, "We need these data centers. Like
151. 22:57we need this compute." And so I I just think you're going to have battles, and some communities are going to kick them
152. 23:02out. Um some are just going to have them forced in. Some are going to do things responsibly and get them put in in a better way.
153. 23:09One thing that I would love to see is if those data centers donate a percentage of their compute capacity to the community.
154. 23:20Right? Put them in in a responsible way, but also donate that compute capacity to that community. So that community has
155. 23:29you know, AI capabilities. The interesting thing is well, Nvidia and home builder Pulte Group are partnering
156. 23:34yeah, an energy startup called Span to install to install mini AI data centers on your on your house. Right. Yeah. And this is
157. 23:44and what's what's interesting about that what's weird is this. So they install a quarter of a million dollars of AI compute on the side of
158. 23:53your house or like in front of your house. Which I assume would be a risk for theft. >> [laughter] >> If If the thieves catch wind of you
159. 24:05know, all those white boxes have a quarter of a million dollars in them, you know, I don't know if they're going
160. 24:10to give you uh round-the-clock security. >> [laughter] >> Um Oh, that's Yeah, that's the other thing. We saw that too, Brandon. Um
161. 24:21But But what they're doing is you pay them $150 a month, which I don't quite know why you pay them. They're paying um
162. 24:34all of your electric bills and all of your water bills. And then they're using your spare electric capacity to to do these. So
163. 24:41they're doing distributed data centers using the spare capacity that that's being generated, billed for, and just
164. 24:47not used. They're using that spare capacity and distributing data centers across neighborhoods, basically. Um
165. 24:56They're also paying you like 2,100 bucks a month or like 12 grand a year or something like that. I don't know. Some amount per
166. 25:04per year. So, you can actually make money by having one of these things attached to your house. Um I haven't seen anything that they're
167. 25:11going to give you any of that compute. Like I would say, you know, pay my electric bill and pay my water bill and then give me
168. 25:19an H1, you know, a fractional piece of one of those H100s to do my own token generation. Kelly Henderson, water is is
169. 25:27almost the least impactful of all the resources of data centers will use, but there it's easy for people to reference,
170. 25:33so it makes sense to the consumer. The real worry are the rare earth matter minerals, the semiconductors, heavy
171. 25:38metal, copper, nickel, silver, and lithium. Yeah, well, the Here's the Here's the thing that I think is
172. 25:52a thing that we should all pay attention to. The demand for AI compute is going to skyrocket.
173. 26:07When demand skyrockets and supply is too low, prices go up. So, our 20 bucks a month that right now effectively buys us unlimited AI
174. 26:18compute, we can just use ChatGPT whenever we want, we can use their biggest models, their latest models.
175. 26:25That's going to likely start to come down. We're not going to have as much access. It's going to get more and more
176. 26:30expensive to use those tokens. The chips themselves, because of the supply-demand things, are going to get more and more and more expensive.
177. 26:42If you're thinking at all about buying a high-powered computer for your home to run your own local AI, now's probably a good time to do it
178. 26:52because I think the cost of computers is going to go up. >> [laughter] >> I think the cost of tokens is going to
179. 26:58go up and anytime you have to rent someone else's chip to generate your tokens, um it's going to get more and more and
180. 27:05more expensive. So, so at least start learning about what it looks like to run local large language models.
181. 27:12When does the democratized rug pull happen? Well, I assume you're you're sort of making a reference to um to kind of the NFT thing. I
182. 27:24The the thing about an NFT rug pull is that's where you artificially inflate the price up, you get all the people to
183. 27:31come in and invest their money and then you crash the stock and take your profits, right? Um I don't think that's
184. 27:37quite the right thing about what's going to happen with AI. I think the AI thing is the demand for compute is going to go
185. 27:45way up and then these big companies are going to have to make bets. Like like Google, this is [ __ ] bonkers.
186. 27:56And it's a really subtle it was a small story no one's really paying attention to it. But Google is right now today selling
187. 28:04futures on AI compute. So, right now today you can buy through Google Cloud one, two, or three year prepaid
188. 28:16or pre-reserved, right? You're going to commit to a certain number of tokens per month or compute per month, whatever it is, compute cycles,
189. 28:25um for a for a hugely discounted rate. So, there are companies right now locking in access to compute.
190. 28:34There are companies right now that are don't even know what's going on with AI, that are not locking in compute. So, so
191. 28:41what's going to happen is you're going to even even with big companies, you're You're to have some companies that have
192. 28:46a huge advantage over these other companies because they have the compute. And then these guys are going to be
193. 28:51like, "Well, we want to buy it." So, there's going to be arbitrage markets and and just all sorts of crazy ass stuff going on.
194. 28:58Mr. IT says hi. Beautiful, thank you. Um and I'm I've There's lots and lots of questions down here. Oh my god,
195. 29:04commodities. I know. I know. Glad it's Friday. Um Okay, I have a question. If you want to learn how to use an AI app,
196. 29:16would you ask ChatGPT to explain how what you were wanting to do and best tool? I'd love to make a rudimentary website for my art.
197. 29:27Okay. Well, >> [laughter] >> this is supposed to be an AMA. Ma- Mary, have you um have you played with lovable? Just like right now,
198. 29:43right now, go to lovable.dev and say that Well, a couple of things. You could go take a screenshot of of a website that
199. 30:00you like and you could upload that image to lovable. And then you could say, "I'm an artist. I want to make a a portfolio
200. 30:08website. Here's a style that I like. Go make my website." And no more than that. Picture, "I'm an artist. I want a portfolio website
201. 30:19inspired by this one. Go." And just watch what it builds. So, the answer is yes, you can ask ChatGPT to tell you how to use one of
202. 30:29these programs, but something like lovable, you can just go ask it to build you a website and it will probably build
203. 30:35you something that is shockingly better than you expect. So, go do that right now. >> [clears throat] [laughter]
204. 30:42>> And then yes, and the and the the other thing is this, like when I when I decided I wanted to install Open Claw, >> [clears throat]
205. 30:51>> I opened up two windows. I had the terminal window, which is where you type in all the command line [ __ ] which I hate,
206. 30:57and I had Claude or Chat GPT open here. And then I said, "Hey Claude, go find the documentation for Open Claw
207. 31:06and tell me how to install it." And then it went and did that. And it said, "Type in this thing in your terminal window thingy." And I would
208. 31:16copy and paste it, and I would paste it over there, and hit squirt, and it would do [ __ ] And then it would
209. 31:21give me an error, and I'd say, "That screwed it up." And then it would say, "Well, go do this." Now,
210. 31:29if you use that little setup like I just described like I use it, I didn't learn [ __ ] Because I didn't actually have any
211. 31:36interest in learning how to set it up. I just wanted it working. Which is it's the absolute worst way to do it.
212. 31:44Um so consequently, I didn't really learn any of the core concepts of Open Claw. I have them kind of conceptually,
213. 31:50but I don't have them super clear. Um but you can absolutely use any AI tool to teach you how to use any other tool.
214. 31:59It's it's kind of stunning. You can also just take pictures of like literal pictures of you know, what's on your screen into
215. 32:07Chat GPT and go, "What What is this? What does that mean?" And it'll tell you, and then you can go deal with it there. So, that's
216. 32:15absolutely a way to do things. Okay, Camp Catkin. I can't imagine people being okay with mini data centers
217. 32:21on their houses, and neighbor neighbors being okay with it. HOA takes on a whole new meaning. Well, I don't I it's not
218. 32:27really a mini data center. I mean, it's they effectively look like little mini air conditioning units. And they probably sound like mini air
219. 32:35conditioning units, right? They probably have exhaust fans. I'm sure they have local internalized cooling.
220. 32:42Um and it's just like another utility box sitting next to your house. So, I don't I don't think that these things are going to be
221. 32:49you know, like like massive data centers with big industrial fans, they make a lot of noise because they're big industrial, right?
222. 32:58Scale things. But, if you're just putting a quarter of a million dollars of chips is probably what? It's probably 10
223. 33:06chips. You're You're putting 10 GPU clusters, which are probably like this, right? So, they probably got 10 of
224. 33:15them. They probably got five stacked on each side with some cooling [ __ ] around them. And then they're tapping into your
225. 33:22they're tapping into your on the on the inside of your um meter, right? So, so your your your house is pulling what your house is pulling.
226. 33:34Your house is using usually about a quarter of what it's pulling. So, you're paying for what you're pulling, but you're only using about a
227. 33:41quarter of that. They're taking that excess that is already paid for, and they're using it for that data
228. 33:48center. So, I don't I don't imagine that neighbors would have a problem with them. Um what is Google charging for prepaid
229. 33:56compute? I've no idea. I They I know they say at a massive discount. So, maybe it's 50%. Okay. So, if you're new here, welcome.
230. 34:11Welcome to the AI Salon. If you have not introduced yourself to the community, um there's a couple of things you can
231. 34:18do. One is you can take this live video window, you can pop it out in the lower uh right-hand corner. There's a pop-out
232. 34:25thing. You can create a new tab. So, down here you can pop out this window so you can keep watching this,
233. 34:31and you can create a new tab and go to the AI Salon. One of the top spaces that you'll see is introduce yourself. If you haven't
234. 34:39introduced yourself to the community, go do that now. Just say, "Here's who I am. Here's what I'm doing."
235. 34:45If you know nothing about AI, you're like, "I really don't know anything. These people are going to make fun of me." They're not.
236. 34:51Um if you're in here and you haven't been into introduce yourself in a while, go over there and introduce yourself. Um
237. 35:00the other thing you want to do is check out the events. The other thing you want to do is check out the mastermind. So,
238. 35:04the mastermind is a subscription area of the AI Salon that has things like the AI Salon, the mastermind practice lab. Um we've got
239. 35:15things for the um for the great repurpose. So, so basically the practice lab, if you're feeling overwhelmed with trying to keep up with
240. 35:27AI, right? And frankly, it's hard not to, right? If you're feeling overwhelmed and you're you're you're you're just like
241. 35:37"Where do I start? What do I learn? How do I do What do I do?" That's what the members of the practice lab are on a weekly basis getting
242. 35:46together and working on. How do I be more intentional? How do I How do I lower the anxiety and the the frenetic
243. 35:54pace in myself as the world does all this crazy stuff. That's what that's about. The great repurpose is about
244. 36:01doing things like decoupling your identity from your job. As jobs start to transform and everything starts to
245. 36:07change, how do you navigate that? So, the AI Salon mastermind is is a whole area of people that are trying to
246. 36:14to look at this AI stuff in a in a much more intentional way. Um and it's a it's it's quite a powerful group of people. The programming in
247. 36:23there is really really powerful. Um so, I would encourage you to check that out. Um, the other thing to do is just check
248. 36:29out, you know, the different spaces. There's creative spaces in here, there's business spaces, there's space for
249. 36:34ethics if you're worried about um, advocacy and you know, how do we make sure that our community isn't just trampled,
250. 36:41um, you know, by by all this money, um, you know, uh, participate in the advocacy work that we do. We do a lot of advocacy work. We've
251. 36:49got Daisy Thomas who's helping us establish policies for the AI Salon and positions on how we think about stuff.
252. 36:55So, there's lots and lots of stuff to do here. So, please, please, please participate. And if you've been here for
253. 37:01a while and you haven't been participating in the AI Salon, I would encourage you, just go make a post. Go
254. 37:07make a post in the community feed or in in the irregulars channel, things like that, okay? Beautiful. Other questions. Let's keep it moving.
255. 37:19Kelly Kim, I should probably reintroduce myself in Mighty Mighty Networks. I feel like a different person than I was 3 years ago.
256. 37:25It's actually probably not a bad idea. It's >> [laughter] >> like I could probably do the same, right? Um, yeah, that's it's probably
257. 37:32not a bad idea. If if when you last introduced yourself in Mighty Networks was years ago, probably not a bad idea to just hop in
258. 37:39there and introduce yourself to some of the new people. As we get new activity, um, there's new people to meet.
259. 37:46One of the powerful things about this live session, so I've been doing this, uh, for for 3 years and 3 months now,
260. 37:54these these nightly lives, is that the friendships that are made in the chats and people that show up here on a regular basis, um,
261. 38:06are are very real relationships, very real friendships. And businesses have been started and friendships have been
262. 38:12started and um, so so this is a powerful powerful place to be as we're all trying to figure this stuff out. Okay.
263. 38:21Kyle, have you tried to make a short video movie using any of the new video models? Would like to see a demo if you can. Um
264. 38:29I haven't really So, if you go to flow.google.com, that's the thing that actually lets you use the new
265. 38:38um Gemini Omni model and stitch things together and and make longer form movies. Um I'm no longer a Gemini Ultra
266. 38:46subscriber, so I don't have enough credits to make anything in there. Um the only place that I've made um things
267. 38:54if you um use the Gemini app. So, it's just like the ChatGPT app. Um I can just say in here, "Make me a 16 by 9 movie
268. 39:10of a Stanley coffee mug that is pink and decides that it wants to be a ballerina and go shopping for a tutu."
269. 39:26>> [laughter] >> I feel community here. That's beautiful, Mary. That's awesome. Thank you. Um so, now it's it's generating. Um if you
270. 39:35haven't played with the new video model from from Gemini, I think you need to have a paid account with Google, but if you do
271. 39:43and you download the Gemini app, um you can make a fair amount of uh video. Speaking of wasting water,
272. 39:51meaning I should probably drink some. Is that what you're saying?
273. 40:08All right. So, I'll show you that video when we get it. Um if you haven't heard about the new video model from Google. So, it's called
274. 40:21Gemini Omni. But, the Omni model has actually been incorporated into Gemini 3.5 Flash, which is the model This is
275. 40:35the product marketing at Google has always been a disaster, and then it got better for a while, and now it's getting bad again.
276. 40:43But, anyway, if you go to gemini.google.com, if you go to aistudio.google.com, if you go to flow.google, if you go to
277. 40:51the Gemini app, all of those things are using this new model. Save video. Beautiful This is beautiful. Okay, look at this.
278. 41:26I see it had an extra handle on it there. Um But, how that was what? 30 seconds ago? I said Stanley mug, shopping for a a a
279. 41:36pink Stanley mug, shopping for a tutu, and it it figured it out. >> [clears throat] >> The way that they've got that um
280. 41:46Gemini app videos are a new favorite pastime. Went to a networking event a couple of nights ago, and everyone at the table
281. 41:52was making their own video. Yeah, there you go. Um It's good at at video editing, too. Here's Here's the thing Here's why to
282. 42:00pay attention to this. So, remember when OpenAI had Sora? It was this video model, and you would go to
283. 42:11sora.com, or they had the Sora app. It was this distinct thing outside of chat GPT. What Google has just done is they've
284. 42:21they've taken this what they're calling a world model, Gemini Omni, and they've they've effectively put it inside, I don't quite
285. 42:30know what the architecture is, but they've effectively put it inside the large language model. So, as Demis
286. 42:36Hassabis described it, it's anything in, anything out. So, you could upload a PDF and output a video. You could input a video and output an
287. 42:48image, right? It's anything in, anything out. What's remarkable about it is that when they're talking about a world
288. 42:55model, what they're talking about is that it's a model that understands all of the concepts of the large language model,
289. 43:05like all of the conceptual things, it understands physics, it understands um you know, colors, gravity, objects
290. 43:15colliding with one another. Like, it understands the world. It's a world model. So, when you're producing these videos,
291. 43:24it's kind of like when when Nano Banana came out, it was really good at writing and it could do text really well. The
292. 43:30world model's kind of like that, except it does physics really well. Like, it understands that a glass marble rolling
293. 43:37down a metal track is going to make a certain sound and go at a certain speed with a certain acceleration. And it's not doing that with
294. 43:44programming, it's doing that because it's been trained on enough videos that it actually understands and can can
295. 43:50connect all these different concepts. So, if you haven't played with these things yet, it's worth playing with them
296. 43:55because ultimately where these things go is that you'll be able to do longer and longer form videos, and then you'll also
297. 44:02be able to create worlds that we can enter. Right? And those might be worlds that look like a film or storytelling. Those
298. 44:10also might be worlds that look like here's how to do surgery or here's how to fix the nuclear power plant or here's
299. 44:17how to enter the financial report of your company and understand where the money is bleeding out by looking at it.
300. 44:27So, it's it's it's really significant and I I I think not a lot of not a lot of folks are talking about what a big
301. 44:34deal it is. Um but I think it's actually quite a big deal. So. There you go. Do we need to have a paid
302. 44:43Google account to access Omni or can you just download the Google app and have access? You download the Gemini app. I
303. 44:50think you need to have some sort of paid account with Google. Um but if you just go to Gemini.google.com,
304. 44:57I think they give you three videos a day. This is Remember when the the early days of Sora, you could do like three
305. 45:02videos or one video a day. It's kind of back to that. Um but I get rate limited out on my web-based stuff. I never get rate limited out on
306. 45:13the Gemini app. So, I would go to the Gemini app. Um let's see. Eric Troung has purchased a mastermind
307. 45:24subscription. Awesome. Congratulations. Welcome. Um sign up for the events. One of the things that you get as a member
308. 45:31of the mastermind is you can go back and watch the videos of previous workshops. Brandon uh just finished up or I think he's got one more um
309. 45:41lovable. He did a four-part lovable series that I think finishes up next week. Um Andy Skerentina just did a
310. 45:48four-part series on decoupling your identity from your job. Um I did a quick one on on uh dis- discovering you know,
311. 45:56what's what's the one thing that you can do in the world that no one else can do and why that's important. Um so, there's
312. 46:02all sorts of cool stuff in there and there's there's good things coming up. Sharon Crawford, hubby is asking how
313. 46:07many AI companies there are, OpenAI, Anthropic, etc. Okay. Great question. So, I'm going to I'm going to narrow it
314. 46:14down to what I'm calling the frontier model companies. So, when you're talking about frontier model companies, these are companies
315. 46:24that literally have billions and potentially trillions to invest in developing these AI models. There are tens of thousands, probably
316. 46:35hundreds of thousands of companies that are using the things that they're creating. There are three biggies right now. Those
317. 46:43The three biggies are Anthropic, OpenAI, and Google. Google's the one that invented the transformer architecture that all of
318. 46:52this generative AI stuff is based on. They were the inventor of it. They kind of dropped the ball on it and OpenAI
319. 46:59took over and launched ChatGPT and took all the thunder. Um, and then Anthropic is a bunch of ex-OpenAI employees
320. 47:08that they didn't like the way that OpenAI was doing safety, so so they're doing that. So, those are the big three.
321. 47:14Then you've got xAI, which is Twitter, Elon Musk's company. It's now part of SpaceX, so xAI rolled into SpaceX. They're about to do a
322. 47:25some amount of trillion IPO. I don't know how much it is. 1.4 trillion? I don't know. They're about to do an IPO. Um,
323. 47:36they're not quite in the mix yet. They're playing catch-up and and Elon sort of went for about a year doing kind
324. 47:42of AI porn and stuff like that. All four of these right now are working on these agentic platforms. Like OpenAI has Codex,
325. 47:50um, Anthropic has uh, Claude code, um, Gemini has anti-gravity and Spark, and then Grok has Grok build. So those those
326. 48:00are the kind of the big four right now. The one that was there but but lost their place was Meta. So Meta was was
327. 48:08keeping up with these guys until Lama 4 and it just crapped the bed and now they lost what's his name? Yann LeCun
328. 48:19was their head of AI and they basically brought in a 23-year-old punk and told Yann LeCun to report to the 23-year-old Yann LeCun was like
329. 48:30peace peace out [ __ ] off. So Meta's trying to play catch-up. They will probably do it. Then you've got a bunch of Chinese
330. 48:42companies that are also in the mix but what the Chinese companies are doing is is they're basically copying everything
331. 48:48these guys are doing turning them into open source and then flooding flooding the AI market with their open source
332. 48:55models. So Alibaba has Qwen there's Deep Seek there's uh I don't know there's a couple of them. But they're doing very very powerful
333. 49:05models that you can run on local computers. So So those are the those are the big five. Um you've then got
334. 49:12companies like Perplexity on the image gen and film gen there's a whole other set of companies but but those the majority
335. 49:21of where all the activity is with AI compute and leadership is in those three companies Anthropic OpenAI and Google.
336. 49:30And and I would I would probably we're we're close to including xAI into that I would say we're pretty we're pretty close to that.
337. 49:40All right. Probably overthinking it. When people say agentic AI do they just mean bots that can do stuff for you? Um
338. 49:50What's Alibaba AI? Alibaba AI is the the Quen models are from Alibaba. Alibaba's the like the Amazon of China. Um
339. 49:59they're investing in this stuff. Amazon's another one that that could show up and I think they're starting to make their own chips and
340. 50:07they might be doing their own model. I don't know. They could show up but they're not in it yet. Um Okay. Agentic AI, so
341. 50:24until this year AI agents, people have been calling [ __ ] AI agents for decades. For decades. Basically, they would design chatbots
342. 50:37that were like, you know, logic trees and they would call them AI agents because they used AI in the voice-to-text or whatever. Um
343. 50:49AI agents in the proper definition are are tools that can take your goal, your imperative and work independent of you
344. 51:00over long periods of time doing multiple steps to achieve something for you and and essentially bring it back to you. Some
345. 51:08of them even will create an orchestrator agent that then spins off sub agents. Um so you'll give it a goal like go
346. 51:16start me a business and it'll say, "Okay, we need a researcher and we need a business strategist and we needed this
347. 51:20and we needed that." And it'll dynamically go generate all these agents that'll go do stuff for you and come back.
348. 51:27That's what these platforms are are all doing to varying degrees of good. Um they are still >> [clears throat]
349. 51:41>> not anywhere close to ready for prime time. They're just not. They still require a lot of human oversight. They require a lot of
350. 51:50configuration. They require um If you're going to set up one of these agentic systems, it requires you
351. 51:58to understand here. Let me I'll show you a graphic. So, in in the AI Salon Wait, let me share my screen first.
352. 52:12In the AI Salon in community feed you can see this image. Am I sharing? Let me make sure I'm sharing. I am. Okay, cool. Um
353. 52:26is this image. So, these agentic platforms, so these are things like Codex from Open AI Claude Code and Claude Co-work. They're
354. 52:38they're the same app. They're just different tabs in the same app. Claude Code, Claude Co-work. Um Perplexity computer um
355. 52:47Antigravity by Google. There's also a new thing called Spark, which is sort of a consumer front end sitting on top of Antigravity.
356. 52:56All of them have lots and lots of different components to them. And the broad grouping of components are the
357. 53:05brain. So, how does how does this How do your agents know what to do? Um So you need to figure out which AI models
358. 53:15you're going to use. Then you need to figure out are you going to use local models or cloud models? And a cloud model would be like using
359. 53:22something um like ChatGPT or or GPT, you know, 5.5 or Claude Opus, right? You're you're using the models that
360. 53:33that the computer is happening in the cloud or you're running a local LLM like Qwen 3.6, um which can run on a desktop machine
361. 53:42that's got a decent amount of RAM in it. Um, and and so you need to know about that. Then there's model routing. How do you
362. 53:48choose which of those to use and when? Um, then there's reasoning. Are you turning on reasoning or are they just
363. 53:53moving really quickly? So there's there's all sorts of things about that that you need to understand about designing um,
364. 54:00the cognitive part of what it does. Then you've got memory and context and files and like how does it know who you are?
365. 54:07How does it know know how you write? Well, you provided access to your data. Here's how I write. Here's how I think.
366. 54:13Here's the projects that I'm working on. Um, and then it will do things like, you know, uh, create memory over time. So the more
367. 54:22you use it, the more it understands what you're doing and what your preferences are. Then it's got access to tools, the hands. Um,
368. 54:30things like its ability to surf surf the web and its ability to use what are called MCP connectors, model context
369. 54:37protocol connectors, which is how your agent can use other tools. Like it can use the um, the tool for Figma or Canva, the MCP
370. 54:48connector for Canva. So you can just have your agent go off, um, come up with a marketing strategy, come up with some design documents, and
371. 54:57then use the MCP connectors to go over to Canva to generate all the ads, right? So if if you set it up correctly, it can
372. 55:04start going to do things in the world. It can also run things locally on your machine. Right? So Codex right now and and Claude
373. 55:12co-work can actually run applications on your local computer and and use a mouse. Like literally click around them and use them as if you
374. 55:21were using them. Right? And then there's APIs and there's file editing and you've got to figure out what permissions to
375. 55:27turn on. So that's just the tools that it has access to, right? So it can think, it can remember, it can use tools, and
376. 55:35then it's got things like skills where it can manage and orchestrate all of that stuff together. So just to be able to use one of these
377. 55:45agentic platforms effectively, you need to understand that there are all these components there. Now they all
378. 55:52default to something, right? But if you just use the default setup, it's going to give you generic crap. So the way these things work is you
379. 56:03modify them and improve them and personalize them over time. That requires you to to to actually know the answer to the
380. 56:13question, what am I trying to accomplish here? So if you're just thinking, oh I'm just going to go set up Codex, you you can,
381. 56:23but what you'll end up doing is just burning through lots of tokens and potentially running up your AI bill
382. 56:29unnecessarily cuz you're just messing around with stuff. So over the next 3 years, these systems will get easier and easier to use and
383. 56:39all of this complexity will drift into the background and and you'll just be able to talk to likely an application on your phone or
384. 56:47whatever the device of the day is and then it'll go and configure all this stuff and do it for you. But but right
385. 56:54now if you want to play early, you kind of have to understand a lot of what's going on here. Kind of crazy, right? Chef Kelly's here.
386. 57:03>> Chef Kelly is here. Oh, Chef Kelly. Uh what before we bring Chef Kelly up on stage, um I just wanted to point out um that
387. 57:13uh Sharon is sharing that now her husby hubby is asking more questions. >> Good. That's why That's why we're That's
388. 57:19why we're here. Beautiful. Chef Kelly Anderson, what's happening? >> Hello, hello. What's up, guys? Hello. Good to see you all.
389. 57:30>> It's ask me anything night. How are you doing? is How is How is your your business going? How is How is life as an entrepreneur?
390. 57:38>> [laughter] >> Oh god, life as an entrepreneur. You know how it is. It's up and it's down. Um no, it's it's good. It's really good.
391. 57:46Um I'm getting a very warm reception from the industry. Um but you know, it's >> Well, what what do you tell tell tell
392. 57:53folks what the industry is and what you've created? >> Um well, I'm Chef Kelly and I'm a mechanic. So no, I'm a chef. And so I have um I've
393. 58:03I've I've been um a proponent of a a healthier better food system for everybody for a really long time and
394. 58:09um I when last year when the FDA and the HHS announced that uh they were going to do away with the artificial food dyes in
395. 58:20food, all American food brands had to to to go without by the end of 2027. Um I decided to build something.
396. 58:28And um so wait, I did. It took me almost a year, but um I started I started figuring out what what the crops were
397. 58:35going to be, the natural crops that were going to um become the new food colorants and started aggregating the
398. 58:42data and cleaning it and parsing it and all that computer speak. Um but the the rub here is that artificial dyes are
399. 58:50cheap, but they're really sturdy. You know, they won't break um because of heat or acid or carbonation,
400. 58:56right? And that's why the industry loved them for so long, but they're horrible for us and natural dyes are really
401. 59:02fragile. And so what a lot of food brands are finding is that they can't get the formula right. They're they're
402. 59:07failing on the bench testing. They're wasting time and time is money right now. >> Yeah. >> And so what that my my platform does
403. 59:15basically is it's a matchmaker. So the food company puts in all of their, you know, data, their specs on pH and and
404. 59:23and thermal heat and all all stuff and then we run our database of all the FDA-approved natural dyes that can stand
405. 59:32up to those uh metrics, basically. And that's it. We don't sell the dyes, we don't manufacture the dyes, we just
406. 59:37match them up so that they can >> doing the matchmaker, and then and you a month you charge them a monthly fee to
407. 59:43be able to have access to that data, right? >> Yeah, it's it's an actual annual license, um but we charge It's great.
408. 59:49It's We charge the the food brands to use the platform, and we charge the dye brands to put their empirical data there so that Yeah.
409. 59:56>> So you're you're pulling it on both sides of the equation. That's great. That's really smart. That's awesome. That's awesome.
410. 1:00:04What's um So so so what's going on? What what Where's your head these days with with all the AI stuff? With like like I
411. 1:00:10like I assume that the balance right now is mostly running your business, but it like where are you with trying to keep
412. 1:00:17up with AI? Or how are you doing with your with your app? And as you know >> Well, one thing I'm doing Yeah, a couple
413. 1:00:24things I'm doing is I'm taking that Claude certification course. So um I think we all know and love Sabrina
414. 1:00:31Romanoff, and she's got a female-only uh school community. And in that school community, you can team up with They're
415. 1:00:38calling us pods, cuz I'm not sure exactly how it works, but like you have to have a team of 10 to push through or
416. 1:00:44something, right? And so um it's completely free. A lot of people out there, um if you're listening, people are already charging like two
417. 1:00:54grand to host these certification, and Claude and Anthropic are giving it away for free. So don't um don't don't be
418. 1:01:02fooled. Um so I'm going to do that because I use Claude when I code, and I want to get better at it. And you know,
419. 1:01:08I want to I want to put a a bigger, better agent inside of my platform. Oh, I just applied for uh I'm my patent
420. 1:01:15today. I sent it off to the to the patent office. So my provisional patent. >> Miss Deal, congratulations. Awesome.
421. 1:01:22>> the IP, you know what I'm saying? So, um >> That's beautiful. Did you Did you use AI at all to help you draft a patent or do
422. 1:01:28research or anything? >> 100% So, I I used Claude because I had actually um applied last year, but something happened and then I found out
423. 1:01:37earlier this week it was a bad week that um it was now abandoned. And so, I had to reapply all over again. So, I gave it
424. 1:01:44all to Claude code and I was like, "Update me." You know, and even like the um the drawings, the figure drawings
425. 1:01:51that you have to do of like how the algorithm works, right? And so, Claude did all that and then I had, check this
426. 1:01:56out, cuz the US Patent Office.gov, whatever it is, is not easy to navigate. They don't make it friendly for you. So, I had Claude code um
427. 1:02:07uh what's it called? Plug-in >> Uh-huh. >> open in and it was navigating all through the site and clicking all the
428. 1:02:15places I need to click. Yes, and then it would be like, "Okay, go ahead and upload this file." And I would upload it
429. 1:02:20and then it would click to the next place and say, "Go ahead and upload it." >> god, that's so cool. >> [laughter and gasps]
430. 1:02:24>> Oh my god, that's amazing. >> That is I've I've found that the plug-in, the um for for for co- uh Chrome, it has been
431. 1:02:33one of my favorite tools to use. Like, I use it probably a dozen times a day for certain >> it is it connected to Claude Like, I
432. 1:02:39haven't used it a lot. Is it connected like if you're at Claude and you're doing stuff there, is the plug-in aware
433. 1:02:45of it? Like, can it send out It's kind of a separate >> Yeah, it is. Like, I I've heard that I can have data sent from the plug-in to
434. 1:02:53Claude, but the plug-in can't read anything like any of my projects or any of my chats with with Claude. And in fact, it
435. 1:03:00doesn't even have a memory, which is good and bad for you know, for certain things. But, you can um like you can
436. 1:03:07have it screen record what you're doing to train it to do a task or you can just you know, just after you're done with something like um
437. 1:03:15LinkedIn ads. Again, another very confusing platform. And you know, once once we went and created an ad campaign
438. 1:03:23for me, and then I said, "Okay, everything we just did, put it into a skill format for me." And then I just clicked three buttons, and
439. 1:03:30there was a skill. So, anytime I want to go back into LinkedIn ads, I just hash or slash type in that um task, and it'll
440. 1:03:38create an ad for me. >> Does it Does it Are you using Claude uh Claude Code and Claude Co-work at all?
441. 1:03:44>> Co-work a little bit because Co-work also has the capability, sort of behind the scenes, to go and click around for
442. 1:03:50you if you need to. Um but mostly it's it's the plugin and Claude Code, both on desktop and on uh V V S Studio terminal.
443. 1:04:02>> Yeah, yeah, terminal. >> Yeah. >> Absolutely wild. >> I know. >> Absolutely wild. >> I know. And And prior to to your >> [laughter]
444. 1:04:12>> your B2B SaaS platform startup company, had you ever done coding before? >> Bro, come on. No. >> [laughter]
445. 1:04:23>> No, I don't even think I passed high school algebra. Like, no. >> [laughter] >> I I mean, this is all pure curiosity.
446. 1:04:34And you know, I say it to Anne a lot, too. It was She Leads, it was the AI Salon, and you know, these communities
447. 1:04:42that I've been lucky enough to stumble into because um there's a lot of really bad ones out there. >> Mhm.
448. 1:04:48>> And if you find good ones like this where you're supported and people want the best for you and they're not
449. 1:04:52constantly trying to sell you, you know, get rich quick schemes, >> Yeah. >> um that's where you're going to find
450. 1:04:58your footing when you're unsure about where to be. Yeah. >> Yeah. >> [laughter] >> Put on THE PINK HEADBAND.
451. 1:05:08WE HAVEN'T SEEN HER IN A LONG TIME. >> I I right? Have I got a course for you. >> Exactly. >> [laughter]
452. 1:05:14>> The wheel's on, honey. The wheel's on. >> I really got to learn much more about Claude. I feel like I'm getting behind. I you know
453. 1:05:23>> course. Take the >> Yeah, there you go. Jump in the cohort. There you go. Beautiful. Awesome. >> Well, thanks for having me on again.
454. 1:05:31>> Great to see you. I'm I'm glad you hopped up here. Yeah. >> Hey, are you going to tell everybody about the ambassador thing? You want to
455. 1:05:37spill the beans >> yeah, so so I'm trying to think where we can can point people. Um so, we're we're putting together um an
456. 1:05:48ambassador program. So, if you're a member of the AI Salon we're looking to find people in local communities to find groups within their
457. 1:05:56communities, not AI groups, but just groups that are curious about AI where our AI Salon ambassadors can go
458. 1:06:04you know, educate them, talk to them, just you know, show up there at their events and answer questions and things like that. Um
459. 1:06:11what what Brandon just said is if you're potentially interested in being an ambassador, DM Andee, a n d e or
460. 1:06:18Brandon, producer Brandon, in the salon. Just DM them and say I'm interested in being an ambassador and they'll give you
461. 1:06:24a link to the to the form, to the application form. And >> Hey, I have a question. I have a friend who's actually an ambassador to HeyGen
462. 1:06:32and I'm going to his um his in-person thing tomorrow in Venice Beach. What can I do while I'm there? What do you What can I say? What on
463. 1:06:42behalf of you and the AI Salon? >> I I think you know, what you just said is that is that I mean, especially
464. 1:06:49because you've got experience like there's it's really hard to learn this stuff right now because it's moving so fast.
465. 1:06:55There's lots of groups. There's lots of charlatans. It's hard to find groups that are you know that don't
466. 1:07:03And like I found one and it you know, just share, just tell people to come to the salon.ai. Um and yeah, just just do
467. 1:07:10that. And then as an as an ambassador, as we're putting the program together, we're going to provide, you know, things
468. 1:07:16like PDFs and print outs and you know, stickers and things to talk about and we'll talk about AI readiness and play
469. 1:07:22first and create excellence and generously lead and all you know, all the stuff that we do. Um so so the
470. 1:07:29ambassadors will be trained up on on how to do that. Um but but like what you just said here is is exactly yourself,
471. 1:07:36right? That's the whole thing here is we're not trying to sell anything. It's like we just want people that are
472. 1:07:41curious about trying to figure this stuff out. And >> I think yeah, that's one of the things you guys have been really great about
473. 1:07:47over the past couple years is that it's really about education. It's about safety. Um it's about being thoughtful in what
474. 1:07:55you do and um not about a get rich quick scheme, which is great. I mean, people need to make money. I get it. But um you
475. 1:08:02got to learn to use the tools properly first. And there's no such thing as a YouTube video that's going to make you
476. 1:08:08$5,000 this week. It just doesn't happen. So. [laughter] And and only word of advice for newbies out there, don't get FOMO.
477. 1:08:16Don't get FOMO. Trust me, you have plenty of time. Play around, have fun, stick with the tools you like.
478. 1:08:22>> You're you're yeah, you're way earlier than you think and and uh and it it will seem like everyone knows
479. 1:08:29way more than you. And here here's the thing, it's like at every party, they're thinking you know more than them. >> Yeah.
480. 1:08:35And tomorrow you will. Tomorrow you'll know more than you knew yesterday and so and and so and and so on. Yeah. >> Exactly. Awesome.
481. 1:08:41>> well, peace out. Love you guys. Talk soon. All right. >> That's awesome. Awesome, awesome, awesome. Cool. Ambassador, uh
482. 1:08:53Sharon Crawford, I live in a small community and I think it would be impactful. I can't begin to tell you how impactful
483. 1:09:01if So, there's so much anti-AI sentiment right now. One of the things that I experience when we do these fly-ins into Washington, D.C.
484. 1:09:16I'll share stories of people, you know, like Chef Kelly. Like Chef Kelly, you know, was in the business and this this
485. 1:09:22new legislation came out and she didn't know anything about programming and now she put together this thing
486. 1:09:28and and she started this whole new business. When people hear stories like that, very often it's the first time
487. 1:09:38that they've heard anything positive about AI at all. That the only thing that they've heard is the robots are going to kill us, it's
488. 1:09:45going to take our jobs, it's stealing from artists, it's in and in and in and in. Doing a small thing like showing up at a
489. 1:09:51local library and saying, "Hey, do you want to figure you you want to understand what this ChatGPT thing is?"
490. 1:09:58Or or whatever, like literally whatever. It will be revelatory to people who've only heard about AI as this nemesis.
491. 1:10:10Um and what will happen is someone will go, "Oh, I didn't It can do that?" And then could I do that? Oh, yeah, you
492. 1:10:19could do that. What are you interested in? I'm really passionate about beets. What kind of beets? I don't red beets
493. 1:10:25and white beets. I love beets. Okay, let's let's go let's go see what ChatGPT knows about beets. >> [laughter]
494. 1:10:32>> And then it will ChatGPT will do what it does and then you can say, "Make me an image of a beet salad." Right? It'll do
495. 1:10:39that with the recipe. >> [laughter] >> And then you'll have a beet lover that's now, you know, at least open.
496. 1:10:47Oh my god. Bears, beets, Battlestar Galactica. Um Brandon Built Stuff, is Codex useful if I'm not a developer, but I but I'm
497. 1:10:58trying to build apps. Yes. Yes. >> [laughter] >> If So so here's the thing, Byron, is
498. 1:11:12the simplest vibe coding tool right now is lovable. Largely because what they focused on is those people building apps
499. 1:11:28that weren't developers. So all of the agentic platforms, so it started with Claude Code. So so when when OpenAI was going running down the
500. 1:11:38Sora path and Elon Musk was running down the porn path and Google was just all over the place trying to figure out all
501. 1:11:45the different things they could do. Anthropic put out Claude Code. They focused on, let's create this thing to allow coders
502. 1:11:54to really amplify their development tools. And they they they worked on the connections between the the code generation from Claude and the
503. 1:12:04development environments and the tools, the MCP tools and skills and all the things that that could augment this generation capacity.
504. 1:12:15And so they built this kind of framework. Well, everyone else, once that thing took off, everyone else copied that. So all of
505. 1:12:23these platforms, Grock Build, Anti-Gravity, um Codex and Claude Code, all four of those and and uh Perplexity Computer,
506. 1:12:35that one's a bit of an anomaly, but all four of those main ones started as coding environments and then
507. 1:12:43they realized, oh [ __ ] they're also good at like file management on the local computer. So we can create this standalone
508. 1:12:51application that's going to run on someone's computer that can do all this coding stuff and interact with all the
509. 1:12:57development stuff, but it can also do things like file management and marketing campaigns and and you know,
510. 1:13:04you know, uh document generation and things like that. And so they now work nicely together. So Codex can absolutely
511. 1:13:11do the coding thing. It can do lots of other things. The downside of those environments is that they are configuration [laughter] heavy.
512. 1:13:22Like you have to set them up and you have to be curious enough to kind of understand the components of them and
513. 1:13:28how they talk to one another and and just just it's taken me three or four weeks just to conceptualize all the different piece parts.
514. 1:13:36And then learning how to make them talk and do that really well is a whole 'nother level. So if you just want to
515. 1:13:42build something, start with something like Lovable where you can actually build it, launch it, point a a custom
516. 1:13:49domain to it and you know, have something live in the world in an hour or two. Um or you can set up Codex as,
517. 1:13:57you know, a more robust environment. All right, we've got about 15 minutes left. What other questions do we have? Let's see.
518. 1:14:08I'm building a dashboard in Lovable, but also building the same in Codex and Claude code just to see the difference. Great.
519. 1:14:16Great. Wondering how good the graphics will be in Codex and Claude. Well, so here's the thing, Kelly, about This is
520. 1:14:24one of the things that I'm starting to get. Claude does not do image generation. And I saw about a month ago, I saw an
521. 1:14:35interview with Dario Amodei where they asked him about that and he said, "I don't think it's all that interesting." And my initial response
522. 1:14:43was like, how could you not be interested in having Claude generate images like like if someone wants to do
523. 1:14:49marketing using Claude, which is good at writing, why would you not want to be able to generate the ad? Here's what hit me.
524. 1:14:57They invented the MCP, the the model context protocol, so that Claude didn't have to make images. Claude could use MCP to have
525. 1:15:07other sites make images or videos or things like that. So, Claude code out of the box may not be able to do image generation or it might
526. 1:15:20be crappy at it. Um but with MCPs, if you have it configured correctly, you could probably do some pretty sophisticated stuff with
527. 1:15:29Claude code and with Codex. So, again, that's why I'm saying I think Codex and Claude code are heavier lifts to set up
528. 1:15:37cuz you have to know what you want to do and then you have to understand, well, what are the applications that I could tap into to do that?
529. 1:15:43Um Lovable's always had beau- beautiful user interface and Lovable about 3 weeks ago just added a whole new interface builder
530. 1:15:54um that that it uh Lovable now can do much more sophisticated design than it could 3 weeks ago. So, if you haven't used
531. 1:16:01Lovable in a month or two, um it's better now than it was 3 weeks ago. All right. Stitch from Google also syncs
532. 1:16:08to Lovable. Okay, cool. Yeah, all Here's the thing. This is the thing that's just absolutely trippy and and is going to be confusing.
533. 1:16:17All of these tools can all talk to one another. You could conceivably be running a local large language model
534. 1:16:26in, you know, just some local tool tool, Open Claw or Hermes, talking to Claude code, talking to Codex, talking to some other thing.
535. 1:16:37I think that is complication for complication's sake. Uh so, I'm just going to pick one. My My instinct right now is Codex. I think I'm
536. 1:16:45going to pick Codex. Um but I might be convinced to do Cloud Code, but it's between one of those two for me.
537. 1:16:53Oh Oh, yeah. And Codex for Windows just dropped today. So, if you're on Windows and you've been bummed that you couldn't
538. 1:16:58use OpenAI's Codex, you can as of today. Um okay. Getting ready to wrap things up. Before we get out of here, next Tuesday is the
539. 1:17:08AI Salon presents Um Marta is going to be talking about the capacity gap. So, nervous system activation when you are
540. 1:17:21amped up trying to keep up with AI and your nervous system gets all jacked and it creates a gap between what you
541. 1:17:28want to do and what you can do. That capacity gap is something she's going to be talking about. It's going to be quite powerful.
542. 1:17:37So, please go to the event calendar and RSVP for that event. So, that's next Tuesday. Um weekend homework.
543. 1:17:50I would say that there's a couple of things. I would I would if you're into research, I would start to research what's going on with
544. 1:18:03the available compute with resistance to building data centers, how that's going, with what's going on with chip fabrication in
545. 1:18:14Taiwan, and how essentially 100% of the world's high-powered AI chips are built in Taiwan. And that could be an issue.
546. 1:18:26Um so, if you're into research, I would research, you know, that side of it. If you want to start exploring tools, I
547. 1:18:33would educate yourself about what Cloud Code is, what Claude Co-work is, um anti-gravity from Google, any three of those.
548. 1:18:46I wouldn't expect over the weekend to totally grok them if you if you haven't quite figured out what they are yet, but
549. 1:18:55you can use Chat GPT or Claude or Gemini to explain them to you. Tell me what they are. Explain it to me like a fifth grader.
550. 1:19:03Um and you can listen. You can download them. What one of my favorite things to do is I've got a couple of my stock prompts, right? For
551. 1:19:12image ones, it's it's a '70s muscle car in an abandoned factory. For these agentic tools, the the the
552. 1:19:20prompt that I've been using is I'll just download the the tool. I don't do any configuration. I go into the little prompt box
553. 1:19:29and I say, "Make me a clone of the '70s video game Asteroids." And and make it give it CRT effect so it looks like the authentic thing.
554. 1:19:40And just see what it builds. See if it builds you something good. It may surprise you. Um and but but with those things you can go
555. 1:19:49deeper and deeper and deeper and deeper. Um if you haven't learned about skills from Claude, Claude skills, those are now
556. 1:19:56being adopted as an industry-wide standard um that can take those agentic platforms and dramatically personalize them to you. So,
557. 1:20:10those are two areas, right? What's going on with the chips and the amount of com- compute versus the amount of demand that's coming. One is
558. 1:20:21understanding what these agentic tools are. And then the third area that you might look at is really do some research about
559. 1:20:30high-powered desktop machines like the MacBook Pro M5 with 64 or 128 gigs of RAM or the the DGX from Nvidia or the new
560. 1:20:42AMD just came out with 128 GB unified RAM thing that you can run local large language models on. So, you're not
561. 1:20:51actually paying someone else to use their compute. You're actually doing it on your desktop. The the investment range of those
562. 1:20:59computers is somewhere now between $2,000 and $10,000. It seems like the like the AMD one is somewhere between two and three grand.
563. 1:21:08The Mac ones I think are probably going to land in the three to $5,000 range. And then you can get higher-end ones for more.
564. 1:21:18And then why would you invest in that? Well, if the chip thing happens so that we don't have as easy access to tokens
565. 1:21:26as we used to, being able to do your own AI inference on your desktop is going to be super valuable. Right? That's not this week.
566. 1:21:35That's not this month, but it's probably within a year. Claire Jacobs, sparked by your question at Mary, I just had ChatGPT make a
567. 1:21:44prompt for lovable lovable to make me a website. It's done. There you go. It's Claire Jacobs claire-jacobs-creative.lovable.app.
568. 1:21:54That's so cool. Can I I don't think I can copy that anywhere, can I? Oh, wait. Yes, I can. Brandon gave it to
569. 1:22:03me on a sticky note. Cuz Brandon's the best. All right, let's go look at this. Let's go Let's go look at >> [laughter]
570. 1:22:09>> Let's Let's go look at Claire's website that she built while I was rambling. So good. Look at that.
571. 1:22:25Can you all see this? Am I sharing correctly? Yes. Artful vibe by Dr. Claire. Art, music, and imagination created at the edge of
572. 1:22:36human feeling and AI possibility. Look at this. >> [laughter] >> This is great. Here's songs. It's so good. I don't think these are
573. 1:22:48playing right now. Um amazing. Look at that. There you go. Artwork. Oh, here's a whole artwork page.
574. 1:23:04Here's a whole music page. It's so good. Custom songs. Books. Creative notes. Yeah, there you go.
575. 1:23:19There you go. So, like I said, we do now live in a world where you can just do stuff. And so, if you haven't been playing,
576. 1:23:29one of the the stage one of the cycle of AI readiness, play first. Just go play. And it it will blow your mind. Um
577. 1:23:37>> Speaking of doing stuff, uh a couple of folks in the chat while you were rambling, uh were asking for the QR code for the
578. 1:23:47AI Salon so they could take it and share it with folks in the real world. And I will follow up in with a post in the AI
579. 1:23:54Salon, but for anybody who is quick with a screen share or a a screen grab, we'll go ahead and throw that up there. Uh screen share
580. 1:24:03just grab that screen grab and that'll take you right to community.thesalon.ai. >> That is beautiful. Beautiful, beautiful,
581. 1:24:11beautiful. Yeah. And you can zoom in on that. Um or just grab the whole thing. And feel free to print that out if you
582. 1:24:19want to, you know, print that out as like little little handouts, you could do that too. Um, that's awesome. Yeah. Listen, we've
583. 1:24:26got an amazing community, right? We've got an amazingly generous community. We've got an amazingly smart community
584. 1:24:31and passionate community. Um it is not common. You know, what we're learning is there's lots and lots of charlatans out there.
585. 1:24:40There's lots and lots of communities out there that are more rigid or more technical or whatever. This one is This
586. 1:24:46one's got some heart. Um and and that is is something that I'm really happy about. So let's get more people that value those
587. 1:24:54kind of things and and keep this thing rolling. Kelly Kemp, I'll share it with my class on Tuesday. Beautiful. Um can I use that
588. 1:25:02in my live streaming? Absolutely, Ivan. Please do. Please please send people over to the salon. Anything we can do to
589. 1:25:08get more people in here is great. Um let's see. I'll post my chat prompt for you. Please give me a complete
590. 1:25:15prompt for lovable to create a website for me to showcase my mostly AI artwork and Suno songs. There you go.
591. 1:25:25Mary Chester, dreams can come true. We've even got Mary's husband now asking about AI. >> [laughter] >> For the for the past few few months,
592. 1:25:34he's been like, "What's she doing over there with that AI thing?" >> [laughter] >> Now he's curious. That's awesome.
593. 1:25:44Um all right, everybody. This has been fantastic. Thank you as always. Thanks for hanging out on a Friday night.
594. 1:25:50Friday night date night. Asking really good questions, thoughtful questions. Um I hope I did I you know, did them did them some some service
595. 1:26:00um tonight. But yeah. Have a beautiful weekend. Summer's started. Get out the short-sleeve shirts. Easy peasy. Exactly.
596. 1:26:12And we will be back on Monday night. All right. Great show tonight. Thank you, Kelly. I appreciate that. All right, everyone.
597. 1:26:21Take care. Have a beautiful weekend. Bye.