9/23/2026 - Building a Local AI Image Generator Without Ever Reading the Manual

Transcript

1. 0:00[music] [music] [music]
2. 0:18[music] [music] Good evening, good people. Welcome to
3. 0:33the AI learning lab. Uh, welcome to the AI salon. If you are new here, welcome. If you're on TikTok, just go to my
4. 0:41profile, click on the link, that'll take you over to live stream. Uh, it's Wednesday night. That means it's about
5. 0:46creating excellence tonight. So, we're going to create some excellence. Um, there's also been uh more news.
6. 0:54the the news she is accelerating. Um, and I know we really shouldn't gender the news. [laughter] The news they are accelerating.
7. 1:14Ah, good people. Um, if you are here, welcome. Uh, the the chat is down below. if if you you can chat with one another
8. 1:23down there, but I can also see the chats so we can we can pull comments up on stage. If you have any questions, feel
9. 1:28free to pull those up. Um, if you're new to the AI salon, go down the uh the the little spaces here, the little areas to
10. 1:35the left hand side of your screen and um you'll see things like the events that we have. You'll see spaces to learn more
11. 1:44about the mastermind which is our subscription area only uh of the AI salon that's got really remarkable
12. 1:50programming in it and people in it and you should be a part of it. In fact, if you uh want to go to producer Brandon's
13. 1:57codeex workshop, tomorrow's the fourth uh workshop that's inside the mastermind. There's also spaces in there
14. 2:04to, I don't know, show off your work and just have chats and and have deeper conversations and all sorts of stuff
15. 2:10like that. Um, lots of different spaces you can almost think of. I was talking to Andy today about the the future of
16. 2:16the AI salon and and what it is and we're talking about it being like it's kind of like New York City where New
17. 2:22York City is this thing, but it's like there's all these different neighborhoods and that's kind of what
18. 2:26the AI salon is. So, what's cool about that is that you can find your peeps in the different neighborhoods and there's
19. 2:33starting to be, you know, sort of different kinds of neighborhoods evolving. Uh, and it's it's really powerful. So,
20. 2:40so that's that. Um, I'm going to definitely go look at some news. Uh, one of the things if you're new to the AI
21. 2:47salon that we talk about is this thing that we call the cycle of AI readiness. And if you haven't seen that, kind of
22. 2:53looks like this. It's on the the welcome screen. When you first log into the AI salon, you land on a welcome screen and
23. 3:00you see this graphic. And the idea here is that AI is whatever. But when you put yourself at the center and you use AI to
24. 3:10amplify yourself and and you learn through play and then you apply what you learned by building interesting stuff
25. 3:17and then you share what you learn with other people. Play first, create excellence, generously lead. that will do some remarkable things in
26. 3:25your life. And that's one of the things that we see in the AI salon a lot. Um, yeah. By the way, tomorrow I'm going to
27. 3:34be on I don't know if Linda Sherman's in the house tonight. Is she? I don't see her yet. Um, tomorrow I'm going to be on
28. 3:42Linda Sherman's podcast. I think it might even be her inaugural, her kickoff. It is Rethinking Aging Club.
29. 3:53and it's at noon Mountain time, so two o'clock Eastern. Um, and I don't know if it's going live right away. She might be
30. 4:02producing it, so we might just be recording it tomorrow, but I don't I don't actually know now that I think
31. 4:07about that. Anyway, that's that. Mel's in the house. What's happening, Mel? How did your uh how did the LOL go? I hope it went well.
32. 4:17City in the house. Lab five, applied futures lab. Listen, listen. We got some talking to do all of us. The world is weird.
33. 4:31The world is weird. There's weird stuff going on. The stuff that we did in here two years ago just doesn't seem to matter as much
34. 4:43anymore because there's so much happening so fast. I was talking to Cindy today and we we were talking about
35. 4:50the hell is going on. Everything's about to change. Mel's LOL was excellent according to uh Silver Fox. Fantastic. Unsolicited testimonial
36. 5:02for Mel's fine work. Taught people how to make their own digital selves and use these AI agents in a more powerful way.
37. 5:10Remember that thing I just showed you, the the cycle of AI readiness? This thing, the you at the center of this,
38. 5:16what Mel taught everyone how to do was how to put you at the center of a powerful AI agent where you're in
39. 5:23control of your own data, your own sovereignty. Yes. Yes. Do that. Be like Mel. All right. Uh
40. 5:37let's see. We were talking about the live episode recording tomorrow. Yes. Oh, there we go. LinkedIn. So, go look
41. 5:45up Linda Sherman on LinkedIn and you can see there's a post right there about our live episode recording. So, it will be live. Yay.
42. 5:55Fantastic. Oh, thanks, Lori. It was a blast. Looking forward to two and three. See, here's what happens with these
43. 6:03people doing LOL's and doing workshops in the AI salon. They're like, "Oh, I don't want to do it. What if it's not
44. 6:08good?" And then they do it and people are like, "That's awesome. Can we have more?" So, so I hope there's two and
45. 6:13three. Is that true, Mel? Are you going to do parts two and three? It's the thing about that the digital self stuff
46. 6:19is it's actually a big and deep topic. So, you could probably you could probably do those for the next couple of
47. 6:24months and have enough content to do it. That's awesome. Cam Cactton's in the house. We got lots of people in the
48. 6:30house. I'm really excited. Um, so one of the things I'm going to play with tonight is one of the things that I've been
49. 6:38reluctant to play with on the channel are agentic sites like Chat GPT work, Codeex, Claude Code, Claude Co-work. Now today
50. 6:51we've got Muse from Meta. Um, we've got Grockbot from SpaceX AI. So So there's lots of these things out there. The
51. 7:00reason I don't really play with them on this channel is that they don't demo very well. They're kind of boring.
52. 7:05You're like, "Hey thing, go do me a thing." And then it's like, "Sure, boss." And it just starts, "I'm thinking, I'm thinking." And then 20
53. 7:14minutes later, it produces a result that disappoints you. Like your children. [laughter] [laughter] Hey, Tik Tockers. If you're on TikTok
54. 7:30and you want to join on the uh in the AI salon, just click on my my profile there and the hyperlinks there and we'll pop
55. 7:36you right over. Uh so come join us over here because I'm going to be showing some stuff on screen and if you want to
56. 7:40see the stuff I'm seeing doing with my fingers and mouse parts, mouse parts like I'm a serial mouse killer. [laughter]
57. 7:53I've got their brains over here. [laughter] Um, okay. Newly found [clears throat] treasures in the house. [cough] So, anyway,
58. 8:07those agentic platforms are powerful, but they're kind of boring unless you're kind of in the flow and working with
59. 8:13them and you can talk to them now. And so, there might be something we can play with there. Um, codeex today, um, chat
60. 8:20GBT work, you can now, um, put it in voice mode. and voice mode now has access to all of the tools. So
61. 8:28basically, if you've hooked up GPT work to your Google Drive, for example, or to your Gmail, now when you talk to it,
62. 8:38it's got access to those tools as well. It didn't before. You just had to do other with it. Um, so now you can
63. 8:44kind of do everything that you were doing by typing with voice. And so, um, so these agentic tools are are probably
64. 8:50going to get a bit more demoable as they get better. But I I found a loophole. Let's see. Kyle, please use this link to
65. 8:58join. Uh yeah, do me a favor, Linda. Don't put that here. Don't put that here just because we're recording live.
66. 9:10Yeah, don't put that here. Just send that to me in email or or or whatever. Um yes. >> Hey, Kyle. She I I've got it. Um we can
67. 9:20I'll I'll get it to you. >> Okay, cool. Perfect. Awesome. Um, so here's the loophole I found. There's
68. 9:31a new um, openw weight model from Quen. I think it's Alibaba is the company behind the Quen models. There's a new
69. 9:42image generation model called Quen 2.1. And you can run it on a laptop and it generates pretty decent results. They're
70. 9:52pretty good. I've done I did I did a classic uh uh resto muscle car resto mod in abandoned factory and it it was not horrible.
71. 10:03Wasn't great, but it wasn't horrible and it did like a banner and it did it in 70s font. So, it did it did all the
72. 10:09stuff decently well. And then um and so I'm using chat GPT work to install it just because I'm lazy. I didn't want to
73. 10:22go to hugging face and actually have to find the thing to download and then download it and make a folder and right
74. 10:29cuz then you'd have to read the read me. You want you want to honest to God's truth why I never got into open source
75. 10:37is that whenever you went to a repository you you'd have to read the read me file and you know I know I just I don't read manuals and so I
76. 10:49just I would get to that step and I'd be like I don't care. I don't care. I would rather not have free software than have to read
77. 11:00the readme files. So, [laughter] honest to God, my my my entire career is based on me being too lazy to do basic
78. 11:15[laughter] like StoryVine, my my nearly 15year-old automated video storytelling company exists because it took me like a week to edit a
79. 11:29two-minute video of my kids. And I hated it so much. I thought I would rather start a company to figure out how to
80. 11:35automate video editing than read the manual. [laughter] Getnormal.lovable.app. What is that? This is something that I
81. 11:48built to answer that question for anybody who didn't like reading the readme files. It takes a readme file and
82. 11:53a repo and turns it into like normal language. [laughter] >> That's awesome. That's really good. Get normal. Get normal.lovable.app.
83. 12:03I love it. That's awesome. Um, yeah. So, exactly that. So, what I said is I took it a step further and I said, "Dear chat
84. 12:10GPT work, just go here's the repository for Quen 2.1. just go get it. Go get it. Get get it and stick it on my hard drive
85. 12:22and then make it so I can make images. And then what chat GPT work did was it actually built me a custom app. It installed Comfy UI,
86. 12:34which I've downloaded Comfy UI a bunch of times, but then when I realized you actually have to install things and
87. 12:40connect them, and you kind of have to read the manual to figure out how to connect and use it, I was like,
88. 12:44"Eh, I don't care." So, it installed Comfy UI. It installed the model, and then it wrote a custom web front end for
89. 12:52it. And so, what I thought we could work on tonight is I'll show you the web front end. We'll generate some images
90. 12:58locally. I got two versions of the model. One version of the model is the uncensored version. So you can, you
91. 13:09know, make naughty pictures. And if you're into like horror films and with chat GPT and and with, you know, the
92. 13:18other commercial things, you can't make pictures of gory scenes. You can with this one. So I installed that. And then
93. 13:25I also installed a turbo version which which makes the images faster. And so and so they're both in there. I've got a
94. 13:31pull down menu where I can choose between the models. Uh, and it's pretty cool. So, I figure we'll work on that
95. 13:36tonight. I also want to go look at some news. Um, there's it it it seems clear that OpenAI's new models have um solved
96. 13:49more than a hundred mathematical problems. They are now working with the mathematical ivory tower community to
97. 13:57figure out how to announce this in a way that doesn't humiliate mathematicians. [laughter] I I don't I
98. 14:07there there there are a lot of knowledge work is confronting this ex exact same thing where as the tools get better and better
99. 14:18and better those people that were like not you know not going to happen on my watch well it's happening on their watch
100. 14:24and so they've got to figure out how to do that so it's not a PR nightmare so so that happened a really interesting thing
101. 14:31happened with claw OD's wet lab today, which this was last week they announced it. If you don't know what a wet lab is,
102. 14:40Kim Ken will explain it to you. We're sticking cells in petri dishes, people. Um, they made a big announcement today,
103. 14:49so I'll talk about that. Um, we got new models from Claude, we got new models from OpenAI, we got Muse today from Meta.
104. 15:01Um, I want to actually show you. I think the first thing I'm going to I think the first thing that I'm going to show you is an animation
105. 15:14that I think is really quite something. Okay, you're all over here. I can see you. There's Cam Catkin. Okay.
106. 15:30Get normal.loable.app. Okay. Good. All that's happening over there. I'm sharing my lovely calendar. And then we're going to go here.
107. 15:43And then we're going to go here. Hide this because that's a pet peeve of mine. People that don't hide that thing annoy me.
108. 15:59Wait, where is my Did I miss something? Oh, I know what I did. It wasn't I
109. 16:14didn't send it to Brandon. This is one I didn't send to Brandon. This is one I did. If I go to my profile, I do a post. here this okay
110. 16:29let me make this full screen okay so we have talked so where are we now we are effectively two months out
111. 16:42from the fourth year anniversary of chat GPT two months out so so we're at three three years 10 months of generative AI And we're
112. 16:58this community, the AI salon, but particularly the people that show up here late at night and hang out have been doing a pretty
113. 17:07damn good job of keeping up with technology. And I would say I would argue that sometime around 18 months ago,
114. 17:14our ability to do that essentially started failing. and people have gotten overwhelmed and we've created things
115. 17:22like daily practices in terms of how we manage through all that overwhelm and to varying degrees of success. But
116. 17:34the charge is relentless. Um Elon Musk describes it as a supersonic tsunami. What's happening? What's coming? And so what this
117. 17:46animation is, I don't know if you've seen it today, what this animation is is from November 2022 until today basically
118. 17:59um and it's just showing the major releases from OpenAI and Anthropic. So it's only two companies, the major
119. 18:07releases, but and it's 30 seconds long. So just pay attention to the acceleration.
120. 18:22So days between releases back here is like 100. Now it's dropping. So this is 25. So So it's just crazy.
121. 18:40you know how tight this is getting, right? And so, so it's accelerating and it's it's likely going to keep accelerating.
122. 18:49So, so what was that that end stat there? Here, we'll watch this again. Boom, boom, boom, boom, boom. 01 preview
123. 18:5901. That was when I realized AI was smarter than I was. The 01 models
124. 19:142023 once every 73 days. 2026 once every 18 days and that's likely going to drop before the end of the year. So by the
125. 19:22end of the year we'll probably be having major releases and this is just from two major labs, right? This does not include
126. 19:30Grock or Meta or Google or anything. Right. All right. Fantastic. Bob. Yeah. There
127. 19:45you go. Um, so yeah. So, so that's that. Let's go look at some other news that I think is worth worth taking a look at.
128. 19:58Uh there's a new uh new glasses came out from Meta today. I might go look at those. Um this one I want to talk about. So this
129. 20:09is this is a a post from Anthropic. A week ago, last week, Anthropic announced that they opened a wet lab,
130. 20:20which is a lab where you do actual scientific research using tissues and, you know, chemicals and proteins and
131. 20:29whatever you do in a wet lab to to, you know, bacteria, viruses, that sort of Um, they announced it last week. They
132. 20:39they announced a discovery today that has been confirmed in that wet lab. So, so Claude has discovered a previously unknown en enzyme system
133. 20:51hidden in the DNA of bacteria phase phasages, which if you haven't ever looked into bacteria phasages, they're
134. 20:58actually really fascinating. And the scientific community was heading well down the bacteria phase path when
135. 21:08antibiotics were discovered. And then antibiotics the the antibiotics lobby basically pushed the bacteria phase
136. 21:15people out. Anyway, it's a whole fascinating thing. But anyway, um so they so they said bes let's see they
137. 21:25they found this unknown enzyme system hidden in the DNA of bacteria phasages. Besides the enzymes gene sits a long
138. 21:33array of repeating DNA, a structure that looks somewhat similar to crisper. So, Crisper is the gene editing technology
139. 21:39that came out whenever five or 10 years ago that's allowing scientists to edit DNA and do customized drugs and things
140. 21:46like that. And so, this guy, this this post that I did, this guy wrote an explanation of it. Protham is his name.
141. 21:55449,000 followers. Um, let me explain what just happened. Anthropic built a biology lab where Claude works alongside
142. 22:03real scientists. They told Claude to search a massive DNA database for interesting reverse transcriptases,
143. 22:11enzymes that copy RNA into DNA. And then they let it run. Approximately 950 clawed agents worked in parallel for 21
144. 22:22hours across 200,000 enzymes. So basically in a 24-hour period, um 950 agents went nuts on these 200,000
145. 22:32enzymes. One agent spotted a repeating DNA pattern next to an enzyme. The same kind of signature that originally led
146. 22:39humans to discover Crisper, the gene editing tool that let scientists precisely cut and edit DNA now used in real medical treatments. It
147. 22:48cross-checked the pattern against known systems, searched the literature, confirmed that nobody had found it
148. 22:54before. So, so it discovered a new novel pattern, then confirmed that it it indeed was new and novel. Um, it then wrote up a full scientific
149. 23:04report like a human researcher would. Um, anthropic scientists verified it in the lab, named it ART. Uh, and it but
150. 23:13its exact function is unknown. It's still early sign. So, they don't actually know if it's programmable like
151. 23:18crisper is. Um, but this is a new DNA editing mechanism that they hereto didn't know about. Why it matters.
152. 23:27Finding tools like crisper used to take human experts months of manual searching. Here AI did that work and
153. 23:33found something real at the speed no human team could match. Um so don't misunderstand it. As Claude as Claude
154. 23:42curing cancer, Claude spotted something new in nature that humans missed, but it's still crazy. Um yeah, crazy, right? Bonkers.
155. 23:52All right. So, is the wet lab also where they're doing biomputing or biochips? Uh, what were we talking about, Cam?
156. 24:00That I don't know. It looks like Cam Katkin says yes to that, Mel. Um, but but that's that's bonkers. Um, let's see what else we got here.
157. 24:11Bunch of meta stuff. Video creation in Google Workspace. We've added creative controls to video to to Google Vids.
158. 24:22Cool. Because their model is actually really good. Google AI 10 hours ago. We're launching Gemini 3. 3.8 Flash text to speech and
159. 24:35Gemini 3.8 flashlight text to speech. Cool. The Open AI Eon link. Eon is going to be more than a chatbot.
160. 24:47Wait, who's this person? Is this person a real person? Probably not. Maybe. It could have its own profile DMs and groups designed to
161. 24:58compete directly with Grockbot. Oh, yeah. Okay. So, I guess Eon is going to be OpenAI's Grockbot competitor or Muse
162. 25:05competitor. Okay. This is something I called here. Um, let me let me pop back up here. Um, remember when when um we started out
163. 25:19with with Replet and cursor and um there was another vibe coding tool. I forget what it was. Um yeah, I forget what it was. Doesn't
164. 25:32matter. And then we got Claude code and then we got um codeex that both all all of those were designed for programmers
165. 25:44and then claude code created claude co-work um chat GPT created chat GPT work they they basically did renamed
166. 25:52versions of their programming tools as productivity tools right work tools and and then we see things like agent
167. 26:02agents come out like Grockbot where it's just really focusing on the agent. You're having a relationship with the
168. 26:08agent. You can create multiple agents. Muse came out from meta. It's now a single agent. Where these things are going is
169. 26:18increasingly simple. Hide all the complexity behind, you know, let the AI do all of the thinking. Let the AI do
170. 26:26all the routing, figuring out which models we use, figuring out how to spin up sub aents. All of that stuff is not
171. 26:33going to be anything we have to deal with. As early adopters, we like to tinker and we'll go in and play with stuff. You
172. 26:42know, I would consider myself I've got the the enthusiasm of an early adopter, but I don't have the patience to truly
173. 26:49geek out. So, I like to be right at that cusp where it's technical enough that I still, you know, it's still a challenge,
174. 26:56but it's easy enough that technical people make fun of it. What's coming is a stage more simple than that, right?
175. 27:03Which is just gonna have you're gonna have a single agent. You're going to talk to it and it's gonna go do
176. 27:07That's coming, right? And it's coming pretty fast. I think we're going to see it. We will likely see something from
177. 27:13Apple. We're going to we, you know, we just saw the one today from Meta. We've got Grockbot from XAI. And then um it
178. 27:20sounds like Project Eon Aon from Chat GPT is going to be one of these. And then we'll probably see something
179. 27:27from Claude as well. Um, and then we will likely see custom versions of this. Like there's one out right now called
180. 27:38you only use it in iMessage and it begins with an I. Is it called Instinct? I forget what it's called. Instinct? No,
181. 27:45I don't know. There there there's there's a couple of them out now that are just there there's Bordy uh 2.0
182. 27:52launched today. So, that's an agent you just talk to in iMessage and it'll make intros for you and do do all sorts of
183. 27:58things like that. So, so that kind of stuff is coming. Okay. Um, let's see. Digital Ocean managed agents. Here we go. RIP to
184. 28:07anyone who bought a Mac Mini. Okay. So, Digital Oceans, what this does is rather than buying a Mac Mini and having to sit
185. 28:13it on your desk to run 247, this thing uh just does it in the cloud. So, you can do that in the cloud. What is
186. 28:20project AON? project. This type of AI wet lab helps science and medicine leap frog ahead. Exciting. Chef Kelly, yes, I have Instinct. It is
187. 28:30called Instinct. Okay. Yeah, it's a new San Francisco startup. It's an agent that just works in iMessage. It's called
188. 28:36Instinct. It's by Invite Only and I forget someone in here gave me access to it. Who was it? Anyway, I haven't played with it yet.
189. 28:48Um, but that's coming. Uh, AON is OpenAI's rumored Grockbot competitor. That's what that is. Um, Digital Ocean
190. 28:58is uh, basically a Mac Mini in the cloud for lack of a better term. [snorts] Jev is now in lovable. Uh, I talked a
191. 29:05little bit about Jev in here. Oh, [laughter] so so this this one's worth going to look at. I'm not gonna play the audio
192. 29:14because his voice drives me nuts. But um so Trump at the UN basically said the UN has an AI treaty that that basically
193. 29:25says we're going to slow things down. We're going to cool out. We're not g, you know, we're going to cooperate, not
194. 29:29compete. And Trump basically said, we're not going to sign your AI treaty thing because we don't call it AI anymore. It's not artificial
195. 29:39intelligence. It's now called super intelligence. So, so we now live in the world of super intelligence. So, I
196. 29:46thought I thought you'd like to know that [laughter] we are living in a South Park episode. We are living in a South Park episode.
197. 29:58Um, Google just introduced a new line of laptops called Google Books, which I assume are like Chromebooks, but googlier. Um,
198. 30:12Jeb is now in there. Let's see. >> Jeb Jeb is now in lovable and you can use um it for unthrottled this week.
199. 30:23>> Oh, wow. Okay, cool. Maybe we'll go play with that. Although, I don't quite know what to build with it yet. Maybe they'll
200. 30:28maybe they'll have some some app ideas in there. Um, Muse connectors. So, Muse can now connect to other apps. So,
201. 30:36that's going to be similar to Grockbot and and ChatgBT work. All right, I think that's it. That's all
202. 30:44the big ones we can go. Let's go. Let me just do a quick spin through raw Twitter. Let's see what I see. Zuck has re revealed a muse powered
203. 30:58tomagotchi muse charm [clears throat] keychain. We packed the whole Muse experience, >> including the whole real time voice and
204. 31:11avatar stack into something that fits keychain. >> So, I I I watched this live. Uh they they've basically got their new Muse
205. 31:25agent and it can be down in this keychain. They actually haven't built this yet. This is like a prototype that he's holding. Um,
206. 31:34they're anticipating it to ship in time for the holidays this year. So, you has a thumb print or fingerprint sensor that
207. 31:43you activate it with for biometrics. >> So, you don't need your phone. You don't need your glasses. You don't need your
208. 31:49Facebook account. You just thumb press and tell it to do whatever you need it to do and it'll go off and work in the
209. 31:56background and and do your things. >> I assume it's connected to the internet though, right? >> I would assume so. either through uh
210. 32:03Bluetooth, Wi-Fi or or >> its own 5G. They were very scant on the details. He basically just showed it and said it's coming soon.
211. 32:13>> Yeah, that makes sense. And and this is this is probably some version of what Johnny Ies is working on for for for OpenAI.
212. 32:22Um real time voice with 11 Labs. Um, oh, here's the new glasses from Meta Spring 2027.
213. 32:42Why is there no audio here? Cuz someone pirated this. >> No, this is just a teaser. It's not actually uh >> doesn't have audio.
214. 32:51>> Doesn't have audio. But this one, this one should be good. Are you not sharing correctly?
215. 33:05>> No, I'm not hearing audio. I think I think someone just stole that video. Okay. Voice. So, Tibo from from Chat GBT
216. 33:17voice. I've gotten so used to to just call chat GBT and talk about all the work, check on emails, do some things.
217. 33:23Okay, we heard you loud and clear. Chat GPT voice can now use plugins like your email, calendar, and Slack. Be powered
218. 33:30by GPT Astro Soul and Luna. So, they've got this campaign with these these cute older ladies. So, let's uh let's
219. 33:39see what they got going here. >> Can you go to my calendar and find me a half an hour tomorrow so I can mail some
220. 33:46of these cookies to my grandson? >> I've added a post office trip at 10. >> All right, SJ, we're ready. Chaty PT Boys.
221. 33:53>> ChatchyBT boys got a whole lot more useful. It's no longer just answering questions. It's better at getting things done. [music]
222. 34:06>> Let me show you an example. Can you go into my finances and let me know if you see any duplicate charges?
223. 34:11>> Sure. Checking your accounts. Okay. It looks like you're paying twice for a magazine called Sound and Stereo.
224. 34:17>> Well, that's odd. Can you cancel the duplicate and send an email asking for a refund? >> Of course.
225. 34:23>> Can you make me a website where I can sell my pottery? >> Absolutely. I'll start building. >> This looks good. Can you add just a
226. 34:31simple check out page? >> Yeah, adding now. >> Can you send an email to the Fairfield Marina requesting a sailboat rental for
227. 34:38this afternoon? >> Sure thing. I'll send my email. [music] >> What time is my appointment at the salon?
228. 34:44>> Checking now in 10 minutes. Oh, got to go. >> Did the Marina boat rental email me back yet? >> Yep. They have a sailboat available at 4
229. 34:53p.m. >> Great. Book it. >> All right, sir. >> Can you check if I got any orders on my pottery website?
230. 35:00>> It doesn't look like you've gotten any orders. >> Darn. >> Can you go to my friend Lavel's website? I think she emailed me the link and I
231. 35:07want to buy a cookie jar. [laughter] >> Can you check the weather at the Fairfield Marina? Winds are around 8
232. 35:14knots out of the west with gusts up to 12. >> Like you know how to sail. >> Of course I know how to sail.
233. 35:20>> Right. I feel like we're in a farmer commercial. >> Say something nice to me. >> As you can see, you can now use chat GPT
234. 35:32boys to get things done on the go. >> Good. I think we got it. >> Thank god. Now, how do you sail a boat? Oo, great question. [music]
235. 35:45>> Cute. All right, my prediction. You're going to see those three ladies on the Super Bowl in a Super Bowl commercial
236. 35:51this year. Um, they've got the money. It's a good campaign. And I bet that makes it to the Super Bowl. You heard it here first.
237. 36:01She's on her way to the salon. I thought she called called it chatbt boy. Uh, not a lot of love for Zuckerberg amongst the uh amongst the
238. 36:10irregulars. [laughter] That's tough, Zuck. Oh, man. All right, let's let's go through a few of these and then we'll
239. 36:21head over to uh chat GBT work and I'll do a little building. Okay, I wish we didn't need these again. Okay, there's that.
240. 36:48Oh, the new trailer for the Elizabeth Holmes documentary is out. Let's watch this. This is disturbing. This keep you up at night.
241. 36:59>> Really feel you did nothing wrong. >> I did not. >> So, where did things get lost? Like, how are you where you are right now?
242. 37:09I I wish I knew, you know. I I don't know. I don't I don't feel like I understand that. So, I know you guys invited me to to stay here.
243. 37:26Does any part of you worry about me being here 24/7? >> I don't think so. >> What if I see something that makes you look bad? [music]
244. 37:42I think this is the moment in time to capture this story before I'm gone. >> I fell in love in that first conversation.
245. 37:54>> Before or after that she revealed she was the founder of Theronos. >> Probably before. >> I mean, weren't you worried she was uh
246. 38:01sort of scamming you? >> 100%. But what did she [music] have to gain from me? Well, isn't your family rich?
247. 38:07>> Yeah. I think that there's like a lot of different levels of wealth. >> Come jump in with me. >> You're asking me to go? >> Yeah.
248. 38:21>> You've [music] never been suspicious that she's ever lied to you. No. Ever. >> I still have this weird feeling that
249. 38:29there's some things that I'm not seeing. It's because I don't have anything to hide. [music] >> I know what you're feeling. [music]
250. 38:44>> No, you don't. I mean, it took me a really long time to
251. 38:58be okay with um with letting someone see me cry.
252. 39:24She's not too weird. [laughter] Muting your mic when playing audio. It's creating an echo. Oh, shoot. I wish I
253. 39:32just seen that earlier. All right. Well, I don't think it's muting my mic. I think it's I've got to change my audio.
254. 39:39>> No, I I muted your microphone so it cut the echo. There was like a a clapback. So, you're good. I just wanted to let you know you were on
255. 39:47>> I have e echo cancellation turned on. Okay, cool. All right, cool. Uh anyway, that looks wild. Dr. Singularity,
256. 39:5799.9% of people are not ready for 2027. So, listen, this is something that I think that we um I want to talk about something for a
257. 40:09minute here. I don't know if if uh if Cindy's still here. Um, I think we need to
258. 40:31learning about AI in the early days was like this magical mystical. there's this there's this new magical mystical tool that we
259. 40:44get to learn about and we learned about it and then you know we we dug deeper. We started have having critical conversations about you
260. 40:56know ethics and policy and is it good is it bad and how do we reclaim our agency and how do we work with it in such a way
261. 41:04that it amplifies who we are and what we do and makes new things possible for us. And that's a big part of what the AI salon's about.
262. 41:15But if you look at the the the definition of what a salon is, it's about curating conversations, you know, with people having thoughtful
263. 41:25conversations about, you know, some topic of the day. The acceleration that's happening right now, like it's just now starting
264. 41:37it. It's almost like what's happened in the past year, 18 months is is like a a tsunami where, you know, the water's
265. 41:46normal. We're sort of playing in the water and we're like, "Oh, there's we hear there's going to be, you know, we
266. 41:53hear there was an earthquake 500 miles away and we should probably learn to surf." And so we were playing in the
267. 41:59waves and we were learning to surf. And the past 18 months have felt like this this retreat where the you know the
268. 42:07ocean goes away. It just leaves the scene and we're like how do what what's going on? It's it's normal but it's not
269. 42:17normal. I haven't seen this much beach before. And the acceleration that's starting to happen right now and all of the chaos of
270. 42:29anthropic and open AI screaming to the government, our the things that we're building are going to kill you.
271. 42:39Save us from our competition. That whole narrative is part of the acceleration because I think what they recognize is
272. 42:50what they're build building is moving fast enough that it's going to do things that they're going to be held liable
273. 42:58for. This is this is my total chat TMZ speculation that this is all about them limiting their competition and limiting their liability.
274. 43:11And it's freaked everyone the out. And I got a letter from someone in the salon who's been part of it for a long time
275. 43:23about 10 days ago that said, "I'm piecing out. My artist friends hate AI and the data centers are tearing my
276. 43:30other community apart and I can't be a part of this in my right mind." And I think that one of the things
277. 43:42that I want to do in the AI salon is to start to have bigger and harder conversations about some of the bigger things that are happening.
278. 43:57One of the things that makes me nervous about that is a lot of that stuff is out of our control. The one thing I love
279. 44:04about the salon is what we're doing is we're managing the things that are in our control, but we're doing it in an environment
280. 44:13right now that is incredibly noisy. It's incredibly chaotic, and it's about to get noisier and more chaotic. And so I feel
281. 44:22a weird irresponsibility if we don't shift the dialogue in the AI salon or at least create significant areas for that kind of
282. 44:36conversation to happen. I don't think it means that things like, you know, the AI learning lab goes away or, you know, workshops that teach
283. 44:45people how to do stuff go away. I think it's, you know, I mentioned earlier about the the AI salon feeling like a
284. 44:51big city with all these different neighborhoods. I feel like especially for the people who've been here for a while and have
285. 45:00been through these cycles of excitement and optimism and, you know, doom and dread and then I I don't know if I'm
286. 45:08going to work again and then maybe I am and maybe I'm not. Now I'm too busy and now I'm and I but I still can't keep up
287. 45:13and I thought I knew what I was doing and now I don't know what I'm doing. There's a level of conversation that I
288. 45:22think is important for those of us who've been around for a while to have that's that's just not happening right
289. 45:27now because they're hard conversations. And I don't know how you have those conversations them without them being
290. 45:34just masturbatory overindulgent conversations that accomplish nothing. So, so one thing I want to do is I want
291. 45:43to gather a bunch of my favorite people and have kind of an all hands meeting about the AI salon and and what we can
292. 45:49and should do about what's here and what's coming. Um, so anyway, so I just put that out there. I I would encourage you to just start
293. 46:00thinking about that. Um, what are the kinds of conversations that we should have more of or what are what
294. 46:09are the kinds of activities we should do that create space for the kinds of conversations that allow people to process what's happening. the
295. 46:23like one of the canaries in the coal mine for me of of the magnitude of what's happening the magnitude of the acceleration is what's
296. 46:33happening in mathematics where a new model from open AI has apparently solved more than a hundred unsolved mathematical problems across
297. 46:46all areas of mathematics. a hundred, not Navier Stokes, which that one may or may not be solved, but a hundred of
298. 46:54them. And they're now working with mathematicians to figure out how do we confirm that this is true? If it is, how
299. 47:02do we announce it? How do we announce it in a way that supports the mathematics community? There's a tremendous amount
300. 47:09of energy being spent talking about how we talk about what this model has apparently already done. Like it has solved problems
301. 47:23and why we don't know about it yet is they're trying to figure out how to talk about it. How do we as a community,
302. 47:32you know, we've got an ambassador program and the ambassadors are charged with go talk about AI in groups that are
303. 47:38trying to figure this stuff out. How do we have conversations about what AI is and makes possible in an environment
304. 47:46where people are literally trying to figure out, you know, what's happening and how do we keep up with this thing
305. 47:53and what is all this stuff and other people are like, what's AI? So there's there there are there are are deeper
306. 48:04non-trivial conversations to have that could very easily just turn into waste of time conversations. And so how
307. 48:16do we identify what's worth talking about and then how do we talk about that in such a way that it creates a safe environment
308. 48:29for people in the salon that really want to understand this stuff and move forward with it. Anyway, AI equals emotional whiplash.
309. 48:39Did you pose that question on the salon? No, this is this is two-day old thinking, Lori. This is I'm I I I am
310. 48:49I'm looking at I'm we're we're coming up on the fourth year anniversary of the salon and one of the things that we're
311. 48:55looking at is what do we want to be when we grow up, right? We we looked at what we've done over the past year. Um Andy
312. 49:02joined us about a year ago. Brandon officially joined us. Brandon's been here for a long time. Uh but officially
313. 49:08joined, you know, as part of salon staff in 2026. And so we're looking at here's what we've accomplished so far this
314. 49:16year. What do we want to do in 2027? And for me, the what we currently do are expansions and remnants from what we
315. 49:30started in 2022. Um, and we're in a completely different environment. And so that's what it's about. New new viewer Tik Tok pin. I
316. 49:40think the majority of people are genuinely afraid or simply don't care enough to learn it. I think that's right. Um
317. 49:52my problem is they're genuinely afraid of a thing that they've never tried. The whole point of the salon is to create a safe haven
318. 50:01for people that are curious about AI, people that are trying to figure out AI. And one of the things we consistently
319. 50:07hear from people in here is that in the AI salon, they can talk freely about what they're doing with AI, what they're
320. 50:16learning with AI, what they're frustrated about with AI, what they hate about AI, and they can just have a a
321. 50:22conversation about it. And that they don't have that anywhere else in their life. My problem is if we're just
322. 50:29talking about the tools and how to do automations and work stuff and creative stuff, we're missing this bigger thing that's
323. 50:39happening. Um because I've had this fantasy of when AGI gets here, it's
324. 50:53going to be this cool magic moment. And um I'm looking at what's happening with the mathematicians and it's not cool and magic.
325. 51:06their identities have been threatened and they're not responding well to it. And I I you know that's going to
326. 51:12continue to happen, but it's going to happen in accelerated and amplified ways as the wave starts coming on shore. This
327. 51:19supersonic tsunami where right now there's this pullback going. You know, the pullback has been well, it's not
328. 51:26really real. It's not real. It still sucks. It still sucks. It's not really real, right? It's I know that model's
329. 51:30better, but it still sucks. It still sucks. It still sucks. Right? This is like a a a really tight rubber band
330. 51:36that's being pulled and pulled and pulled and pulled and pulled and and we're I think we're all to some degree living in the fantasy
331. 51:46that this technology isn't going to get dramatically good dramatically fast. And when it does, it's going to come in like that.
332. 51:58Most people don't learn anything anyway. I don't know if I agree with that. I think people learn stuff. People learn
333. 52:02what they need to learn to survive. I think if if you're saying people don't proactively learn stuff, there's not
334. 52:08enough curious people in the world. I would agree with that. Um it's it's difficult, but talking about it is important all aspects. Yeah,
335. 52:18exactly. I'm just a little fish trying to barely understand the conversation. I know. So am I. I like We all are. We all are.
336. 52:31Our break is over. Here's Johnny. [laughter] Oh my god. All right, let's see. It's 8:22. Let Let me hop over right now and go share.
337. 52:46Oh, did I not? Oh, there it is.
338. 53:04Hang on. Lord, digital God just wrote us a book. >> He wrote us a book. Oh, boy. Okay. Hang on. Okay. The Lites. Oh, here we go.
339. 53:22The Lites were skilled textile workers in England who destroyed machines from 1811 to 1816. They were protesting
340. 53:28employers use of those machines to cut wages and replace skilled work rather than opposing every new invention. Um
341. 53:36the British government sent troops to suppress them. Some leites were arrested, executed, transported to penal
342. 53:42colonies. The movement faded, but the mills and machines remained. Many textile workers had to adapt to lower
343. 53:48paid factory work or find other livelihood. Yeah, exactly. What's this? Regardless of how you feel about Tim Balu, I don't know him. He's a
344. 54:02wellressearched Here's a This is a wellressearched commentary on following the money far beyond s and Dario. Yeah,
345. 54:09there there's just there's weird stuff going on right now. So anyway, um I think we need to find a way to elevate the dialogue.
346. 54:24um in a way that's not a waste of everyone's time. Yeah. So anyway, for what it's worth,
347. 54:41I don't know what that means yet, but you heard it here first. Okay. So,
348. 55:00okay. All right. Cool. I got that. Okay. So, here's here's where I am. I am in um I'm sort of uh lpped over into my um
349. 55:14my MacBook that's that's running local models. And so it's running an LLM right now. And now it's now it's got an
350. 55:23application on it called Local Image Studio. Now this application doesn't exist anywhere except on my hard drive. Um,
351. 55:31all I did was I went to um Claude code or I mean Claude or uh chatgbt codeex. Can I get how do I get rid of this? Get
352. 55:43rid of that. And where did I start this here? Here was my prompt. Go find the latest best Quen 2.1 image model that is
353. 55:53uncensored because I heard there was an uncensored one. um and download it to make it easy to use here. Not sure if I can use it with
354. 56:02LM Studio or if you need something like MMX, but I want it to be easy to make images. Now, for someone on this call
355. 56:11like Lord Digital Gods, a prompt like this probably makes him a little sick to his stomach [laughter]
356. 56:17because he's like he's like, "Kyle, you could just go read the you could just go read the readme file and install the thing with lm-gpack
357. 56:26slash4112.6.7.trz enter twice." Make sure you hit enter twice. So, and I'm not making fun of Lord Digital Gods. Like, this is how his
358. 56:40brain works. That's not how my brain works. My brain works like this. Go find me the latest stuff and stick it on this
359. 56:46machine, [laughter] which is why I like AI. AI AI is here for the impatient uh add friendly people amongst us. Um, and
360. 56:58then and then um Codeex went off and did it. It it found it. It found the Comfy UI um official file. So, it installed
361. 57:06Comfy. It installed the model. Um, it made me a website. And then, let's see. I'd like a website where the images
362. 57:21wait where I make the images to have a bit more. I'd like a progress indicator. So, basically, I asked it to improve the
363. 57:28website and it did that. Um, what else? And then I said I found a new model. There's a new turbo version of this model
364. 57:43and then it improved it. It went through all that and then it ran out of then I ran out of usage. So I've been using the
365. 57:49GPT6 sole model. I just have it in medium because this isn't really there's not really a lot of uh of big
366. 57:57functionality going on here. But let's let's go look at the app that it built. So, this is LM Studio or LM, what is it?
367. 58:09Local image studio. There's my uh raccoon skateboarding. Skateboarding raccoon flipping tricks. Oh, I also had
368. 58:17it I wanted it to improve prompts. So, it did that. It It added an improved prompt button. So, I can if I click on
369. 58:24that, it actually uses my LM Studio um Quen 3.6 six model uh to do some some large language model magic.
370. 58:34Um here's the pull down menu for the two different models, right? The Quinn 2.1 original, that's the one that's the
371. 58:41uncensored model. And then here's Vigle Turbo. This is makes images in four steps that that are pretty good, but
372. 58:48they're fast. Um I said that I wanted to be able to upload images. So, I can upload up to four images and then I can
373. 58:56choose whether those images are subjects or styles or both. Um, it's got u aspect ratios in here. It's got standard resolution or 4
374. 59:09megapixel. It's got number of steps, so 25 steps or 40 steps. Um, or if I use the turbo model, it takes that away because that's just four
375. 59:18steps. And then I can do transparent backgrounds or normal backgrounds. and then it puts a seed number in there and can generate images. Um,
376. 59:28so let me I'll I'll go ahead and make an image. Let me I'll flip this over to the turbo model. And then there's its improved prompt.
377. 59:42And then I'll just do another 3x4 portrait. And then we'll say generate image. And so it it pops us up to 6%.
378. 59:50and then basically waits for the steps to start loading. So I think what it's actually doing now is loading the model.
379. 59:58So it's counting the elapse time down here. It's showing me the the size of the image it's going to create and then
380. 1:00:05it's doing its thing. Now one of the things I don't like about this is I've got to scroll to see the whole interface. So we can actually go
381. 1:00:14fix that. So that's what I'm going to do. I'm going to have I'm gonna have chat GBT go fix it. By the way, if
382. 1:00:20you're on Tik Tok and you want to see what I'm working on, it's we're over in the AI Salon. If you go to my profile,
383. 1:00:26click on that. That'll bring you to into the AI Salon uh webstream. So, there's uh steps one, two, three, and four.
384. 1:00:35And then it'll finish it up and generate an image. This takes, I don't know, five or 10 seconds to get to the end of this thing.
385. 1:00:56And if you're like, Kyle, this is really slow. Well, yeah, but it's running. There's there's our there's our
386. 1:01:05threearmed raccoon skateboarding. [laughter] Cindy you still here? [laughter] We have we have genetic anomaly. Roger.
387. 1:01:21[laughter] Oh my god. Okay. But the the whole point here is the this image is being generated on my local laptop, right? I'm
388. 1:01:33not having to go to Midjourney or OpenAI or anything like that. It's being it's being generated locally. So the the
389. 1:01:41local models are starting to get decent. Um, and that and that's just going to continue to get better and better. Okay.
390. 1:01:48So, now I'm I'm back over at chat GPT work. So, I'm going to go um let me just make sure my usage isn't destroyed. No, I'm good. Okay. Um,
391. 1:02:02okay. Uh, I'm gonna go take a I'm going to go take two screenshots. I'm going to take this screenshot.
392. 1:02:27Paste that there. Oops. Do this to the clipboard. I did do it to the clipboard.
393. 1:02:47Okay, there's that. And then I'm going to scroll down here and show the bottom of it.
394. 1:03:02Okay, let me say I Oh, wait. I can't do that because it's on a machine not by me. So, I'm
395. 1:03:16just going to say um the interace wastes a lot of vertical space
396. 1:03:32by having all of the interface elements in the left column. And by having so much
397. 1:03:49padding left and right, why don't you design the inter base thoughtfully? In fact, you know what I'm going to do? I'm gonna I'm
398. 1:04:03gonna up my effort here to high. Medium. Medium isn't going to cut it. I actually want to think it uh why don't
399. 1:04:12you design the interface thoughtfully um to better use the space organize.
400. 1:04:26>> Hey Kyle. >> Yeah. >> Before you hit enter, do you want to see a cool trick that we learned about in my Codeex mastermind course?
401. 1:04:34>> You mean the one that I didn't go to or I would have learned this if I had gone? >> Uhhuh. Uh well, you were traveling so
402. 1:04:40we'll give you a pass for last. But uh if you do a if you follow this up with a slash plan, it will put it into planning
403. 1:04:47mode and it'll actually come up with a plan before it executes. >> Okay. At the end of this >> anywhere in there, if you just do slash
404. 1:04:55plan, it'll add a little planning badge to the bottom. >> Oh, cool. I like it. Okay. Um, better use space. Organize options intuitively
405. 1:05:07and make it more like world class software flash plan. Did it do it? It did not.
406. 1:05:21>> You have to hit either press space or select it. >> What's going on? Oh, did I record? What's going on here?
407. 1:05:38If you just hit space, it should lock it in. >> Maybe not. >> No, it's I've got this weird sort of gray box where the
408. 1:05:54Here it is. Let's see. skills. I don't have plan in here. >> It's right at the top. It's just under the gray box.
409. 1:06:10>> There you go. No, it's plan deal strategy, but >> I don't know what that is. >> I don't have slash plan.
410. 1:06:24>> Uh it's probably above the scroll point. That's all right. It's it it's there somewhere, but >> All right. Well, I learned it, but
411. 1:06:33I'm not gonna do it just because I can't find it. [laughter] Okay. Let's see what is going on.
412. 1:06:50Okay, that's off. The interface wastes a lot of space. Okay. And so now it's going to go off and do stuff.
413. 1:07:06What is happening? Why do I have this? Am I recording something? I might be recording something. See this gray this gray box?
414. 1:07:30Yeah, it's still there.
415. 1:07:52All right, let's see. Current page make the controls feel like a long form, leaves too much of the canvas unused.
416. 1:07:59I'll reorganize it into a wider workspace. So, let's flip over to here, and it should just start editing this and fixing this.
417. 1:08:10Capture selected position. Oh, I bet I know what I did. No.
418. 1:08:46loaded tool, read the files, read impeccable skill, read layout.md, init.md, product.md. So I went and looked at some skills.
419. 1:08:59I have the new layout mapped out. I'll keep the existing generation behavior and move the controls into a compact horizontal strip
420. 1:09:07with the prompt and references side by side and the image on a full full width stage. All right. So it hasn't fixed anything yet.
421. 1:09:26So, I accidentally escaped out of this. And this is what I meant when I said when I said showing these agentic tools
422. 1:09:40is tough. Part of the part of the reason is because it just because they go off and think and do work and there's
423. 1:09:46there's not a lot lot of stuff to do there. Oh, and I just disconnected from my from my machine. So, let's fix that.
424. 1:09:56Oh, and now it got rid of my that weird gray thing. Okay, so that makes me feel better. I guess while that's doing that, we
425. 1:10:10could do another We could do another image. Let's do a uh um kid at a carnival
426. 1:10:24dressed as a zombie that says I like turtles. [laughter] We'll improve that prompt.
427. 1:10:40What's going on in the comments over here? All right. You should know better than post stuff.
428. 1:10:55I got Quen 2. I'm using Rockspot to run it for me. No interface at all. Oh, that's cool. Charles Adams, I knew that trick
429. 1:11:06already. That makes me feel excessively competent. Posted a live posted the live podcast recording with Kyle tomorrow. Very cool.
430. 1:11:20Gareth's coming back. Oh, look. We got a new interface. [laughter] That's pretty cool. Do I Do we like it like that?
431. 1:11:43I don't know if I like it like that. Well, let's let's try to we'll we'll let's make something. Okay. So, Vigle Turbo.
432. 1:11:54We're going to go landscape. We'll do 16 by9. We'll do standard 1 megapixel. The quality is four steps fixed. Seed is
433. 1:12:05random. Background is normal. And we'll say um a pan of brownies.
434. 1:12:21Oh, what did it just do? Oh, it's still working. This [laughter] is this is sometimes sometimes I'm a very bad uh I'm a [laughter]
435. 1:12:35I'm a very bad agent boss because I I have immediate gratification syndrome and if it ain't giving me the good
436. 1:12:44stuff, uh that's bad. Let's go look at our usage. Uh speaking of ADD, Why can I not do that? Oh, my whole my
437. 1:13:00whole uh viewport is hung. Okay, I redesigned the local image studio and it's live. I also installed Vigle's recommended
438. 1:13:12turbo checkpoint, verified its SHA265, and generated a test image successfully. Oh, did it did it generate an image? No.
439. 1:13:26Okay, that's okay. I guess it did it locally. All right, you can switch models on the page. The uncensored LM Studio
440. 1:13:34model powers the improved prompt. The original image checkpoint itself is the standard Quen image 2.1.
441. 1:13:42Um, okay. So, it's done. Let's go look at our usage. So that thing that we just did where it redesigned the website when I was using
442. 1:13:56Astra, I was blowing through I don't know 40% of my 5 hour usage. That that only used 8% of my five hour usage um to
443. 1:14:06do to do that change to the website. So that's not bad. a pound of brownie. Pan of brownies steaming um just out of the
444. 1:14:16oven in a lovely Amish kitchen. Uh and then we'll have Oh, here's the improved prompt but button up here. So, there's the improved
445. 1:14:31pro uh prompt. A square baking pan filled with rich dark chocolate. I got all my settings set. I could do a style.
446. 1:14:43Actually, let me remove that style. And we'll generate this image. So, this is the we're using the turbo model.
447. 1:14:58And then I think what I could do, let's see what I want to do here. Do I want to go
448. 1:15:13Oh, there's my fly. I could go to How we doing on time? Few minutes left. Okay, I'm going to go to midjourney. Midjourney.
449. 1:15:30I got to log in. This is going to be a pain in the butt. Go Google. What's wrong, champ?
450. 1:15:49All right, we are going to go find a style. That was a bunch of the new great repurpose branding. Let's go grab
451. 1:16:09something very specific. All right, let's grab grab this. That's quite specific.
452. 1:16:24So, I'm going to download that and then I'm going to come back over here. All right, there's our there's our brownies.
453. 1:16:37That looks like that could be an Amish stove. That's not bad for a four-step image. So now what we're going to do is I'm going to upload
454. 1:16:54this really specific image. And then I'm going to say style. I'm going to choose this as the style. And
455. 1:17:09I'm going to get rid of the seed. And I'm going to generate an image here. And we'll see if it actually uses the
456. 1:17:18style. If it doesn't, then we can go tell um chat GPT work. What's going on?
457. 1:17:33Thoughts on Muse? Um I haven't played with it yet. I I Well, no, I played with it briefly early. One thing that you should know, Daisy
458. 1:17:43Thomas, Daisy, Daisy and I both got early access to Muse Meta has given us if you sign up using Daisy's code, which she posted in the AI
459. 1:17:57salon. I don't know if you found that, producer Brandon, but you'll get a billion tokens uh to use with Muse. So,
460. 1:18:04if you want to use one of these agentic platforms, I have a bunch of, um, tokens to play with, um, go sign up for Muse
461. 1:18:13via via, um, Daisy's post in the salon. And if you can find that, Brandon, and and put that somewhere people can find
462. 1:18:22it, that would be awesome. Okay, so that did not work. Okay. So, what I'm going to tell yield, I'm going to go back over to chat GPT.
463. 1:18:38I'm going to say I just uploaded an image and set it to use as style.
464. 1:18:54and the image gen. Oh, you know what? Wait, we got to check something. It might be that you can't do a style with the turbo model.
465. 1:19:07Let me do a 25step version of the uh the original model. I'll be right back. Champy's whiny. Hang on.
466. 1:19:47All right, on step four of 25, five of 25. All right, so our image is generating. Anyway, um I thought this was pretty cool that
467. 1:20:01you can just tell these agents to install something for you and they just make you apps that are kind of cool and
468. 1:20:08then you can go fix the apps. Fabiana, my son is a computer science student and we're always talking about what is
469. 1:20:14happening and coming up and he's still excited for the future. So am I. Yeah. Listen, I mean, here here's the thing.
470. 1:20:22With all the doommongering and all that sort of stuff, it's easy to lose sight of the fact that holy a 100
471. 1:20:29unsolved math problems are are being solved in 24 hours. Um, they discovered a crisper-l like gene editing tool in
472. 1:20:37nature that, you know, could have been discovered but but hadn't been. Um, there's just like my ability to even do
473. 1:20:46this that there's a model out there that can generate images that you can just run on a local computer. You don't have
474. 1:20:52to pay anyone else to do that. That I can just do this and that I can just ask some other thing to install it and set
475. 1:20:59up an interface for me to be able to play with it. It's pretty flipping cool. So, we we are living in wild wild times.
476. 1:21:07Uh uh uh uh uh uh uh. Let's see if this used our style. And if it didn't, then we've got some Yeah, it didn't. Okay. Right. That's annoying.
477. 1:21:20Oh, but I had this seed in here. That seed could have it up. Well, doesn't matter. It's unexpected behavior. So, we're going to go over
478. 1:21:28here. I'm going to say um I uploaded an image and selected style and the image had
479. 1:21:46zero impact on the image generation. Did you just give me a useless interface element
480. 1:22:03or is there something else we need to connect? Okay, let's challenge this bad boy. Let's see how good GPT6 Soul does. I'll trace the
481. 1:22:20uploaded uploaded image through the workflow that actually ran. If it reached the model but didn't affect the
482. 1:22:26result, I'll fix the reference handling or make the limitation clear in the interface. Good. That's a good thing.
483. 1:22:35Kyle cools his local AI system using his neighbor's pool water. [laughter] I've got my own data center here. I've
484. 1:22:45got I've got a Costco purchased gaming PC that I don't really use anymore. I think I'm probably going to put
485. 1:22:53something some small model on that because it's I've only got eight gigs on that graphics card. And then um and then
486. 1:23:01I have my Mac. So I I have I have a two I have a two computer data center over here. And I I've I've I actually cool it
487. 1:23:12in the morning by spilling orange juice on it. >> [laughter] >> Oh my god. Oh my god. Okay. The upload did reach Quen. The saved workflow
488. 1:23:26contains the image node. Yeah, because it's using uh Comfy UI in the back end. It connects it to the reference encoder.
489. 1:23:33Includes image one in the prompt. You're still right about the result. The note reference and style reference images are
490. 1:23:41nearly the same. I'm checking whether the workflow needs a different edit setup and whether style promises more
491. 1:23:49than this model can reliably do. All right, good. Maybe you could have figured that out the first time I asked
492. 1:23:55for this, but you know, whatever. Uh, by the way, if you are on Tik Tok and want to see what I'm doing over
493. 1:24:07here, uh, I'm in the AI salon. You just go to my profile, click on the link there, and that'll bring you over to the AI salon. Um,
494. 1:24:19tomorrow we've got the mastermind practice lab at 10:00 a.m. Mountain time. We've got producer Brandon's um
495. 1:24:29codeex workshop where you can learn things like slash plan, which I didn't learn because I didn't go there uh
496. 1:24:36because I was out last week. By the way, my trip to Philly went really well. Um, ended up speaking on stage with one of
497. 1:24:44our clients, one of our StoryVine clients, and we talked about the project that we're working on, and she talked
498. 1:24:49about the results they're getting, which are pretty impressive. Okay. I found the gap. The image was connected, but Style
499. 1:24:57used the uploaded image as a loan reference while asking Quen to create a new picture from scratch. Quen's
500. 1:25:04documented style transfer pattern uses two images, the picture to restyle and the style example. I'm changing this to
501. 1:25:15gen to generate the requested picture first, then restyle it using the uploaded image. This will take an extra pass. So
502. 1:25:25I'll test test whether the results earn the added time. Well, that seems stupid. I'm gonna stop it. I'm gonna tell it. I
503. 1:25:37think that seems stupid. Um, that seems stupid to me. As a user, I don't
504. 1:25:52want to have to wait double time for a good result. Is there another way or a more clever
505. 1:26:06use of comfy? Or maybe your big trillion dollar brain brain.
506. 1:26:23Get smarter. [laughter] Oh,
507. 1:26:37is it Daisy who posted a Muse invite? It was Daisy that posted a Muse invite. If you use her invite, you get lots and lots of tokens.
508. 1:26:47Lots and lots and lots of tokens. A billion. That'll probably at least take you through Friday. Reconnecting.
509. 1:27:02You're right. The second image generation pass is too expensive for a style reference. I'm stopping that test
510. 1:27:08and looking for a single pass approach that turns the uploaded image into a concrete style instruction before Quen
511. 1:27:15generates. Oh, that's going to be a disaster. AYA22U is the invite code for Muse that if you
512. 1:27:29use it, you'll get a big old pile of tokens. All right. All righty. Check the running test. Stream disconnected before completion. Error
513. 1:27:43sending request for URL. I'm going to go try again. What happened? Blow through all my tokens. 85%. We're down 15%. Try.
514. 1:28:00All right. Well, we'll see how that goes. Anyway, I hope you see that, you know, with just a little bit of um resourcefulness and
515. 1:28:15the ability to ask stupid questions to things like codecs, you too can make your own image generation tools that run
516. 1:28:23locally on your uh on your computer. Yeah. Right. So, so here's what we're going to do. Danielle, thank you. Code
517. 1:28:33Daisy, the code worked for me. One of the reasons to come to the AI learning lab is you get to benefit from generous
518. 1:28:43people like Daisy, who was willing to share her invite code to get you a billion tokens. And we all now know that tokens are
519. 1:28:54expensive. Mark Zuckerberg is trying to seduce you, ply you with free tokens. [laughter] All right, everybody. Uh, I'm going to
520. 1:29:06go on and get the hell out of Dodge. Um, I appreciate y'all. Appreciate you hanging out. Um, I will see you tomorrow
521. 1:29:14if you're a mastermind member. I'll see you in the practice lab and at Brandon's Codeex workshop tomorrow. And then
522. 1:29:21Friday, we've got office hours at 11:00 a.m. Mountain time and Friday night date night. Friday night right back here, Mr.
523. 1:29:30>> and your uh interview with Linda tomorrow. >> Oh, yeah. Yeah. I'm gonna be on. If you go on LinkedIn and look at my profile or
524. 1:29:37look at Linda Sherman's profile, you'll see the link to my interview tomorrow with her talking about aging and AI and
525. 1:29:46keeping yourself as bright and shiny as as this cognitive large language model is. [laughter] Thank you, Source Camp. I appreciate the
526. 1:29:59kind words. All right, peace out everybody. Hope you have a beautiful night. Bye.