8/28/2025 - The Power of the Creative Generalist in the Age of AI

Transcript

1. 0:12[Music] Every time I say it now, get that off your mind. Every time I see your mouth, I hear that smile.
2. 0:28Early misty morning light heard the engine turning the old fall outside. [Music] You would even me
3. 0:45again today. You will convince me again today. You're leaving this hot tail looking for someone else's cold and ring.
4. 1:01[Music] should have said so long, Suzan. [Music] Hush now. Don't you cry. [Music]
5. 1:16So long as she was head, [Music] don't you cry for me. [Music]
6. 1:30Sharing jeans and cigarettes and keeping warm out on the road. Chasing down a lifestyle out on Highway 24.
7. 1:41New York states a rolling breeze and the sunshine with a blue sky falling chill on September dreaming. You believe in me
8. 1:58again today. You will convince me again today. Leaving this hard time looking for someone else's golden ring.
9. 2:15to say so long Suz. [Music]
10. 2:48Oh, that champ Shannon. That Champ Shannon. He's singing so good. [Music]
11. 3:27[Music] Good morning. Yay.
12. 3:42Woohoo! [Music] [Music]
13. 4:07Hey, [Music] hey, hey. [Music]
14. 4:30Heat. Heat. [Music] [Music]
15. 4:45[Music]
16. 5:12All right, it is a Th Thursday th Thursday. Is it Thursday? Get it? Like you're thirsty. It's Thursday.
17. 5:29I feel like I'm so clever sometimes. [Music]
18. 5:53Encourage your viewers to share your live. Hey viewers, I encourage you to share my live. I wouldn't say that if Tik Tok hadn't
19. 6:06suggested it. They really did encourage. See Danielle sharing away. She's been sharing away the whole time. All you other slackers,
20. 6:14you're on there like, "Where was I don't know where the share button is? I ain't seen the share button. Is there a share button somewhere?
21. 6:23My eyes aren't so good anymore, you know. But I'm not judging you. I'm not a little just a little judgment.
22. 6:32Just small pico amounts of judgment. Nano. Nano banana amounts of judgment. [Music]
23. 7:04[Music] Oh, good lordy good people. God lordy good people.
24. 7:21[Music] The boys were singing shingling the summer that we met.
25. 7:37You were 10 and 17. How could I forget? All the stars from near and far were watching from above.
26. 7:52too little da fall in love. I don't know what those words were. The way we danced was not a dance but more a long embrace.
27. 8:07We held on to each other and floated there in space. I was shy to kiss you while the whole wide world could see. So shingle said
28. 8:22everything to me and all the poor old old folks. They thought we'd lost our minds. They could not make heads or tails of
29. 8:37the young folks. Funny rhymes. You and I learned all the words and we always shang along to old shaming dong ding. Shamboling
30. 8:53dang. [Music]
31. 9:10That's no Jesse Winchester number. Do I know any songs from the dog days of summer? I do not. I need to learn some new songs,
32. 9:24as anyone who watches this channel regularly knows. I will occasionally remember one like that one. came on him like slow moving cold
33. 9:39fronts. His beer was warmer than a look in her eyes. [Music] Sat on a stool and he said, "What do you
34. 9:55want?" [Music] She said, "Give me a love that don't freeze up inside. [Music] [Applause] [Music]
35. 10:11He said, "I have melted some hearts in my time, dear. But to sit next to you, well, I shiver and shake.
36. 10:26And if I knew love, well, I don't think I'd be here. Asking myself if I've got what it takes. [Music]
37. 10:41A two. A two. Your icy blue heart. Should I start? Turn what's been frozen for years
38. 11:00into a river of tears. [Music] Oh man. All right, let's get going.
39. 11:17Let's do something. Let's do something tonight. I know. We'll do Nano Banana. Okay. I just made a video. So, I think I
40. 11:24think I'm in some sort of Tik Tok [ __ ] list. I don't I don't know what's going on. I I guess they've got some new
41. 11:30algorithm and they don't like my [ __ ] Um, but I put out a video tonight called I'm an idiot. Maybe they didn't like me calling myself
42. 11:41an idiot, but I called you idiots, too. Um, one of the things that's been striking me the past couple of nights, few
43. 11:50nights, three nights, four nights, is as powerful as the new Gemini image generation model is, I don't know how to use it.
44. 12:02Like, I'm getting really inconsistent results. Sometimes I get results that are like mindblowingly good. Sometimes I
45. 12:10get results that are like it didn't even try to give me what I thought it would give me. Um, I feel similarly with chat GPT5
46. 12:20right now. Still, I feel like a clueless newbie. Feel like an idiot. And I don't listen, I idiot might be a harsh word. I
47. 12:27don't know where your uh sensitivity is on the triggering list with the word idiot, but but uh but um you know, just
48. 12:37feeling like a dumb dumb, feeling out of it, like not connected, like don't have a handle on it. And the point I made in
49. 12:44the video was if if every time a new model gets released, you're like, "Oh god, I just I just don't know what I'm
50. 12:51doing." It's not you. Like that's just I think it's the natural it's the natural shape of things as as especially as we're living in an
51. 13:03exponential technology development curve. These tools are going to get better and better faster and faster and
52. 13:09our ability to keep up with how to use them is just it's going to be bad. So give yourself some grace. But I figured
53. 13:17we'd play with that some more. We love anyway. Mr. He loves us anyway. Even though we're clueless. Even though we're clueless. Dude, give
54. 13:27yourself a break. How many irons do you have in the fire right now? About 462. I hear you. I'm overwhelmed with the onslaught of AIS.
55. 13:39Hi. Sensitivity is high. Yeah, exactly. Sounds like midjourney. I know. Like even midjourney, just keeping up with
56. 13:47midjourney. just keeping up with what can it do and how do you actually use that here. Let me show you another let
57. 13:54me show you something right now. Am I sharing this tab? Share this tab instead. That's not it. That's not it. That's it. Okay.
58. 14:06Wait, what did I do? Oh. Oh, producer Brandon and I were were dual clicking. Um, so this is Korea's new real time video
59. 14:19model which you can get on the wait list for um, but do you remember do you remember a
60. 14:33year and a half ago when Korea Realtime Image came out and I was making realtime images and then it made a painting of
61. 14:40boobies and I got perma banned from Tik Tok. Um, this is like that only worse cuz this is video. What do you want, champ? You want to get
62. 14:53up on the bed?
63. 15:16So, so a couple of thoughts on this. My first thought is, well, what the [ __ ] would you use this for? Right?
64. 15:28a thing that you could use it for, which would actually be pretty cool, is real-time performances. So, imagine like a theater improv group.
65. 15:39Oh, great. Another wait list that I'll forget I'm even on. So, I don't check until 3 months after find out I was let
66. 15:46in. I know. That's totally what it is. You know what's funny, Tom? I was just thinking tonight, um, I went and looked at my at my Discord,
67. 15:57my list of Discord servers where I've got like the official Discord server of the thing. I've got some creative
68. 16:04partner program versions. I've got I don't know 30 different discord servers for all for stuff I'm interested in or am in the
69. 16:17creative partner program of it absolutely reminds me of when I was following NFT projects like it's like oh
70. 16:27my god and and they're just relentless and I and I feel like it's the same way with these things like they just keep
71. 16:34you know releasing new things. So, you know, anyway, give all yourselves some grace. I'll give myself some grace.
72. 16:4430 lightweight. Yeah, I know it. It's probably more than that, Vicki, but I'm trying to not seem like a total sociopath. Um,
73. 16:57you know what? Let's um let's cruise through. Yeah.
74. 17:12Let's just cruise through X. I'll be the sociopath here. That's just your Discord horde. Um let's just cruise through X.
75. 17:20So, so one of the things people often ask me is how do I keep up with keep up with stuff? Um, one is just really like
76. 17:28who who I'm who I'm following. So, one of the ways that I'll do this is if I find someone I like, like if I like this guy Mark uh Cretchman,
77. 17:40one of the things I'll do, like he's got 14,000 followers, but he's only following 493 people. Let me make this a little bigger.
78. 17:50He's only following 493 people. So you can click on that and you can now go see the people that he follows. And
79. 18:00sometimes you'll find people in here that you're like, "Oh, I should follow them." So So one of the ways I tweak my
80. 18:06feed is is to follow people who who I like what they do and I want to find out who they're following,
81. 18:14especially if their ratio is really low like that. Like I think my ratio is stupid because I followed a bunch of
82. 18:20people dur during the NFT days. Um yeah, my my my ratio is I've got 3,500 followers. I'm following 2,000. That's
83. 18:32useless. Like my list my list is useless. Um although I suppose if you look at my most recent ones, my most recent I don't
84. 18:42know 200, those are probably decent ones. Um, OpenAI today did drop um a new version of speech to speech for
85. 18:53developers. Um, producer Brandon also posted on the Irregulars channel in the AI salon um that his his voice in chatbt
86. 19:04is glitching a lot. It sounds like it's had helium. It's all crackly and and sounds like compression artifacts. Um,
87. 19:11I've had that when I've had low bandwidth, but he said he's on really high bandwidth. So, this could be a
88. 19:16thing where OpenAI is throttling access because they have too many too many concurrent users. I don't know. That's total speculation.
89. 19:26Um, I'm starting to see a lot of um, you know those annoying videos like I I I've been noticing some recently that
90. 19:37are like um, moments before disaster 10 wild examples and it's video of like the the one that I've seen is a it's like a
91. 19:46mall with a giant fish tank and the the wall falls out of the fish tank and smashes a bunch of people. It's total AI [ __ ]
92. 19:55So, I'm but I'm starting to see a lot of those, you know, shitty shitty accounts get even shittier. Uh, so that's a thing
93. 20:03I'm noticing on on X right now. Um, CES is is coming. People are talking about that again. I don't know why planes are crashing.
94. 20:25[Music] Oh, here's a fake a fake doll. This isn't creepy. Good luck sleeping.
95. 20:40Oh my god. >> Hey, Kyle. >> Yes. >> Just a quick question for the X uneducated. You've got a tab there that
96. 20:49says generative AI and another one that says AI stuff. How did you create those? Because I've tried and not been
97. 20:57successful. And I could Google it, but you're right here. So, >> I I don't remember. I think I think I go to
98. 21:14>> I thought they were lists, but they're definitely not lists. >> They're lists. They are lists. your lists.
99. 21:25Let me just make sure. Yeah, here's my generative AI list. See how >> you pinned that? >> I've pinned it.
100. 21:32Okay, so I definitely know how to do this now. Thank you, Brandon, for for reminding me to remind myself how to do
101. 21:38this. So, you go to your three dots and then you go to lists and then you can make a list. And basically when you make
102. 21:46a list, you just add a bunch of people that you're following or you can like like uh Robert Scoble, if you follow
103. 21:54him, he's got amazing lists of like 2,000 people. So you can just go grab their lists and add them to your lists.
104. 22:02So that's that's a way that I'll do this. Like AI artists, this is this is not a list I made, but this AI artist is the one that has
105. 22:1218 members in it. I made the one that has 3,500 members in it. I didn't. That might be Robert Scobble's list.
106. 22:20Um, yeah. So, here, AI consultants by Robert Scoble. And I've got that pin. And so, then when I'm when I'm here, I
107. 22:28can just like one of What's weird about the for you page lately is that it seems to have during certain hours of the day,
108. 22:37it's getting different. Normally, what I'll get is mostly AI stuff. But it seems like at 3:00 or 4 in the
109. 22:45afternoon, I get like it's just all videos and it's all videos of stupid [ __ ] Like stupid [ __ ] on Tik Tok.
110. 22:54People falling, tripping, you know, smacking their head on garage doors, burping after they take the Sprite
111. 23:00challenge, you know, like Tik Tok. And so if it gets like that, then I'll flip over to one of these lists. Um,
112. 23:12yeah. So, that's what's that's what's there. And then the other thing I do is, um, a lot of times you'll see headlines that kind of
113. 23:25catch you because they're cleverly written. Let's see if I can find one. Grock code was officially released.
114. 23:58But anyway, I don't I can't find one now. Um, just forward it to Manis. Like, follow the tools. Like, if you like Manis or
115. 24:08GenSpark, follow the companies that make them or their official pages. That's how I'll find out there's announcements. Um, pay attention to
116. 24:18people that you like. And then the thing I do is if I if someone I like posts something about here's a thing coming up or here's a
117. 24:27rumor, then I'll just keep scrolling and if if I don't see it come up again, then I just assume that it's [ __ ] And if
118. 24:36but if I see something show up four or five times from four or five different people, then I have a pretty good sense of what's happening.
119. 24:45Uh, one other thing while we're on X, uh, we have, um, a DM from me to you for Grock jumping into the vibe coding space for free.
120. 25:01>> Strong model, but it's a strong model. Do is there a place where you can actually just go play with it?
121. 25:06>> Uh, here say show the expanded out. because I noticed it was on like some vibe coding tools like wind
122. 25:21wind whatever wind surf or wind sock or whatever the [ __ ] that thing is. >> Yeah, but it but free uh is always a
123. 25:31it's on co-pilot. I'm not sure how you get to it in Copilot, but >> yeah, cursor co-pilot. Yeah, it's it it's living. I mean, if if you use
124. 25:41cursor or copilot or things like that, then I would go play with it. It's a new coding model from Grock, it looks good,
125. 25:47but I didn't see anything where it's like easy to just go play with it. Um like like clawed or or lovable, uh
126. 25:56things like that. So, um it might be really good. So, okay, cool. Beautiful.
127. 26:10There's a lot of Nano Banana uh prompts. People are liking it. It ain't perfect. Nano Banana, if you don't know what Nano Banana is,
128. 26:24it's Google's new image generation tool in Gemini 2.5 Flash. There's another weird thing like you've got 2.5 Pro which has Imagine in it and
129. 26:38then you've got 2.5 Flash which which is like a smaller less powerful version of 2.5 Pro but it's got a more powerful
130. 26:45image generator in it. How are you supposed to keep track of that? You're you you can't you're not you
131. 26:52can't it's not you you're not the idiot. I'll tell you what, here's a billion dollar opportunity. Make an AI to make
132. 27:05sense of all this stuff to just track it to just be able to just go um you know I need something that does this and it's
133. 27:12always got the latest one and you don't have to link to it. It's just right there. Here you go. Start prompting. Go.
134. 27:18That'd be nice. Someone go build that. I've been working on visual poems.
135. 27:37If AI became unavailable today, would you still continue building? Interesting.
136. 28:01XI is cooking clever and creative ways. Oh, aren't they doing something new with uh Oh, there was a post from Notebook LM where
137. 28:15they're going to put um they're going to put the new image generation model into Notebook LM in their video generation tool. So, you'll
138. 28:25be able to prompt their video generation tool to do to do things in a more predictable way with probably better image generation.
139. 28:36I hate it when I'm led in and pay the sub for 6 months and then do absolutely nothing with the tool. That's archetypal. That's that that's
140. 28:52I would say that 90% of the tools that I pay for if I were to divide the monthly fee by the number of prompts that I give
141. 29:01it in a month on average, it's probably about 50% of the monthly fee. So So I'm paying $10 a prompt for most of the tools that I that that I
142. 29:13use. I have completed not one single thing in manners. Well, okay. So, here's the thing. So, I said this last night, but
143. 29:27I'm going to say it again because I I want to find a way to work this into into my more organically into my thinking. I saw a LinkedIn post
144. 29:38yesterday that said 10,000 prompts is the new 10,000 hours. And basically meaning like if you want to get good at a tool, you just got to
145. 29:48[ __ ] use it, right? Play first. You got to play with it. You got to play. You got to push it. You got to put in your 10,000 prompts. And
146. 29:58I think the expectation is, oh, I should be able to do that for lots of different tools. I think what's happening is the tools
147. 30:06are getting sophisticated enough that it's like it's kind of like pick a lane. Pick a lane. Pick a couple of tools.
148. 30:14put in your 10,000 prompts and just just get good at those things. I think that's where it's where it's going because
149. 30:22otherwise you just end up with like like like the like chat GPT at this point I would say if you're starting from a
150. 30:30standing start with chat GPT and you really want to master it, you really want to understand all of the
151. 30:38settings and all of the personalities and and custom prompting and custom GPTs and codecs and uh canvas
152. 30:49and deep research and deep thinking, all that [ __ ] image generation, voice. Like, if you did it eight hours a day for a month,
153. 31:04you might feel like you've got some decent mastery of it, but it's probably going to be more like three.
154. 31:13So, just kind of where we are right now. The tools are a bit more sophisticated. Um, [Music]
155. 31:27do I have a good recommendation for a HubSpot AI alternative? I don't, Mark, I haven't done I haven't t done a ton of
156. 31:35pro programmatic marketing. Um, it's something I want to get into. Um, Vicki in here is a huge fan of Blot from
157. 31:44Sabrina Romangh. Um, I don't know that it's apples to apples to HubSpot, but I know it's an AI native tool. Um, it it's
158. 31:54not cheap, nor is HubSpot. Um, yeah, I don't know. I I just I don't have a good answer for you there, unfortunately. Um, I just don't do
159. 32:06enough of it. Yeah, Vicki says she loves Blot. I get loads of use out of pseudorite. I don't know what pseudorite is. Is that
160. 32:20an LLM? All right. Um, I think that's true. 10,000 rule. My prompts are really
161. 32:35getting next level. Oh, that's cool. GPT5 hates us. No, GPT5 doesn't hate us. GPT5 thinks we hate it because we haven't taken the time to
162. 32:48learn it. You know what it is with GPT5? It's like we went out on one date with it and we're we're expecting it to it's like on
163. 32:58the second date we're like you want to get married and chat GBD5 is like whoa whoa whoa. Why don't you learn to prompt me first?
164. 33:08You know, I've got different models and I've got this routing model that will make some choices for you. Why don't you
165. 33:13learn what's good about that before we get hitched? I've been using Brievo, but it's falling short.
166. 33:28I mean, you know what's funny, Mark, on on the on the marketing automation stuff is I I've never even heard of Brevo. Um
167. 33:39SEO is totally [ __ ] right now and changing. You've got to you've got to do AIO, right? So AI optimization. So how
168. 33:46do you get into search into you know chat results? Um so that whole world's changing. The marketing world is so
169. 33:56fragmented. Like if you just think about short form video, Tik Tok, Instagram, YouTube shorts, things like that. Like
170. 34:04where do you put your attention? And you know the answer is like you put your attention on all of them you dummy.
171. 34:10You just what you're going to do is you're going to make one piece of content and you're going to leverage
172. 34:14that content across multiple channels. And that's what Vickiy's doing with Blot. But it's like just getting your
173. 34:23head around where you're going to put all this [ __ ] and what you're going to put there and what your voice is. It's
174. 34:30it's a lot. And then and then you have to create the content to fill that machine, right? It's this big circular [ __ ]
175. 34:40nightmare. Just moved from a sauna to ClickUp. Yeah. You're you're in a world that I just I haven't I used to run one
176. 34:47of the biggest agencies in the world, but we were more of a creative agency than a media agency. And then I I got
177. 34:54out of that in 2002. So basically from 2010 on from all the social media stuff on I I'm kind of clueless with with what
178. 35:02to do with this stuff now. Vicki, it's not quite HubSpot but fills some of that capacity. There you go.
179. 35:11That's been amazing having all the AI integrations. I really just need a CRM and an email tool. Really? Yeah. Yeah. So, um,
180. 35:25good question. Keep coming back here and when you figure it out, let us know. Join the AI salon if you haven't.
181. 35:31Brandon, if you would pop the uh the address for the AI salon up on screen. Yeah, I've dropped my blot sub, too. I
182. 35:39just don't have anything that I want to say that badly or that much. I know. I I'm I'm the same way right now. Like
183. 35:46even though I put a fair amount of content out there, I'm like, "Wait, what am I going to talk about? What do I want
184. 35:50to talk about? What's important right now?" And I don't know. I'm I'm I'm still in this weird place where I don't
185. 35:56I don't have a clear point of view right now. There you go. So, the AI salon, if you haven't joined the AI salon, please
186. 36:03go do that. The salon.ai. If you go to community.thesalon.ai, that'll take you straight into our community website. Sign up there.
187. 36:12Introduce yourself. Step two is introduce yourself. just go introduce yourself. Um, I need an AI that paints
188. 36:19my 3D models for me in Bamboo Studio. I can't believe someone hasn't come up with that by now. That seems vibe codable, Vicki. Come on.
189. 36:33That Come on. You got to be able to vibe code that. That's all just that's all just geometry and and pixel colors.
190. 36:44and normal maps. I don't know enough about 3D. I know they got these things called normal maps and shaders. [Laughter] Oh, man. All right.
191. 36:56Is he reading this chat or is he just plugging his other stream? Uh, I was not reading this chat. What's your question?
192. 37:07Oh, here we go. Do you feel AI has an honest case to replicate human output like full replacement? Does it have an honest case? I don't
193. 37:22quite understand the question. What's your what's your name? Aha. Ahaz a sh a sha a shush. Um,
194. 37:37do do wait, how did you how did you phrase the question? Do I feel AI has an honest case to replicate human output? Oh, do do I
195. 37:49feel that AI is going to get good enough that it's going to be indistinguishable from humans? Is that it? Is AI output a
196. 37:59replacement for human work? Oh, yes. I not only do I think it has an honest shot, I think I think what's going to happen is this.
197. 38:12Right now, all of the AIS are still bad enough. Like Sam Alman talked about GPT4 as being slightly embarrassing. like he knows
198. 38:27that it hallucinating that much and it being that imprecise is not acceptable, right? And he called it slightly embarrassing. GPT5 is
199. 38:42better, but one of the things you'll notice about GPT5 is that there's no clear there's no clear consensus on is it
200. 38:53good, bad, or indifferent because all sorts of different people are using it. People that are using it for high-end
201. 39:00thinking stuff are saying it's absolutely remarkable. people like me that are living down in more mundane
202. 39:08things like, you know, marketing and communications and storytelling. It's just different. It's not better.
203. 39:14It's just different. And I don't know how to prompt it yet. And so, so all of the tools right now are for
204. 39:23the most part non-automated. The ones that are automated are janky as [ __ ] But I think what's going to happen is
205. 39:29this. I think what's going to happen is someone's going to crack the code. Yes. >> Yeah. So there's I want to call it it's
206. 39:40an important distinction here that you know for you know the answer not to be an immediate no on replacing humans. It it
207. 39:50will replace most human work in terms of the work as we know it today. But I I think if I if I know you well enough to
208. 39:57know that that's not going to replace human drive and innovation and I want to make sure we get to that point as well.
209. 40:05>> Yeah. Yeah. Yeah. No, I No, like I don't No, I don't think it's going to I I'm talking about human human jobs. Um will
210. 40:13will it replace human jobs? I think it will it will replace human jobs of today. I think it will enable new jobs
211. 40:23tomorrow that we can't even imagine right now. That's what new technology always does. It automates the things
212. 40:29that we know now. We can't imagine what the future would be like without those jobs. And the fear is always, oh well,
213. 40:38what are humans just going to sit around and rot? No, we're humans. We we seek to make meaning. We seek to connect. We seek to
214. 40:48make a difference. So, we're still going to do that. And I think I think what happens is what AI does is it it
215. 40:57automates it replicates the tasks that we do today. But the tasks are not the important part. The important part is what's the
216. 41:06idea like what's the idea of the business that you work for? You know, what's the idea of how you want to um
217. 41:13impact a customer? I think the human role is going to elevate into well what difference do I want to make in the
218. 41:21world and then the AIs are going to do all the tasks. So I think that's that's what shifts and so I think our roles elevate or
219. 41:31you could choose not to live in that more elevated highinking world and you could just say well [ __ ] all the [ __ ]
220. 41:38all the tasks are taken care of the thing I'm really excited about is LED macra and so I'm going to master LED macra and
221. 41:52at some point that's going to become a job like stuff that we just like that's not a job. If anyone from 1895 saw what we do for a
222. 42:02living right now, they'd go like that's not work. Work involves muscles and grain. [Laughter] The tech intakes language and outputs
223. 42:14the average of that language, nothing more. That's not true. That's not true. That's not true. That is a trope from if you look at how
224. 42:27a large language model predicts the next possible token, then what you say is true, right? If you're only looking at it at the token level,
225. 42:41right? You're saying we take all this information, we embed it into thousand-dimensional mathematical space,
226. 42:46we put all these tokens in in this weirdass math space, and when you type a prompt, it creates a probability of
227. 42:55what's the next most probable token. And and you could argue that that's some amalgamation of all the [ __ ]
228. 43:03But when you start to do things like create chain of thought like the the reasoning models, the thinking models
229. 43:10that are not just doing next token prediction and if you start taking the agents, the autonomous agents and then if you also start taking
230. 43:20human contribution where so like like one of the tropes of AI and and the the one that you just said is that it just gives you an
231. 43:31average answer. If you give it a shitty prompt, it will give you a shitty answer. So if you say, "Write me a
232. 43:38LinkedIn post about marketing," you will get the amalgamation of all the LinkedIn articles and every other article of
233. 43:45marketing, and it will be boring ass repetitive dril. But if you spend 15 minutes going, "Okay, I've got a point of view on
234. 43:56marketing and I've got some ideas here and I'm going to put my ideas into the prompt and I'm going to see what it
235. 44:00gives me back and then I'm going to go, "Ooh, I like that idea. I don't like that idea." I I'm still actively
236. 44:06involved in the curation and refinement of what the model's doing. So, it's not a oneshot thing, right? I think that's where you where
237. 44:17you're like what I would encourage you to do is like use these things. Chain of thought is still average. It's not
238. 44:26because it's not because you're you're missing the human component here. If if what you're asking is might an AI
239. 44:35be so good at some point in the future that we don't need the human involved in it, it might be. But to say that you can't get high
240. 44:45quality original content out of large language models is just not true. It's just not true. Um
241. 44:59it's like that for using AI. Let's see want to write. I work with AI daily. I use it constantly as a staff dev. It does. Okay. But
242. 45:13it's not original. Let Let me ask you something. What is an original thought out of a human?
243. 45:27How do our brains work? How our brains work is we get some inputs and those inputs go into what is effectively a large language model.
244. 45:39that's had a bud a bunch of information encoded in it, right? The original source material ceases to exist, but
245. 45:47we've got, you know, synapses and neurons that are firing off with chemicals. And when an input comes in, how do we
246. 45:57have an original thought? Well, we have an original thought like this. We go, "Huh, that kind of reminds me of Andy
247. 46:03Warhol. Oh, and that makes me think of Tom Robbins, that author. And oh, that thing my mom told me when I was eight.
248. 46:10And and we mash together those disperate thoughts in this crazy straw of a [ __ ] brain we have, and we go, "Huh, here's a new idea."
249. 46:22Which is some weirdass combination of the thoughts that got put in our head over our lifetime. How is that [ __ ]
250. 46:29different than what a large language model does? If it's not generating original thoughts, then you're not prompting it
251. 46:37in a way that's taking inputs from from, you know, in congruous things and mashing them together and putting them
252. 46:46out in such a way that you go, "Oh, that actually seems interesting and different." Two nights ago in here, I
253. 46:53was I was doing something. We were talking about writing kids stories, and just on a lark, I said, "I don't want a
254. 46:59kids story like every other kid story. I want something that makes them, you know, question thought and I I want
255. 47:06something that shakes up the kindergarten. And what Chad GPT came up with was really interesting and it was
256. 47:14new and novel from where I sat. Now, was it mathematically new and novel? I don't know. Maybe not. But I don't see how
257. 47:24how our brains create an original thought. I don't see how that's different than a wellprompted large language model.
258. 47:33So for what it's worth, okay, you're talking about context size then it cannot replicate a human buffer size. I I
259. 47:44agree. I said at the beginning of this these tools suck right now. The context windows too small. The memory is too
260. 47:52imprecise. The hallucinations are not predictable. or or or manageable. Although it looks like five has gotten
261. 48:01that much better and it looks like memor is about to get much better. But I agree with you. Like yes, human brains are
262. 48:11better than large language models right now. But to say that a large language model is not capable of generating an
263. 48:16original thought, I just don't agree with it. Um a million tokens cannot speak to human experience. I'm not talking about
264. 48:25speaking to human experience. It doesn't have the attention head space to differentiate actual content. No, but we do. But we do.
265. 48:40Here's what you're missing. We're using these tools. If you want the large language model to suck less, then find more creative ways
266. 48:56to prompt it. Find different ways to come come at things. Find different ways to synthesize the world so that these
267. 49:04things are reflecting better answers back to you. But anyway, like I I I think you and I are actually
268. 49:11agreeing. I get what you're saying and I actually like like it. Not going to lie. But but but here's here's my point. I'm
269. 49:19I'm not saying that you're wrong. I think you're actually right that from a from a definitional place where large
270. 49:31language models are right now is like child's play to the sophistication of our brain. Absolutely agree.
271. 49:40My point is the tropes of things like it can only do predictive outputs and it can't it's only going to give you the
272. 49:46average of things is just not true. if you've got a human collaborating with that machine because then what h so okay
273. 49:55so so I'm writing a book right now called feed your prompt and and the whole thesis of the book is that you if
274. 50:00you feed your prompt you what the large language model does is amplify your ideas and I think the the the the sort of area
275. 50:11that you're treading on is I think one of the mistakes that people make and it's it's one of the things that the
276. 50:18Silicon Valley bros have have precipitated this idea that the AI is the genius, right? They're like the AI
277. 50:28is going to have AGI and it's going to be smarter than all of us. And so they're saying the AI is the genius. And
278. 50:32so if you treat AI like this thing where you give it a prompt and it's going to give you a genius answer, it's horrible
279. 50:40at that. It doesn't give you a genius answer. It gives you a shitty answer. But if you shift the framework to say,
280. 50:49"No, no, no, no. The AI is not here to replace me. The AI is here to augment me. It's here to amplify my ideas." Everything changes
281. 50:59because now you take what it's good at, and you let it amplify your stuff, and then you can take your brain, which is
282. 51:07better than it at some things, and you can go, "Ooh, that's a good idea. Let me take that one. Put that in my crazy
283. 51:13straw. spit it back out to it. See how it amplifies that. See how it amplifies that. And so within an hour, you can do
284. 51:20in an hour that might what might what might have taken you 3 months without something like AI, but it's a very different mindset. So
285. 51:30anyway, good question, good banter back and forth. Thank you, bro. Truth, Lisa. It's happening. Um, all right.
286. 51:43I misjudged this and I do respect this te take actually. Thank you for entertaining it. Thanks for the qu. Like
287. 51:49seriously, I love these conversations because for one thing it's like like when Brandon said, "How do you make
288. 51:56those things across the top?" I'm like, "Uh, I don't know." The whole purpose of this channel, the whole purpose of this channel
289. 52:05is for all of us to be in this conversation. You're like, "Well, what is this conversation?" I don't know.
290. 52:12It's it's it's like it's questions like that saying, well, wait, is it theft? Is it copyright violation? Can you
291. 52:20copyright something? Can it have original thought? Like it it is so important for us to to start to get our
292. 52:28head around what these things are, what they make possible, what they can't do right now, but the trajectory that
293. 52:35they're on. And that if they can't do this right now, are they going to be able to do it 6 months from now or 3
294. 52:41years from now? And if they can do it 6 months from now, well, we better start [ __ ] preparing for that, right? And
295. 52:48that like, and the only way you can know that is like I've been on this planet for 60 years. The only thing I know is when you're in
296. 52:58a a burgeoning um primordial soup stage of a technological innovation, the only way you can learn it is to be
297. 53:10in the soup. And that's that's what this channel's about. Tik Tok question. Um McKenzie Sim, how much coding should I
298. 53:18learn? I'm learning Python. Uh might try uh wolf language. Okay, great question. Not a simple answer.
299. 53:31Do you technically need to learn coding right now? You don't. I would argue that learning philosophy is probably as
300. 53:40valuable, if not more valuable, than learning coding. And I' I've got lots of reasons for that. Um, but that said,
301. 53:49people who are competent with coding, who also understand vibe coding and a lot of other things that AI makes
302. 53:56possible are going to be way more valuable than just people that can code. So, and and it also has a lot to do with
303. 54:04what you want to do. If you want to go work for Meta or Open AAI or Google, um, yeah, they're probably going to want you
304. 54:13to not only, you know, learn coding, but be really [ __ ] good at it and be able to use all these AI amplification tools
305. 54:21like cursor co-pilot and, you know, cursor and copilot and, you know, all all of the replet and things like that.
306. 54:28Um, if you want to be building models and fine-tuning models, if you want to be digging deep into um the open-source
307. 54:35world, if you want to develop your own applications, even if you're going to vibe code them, really understanding
308. 54:42databases and security and O and all of the stuff that you need to do to build a worldclass application. Right now, these
309. 54:50vibe coding tools suck at all that stuff. But here's here's the butt. Here's where you don't have to necessarily learn coding.
310. 55:01It depends what you want to do. One of the most important roles in the next five years is is going to be what I
311. 55:08call a creative generalist. And that and that is someone who can look at the entirety of a project, the entirety of a
312. 55:16of an idea. So you're working for a company and you need to revamp your I don't know your sales outreach.
313. 55:25One way you could do that is you could go into application development and you could get with a UX designer and you could
314. 55:33you could run down that deep rabbit hole and that's going to take you a long time to build. The other way is to be a
315. 55:38creative generalist that sits above all of the tactical execution but can actually understand all of the different
316. 55:44things that are required to do that. Okay, there's going to be some change management required. There's going to be
317. 55:49some systems development required. We're going to need to really understand our customers. We're going to really need to
318. 55:55understand how our current sales um approach isn't working. Where's it breaking down? We're going to need to do some data analysis.
319. 56:04So, imagine being so fil with AI tools that you could just sit there like the the Wizard of Oz or like the the conductor of a symphony
320. 56:18and just go, "Okay, I'm going to pull in this tool and do the data analysis. Then I'm going to take the answer of that.
321. 56:22I'm going to put that into a presentation. And this afternoon, I'm going to give that presentation to the boss. I'm going to get the boss's
322. 56:28blessing. And then I'm going to go do all this. And then and then I'm going to write scripts. And then I'm going to
323. 56:33create training materials. Like that kind of person is going to be they're going to be like [ __ ] magicians. It's going to be it's
324. 56:41literally going to be like the Wizard of Oz. They're going to have that kind of power that doesn't require coding,
325. 56:50right? That requires broad thinking. That requires understanding lots and lots of elements. I think the day of the
326. 56:59hypers specialist is numbered. There will still pe people that hypers specialize, but they're going to have to
327. 57:06get better at going horizontal. Um, that's me, the creative generalist. Uh, there was a Tik Tok pin, but I
328. 57:12missed it. Uh, uh, uh, I agree with you. to understand the basic code even blender but limited all obsolete. Yeah,
329. 57:22exactly. Okay. Communication skills will be really important. Yes, communication skills are going to be critical, right? Empathy,
330. 57:32um critical thinking, creative thinking, curiosity, adaptability. Holy [ __ ] You want to know you want to know what to study
331. 57:43rather than coding? Learn enough code so you understand the components of coding. But you want to really master something, get
332. 57:53really masterful at something, become the most adaptable person on on the block. Become someone that the minute something
333. 58:02new comes out, you've got curiosity enough to go, "Oh, I'm going to go learn that." And then you can immediately go,
334. 58:09"Is that better or worse than what we have today?" If it's the same or worse, I don't give a [ __ ] about it. If it's
335. 58:15dramatically better, I adapt all of the other stuff that I do and how I think about stuff to this new reality.
336. 58:23And that's just going to that's going to keep rolling for the next five or 10 years. Those people that have curiosity and
337. 58:31hunger and ambition and adaptability, holy [ __ ] you're going to be valuable. like deeply valuable like you know
338. 58:47won't ever be looking for work valuable and there going to be a lot of people looking for work um new pin I don't see a pin okay
339. 59:01hit you on LinkedIn great this is a refreshing take from sea levels expecting raw 40% increase in output
340. 59:09yeah so listen I'm glad you hit me up on LinkedIn. I'm happy to chat. Um and and excited to I'm I'm I'm excited by your
341. 59:16things. Um the MIT study that just came out that that said 95% of AI projects have failed 0% return 0% value. They said 95% of of
342. 59:33the things you've got to look at the methodologies of that study like how they're defining success is increased revenue.
343. 59:43So like there could be no other value from an AI implementation than increased revenue, right? What no they also they
344. 59:53didn't define they like it's not based on quantitative data. It's based on 54 qualitative interviews where they said,
345. 1:00:02"Did did you make more money or not?" No. Oh, okay. So, that's a zero. Like, that study is [ __ ] ridiculous.
346. 1:00:11And so, CEOs that expect a 40% bump in revenue because we gave people access to J Chat GPT is [ __ ] [ __ ] Nobody knows how to use it.
347. 1:00:24That's what this channel is about is nobody, nobody, Hang on, blowing my vocal cords out here. Nobody knows how to use this [ __ ]
348. 1:00:41One of the things I'm doing in consulting right now is anyone in the sea suite. If I'm going to work with
349. 1:00:46anyone in the seauite and they don't [ __ ] use generative AI, I'm not going to work with them because this is not the kind of thing
350. 1:00:56that you can go I'll just delegate it to the IT group. I'll just delegate it to the people that handle that. I I'll just
351. 1:01:03delegate it to the innovation group. No. You've got to understand as someone who's trying to drive change in an organization, you've got to
352. 1:01:13understand that what generative AI is doing to computers is making them behave in ways that they've never behaved before.
353. 1:01:22This is not an efficiency play. Can you use AI to make more efficient ways of doing what we've already done? Yes.
354. 1:01:37It's not all that interesting. We've had Zapier for years. We've had automations for years. We've had specialized AI for years. That's not
355. 1:01:48what we're talking about here. We're talking about generalized intelligence that does stuff. It generates.
356. 1:02:00No one in an organization knows exactly what that means for their department, for their job. No one. So, the only way
357. 1:02:08we're going to get through this is to collaborate and to mix ideas together and be in conversations like this. And
358. 1:02:14Murphy in the house was happening just in time for my favorite topic. Um, what's we're we're going deep here. We
359. 1:02:23are going deep here. Um, we paid 240 bucks for the tool. source camp picked up three new clients in 24 hours
360. 1:02:31because I can adapt to my clients pain points. Yeah, this is this is the [ __ ] thing about AI right now is
361. 1:02:45h it's it's it's so [ __ ] exhausting. It and it's only exhausting because we've been through this before, right?
362. 1:02:52Every time there's a new technology, we go through this. In the mid 90s, the number of conversations I had with
363. 1:03:00seauite executives that were saying, "We need a website and I would ask,"Wh do you need a website?" And they go, "Cuz I
364. 1:03:07was golfing with my buddy and he's got one and I got a I'm" I'm like, "What do you want it to do?" I don't know. Well,
365. 1:03:14then how do you know you need a website? You might need a TV ad. You might need a better sales force. You
366. 1:03:21might not need a website. And it's the same [ __ ] with AI right now. We need AI. Why? What do you want it to do? Do you
367. 1:03:28know? Do you even know what it makes possible? Well, I just need it. Okay, I got chat GPT for the staff. They're going to go
368. 1:03:37figure it out, are they? It's like, [ __ ] What source camp's doing is going in going, got a problem? Oh, do you want me
369. 1:03:50to show you how we can solve that problem? You know the problem that you haven't been able to solve for two years
370. 1:03:57and you think it's not solvable? Do you want to watch me solve it in an hour? That's what she does. It's remarkable.
371. 1:04:08And then she and then she's she like solves their problem in real time in front of them. And they're like, I how's
372. 1:04:15that possible? She's like, "Well, keep paying me and and we'll knock a bunch of other things down like that."
373. 1:04:26That's because she's living in this creative generalist role where she can look at any problem in any business and
374. 1:04:35go, "Okay, um, if I use that tool and that and I put those together, right?" She's like the conductor.
375. 1:04:43So those people that that get fil with lots of tools and even tools that are way outside their expertise like coding or images or music
376. 1:04:56they're going to be the ones that going to be able to walk into any situation and go what's your problem? Oh we can we can crack that nut.
377. 1:05:03All right seuite is uniquely unqualified to strategies for AI. Yes, I have to educate them under the cloak of
378. 1:05:12darkness. Actually, you know what, Anne Murphy, that's a really good point. I'll tell I'll tell you something fun.
379. 1:05:18When when when we first started agency.com, the the original name of the company was the agency. And the reason
380. 1:05:26we called it the agency was we said we understand what websites are. We really didn't like, you know, we understood as
381. 1:05:33much as anyone because it was just being born, right? But but what our hypothesis was was there's going to be a lot of seuite
382. 1:05:41executives that are going to want a website, but they're not going to want to admit that they don't know what a
383. 1:05:46website is. And so we talked about the agency being like this, you know, cloak and dagger agency that we would sneak
384. 1:05:54in, you know, the the service entrance of the big building, you know, go up to the sweet sea suite, teach them what
385. 1:06:01websites were, and then leave. You know, we'd come build your website, then go out the back door. Um, I think there's
386. 1:06:08something there. I think with generative AI, what what you've got, especially in big companies, is you've got executives who their
387. 1:06:18organization has had AI for decades. They're like, "Oh, yeah, we got AI. We got people for that." They don't
388. 1:06:27actually understand at all that what generative AI is is different and and and that the implications of it are are profound.
389. 1:06:38Um Boopsy, good to see you. I haven't seen you in a while. I use chat GPT to prepare for an appeal with the country.
390. 1:06:46Oh, with the county. I was able to make them admit fault. Awesome. Yes. Oh, you want to know something cool today?
391. 1:06:57So, um, so my wife's an artist and she got, um, selected to be interviewed for an article and she had a brain injury at some
392. 1:07:06point. So, she gets overwhelmed in conversations and things like that. And so, when this reporter was interviewing
393. 1:07:12her, she was getting overwhelmed and she gets, "Can you just send me the questions?" And so, the woman sent the
394. 1:07:17questions and there was a lot of questions and a lot of them were repetitive and things like that. So what we did was I said, "Oh, this is a
395. 1:07:26perfect use for chat GBT, not to write not to write the questions, but just to make sense of let her sort
396. 1:07:33of ramble, right, in in all sorts of order and then let chat GPT make sense of it." And so I would read the question
397. 1:07:41and she'd think about it for a bit and she, you know, we'd go back and forth a little bit and then I'd say I put on
398. 1:07:47chat GPT and I go, "Okay, Quinn, you know, here's the question." and then Gabby's gonna answer and then I want you
399. 1:07:52to make sense of it. And so I would ask the question, then Gabby would answer and she would say this idea and that
400. 1:07:57idea and and you know she'd go all over the place and then Quinn would go, "Well, here's how you might say that."
401. 1:08:03She would write a perfect like retelling of what Gabby just said. Well written, wellcrafted. I would literally copy and
402. 1:08:11paste that, remove m dashes because there were too many of them. And it was good to go. It was amazing. It was
403. 1:08:19absolutely amazing. Um, yeah, that that ability to to take the thing you want to do, right? You want to get the the
404. 1:08:28county to admit that they [ __ ] something up and and they've got all sorts of ways of avoiding responsibility or admitting
405. 1:08:36that they [ __ ] something up. Well, all of that's been documented somewhere. So, that's all in the learning model, right?
406. 1:08:43So, you can go into it and go, "Okay, here's my situation. Here's what the countyy's saying. Here's my evidence.
407. 1:08:49Give me an outline. Give me a speech. Beautiful. Brilliant. Love it. Tik Tok pin. Um, no pin there right now. Repin.
408. 1:09:08Cavuno. My VP is meeting with our CHRO and I prepared her for their AI conversation. So, so Kuno works for a
409. 1:09:19big company and and uh there's a lot of people. So, so if you're new here, it seems like we've got some new folks
410. 1:09:26here. The people that show up to this channel night after night. I'm here five nights a week. I do this for one and a
411. 1:09:32half, two hours, five nights a week, starting at 8:00 p.m. Mountain time. It's a little late, but [ __ ] it. What
412. 1:09:37else are you going to do? Um the people that show up here night after night, they're called the irregulars because they're just not normal.
413. 1:09:46Because why would you watch this, right? Why would you watch an old man rambling about [ __ ] AI? Because what they
414. 1:09:54understand is if they're in the conversation and they can be just a little bit smarter about AI than the people around
415. 1:10:02them and if they can talk intelligently about it, then good things are going to happen. And and a number of people on this
416. 1:10:12channel started out as what Ethan Mullik described as the secret cyborgs. They were interested in AI, but they were in
417. 1:10:19the shadows. And one of the things we talk about in the AI salon is play first, mindfully create, and generously
418. 1:10:26lead. And the generously lead part looks like talk about AI. Talk about what you're learning. Talk about what it's
419. 1:10:34good at. Talk about what you love. Talk about what you hate. Just talk about it. Talk about it at work. Talk about it at
420. 1:10:41the dinner party. Talk about it on [ __ ] Facebook. You want to get some hate mail? Talk about AI on Facebook. I
421. 1:10:49don't do it. It's bad enough when I get my ass handed to me on LinkedIn. But just talk about it because what it
422. 1:10:57forces you to do is it forces you to think about, well, wait, what is my opinion here? And then you put it out
423. 1:11:03there and you establish yourself as someone who is thoughtfully engaging with AI. You establish yourself as some kind of expert.
424. 1:11:12None of us are experts right now. Zero people, zero percentage of us because the stuff's changing too fast.
425. 1:11:23But you can be ready for AI. Ann Murphy who's here, she and I put together with Vicky Baptiste a thing called the AI
426. 1:11:30readiness training program. It's a fivemodule comprehensive training program. Yeah. If you go to are you readyforai.com,
427. 1:11:42it doesn't teach you any tools. It's not about tools. It's about shifting your mindset. How do you think about being creative
428. 1:11:49with AI? How do you think about business with AI? what you know, how do you think about the ethics and security and
429. 1:11:55privacy? How do you apply this to business? And so there's all this remarkable video content in there from from last year's
430. 1:12:03AI Festivus. And then we pulled the lessons out of, you know, all of those videos that were kind of consistent
431. 1:12:10across them. And that's what that training is about. And it's amazing. And so if you do that, what happens? Like one
432. 1:12:21of the things that happened with Kuno is, you know, she got asked to do some talks and then she got asked to do more talks
433. 1:12:29and then she someone at one of the talks said, "Oh, you should talk to the VP and then she gave a talk to the VP and then
434. 1:12:35she gave and then she was on the AI steering committee." I think I think is that right? And now there's all sorts of
435. 1:12:41things where she's training up people high in her organization because I guarantee you this, she is probably one
436. 1:12:47of the smartest people in her organization about AI, about generative AI, you know. I don't I don't even know your
437. 1:13:01background career. I don't I don't know what you did before. Was it was in instructional design? I don't know.
438. 1:13:10But why she's expert right now is is because she's thinking broadly and thoughtfully and critically about what
439. 1:13:17AI is good at, what it's not good at, and how you apply that to a business in a responsible way. Very few people are
440. 1:13:24doing that right now. So if you do that in a loud and visible way, you will get noticed. Okay. Um
441. 1:13:47Okay. Um, the other thing, um, since Ann Murphy's here, so I mentioned AI Festivus. If you were not here last year for AI
442. 1:13:57Festivus, shame on you. And if you were not here the year before when we did GPTs for good, shame on you. So, every Christmas
443. 1:14:08holiday, Ann Murphy and I seem to lose our mind and we say, "Hey, wait. We've got two weeks off. Why don't we fill that with a
444. 1:14:18high stress live event that covers 24 hours?" And so, so two years ago, we decided to do 24 hours of GPT building for
445. 1:14:32nonprofits. And we created more than 500 custom GPTs for nonprofits. And last year we did this thing called AI
446. 1:14:39Festivus. And it was 34 speakers from four different communities that just came and it's the Friday and Saturday
447. 1:14:48between Christmas and New Year's or Hanukkah and New Year's or whatever you celebrate there and New Year's.
448. 1:14:55It's like like you got you got to include them all now. Um so that Friday and Saturday. So this year
449. 1:15:04it's December 26th and December 27th. We start at 9:00 in the morning, we go till 9 at night, 12 hours on Friday, 12 hours
450. 1:15:12on Saturday. And you're like, "What are you [ __ ] insane?" Yes. And we're going to do it again this year, and it's going to be bigger and
451. 1:15:20better than last year. Last year we got some boosts from some some people in other communities and we thought we might be doing this
452. 1:15:32for about 400 people if we were really lucky and got our [ __ ] together. We ended up having 3,500 people um sign up
453. 1:15:40for festivists. We had 200 people show up and we averaged 540 concurrent sessions every hour for 24 hours over a
454. 1:15:5036-hour period. It was absolute insanity and it was amazing. And people said it changed their lives. And like we we uh
455. 1:15:59we on our podcast on the AI readiness project podcast um this week we interviewed someone that said, "We have
456. 1:16:06a question. How did you get into AI?" He said, "I got into AI because of AI Festivus. changed his life.
457. 1:16:14So, we're doing that again this year. So, save those dates. If you know someone that should speak at it, let us
458. 1:16:20know. Um, and if you know someone who wants to sponsor it. The the thing that we're committed to doing, which is what
459. 1:16:27we did last year, is that it's free for everyone. And so, we're committed to doing that. It's not free for us because because
460. 1:16:36what we found out, we did this on Zoom last year and we found out if you want to have more than like 20 people on your
461. 1:16:41Zoom call, you got to you got to [ __ ] step up. You got to give up some coin. And luckily luckily we had some people
462. 1:16:49come in and sponsor at the last minute. But um but yeah, that's that's coming up. All right. It's so cool.
463. 1:16:59What time is it? It's 9:18. I I do want to go do some Gemini um some Gemini banana time. And here's why. Um you could absolutely argue
464. 1:17:16Gemini, go to gemini.google.com. Okay, you can flip the screen up there. There you go, Brandon. Um, you could absolutely argue
465. 1:17:28that image generation inside a large language model, you can do it inside chat GPT, you could do it with image gen
466. 1:17:35inside Gemini before and now we can do it with this new model called nano banana. Why take any time on that? Why why spend
467. 1:17:44I've now spent this is now my third night where I'm spending time working on this. Why am I doing that? Because it's
468. 1:17:51different. There's something different about this one and I don't know how to use it yet. And that drives me [ __ ]
469. 1:17:58crazy. And like that's that's an attribute if you can if you can develop that thing of like I got to I got to
470. 1:18:05know. I got to know how does why does this how do you make it do the good stuff? Why why is it not doing what I
471. 1:18:10want it to do? Why is it not wanting to Right? If you can develop that hunger for understanding these things, you're
472. 1:18:18going to be in really good shape. Like that's the core skill for the next 3 to 5 years. Just that curiosity and that
473. 1:18:24drive to be like, "How does this [ __ ] thing work?" Um, I'm clueless on how this thing works. I can do a little bit with it. I
474. 1:18:32understand what it's doing. I just don't understand how to make it sing yet. And so that's why I want to play with it. All right.
475. 1:18:39Um, why can't OpenAI come up with fun names like Nano Banana? Well, but you know what's funny, Source Camp?
476. 1:18:46The name of this is not Nano Banana. The name of this is 2.5 flash image generation. Like nowhere does it say nano banana.
477. 1:18:57That was the code name of the project that everyone on the internet said this nano banana model is [ __ ] amazing.
478. 1:19:05And because it had a fun name, people are sticking with that. And now Google's leaning into it a bit and they're making
479. 1:19:10banana images and things like that. That's not the name of it. The name of it is yes. Um it uh the we
480. 1:19:19talked to the the branding professionals here here at Google. It's that's at google.com and uh w within within the the branding
481. 1:19:28framework of of the Gemini artificial intelligence suite. uh the uh the the image generation well it's natively
482. 1:19:37within uh the Gemini 2.5 flash which is the model of course and it's a multimodal model and within that is the
483. 1:19:43image generation and and that's what was the image portion of that is is is what was codenamed nano banana for testing
484. 1:19:50purposes of course that's not the official the official name is is just it's Gemini 2.5 flash no call it [ __ ] nano banana Hannah
485. 1:20:04idiots. Okay. [Laughter] Oh my god. All right. Um, okay. Let's see. So, if you're at
486. 1:20:19gemini.google.com, you have to be in 2.5 Flash. This is another confusing thing. If you're in 2.5 Pro, this does not have the newest
487. 1:20:28image generation model in it. So the less capable large language model has the more capable image generation model.
488. 1:20:35Remember that. Put that in your pipe and smoke it. Okay. So what we're going to do here is we're going to say I want you to create
489. 1:20:46a J jso n. If you don't know what jso n is, it's JavaScript object notation. JSON or some people call it JSON, but
490. 1:20:57that just sounds like a dude. So I call it JSON. So I can sound impressive. All right. I want you to create a JSON framework to describe
491. 1:21:10all of the elements that might be in an image that
492. 1:21:28that might be in an image. So let's just see what it comes up with. This JSON framework includes a structured way. So we got a time stamp,
493. 1:21:40overall scene description, camera parameters, elements, person label, pedestrian, segmentation, mask, URL. Good l um
494. 1:21:53gender value female. All right. Good lord, that's a lot. Oh, and here's a description. Now, now we have documentation of our of our JSON
495. 1:22:03framework. Okay. So, then what? How do I get out of this? Okay. Then I'm going to say um I want you to write
496. 1:22:18a JSON prompt for a cinematic shot. from a movie in 16 by9 aspect
497. 1:22:37ratio with a an unforgettable character
498. 1:22:51in an unforgettable place with an unforgettable object
499. 1:23:06and some unforgettable action. Okay, let's see. We'll see what it comes up with. So what I'm doing here, the the
500. 1:23:17the reason to play with JSON is you don't have to do this at all, but it does seem that that this image model
501. 1:23:28does better the more context you give it. So um there's a there's a music generation tool right now called producer.ai
502. 1:23:39that you create songs in the context of a conversation. And it's really amazing because you can just go, "No, make it
503. 1:23:45more boppy and it'll make the song more boppy." Like it just it understands imprecise creative direction.
504. 1:23:55This nano banana thing doesn't. Like you'll get a really good shot and then you'll give it an imprecise creative
505. 1:24:01direction and it'll just make a completely different image. So what I'm trying tonight is is trying something a
506. 1:24:08little different. All right. Shot ID. unforgettable climax shot 01 aspect ratio camera angle low tracking shot. So
507. 1:24:18here's the other thing. If you don't know anything about cinematography and if you don't know anything about
508. 1:24:24movie making and and visualization um oh last night I was oh thank thank god I said that that nano banana is not the
509. 1:24:38actual name of the product. Yeah. No, you can't find Nano Banana. That Yeah. No, what you can find is Gemini 2.5
510. 1:24:45Flash. And if you don't know that it's Nano Banana in there, then nanny nanny boooo on you. So, I'm glad you found it,
511. 1:24:52Chef Kelly. Okay. Um, but large language models know all of the cinematography terms. They know all the photography terms. They know colors.
512. 1:25:03They know lighting. They know positions. They know blocking. All the [ __ ] that's ever been written about visual
513. 1:25:10communications is in the large language models. So if you just say come up with a JSON framework, it understands that
514. 1:25:17what what JSON what a JSON framework is is it's just a structured list of attributes of this image. And what's
515. 1:25:25cool is once you have it, you can just tell it to fill in this framework for any image you want. Okay. Low angle, dramatic backlighting,
516. 1:25:39desaturated blues and grays with with spot red accent. Frame composition rules rule of thirds leading lines toward the character.
517. 1:25:48Great name. Elias the whisperer. Okay. Unforgettable place. The obsidian spires of Athlgard. Great. A vast desolate
518. 1:25:59plateau where colossal jagged spires of black crystallin rock pure. Fine. Beautiful. Okay. So, we're going to take this whole thing,
519. 1:26:10this prompt. And then we're going to do this. So, what was the character's name? Elias. Okay. So, I have I have an idea.
520. 1:26:29So, we're going to do we're going to do new new chat. I'm gonna h maybe I shouldn't do a new chat. Maybe I'll go back. Maybe I'll do
521. 1:26:38it in the context of this other chat. Yeah, let's do that. Okay. So, I'm going to go here's an image prompt
522. 1:26:52that I that I want you to make. The uploaded picture
523. 1:27:07is who I want. who I want Elias to look like. All right. And then I'm going to paste in that JSON prompt, but I'm going to upload an image.
524. 1:27:31Kyle Shannon Dreams. Those are bad. Let's see.
525. 1:27:55Um, let's see. Hold, please. I got to go find something. Go to Kyle Shannon Dreams. Ah, here we go. This is the guy I want right here.
526. 1:28:11So, we're going to export this. Uh, mind your tabs. Am I not sharing the right tab? I am not, but I'll be back. That's okay. Thank you.
527. 1:28:24Guitar desktop. Go. We're going to come here. We're going to go desktop. All right. That's the guy. You can't see it.
528. 1:28:38All right. That's the dude. He's got this sort of crazy split beard. So, we're gonna we're gonna give it this
529. 1:28:47intense, highly structured prompt. And we're going to give it a dude for him to look like. And then we're going to make this. All right.
530. 1:28:56Here we go. Is this gonna be any good? I don't know. But mind your black bar. Shut up. I know. Producer Brandon. All right.
531. 1:29:06I don't know what this banana talk is about, but it sounds cool. Oh, so uh uh nano banana was the code name of the new
532. 1:29:16image generation tool. Wait, what the [ __ ] just happened? What did it do?
533. 1:29:40It just wrote us an application. It just coded us a website. Holy [ __ ]
534. 1:29:57Okay. Do do you remember when I said before I don't know how to use this thing? Okay.
535. 1:30:12Um when companies talk about models being multimodal and and uh OpenAI talked about this today with their speechtoech synthesis
536. 1:30:26model. When all this stuff first came out, all these models were distinct models. You'd have a language model, you'd have an
537. 1:30:37image model, you'd have a sound model, you'd have a code model. code and language went together pretty
538. 1:30:43quick, but you had you had specialized code models. The Gemini series starting I think starting with 2.0 and then it got really good in 2.5.
539. 1:30:55Um, these things are multimodal, meaning if you upload a video into this thing, it can understand the video, all of it,
540. 1:31:04the images, the audio, everything about it. Same with audio. What this did, which I don't understand why it did it, but because I gave it
541. 1:31:16code and I said, I want you to generate this image. It generated the image, but it it it it put the the image prompt
542. 1:31:28inside an application and it created the unforgettable cinematic moment generator. You got to be kidding me. This is crazy.
543. 1:31:51Generate a new image. All right. Same image. Okay. So, what I'm going to say is
544. 1:32:05um right now the prompt is hardcoded.
545. 1:32:21Vicki pin on Tik Tok. I told you about this yesterday. That's what it did to me. I didn't I didn't take it in, Vicki,
546. 1:32:29that it did that. Wait, I I'm I'm mind blown because this is not um this is kind of an amazing behavior. We
547. 1:32:43asked it for an image and it gave us a website that generates images. It generates the same image. That's bonkers. Oh, you posted on
548. 1:32:54irregulars. I didn't see it. That's amazing. That's crazy. This crazy talk. Right now, the prompt is hardcoded.
549. 1:33:12Um, let's make it so you can upload an image and give it a short
550. 1:33:28an image of the character and give it a short prompt. and it
551. 1:33:44will use AI to fill in the JSON framework we created.
552. 1:34:01This is amazing. Let's see if Well, I'm not going to spend a lot of time vibe coding here. I just want to see what it does.
553. 1:34:10Yeah, I got that Brandon, I think. Unless there was another one that I missed.
554. 1:34:26I'll proceed with updating the canvas. So, is it rewriting the code now? Yeah, it looks like it looks like it
555. 1:34:40rewrote the code. Oh, here we go. Dynamic shot. No file up. Okay, choose a file. So, let's put the dude in there. Oh, that's cool.
556. 1:34:51Short cinematic idea. Um, Elias stands in the middle of Time Square
557. 1:35:06in the rain after all the people left.
558. 1:35:24generate a cinematic shot. It's generating a cinematic shot. Oh, and I hit it tw Oh, wait. Gener. Oh,
559. 1:35:39look. Generated JSON prompt. Holy [ __ ] Oh my god. This is amazing. Yeah, this is a Kevin Mallister moment for me.
560. 1:36:06And here's the JSON prompt that it created this big ass JSON prompt. And can I can I do another one? Choose file. Yeah. Let's go get a
561. 1:36:20different. All right. There we go. You don't see that, but okay. So, there's a woman. There's a black and
562. 1:36:29white photo of a woman. So, we're going to say um we'll call it Gabby. Gabby is riding
563. 1:36:43on a horse. in a cowboy outfit
564. 1:36:57in the dusty desert west. Uh uh
565. 1:37:17at sunset with a cloud of dust behind her
566. 1:37:34lit by the setting sun. I I want to actually I'm going to read the I want to read the JSON prompt and see what it does with that. See if it
567. 1:37:47generates cinematic shot. Now I can share this. Copy the contents. Can I publish it? Show console.
568. 1:38:06This is crazy. So what now? We put the we put the image in full description of the scene into cling AI. No, what I
569. 1:38:22would do is well so so we can add to this. We can say uh well I won't I won't add a database to it, but I'll say um I want to add a button
570. 1:38:40to download the file and be able to name it. I also want it to
571. 1:38:59download a TXT version of the Oh [ __ ] Uh,
572. 1:39:16I want to create a txt version of the JSON prompt. Let's try and let it fix the error that it it just had.
573. 1:39:33Okay. You know what I said before that I felt like an idiot and I felt like a dumb dumb cuz I really didn't know how to use um this thing.
574. 1:39:42I feel I feel well I'm excited but like I don't know what this means at all. Like is this expected behavior? If
575. 1:39:58you give it a JSON prompt it it's going to it's going to assume you want code. I don't I don't know. This isn't
576. 1:40:06necessarily a bad thing. It's a cool thing. All right. Let's see. All right. Um, Gabby wears a cowboy
577. 1:40:21outfit in 18 94 in the Mesa Wild West.
578. 1:40:37Riding at speed at sunset. Okay. Create a cinematic shot. This is crazy people. It's crazy people.
579. 1:40:54It's crazy. Crazy crazy. It's very cool. All right. It's I think it broke the app
580. 1:41:08quite frankly. because we we don't have the JSON prompt anymore. Add Gemini features. What's that? Oh, what what the hell did I just do?
581. 1:41:24I enhanced the cinematic shot with two new LLM powered features using the Gemini API to give you more control and inspiration.
582. 1:41:33Here's the updated code. Huh?
583. 1:41:51Oh, look. It added it added uh it added AI stuff. Expand the idea. Generate the shot. Wow. Generate an image. Okay. Go get the
584. 1:42:03image. Got the image. character name Gabby. Um, Gabby. Gabby on a horse
585. 1:42:17in the 1800s. And then I'm going to say expand idea
586. 1:42:37Gabby not just on a horse but one with her horse. A powerful slick Mustang whose main Jesus generate a cinematic shot out of that. Now
587. 1:42:50I think it I think it broke our thing. I think it's broken, but we'll see. This is crazy. If you're new here, if you haven't been
588. 1:43:05paying attention, I'm playing around with image generation within Gemini 2.5 Flash. I had Gemini generate
589. 1:43:16a JSON prompting framework for an image. I then had it generate that image and rather than just generating the image,
590. 1:43:24it generated a website to generate images. So, so I'm trying to get my head my my jaw back up off the floor to understand
591. 1:43:34how do I even like make sense of this? Watch her be a centaur. Yeah, that's true. Can it make a custom ratio like
592. 1:43:452.1? If we add that, I'm going to just tell it it broke it. Um, it looks like the app is no longer
593. 1:43:59generating the JSON prompt and there for not generating the image. It hangs
594. 1:44:19on generation. Fix that. And I want to be able to add a button to download it. End the prompt. I also want an aspect
595. 1:44:35ratio chooser that includes vertical, square, and
596. 1:44:49cinematic wide ratios. All right, let's see what we get here. Upper left says canvas. What? Unfortunately, something went wrong. Fix error.
597. 1:45:22HTTP error. All right, it's fixing it. All right, people. I'm getting I'm getting crunchy. We didn't We didn't do much image
598. 1:45:34generation, did we? Okay, tomorrow night. So, it's Thursday night. I made this announcement last night and I was wrong. But this night,
599. 1:45:43it's correct. Tomorrow night is Friday night date night. If you haven't been to a Friday night date night, do yourself a favor
600. 1:45:52and show up. And you might ask, well, what's different on Friday night date night? I don't know. Well, if you have a
601. 1:45:57date, that'd be different. If I'm your date, well, welcome. Uh, we do this. We'll we'll be I'm going to definitely
602. 1:46:06be playing with Nano Banana tomorrow. Um, so do that. And then tomorrow at 11:00 a.m. Mountain time is AI office hours.
603. 1:46:19So, I have AI office hours. Uh, if you go to my LinkedIn channel, Kyle Shannon on LinkedIn, go to my
604. 1:46:27events, you'll see AI office hours mixed in with these recordings and just click on one of those and the URL there is uh
605. 1:46:39is the one for tomorrow at 11. Okay. Enhanced error reporting, flexible as aspect ratio, but there's no
606. 1:46:53Oh, yeah, we now have aspect ratios. All right. Well, they're shitty, but okay. What we can, but now we have them.
607. 1:47:03So, okay. So, we'll do we'll do this as a square image just to test it. We're going to upload an image, Gabby. We're going to say Gabby.
608. 1:47:17We're going to say rides a horse in the 1800s Wild West. And I'm not going to I'm not going to improve it,
609. 1:47:33but I am going to say generate the shot.
610. 1:47:49Hey, Pate, did you know did you know that Gemini will just randomly build you a website if you ask it to make an image using JSON prompting?
611. 1:48:00Because that's what happened here. This is crazy. Uh, it still doesn't look like it did that. So, um, still breaking
612. 1:48:17before it generates the JSON prompt.
613. 1:48:33I've been using a little JSON and chat GPT and now it asks me if is If I want JSON appendix added to deep research. Oh, that's
614. 1:48:44interesting. I haven't had that problem specifically. I don't know that it's a problem paid. I think it's actually
615. 1:48:50really interesting. Like it it seems to me like it it sensed that I wanted code or that I, you know, that I had code
616. 1:48:58and that maybe I wanted something more. But I don't know. It's kind of crazy. But it's written us a pretty sophisticated app here.
617. 1:49:10and it generates images. Check out QC coder. Yeah, this is crazy. All right, dynamic shot generator. Your
618. 1:49:22cinematic shot will appear here. All right, let's try it again. Gabby. Gabby. Um, cinematic. Uh, Gabby rides
619. 1:49:38a Mustang horse in 1800's Wild West. Yeah, that's it. That's a good prompt.
620. 1:50:11Oh, QCoder writes everything. You can chat and it it does everything in Visual Studio. It's like this. Yeah, like a lot
621. 1:50:17of these vibe coding platforms are they're pretty pretty [ __ ] powerful. Um I think this has failed again. Anyway,
622. 1:50:29um I guess I can come back to this chat if I wanted to work on this. I just want it to work and it's not. So, it's
623. 1:50:36pissing me off now. All right, [ __ ] it. Enough. Done for the night. Um All right. I'm crispy. That was a peppy
624. 1:50:44night. Um thanks for the good interactions tonight and the really good questions. Um it makes these things much uh it makes these evenings
625. 1:50:57I I think these kind of evenings are much more important where we get to really talk about what's going on with
626. 1:51:03AI and what they do and how they're going to affect us. Those are really good talks. So, I really appreciated
627. 1:51:08that tonight. All right, let me jump over to the irregularize and reload. It looks like we've got some activity. [Laughter]
628. 1:51:22Kevin Mallister moment. Yep. I had one of those tonight. That was kind of cool. There's me pushing the button.
629. 1:51:34Thank you, Lord. Digital gods. Very nice. All right, everybody. Um, Givi. So, tomorrow, tomorrow is just a regular Friday.
630. 1:51:48So, Friday night date night, 8:00 p.m. Mountain time here. be there. Be square. I'll give you homework for the weekend.
631. 1:51:56And then 11:00 a.m. tomorrow is AI office hours and you should come to that. Okay. You start making them input to each
632. 1:52:08other becomes interesting what you can make. Yeah. Yeah. There's all sorts of cool ideas. And the fact that you could just add
633. 1:52:17like I just hit the button and it said add Gemini features to this app that I just built you and it just did it. That
634. 1:52:24was pretty [ __ ] slick. All right, we're in new territory, people. When when your image generating tool starts
635. 1:52:32writing you an application to generate images, you know you're in different ter we ain't in Kansas anymore.
636. 1:52:41That was absolutely trippy. So anyway, hope you had fun tonight and uh I'll see you tomorrow. Peace out.