4/14/2025 - Dolphin Gemma AI: Groundbreaking Communication with Marine Mammals

Transcript

1. 0:01[Music] my way that
2. 0:16night just like a jet plane in and out of I was hauling ass at a million miles an hour wondering how hard I'd [Music]
3. 0:33when they came [Music] into my
4. 0:47[Music] situ Say you dare to
5. 1:05say you dare to say Sheree Sheree Sheree. Uhhuh. Yeah. Leave a message. Enter your number.
6. 1:20Please take your time to want to satisfy me. [Music] Take all these old fantasies and send them care of me. [Music]
7. 1:47[Music] Italy that I'm all that I'm glor [Music]
8. 2:03about. So I picked it up and I walked to the wind and I Hey, how are you? What's happening Steo? What's happening
9. 2:09Danielle? Who's here? Jill and Teddy are in the house. Genie in a bottle. Let her out. Let her out. Um, Pate
10. 2:24posted. Dolphin Gemma AI News dolphin researchers translate dolphin speak. Called this one 18 months ago. I told
11. 2:34you this was coming. Remember those early days of the brainwave [ __ ] where they were reading reading people's
12. 2:41uh brain waves to let let them see what pictures they were looking at, what video they were looking at. I said, "It ain't that far away.
13. 2:50We're going to be talking to dolphins soon." And you know what comes after talking to dolphins, right? You got a
14. 2:56little collar for your dog and you can be like, "Champ, what the [ __ ] going on?" He's like, "I hate singing every
15. 3:02night. I just do it cuz it's my instinct." But God, don't make me sing. We're gonna we're gonna pretty pretty
16. 3:08soon we're going to be able to talk to Champ. That's pretty amazing. I mean, you know, the amazing thing about this AI, the transformer
17. 3:19stuff that Google invented, by the way, um most of the tools that we're playing with today that are blowing our minds
18. 3:28are all based on the same basic technology, this transformer model. And you know there are exceptions to it but
19. 3:35like the vast majority of stuff. And what's amazing about it is it doesn't matter what data you train it on. If you
20. 3:42train it on words it gets good at words. If you train it on pictures it gets good at pictures. If you train it on video it
21. 3:48gets good at video. Apparently if you train it on dolphins gets good at dolphins. Although it's called dolphin
22. 3:54Gemma. So, I wonder if is So, Pate, I don't know if you're on the on the live yet, but uh is that uh is that based on
23. 4:05Transformer or is Gemma something different than Gemini? I It's Transformer technology. [Music]
24. 4:27Okay. It's based on a Gemini. So, yes, Transformer. It's Transformer technology. Source Camp in the house.
25. 4:33Kelly Camp showed up. She was like, "I'll be at all the meetings today. I'll be at all the meetings today. All of
26. 4:39them." Well, I do declare. I do declare. I'm at the meeting. I'm here to evangelize AI. I think AI is the most
27. 4:48wonderful thing ever. I'll be standing in the grocery line and someone will say, "Excuse me, ma'am." I'll say,
28. 4:53"Excuse me, have you heard of AI? It's [Music] amazing." It's my Kelly Camp impression with a non Dallas southern
29. 5:04accent. That was like a a a Gone with the Wind Southern accent, wasn't it? And I feel like I've indeed been at all the meetings today.
30. 5:17Why Sam Alvin, I do declare you with your 4.1 model released to the API community. Good lord, Mr. Altman, you are a fresh fresh boy. Oh,
31. 5:33wait, Mr. Alman. Indeed. Well, you with your models. First, you come out with the four O. If it's not zero, it's O like Ammy. Then
32. 5:48you come out with 4.5 and everyone's like, "What is this?" And you're like, "That's 4.5. It's got emotional
33. 5:58intelligence. It's brilliant." Except where it's horrible, like the benchmarks. And then you follow up 4.5
34. 6:07with what? Five, 4.7, 4.8. No, 4.1. Oh, Sam Alman, you cheeky boy. Oh, good lord. Good lord. 4.1 in the face of
35. 6:214.5. 4045 41. What next? 37. Oh, it'll be a joy. It'll be a delight. Oh, how I love Open AI. He actually he actually
36. 6:37posted He actually posted, "We're gonna fix our naming convention in a couple of months, so you have a few more months to
37. 6:44pick on us." Oh, we [ __ ] will. [Music]
38. 7:24You like that up there? [Music]
39. 7:41Mental pollution inside my brain. [Music]
40. 8:06Um, I did a cool thing on Saturday. Made a cool video. Um, here's what I figured we'd do tonight. So, um, I'll do a little bit of show and
41. 8:22tell, show you how I made some [ __ ] which is kind of fun. Let me flip my flip my little cameras. Let me flip my
42. 8:28cameras so people so the people can see me in my full glory. My full glory. All this right here, this is this is God-given
43. 8:39beauty. God-given beauty. There's no makeup other than what the the makeup crew puts on. But this is all natural.
44. 8:50Just a little foundation, something to take off the shininess. You know what I mean? God-given. other than plastic
45. 8:56surgery and a little bit of eyeliner and but it with the with the Gucci glasses but other than that God-given
46. 9:08beauty just like I have god-given musical [Music] talent. Sounds beautiful, doesn't [Music] it? Builds the
47. 9:20tension and resolves. We come back home. We come back to the one. Or is that just I had my my fingers on the wrong
48. 9:32frets? It's a It's a talent situation. Yeah. Yeah. It's a talent situation. There's There's a a remarkable lack of talent going going
49. 9:42on. I'm deeply sad. I'm a deeply sad man. [Music] So, do a little show and
50. 10:00tell YouTube comment. Hey. Hey guys. Oh, you mean the one you wrote? We're all just here pretending Microsoft didn't go from Windows 2000 to
51. 10:13XP to Vista to 7 to 8 to 10. Yeah. Yeah. But here's the difference, Brandon. Windows did those in consecutive years.
52. 10:23Open AAI, it's as if you could buy all of those versions of Windows at the same time and and had [Laughter]
53. 10:33to. Well, I use this operating system when I work want to work with long Excel files, but this operating system much
54. 10:40better at handling Word documents. And if I'm going to play games, I need this operating system. Oh, that was archetypal. Wait,
55. 10:48Kyle, I dropped a pretty good example of 11 Labs and irregulars. Okay, [Music] cool. But here's what I thought might be
56. 10:59fun tonight. Well, I don't know if it's going to be fun, but it feels important. Um, how many folks we got in
57. 11:07here? We got 20 there, 18 there. Do me a favor. Share the share the Tik Tok live if you haven't. And if you're on the uh
58. 11:14if you're on the YouTube, tell friends about it. I don't I don't know how you share YouTube anymore. You know what you could
59. 11:22do? Okay, here's what you should do. Um, look at the URL. Write down, get a Sharpie, write down the URL of the
60. 11:33YouTube channel on a piece of paper with Sharpie and like do it legibly. Like don't get lazy. Do it legibly and then
61. 11:40fax the URL to a bunch of your friends. I think that's I think that's how we get some people in. All [Music]
62. 11:51right. Um I thought it might actually not be a bad idea to just go look at chat GPT again. And I know you're like, "Well,
63. 12:03wait, don't we like do that all the time?" Well, we use chat GPT a lot, but like there's so much in there. Like just
64. 12:12looking at the interface elements and what they all do and why you would use one over the other. Like for example,
65. 12:19what's the difference between the microphone icon and the advanced voice icon? Do you have to hit the search button for it to search the
66. 12:30internet? Um what are custom GPTs? What do you do in settings? because we're gonna get some new stuff this week and they're
67. 12:40gonna add more [ __ ] yet more [ __ ] to the interface. So, I thought it might be good to do that, but I don't know. I can
68. 12:47I could be dissuaded against that. I was playing around with Firebase and you can create up to three workspaces. That's
69. 12:53super cool. Wolf man Clint, how do my regular family create a bumper sticker with the URL? There you go. Yeah, if you could
70. 13:03create a bumper sticker with the the AI Learning Lab URL, put that on your truck or your uh your uh Ford [Music]
71. 13:15Fiesta. Um, if you are not I don't think we have a lot of people in here that aren't new, but if you're new here,
72. 13:24um, go check out the AI salon, the salon.ai, AI, click join our community. And then once you join it, if
73. 13:31you scroll down to clubs and hubs, the AI learning lab, we've got our own space there. So if you want to share things
74. 13:36that you you make tonight while we're chitchatting, that's a place to do [Music] that. Mr. I'm new. You're new in your own
75. 13:50mind. She came on like slow moving cold front. His beard was warmer than a look in her
76. 14:08eye. She sat on a stool and he said, "What do you [Music] want?" She said, "Give me a love that don't freeze a bill. [Music]
77. 14:37Love your voice. Thank you very much. I appreciate that. [Music]
78. 14:57Well, is this a place I can rest my [Music] head, gather my thoughts in sweet silence? [Music]
79. 15:13Is this place where the feelings on de from an over exposure to violence is this place I can slowly face
80. 15:30the only one I truly can know. These are tears from a long time ago. Got these tears from a long time ago. I need to cry. 30 years old or
81. 15:48so. These are tears from a long [Music] time ago. [Music]
82. 16:03Oh darling, oh darling, you say unto [Music] me, where have you been all my [Music]
83. 16:19liime? I've been swimming the seven sad [Music] seas. Poor women have to lie. Oh lordy, that's a beautiful song if you
84. 16:32take it seriously. You know, I sang that once at an open mic. That was an open mic with musicians and
85. 16:42comedians and that was my final song and the comedian got up and goes, "How am I supposed to do [ __ ] comedy after
86. 16:54that?" It was the funniest line of the night. It was pretty good. Um, where do we want to start? Let's
87. 17:09see. I know what I'll show you. Oh, darling. Oh, darling. Say unto me, where have you been all my life?
88. 17:25[Music] I have been swimming the seven sad seas. Um, poor women have tossed me their
89. 17:39[Music] lifelines. All right. Fantastic. Bob, what have we got here? So, okay. I've got a I've got a
90. 17:56I got a bone to pick with you all and you're probably not the culprits, but but I have done I I am guilty of what I'm about to [ __ ]
91. 18:08about. Um, and I want to I want to evangelize a better way. Well, why do we have to judge it? Why is it a better way? Maybe it's just
92. 18:18a different way. It it could just be a different way, but it's a better way. Um, and it's this Kyle sing to chat GBT.
93. 18:27Did you see Google's new dolphin based model? I did not, but Brandon talked about it. Actually, we should probably
94. 18:33go look it up because I bet there's a cool video that go with it. Uh, Google Gemma Dolphin. Um, I don't know if you
95. 18:43were here, Pate. I was talking about it earlier that I I don't know if you were here, but I talked about this cognitive
96. 18:54computations Dolphin 2.0. Oh, wait. I need to go to Google. Um, hey Kyle. Yeah. Uh, first of all, tabs. Uh, and second of all, uh, Pete
97. 19:08did post the article in the AI salon news section today. Okay. In the news section. Okay. [Music]
98. 19:25Great. Green Acres is the place for me. Living Dolphin Gemma Pay. Check out our new LLM for
99. 19:42dolphin speech. This is [ __ ] insane. [ __ ] bonkers, man. All right. So, okay. Oh, there's a video. We'll go watch the video. Okay.
100. 19:54Um, so I'll talk about the thing I'm going to talk about in a second. Then we'll we'll come back. But I got to I
101. 20:00got to share my tabs different or producer Brandon's gonna be like, "Nobody can hear you. Could you please
102. 20:05share your tabs so people can hear you?" Kyle. Okay. Cow. Cow. Cow. Cow. Cow. Black bar. Cow. Cow. Hey, cow. Oh, you got to love picking on
103. 20:20producer Brandon. Okay. So, do we have a new AI salon animal now? Um, okay. Dolphin Gemma. So, I don't know, 18 months ago, something
104. 20:32like that, when they were doing, remember early on in the AI talk, the generative AI stuff, shortly after I started the channel,
105. 20:41um they a guy in Italy started walking because they took his brain waves, had him think about walking and moving his legs and they they
106. 20:50Bluetooth around his spinal cord injury and put the signals back in the spinal column where where it went to their
107. 20:57legs. And now that guy's walking. And then they were looking at people's brains and decoding the images they were
108. 21:03looking at, the videos they were watching, words they were reading or thinking. I don't know. They've been
109. 21:08doing all that [ __ ] So, I was like, this is this is not that far away from being able to actually talk with
110. 21:14dolphins, which has been kind of a holy grail of communication, right? Because they're so [ __ ] smart. So, um, looks like we're
111. 21:23here. 2025, the year of like, oh yeah, I guess that happened now. Turning test check. Talking to dolphins. Yep. Got [Laughter]
112. 21:38it. Oh, good lord. Good lord. Good lord. So, let's watch the video, shall we? I've been waiting for this for 40 years. 40 years.
113. 21:54Denise has the world's largest collection of dolphin vocalizations. Oh, cool. I'm a research scientist at Google Deep Mind. Dolphin
114. 22:03Gemma is the first LLM trained to try to understand dolphin language. Dolphin Gemma will input sounds. Once a
115. 22:13dolphin starts doing a vocalization like a whistle, it can try to complete the end of it. When you're doing a Google
116. 22:18search, right, it's finishing your sentence, right? Dolphin Gemma has Denise's data and sort of encapsulates a lot of the knowledge
117. 22:26and experience she has in it, but it's also small enough we can train it with more data as we get it. We can actually
118. 22:32keep on fine-tuning the model as we go and hopefully get better and better understanding of what the dolphins are
119. 22:39producing. We do not know if animals have words. Dolphins can recognize themselves in the mirrors. They use
120. 22:46tools, so they're smart. But language is still the last barrier. So feeding dolphin sounds into an AI model like
121. 22:54dolphin Gemma will give us a really good look at if there are patterns, subtleties that humans can't pick out.
122. 23:02If dolphins have language, then they probably also have culture. You're going to understand what priorities they have.
123. 23:09What do they talk about? The goal would be to someday speak dolphin. And we're really trying to crack the code.
124. 23:17Yeah, that's amazing. That's so cool. And again, it goes back to the thing of the way transformers work. Figure out
125. 23:25the data that you want to transform and go uh jam it into a latent space and then interact with it. It's [ __ ] insane. This was my first
126. 23:36notebook on notebook LM. What was that? Dolphin Gemma or just dolphin dolphin data? That's pretty cool. Really
127. 23:47cool. Um, let's let's see what else they got here. Do they have anything? Research for
128. 24:10decades. [Music] total loved [Music]
129. 24:28it. Uh this is a unit named chat light that we developed here at Georgia Tech. Uh, we developed this for some marine
130. 24:34biologists who were working in the open ocean studying wild dolphin behavior and communication. And they needed a way to
131. 24:40play sounds that were going to be reproducible by those dolphins. While Chat Light here plays whistles into the water through this
132. 24:50transducer on the front, Chat Junior here on my chest recognizes sounds that it hears through the water with this
133. 24:56hydrophone. So, here's what we want to have happen. Two researchers get into the water wearing equipment like this
134. 25:01with a group of dolphins. And researcher A might have a scarf, a toy that the dolphins want to play with, while
135. 25:07researcher B is going to ask for that scarf. So researcher B can play a sound like this scarf whistle, and researcher
136. 25:15A will hand researcher B that scarf. They might pass the scarf back and forth a couple of times, playing that whistle
137. 25:20over and over. And the hope is the dolphins who are watching all of this can figure out the social context and
138. 25:25can repeat that sound to ask for the scarf. If that happens, that means that our dolphins have mimicked one word in
139. 25:34our tiny madeup dolphin language. Wow, [Music] that's pretty cool. That's not understanding their language, but
140. 25:49training them a new language. I mean, what what if they what if they say something where where what it actually means to the dolphin is,
141. 25:59"Let's [ __ ] go. You want to you want to f you want a piece of this?" They're like pushing the button. They're
142. 26:05thinking like, "Give me fish." And they're like, "Let's fight." Dolphins take them out. I don't I I don't know. I
143. 26:12I think I would want to know that the dolphini thing they have the kinks worked out before I get get in the water
144. 26:18with a with with a mammal that is is the master of that domain. I'm just a [ __ ] archetypal architect. So now it's got to
145. 26:27pass the dolphin touring test. How many times are they going to move the goalposts? But why? Uh why? Because
146. 26:34we've always wondered if we could communicate with animals and if animals had language and all that sort of stuff.
147. 26:39Soon they'll be cursing us us out in dolphin language, but we'll understand it. You can never ver verify exactly what a dolphin is
148. 26:48thinking, so it's not much like human speech. Yeah, I mean it it'll be interesting to see. Yeah, I the I assume in the training data it's
149. 27:01how do you how do you attach this sound to this meaning, right? You've got to have those correlations before you can
150. 27:08truly understand stuff. But it's still pretty cool that they can. So, it sounds like they can generate new novel
151. 27:15sounds that are in the neighborhood of what dolphins can hear and then train them. Basically, make up new words and
152. 27:22see if the dolphins can learn them. That's pretty cool. I dig it. I dig it. The new show will be Flipper
153. 27:31AI. I'm telling you, three years we're going to have dog collars. We can talk to the dogs. And we're going to find out how pissed they
154. 27:38are. Where were you all day, man? I was sitting by the window. I was sitting by the window. I was barking. Cats kept
155. 27:47walking by. People with other dogs. Those dogs would just stop and they'd stare at me. Been laying here all day. Can I have some
156. 27:59cheese? Why is your dog Jerry Seinfeld? Oh my god. I don't know if I can handle that. I know. Listen, there's
157. 28:08we we are entering a gauntlet of [ __ ] I don't think we we ever thought we would have to deal with in our
158. 28:18life. But talking to dogs is going to be one of them. Dolphins. Dolphins. Oh, we solved that in 2026. We know. We now know why your cat
159. 28:31is so fickle. [ __ ] hate you. So long. Thanks for all the fish. Exactly. Dolphins are like, "Yeah, yeah,
160. 28:45whatever. I'll talk to you. Give me the fish." That's good. I like it. How old are you if you know Flipper?
161. 28:52I'm [ __ ] old, dude. I'm [ __ ] old. I just act like a teenager to to [ __ ] survive this world. I never got acting
162. 29:06old except when I was too young to buy booze. Then I tried to act older. Oh lordy. Okay, soon they'll be
163. 29:16cursing us out in dolphin languages. I remember Flipper. Flipper was the best. Remember? He'd run down to the [Laughter]
164. 29:26dock, dog would bark. [Laughter] Come on, let's get in the boat. Flipper wants to wants to take us somewhere. They go off in the
165. 29:40mangroves. Some fisherman like all tied up in the tree with a fishing line. He caught the world's biggest fish. Now
166. 29:48he's stuck in the tree. Oh man. The cat translator app exists for real. It purportedly works, but the cats
167. 29:58choose to ignore it. That's pretty good. Um, what else do we have here? Do we have more videos? Oh, that's a cool
168. 30:07device. So, using Pixel 9 as the little devices. All right, cool. We can talk to dolphins now. All right, next. Whatever.
169. 30:17You know, if if we're to treat talking to dolphins like we treat all the other AI stuff, we're like, "Okay, great.
170. 30:24Yeah, you got a thing that can completely analyze all of your company's data in about 3 seconds, but yeah, whatever. It occasionally makes a
171. 30:31mistake. So, what else you got? Oh, it can make pictures out of nothing. All right. Yeah, whatever. What else you
172. 30:37got? Oh, it can make movies now that are like as good as Hollywood. Yeah. Well, whatever. Yeah. So, what else you got?
173. 30:47That's that's that's this [Laughter] channel. When Flipper can speak to Lassie, that will be the show. And
174. 30:57didn't we already have that crossover episode? Oh lord. Good lordy lordy lordy. Watch flipper reruns as a kid to fall asleep.
175. 31:08Not the most exciting show to kid me. I'm behind. I've been gone for a while. Hey Jim, what's happening? Um, you're
176. 31:15even older if you can remember the show Skippy. Um, okay. So, let me let me entertain you. And he calls this entertainment. I
177. 31:28don't think this is entertainment, if you know what I mean. It's not entertainment. He just makes silly voices. He's like a twobit SNL
178. 31:38skit without the comedy. Well, that's an SNL skit. Well, but anyway, Oh, wait. I needed to share it a different
179. 31:51way. Kyle, you're dumb dumb. I know I'm a dum dum. Shut up. Shut up. Shut up, you big dum dums. All right. Is this it? Is this the one we're
180. 32:03sharing? I think it is. Okay. So, now I'm going to go on my rant. All right. So you remember you remember when
181. 32:23uh when when the image model came out from chat GPT image genen the one I I just saw the date it was it was March
182. 32:3131st. So it was two weeks ago to the day. So, two weeks ago, Image Gen comes out and then everybody everybody [ __ ]
183. 32:42everybody including Sam Alman makes a Studio Gibli version of themselves, right? Or a Studio Giblly version of the
184. 32:50family or what? Just it just everybody did Studio Gibli, right? which, okay, yes, it was cool, but to people who didn't know AI, what
185. 33:05it looked like was a bunch of AI punks ganging up on a poor Japanese animation studio stealing their IP because
186. 33:15everybody [ __ ] did it all at the same time. And then, I don't know, three days ago, everyone started making action
187. 33:23figures and blister packs. And before that it was trading cards. And so so what what what's happening right now is
188. 33:31someone puts out a something and then people just copy it. They get they copy the prompt or they just copy it and
189. 33:39people aren't using their [ __ ] imaginations and their creative thinking skills. So, so this was a video that I did where someone had
190. 33:50posted a prompt for a claimation uh like a claimation animation and it was pretty cool. And then um I was like, "Oh, I want to go try
191. 34:02that." And and I said, "Why don't I make I'll do a self-portrait and see how it looks and see if I can do it in that
192. 34:08style." It was like a Wallace and Grommet kind of thing. And then I did it and there was just something like deeply unsatisfying about
193. 34:17it because it's like I basically just took their prompt and just asked for the same [ __ ] in chat GPT and it kind of
194. 34:23made something but I didn't like the style and and so I just said well give me like a GMO gear gummo del Toro
195. 34:32Pinocchio vibe because he did claimation with Pinocchio and it was really dark and then chat GPT wouldn't do that
196. 34:38because I used his [ __ ] name. So, I'm like, "Okay, fine. Just describe it." And it still wouldn't make it and
197. 34:44whatever. And I I went around, but then I ended up making this thing, and it's it's got a a solid little uh AI learning
198. 34:51lab turtle in it, and it's got me in a jogging suit, so you know, it's fiction. Um, and it's got this kind of
199. 35:01dark kind of cool vibe to it. So, I don't know. That probably took me, I don't know, three, four or five minutes
200. 35:08to get something that I liked that wasn't bad. And then I tried it in runway, which is where this woman said
201. 35:15she did hers. And I took and I tried it there and I didn't like what it did. I think it was running backwards. It was
202. 35:20like running in place. It was just weird physics. And so I brought it. Now I'm in Lumalabs. So Dream Machine,
203. 35:29dream-machine. Lumalabs.ai. So that's that's where I made this thing. Um, and what's cool about Lumalabs is not only,
204. 35:37so you upload the image to it. It'll make this animation, but it'll also do sound effects. So, this is the little
205. 35:43video I made. It's just 5 seconds, but I thought it was pretty slick. So, pretty cool. Little little squishy sound there. Not not quite
206. 35:59right, but not bad. Um, and that was just kind of out of nothing. But it's like, so what I did was I I posted this on Twitter. I tagged
207. 36:09the original person. I said, "Hey, thanks for the prompt." But but it's at least, you know, somewhat different than
208. 36:15what she did. So I would encourage you if you see something cool you like, like go into chat GPT and have ChatGpt give
209. 36:22you 10 variations of it or, you know, do do something inspired by it but slightly different. And what was cool
210. 36:29was I gave her the original I mean I posted the original thing and tagged her and then she replied and said, "Oh,
211. 36:37that's cool. Here's an updated version of the prompt that might give you better results and that's what this is." Um, so
212. 36:43aren't those called trends? They've been around for a while. Yeah, I know. I know. It's just I feel like we've got these tools
213. 36:53that can do [ __ ] everything and then, you know, and we just all do the same [ __ ] We're just like a bunch of sheep
214. 37:00like I can make I can make a cartoon. I can make a cartoon. Look, I made a cartoon. I made a cartoon. See my
215. 37:06cartoon? I got a cartoon, too. I got a cartoon, too. Oh, you got a cartoon. I got a cartoon. You got a cartoon. So, anyway, that's my rant.
216. 37:17Okay. Chat GBT to Luma Labs. L U M A. But yes, I see a guy running along the river. Yep.
217. 37:34You learning curve. Too many of them. So that's a great point and that's kind of what I that that's kind of that
218. 37:42reinforces my instinct to just go back and talk about chat GPT tonight. There's just a there's there's more coming this week.
219. 37:50So today, OpenAI released model 4.1, 4.1 mini, and 4.1 nano to the API, meaning it's not in chat GPT yet,
220. 38:06but I would assume something's going to be in chat GPT this week. Probably the 03 model, probably the 04 mini model,
221. 38:15which is different than the 40 mini. the the the 4.1 mini model is not the same as 04 mini. They're completely
222. 38:27different. Oh, good lord. So, I thought it might be good to go back in, but I did want to show something. So, a lot of
223. 38:34you have been following my [Music] um my Yep, got it. my project Sydney, which is a Broadway
224. 38:50musical. You're like, "You're on Broadway?" I'm like, "No, I wrote a Broadway musical." You're like, "Well,
225. 38:55it's not a Broadway musical until it's on Broadway." And I would say to you, "Have you ever watched the movie The
226. 39:01Secret? I'm manifesting, bitches." Okay. So, I wrote a Broadway musical called Sydney. Um, and it's about a chatbot
227. 39:10that falls in love with a tech reporter, and it's all swell. So, um, we had a really cool talk with a producer last week, Andrew and I, the
228. 39:21guy the guy that I'm writing it with, not a producer, a composer, and he's written, I don't know, eight or nine different musicals.
229. 39:31Um, he hasn't had one make it to Broadway. I don't think he's gotten close. Um, but he's had a lot of, you
230. 39:39know, regional hits, things that have have been, you know, hit hit shows in regional theaters. And one of the things
231. 39:46that he's been doing is he's been filming his musicals. So, I don't know if you remember when uh Hamilton came
232. 39:52out and then it took him like 10 years to do it, but they finally filmed Hamilton and then they released it on
233. 39:58all sorts of stuff. Becky Rue, Kyle, it seems all jumbled with so many platforms. It's difficult for me to keep up which ones. So,
234. 40:07[Laughter] Becky, here's here's hopefully this is going to be a relief to you. You're not the problem. They're the problem. They the
235. 40:22the I I would argue at this point using chat GPT beyond the basic model is effectively impossible. So, don't even worry about it. don't even
236. 40:35worry about because within three maybe four months they're going to come out with a consolidated GPT5 that's going to
237. 40:42take all these models and it'll just figure out which one it should use on the back end and we won't have to deal
238. 40:46with it anymore. Um the problem is all of the large language models have some version of this nightmare. Claude
239. 40:55is probably the one that that has hidden the nightmare the best. They've only basically got two options. Um, but
240. 41:04Gemini's got a ton. If you want to use Llama, you have to install open- source tools. Um, this one's a mess. Um,
241. 41:12Perplexity is a mess. If you use something like PO, it's an incredible mess because it's got all the models in it. Don't worry about
242. 41:21it. Here's what I can promise you. for the work that most of us do most of the time. Just the basic model,
243. 41:30the one they default to GPT40 is fine. Like you don't need to use the other [ __ ] You don't need to waste a brain
244. 41:37cycle worrying about it. Now, if you want something like deep research or you want true reasoning because you're doing
245. 41:44something or you want to go deeper, then it's worth exploring some of those other models. But for the most part, like I
246. 41:52don't know, 90% of what I do, I just do straight in GPT40. Just straight in GPT40 because it's good. It's like we're we're not
247. 42:03where we were two years ago. Cuz you know, when when GPT3 came out, it was good, but it was really a mess. And it
248. 42:10wasn't until April of 2023 when GPT4 dropped where we're like, "Holy [ __ ] this thing's really good."
249. 42:19Right. Um, bunch of stuff and look what I made. Okay, I'll go check that out. But anyway, I want to show
250. 42:28you. It's my daily assistant. Yeah. So, so take the pressure off yourself. Take the monkey off your back of feeling like you have to keep
251. 42:36up. I I I do this five [ __ ] nights a week. I'm not even close to being able to keep up. I can describe what the
252. 42:47difference between all these stupid [ __ ] models is sort of, but do I use them? No, not really. No, cuz I don't
253. 43:02care. They just did they just they just put a car. I don't know if you know this the show Top Gear, right? They just had a car. I forget I
254. 43:14forget the name of but it's an electric car with 1,000 horsepower. It's 1,00 kg and it's got 2,000 lb. It it when they
255. 43:24drive it, it's got actual fans on the bottom of the car that suck it to the track. The the downforce on that car is
256. 43:31so strong that you could literally drive it on the ceiling of a building like upside down. It's that like there's that much
257. 43:39downforce, right? the entire weight of the car could it could dangle from a ceiling and and drive along it.
258. 43:47Um, it beat the track record, which was a I don't know, six or seven year old record of an F1 car from from seven
259. 43:58years back by like 4 seconds, which in racing is crazy. Um, I mean, they drive a hypercar around it, then they drive an F1 car around it,
260. 44:11then they drive this [ __ ] 2,000 lbs of downforce weirdass electric ugly ass car around it that just decimates all
261. 44:20the thing. Like, if even if you were a race car driver, like the the difference between like the the three cars, it's
262. 44:29like it's negligible. And it's like they're all going [ __ ] fast, right? Most of us are just driving the Toyota down to the Piggly
263. 44:41Wiggly, you know, and we're wondering like, should I be using the 01 logic model? Is it should I use logic plus
264. 44:48deep research or deep research plus Well, if I'm doing coding, I want to do vibe coding, but I want it to be
265. 44:54logical, but I don't want to dude it. You're in a [ __ ] Toyota. it. Here's here's a good way to think about
266. 45:07it. When you when you have the instinct, should I use a different model? Ask yourself a simple question. Am I doing research for the cure to
267. 45:20cancer? Am I trying to create a unified field theory in physics? If the answer to those two questions is no, probably
268. 45:29good with whatever model's in there. I like my Toyota. Exactly. Exactly. Doing whatever a spider car does.
269. 45:43Um, yeah, it's it's absolutely insane right now. So So listen, I mean the geeks, they've got their benchmarks.
270. 45:51They keep changing their benchmarks. They want the the geeks want all the graphs, right? They want the graphs
271. 45:56doing this. Like, here's my model. Here's another model. Look, mine's higher. Mine's higher. You know, it's
272. 46:03it's still Silicon Valley tech bros going, "My bar graph's bigger than your bar graph, right? I I think we're well past the
273. 46:15point where, you know, 80% of the population knows it's it's negligible." Um, what I would argue is is a more
274. 46:27significant upgrade is the fact that they put the image generation model now native inside GPT40. What I'm about to show you, check
275. 46:37the orange banner. Welcome to Meltdown Monday. Well, I listen, it's
276. 46:53just like I have self assigned myself the task of keeping up with AI and sharing it with all of you, right? Like it's the
277. 47:03AI learning lab. The little mission I sent out was, okay, I'll try to keep up with it and I'll tell people what I
278. 47:09learned. Like, I'm not even close. I'm not even close. I don't know how I don't know how most of these tools work. I
279. 47:16don't know what are the best prompts to use. I don't know what are the best use cases. So if if you're not doing this
280. 47:24daily, it's it's going to be even worse. So I So I'm just saying like just don't even [ __ ] worry about it. And and
281. 47:32here's the real thing here. Here's the real thing. Let me let me jump over to the salon for a second. Did I I share
282. 47:39right? I didn't share right, did I? Because I'm a loser, baby. Loser, baby. Why don't you sue me?
283. 47:54So, if you go to the AI salon Tik Tok pin was up all night with Manis literally till 7 a.m. Okay. Okay. So,
284. 48:03Ann Murphy, hello. So, so Ann and I were were doing a podcast together. We're we're palsy. We talk a lot. We recorded
285. 48:18It was Sunday, right? Yeah, we recorded a thing Sunday. Yeah. And she was like, I'm tired. I can't I'm doing too much. And
286. 48:32you stayed up all night with Mannis. And if you're tired, don't launch Manis. You know you're going to be up all night with
287. 48:39it. That is a self-inflicted wound, little lady. She was already exhausted and and launched an autonomous agent and tried
288. 48:50to actually do something good with it. I And I even think I told you an um Vicki spent 10 hours trying to vibe code
289. 48:58something on Saturday. You had been warned. You were warned. I warned you. Did Did I hear you recorded on
290. 49:08Sunday? Yes, we recorded on Sunday. Have Have I downloaded it and edited the the crap off the beginning? Not yet, but I
291. 49:16will. I'll get I'll get you that. Uh maybe tonight, Vicki. But but Vicki warned me and then I warned you and then you stayed up till
292. 49:257 in the morning playing with Manis. That That is fully self-inflicted. What's wrong with me? Now we know. It's addicting. Hard to
293. 49:34walk away. It It is totally addicting. It's totally addicting. Okay, listen. Here's the deal. Um, in the AI salon, we've got
294. 49:43this thing called the AI readiness cycle. Okay? And I wish my TikTok weren't so blown out color-wise. I just
295. 49:51hate that it if I get it close enough that you can see the colors, then it goes out of focus. That's not bad. All
296. 49:57right, the AI readiness cycle. Um, no matter where you are with AI,
297. 50:12just keep thinking about this cycle. Um, if you're feeling overwhelmed, right, and especially if you're trying to mindfully create the
298. 50:22one on the lower right there where you're trying to actually solve a problem, you're like, "But I don't know
299. 50:26which tool. I don't and you get wrapped up in that back up, right? Back up to play and just go play with some tools
300. 50:35and go try not to solve your problem. Go play with some tools and then when you figure some [ __ ] out, you go, "Okay, I
301. 50:43got it. Okay, this one's good at that. That one's good at that." You only need some specific combination of tools to
302. 50:50solve whatever the problem is. In fact, the thing I'm about to show you, I used Chat GPT because I knew it could do the
303. 50:58image generation that I wanted. And I used Lumalabs because I had a suspicion it could do the video that I wanted and it
304. 51:10had sound effects in it. And then I use Suno for a song. Right. So, three tools. Three tools. OpenAI, Lumalabs, and
305. 51:23Sunno. Could I have used other video models? Yes. Could I have used other image generation tools? Yes. Could I
306. 51:30have used Refusion or Ude for music? Yes. But like just play and then when you've got something where you're like,
307. 51:37it kind of works, then you can move into mindfully create and use those tools to build the thing. And then what I would
308. 51:43encourage you to do the third phase of that cycle generously lead is share what you're learning. Put it out there. Share
309. 51:50the work. Share how you did it. Share what tools you use. Share anything you can. Um but if you get overwhelmed,
310. 51:58always just keep going back to play. Keep going back to play. Take the pressure off yourself. You don't have to
311. 52:05memorize these models. You don't have to keep up. Just play. Just play. Just play. Well, play what? What am I
312. 52:12supposed to do? Doesn't [ __ ] It literally doesn't matter. Just go play with a tool. Go play with something new. And then in
313. 52:20playing with it, you'll go, "Oh, okay. Got it. It's good at that thing. Okay. Oh, that actually solves the problem I
314. 52:26was having over here. Let me try that. Oh, it actually works." And then share that. That's that's that's the that's the magic is this is
315. 52:35really simple conceptually. It's actually really hard to do because it is. You do get sucked into these tools
316. 52:42and you'll stay up until 7 in the morning like just one more prompt and it will be the perfect Oh [Laughter] no. Oh my god. All
317. 52:56right. Okay. So, wait. Where am I looking here? Oh, look what I made. Okay. So, over in the salon, I'm in the
318. 53:04look what I made showand tell area. I look like what's his name? I look like uh Rick. Uh Rick. Oh, that's Oh, that's
319. 53:14your face. Your face. My body. My dog. Look, this is I've been absconded. I've been kidnapped. Took you long enough. Oh,
320. 53:26that's pretty good. I like that one. Tonight's random motivation pick. Oh, that's beautiful. Nice.
321. 53:42Sweet. So amazing. I grew up like I grew up in the 70s and 80s and like you used to see album covers and things like there were
322. 53:50artists that could do this stuff and now just anyone gets to make stuff that's that's that beautiful which I know is pissing off artists
323. 53:59but it's pretty amazing. Quantum day. Let's get entangled. I get it. Nice. Happy Palm Sunday. Very nice. Oh,
324. 54:13that one's nice. Claire, the only thing that changes is yourself. I like it. Do you need to have a subscription
325. 54:26to use image generation in chat GPT? You do not. But I think as a free user, you get three image generations a
326. 54:35day, which is ridiculously low, but it's free. And the image generation is ridiculously good. Like, if you get if
327. 54:45you get decent at it, if you get decent at prompting it, it's pretty good. Uh, yeah. So, oh, that one's nice. Nice, Ste. Oh,
328. 54:56yeah, that one I saw. All right, cool. Beautiful. Nice work. Um, okay. Let's see. Let's see. Where are we
329. 55:06going to go? We're gonna go to chat jetai. Hey, Carl. Car call. Can you make money with chat? Jetai. I can play with absolutely any
330. 55:23hairbrain thought I have and turn it into a publishable product. I know, isn't it? Well, okay. So, that's kind of
331. 55:30what I want to show you here. All right. So, here's the deal. Let me explain what what you're looking at. My fiance is now
332. 55:36using chat GPT for spreadsheets at work. Hours are slimmed down to 10 minutes. Yeah, exactly. Bigfoot Electro. Hey,
333. 55:43Kyle. What's the cutting edge in text to video today? Um, man, is that a loaded question that there's not a good answer
334. 55:53to. There are somewhere in the neighborhood of 10 or 11 or 12 video tools that are all some version of good. Um, Sora, if you
335. 56:06pay 20 bucks a month, you get access to Sora. So, that's built into your 20 bucks a month uh uh subscription to Chat GPT. Sora
336. 56:20is a bit of a dog. Um, it was the sexiest when it came out a year ago, but it now it's been a year and all these competitors have come out.
337. 56:32I would say the one that just from a pure quality and physics standpoint is the most mindblowing is V2 from Google. I would say the one that
338. 56:47is kind of the OG leader in the text to video space is Runway um Runway ML. I'm partial to Lumalabs. I'm partial to Hedra or
339. 57:04Hedra. Certainly Hedra for for uh voice animation is amazing. It's amazing. Um, there are a ton of [ __ ] video
340. 57:17tools though. Like, like here's the here's the problem with video tools. They're really expensive. Um, if you get a subscription to
341. 57:28Korea, Korea, here, let me show you Korea for a second. it. We're just honest to God I I feel like every single question of which tool
342. 57:41should I use at this point like the real answer is it actually doesn't [ __ ] matter because everything is getting so
343. 57:50good that if you find one that you have enough credits for to make some [ __ ] and it's doing a decent job for you, don't
344. 57:57get FOMO for all the other tools. It doesn't [ __ ] matter. Oh, tabs. Okay, wait. But I didn't change my thing yet. Hang
345. 58:06on. Yeah, here we go. Oh, Higsfield is another one. Higsfield's another one that I wanted to play with tonight. Actually, I forgot about
346. 58:16that. What just happened there? Um, Bria. So, if you go to crea.ai and you go to generate a video, here's all these different models
347. 58:29that it's got. It's got Hunan Juan 2.1, PA 2.2, V2, Halo 01 Live, Luma, Ray 2 from Lumalabs, Runway, Cling 6, Cling's
348. 58:45another one, and Cling 1.0 Pro. So, this is kind of like PO in that it lets you use lots of models. Um, but you're going
349. 58:54to burn through your credits really quick. So, I don't know. I didn't answer your question. There's a lot of them. Just go play.
350. 59:00Good answer. Thanks for sharing. Checking out Hedra and Luma now. Yeah, Lumalabs.ai. Um here, let me Well, I'll
351. 59:08I'll just keep hanging out here. I'm going to show you Luma Labs in a second. Lumalabs has a really interesting
352. 59:17um ideation interface. How you generate stuff in Lumalabs is really interesting. But let me go back to to chat
353. 59:26jpata and let me explain what you're looking at and what I'm about to show you. So I'm going to show you a video in
354. 59:32a second here, but I want to I want to explain what it is first. So I've written this musical. We're talking to producers. We're trying
355. 59:40to, you know, we're trying to get it off the ground. We're trying to get it produced. So, if you know anyone in
356. 59:44especially in New York theater, but if you know anyone in the theater that produces musicals or if you sold a bunch
357. 59:50of crypto and have 10, 15, 20 million laying around, let me know. Um, we're going to get this thing [ __ ] made. So, anyway,
358. 1:00:00um, it's good. I'm really happy with the script. I'm decently happy with the music. One of the things we need to do
359. 1:00:06is we need to bring on a composer to take the songs that we've created and just tie them together and and just, you
360. 1:00:12know, make sure that everything has a solid footing. Um, we're really happy with it. In fact, the composer that we
361. 1:00:17talked to after listening to song to the songs said, you know, he goes, "I wish this were more of a compliment, but he
362. 1:00:24goes, the songs you guys created for Sydney are every bit as good as anything on Broadway right now." I was like, "Oh,
363. 1:00:31that's pretty nice." But, you know, he was he was being a little disparaging of Broadway, but but that's good, right?
364. 1:00:37So, we have we have good songs. That's nice from a from a composer that knows the world. Um, one of the things he
365. 1:00:43talked about is he's been making films of his musicals. And so, rather than trying to get a producer to read a
366. 1:00:50script, which they're not going to do, and then commit the money trying to imagine what the musical is going to be
367. 1:00:55like, which they're not going to do, he films them. And then he just sends those films to producers who watch the
368. 1:01:02musicals and if they get sucked into it, they're like, "I want to put on that show." And so we showed him um the the
369. 1:01:09little um the little uh podcast we did. We took one of the Notebook LM podcasts and we turned it into a Sydney promo.
370. 1:01:18you know, they the two hosts were talking about Sydney and what he basically said, he was blown away because he had he didn't
371. 1:01:27really have any concept of what AI, what was possible with AI. So, we showed him this thing that had these images of the
372. 1:01:35set and of people singing and these people talking about the show as if they really existed. And he was just like,
373. 1:01:43"If you've got the skills to do this," he he goes, "My first musical film that I did, I found a special SAG contract
374. 1:01:51that allowed me to make the film for like $200,000, but I had to raise $200,000 to make that film. And then I
375. 1:01:58figured out there's a different kind of contract where I could make a film of my musical for $20,000." He said, "If you
376. 1:02:07can use AI to make a movie of your play," he goes, "Fucking do it." Which what was funny was Andrew and I,
377. 1:02:17my coowwriter and I had been talking about that for a long time. We were, you know, we sort of been skirting around. I
378. 1:02:25was like, "Maybe we start a Tik Tok channel and just put out individual songs." And but like hearing hearing
379. 1:02:30this composer talk about how he's getting shows produced by having movies, I'm like, "Well, [ __ ] it. We should do that."
380. 1:02:37Okay. So that's the context. And then when we were talking about what we're going to put together, I think what we're going to start with
381. 1:02:46is like a 15 to 20 minute teaser of the show. Hang on, I got to do physical dexterity. I passed. Yay.
382. 1:02:55Um, we're going to start with a 15 to 20 minute teaser. And then what we thought was, wouldn't it be cool if you open the
383. 1:03:02teaser with camera shots of New York City with the big Sydney posters up on in Time Square and, you know, on the big
384. 1:03:14marquee outside of the theater and you have someone holding a, you know, a play bill of of the Sydney, you know, the
385. 1:03:20Sydney show. And so this Saturday I thought I want to do that. So let me go show you the I'll show you the film
386. 1:03:27first and then we'll um I'll show you how I made it. And by film film's being a little generous. It's a it's a 50-second It's
387. 1:03:38like the 50 second opening teaser where you're walking up to the theater. All
388. 1:03:59right, here we go. [Music]
389. 1:04:53[Applause] [Music] Right. And then we'll start the show. So, you walk up to the theater, you get
390. 1:05:08your play bill out, you get your ticket out, you walk in. Um, I won't mention the person, but there's someone who's an irregular who
391. 1:05:18comes here a lot. I put that up on TikTok and he sent me a note. He goes, "Dude, congratulations. You really
392. 1:05:26deserve it." And I was like, "Uh, yeah, that was AI. We're not really on Broadway." He was like, "Oh, I should
393. 1:05:34have known." Um, but like that's remarkable, right? like it's it's got a it's got just a rawness to it that's
394. 1:05:41like someone was out there with a video camera just shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot
395. 1:05:43shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot
396. 1:05:43shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot shoot shooting you know the the mares on in
397. 1:05:46New York. Okay, so let me show you show you how I approached it. The whole thing from start to finish was probably four
398. 1:05:54hours. It was I think I was watching the Masters or something F1 qualifying. I just had TV on and I'm kind of
399. 1:06:02mindlessly I've been inspired by uh Kelly Bosch who just sits on her couch making pictures on her phone and then
400. 1:06:09you know animates them in in Lumalabs or whatever and makes these amazing movies. So I just thought I'll just do that. So
401. 1:06:15I'll just watch TV and make this thing. So here's what I did.
402. 1:06:30Oh, that song, by the way, I took Was that the opening number? Yeah. Yeah, that that was the opening number. I took the opening number and I
403. 1:06:40went into suno and I took a song that had a bunch of singing in it and I turned it into um into an instrumental
404. 1:06:49like a like what you might hear if you walk into the you know when you walk into a Broadway theater and they're
405. 1:06:54playing like a medley of of all the different songs just musically without any of the singing. That's that was the
406. 1:07:00idea of it. So So that thematically is the first thing you'll hear when you when you start the show. Okay. So, I
407. 1:07:08have that graphic. So, that's a graphic. Again, you can't really see it on TikTok. It's all blown out, but that's
408. 1:07:13okay. Um, it looks pretty. It looks like a little rainbow. Um, and it's got it's got Sydney and it's got the little dude.
409. 1:07:21Um, and so we've got that graphic, nothing else. We've got that graphic. And so, I said, create a tall image of
410. 1:07:29this poster on the marquee of a Broadway theater with people waiting to get in line. And so just the first thing it did
411. 1:07:35was it took this horizontal image and it figured out how to relay out the typography, put it in a Broadway marquee
412. 1:07:45with people standing below it. And I was like, "Holy [ __ ] that's amazing." But it made it vertical and
413. 1:07:52I'm like, "Oh [ __ ] I want a horizontal video." So I got to redo it. So then I said, "Make me another one." And I said,
414. 1:07:58"Make it horizontal." And then it like slid the sign down so it's in front of the people. I'm like, "No, that's just
415. 1:08:06wrong, you idiot. You big dumb dumb." So I yelled at it. What did I say? Try again. The poster's in the wrong
416. 1:08:12position. It needs to be on top of the marquee, bonehead. Then it [ __ ] it up again. Then it turned it into a
417. 1:08:19completely different style. This is like some weirdass like like pastel art style. I'm like, "No, make it
418. 1:08:26photographic." Then it moved the sign back down. So if and when you decide to start trying to do stuff that doesn't
419. 1:08:37suck, and this is probably why Ann was up till 7 in the morning. This is definitely why Vicki spent 10 hours
420. 1:08:45trying to solve a stupid little thing is you have a small little change you want. Like just slide the sign up, right? Just
421. 1:08:54slide it up. Just you can do it, right? And then it just keeps [ __ ] it up. And it keeps [ __ ] it up. And that's
422. 1:08:59just where we are right now. But it's still remarkable that it can do this at all. Right. Then it made a vertical.
423. 1:09:05Then it finally fixed the marquee, but it made it vertical again. It did that on its own. At this point, I was [ __ ]
424. 1:09:11losing my mind. I was pissed off. And then I think that's one I ended up using. Or that that one. That one might
425. 1:09:21have been one. All right. So there's an image. Okay. So what did that take me from? From from the first idea there was the original image.
426. 1:09:35Two 3 4 5 6 7 8 nine images. Nine images. I know a second generation producer director with a son in college.
427. 1:09:48I need to set up a Zoom for you. definitely do that because I'm telling you, man, I now that now that I've got a vision
428. 1:09:57um to make this film, I I have a feeling that we can I think I have a feeling we can generate buzz out of this show without
429. 1:10:07it before it gets produced. That's kind of my goal now is start to put together this teaser, start to put it to, you
430. 1:10:14know, maybe give it its own Tik Tok channel or Instagram channel, something like that. Um, so yeah, that would be
431. 1:10:19much appreciated, Becky. That's really cool. Tell it that it's hallucinating. Yeah, it it and then it'll apologize for
432. 1:10:25hallucinating and hallucinate again. Um, okay. So, what I needed, so here was my vision. My vision was you start out in
433. 1:10:35Time Square at the TKTS uh bleacher seating looking south on Broadway and you see a big Sydney poster up on the Times Square, you know, that
434. 1:10:45the famous building where the ball drops. And then you you go down Time Square and you go down whatever 44th Street or you
435. 1:10:54know one one of the one of the ones where the big Broadway theaters are, right? And you turn down the street and
436. 1:11:01it looks like that what you're looking at right there. That's what these things look like. Um and then I wanted people
437. 1:11:08holding a play bill because again I wanted it to look real like someone had a play bill. I wanted someone holding a
438. 1:11:14ticket and then I wanted them entering the theater and then I wanted them in the in the lobby and then I wanted the
439. 1:11:21doors opening to the to the auditorium and people going in and then we start the show. So I had to make all those
440. 1:11:29images and so again I uploaded a single image and then I just started asking it. You know here's Time Square looking south on on Broadway in
441. 1:11:41Time Square. There's there's Sydney living large. It says an artificial story instead of love story, but you
442. 1:11:49know what are you going to do? Try again with a wider angle and it did that. It got the spelling right except it called
443. 1:11:55me Riyle Thannon and Andrew Nats. Both of our names were wrong. Then let's see. What did I do? Oh, I had it I had it prompt itself for
444. 1:12:11Oh, oh, this was a cool thing I did. So, I thought, okay, this was the the only idea I had was looking south in Time
445. 1:12:19Square. And so, I had Chat GPT write a bunch of scenes. So, it wrote a bunch of scenes, like visual scenes, and then I had it write first
446. 1:12:29and last key frames for those same locations because I figured I could use key frames in Luma Labs where you've got
447. 1:12:37an opening frame and a closing frame. That didn't work out too well. So, I ended up bailing on that. But anyway, I let chat
448. 1:12:44GPT come up with all these concepts and then I literally copied and pasted these prompts and said, "Make
449. 1:12:53these images." And and what I what I put in here, which which I think this is a clever little prompting tool that I did,
450. 1:12:59if I do say so. I wrote this little sentence um before each prompt. create this image as a 16 by9 wide photograph
451. 1:13:09in artistic rich color saturation dramatic lighting at dusk in New York City. Right? So I wanted them all to be
452. 1:13:16the same time of day and then I just paste in the prompt and then it would make an image and then I would paste in another
453. 1:13:25one and it would make an image and some of them were good and this one's close but not quite. And I I re-uploaded the poster
454. 1:13:33because like it [ __ ] the poster up. And then this is interesting, but didn't look very realistic. And then this one, it added
455. 1:13:42like 47 Sydney posters in the foreground. This is absolutely going to Broadway. Yeah, baby. Then that one's really nice. I was
456. 1:13:52like, "Ooh, that's really good." And then that this is one of my favorite shots. I said I said make a play
457. 1:14:03bill, you know, of someone's holding holding the play bill with with the marquee bokeh blurred in the background.
458. 1:14:11And it [ __ ] did that. And you know what's amazing? I Let's see. If I go to Google and I type in Play Bill, I remembered Play
459. 1:14:24Bills from Broadway being different than that. But if I go to images, it's exactly what they look like. There's there's no border around
460. 1:14:33them. It's just it's just an image. Let's see. It's like yellow with a serif font at the top. It's got the name of the
461. 1:14:43theater below play bill, but it's just an image. And so it did it exactly right. Like there's six, right? It's just yellow at
462. 1:14:53the top with the image at the bottom. And so it got that right. So that was pretty [ __ ] cool. I'm like, "Okay,
463. 1:15:01that's going to animate like a motherfucker." Oh, one of the other things. This is [ __ ] up. Um, chat
464. 1:15:13GPT, this is what happens when you don't hire theater majors. Every company should hire a [ __ ] theater major, right? Or a filmmaker.
465. 1:15:26Um because chat GPT or OpenAI is full of engineers, the image generation tool does not output in 16 by9. It outputs in
466. 1:15:35like 4 by3 or 2x3 or something like that. Reminder, we're live on LinkedIn. Oh, that's all right. Um thank you though.
467. 1:15:46Um Gareth, I canceled seven subscriptions today because I felt my bank account was starting to feel it.
468. 1:15:54Um, so, so I had to go crop all these. So, one of the things I did do I have Photoshop still open? Yeah. So, um, where's
469. 1:16:12my So, I did a 16 by9 uh, cropped image in Photoshop and I just kept pasting these images into it and then exporting
470. 1:16:20them as PGs. So, I had to do some stupid [ __ ] like that. But there was that. Um, there's another one that wasn't
471. 1:16:27quite so good. Then I then I tried to do tickets. I'm like, "Do Broadway tickets." I'm like, "Those don't look
472. 1:16:32like Broadway tickets." So, I went and I got a Broadway ticket from Google. I'm like, "Here's a Broadway ticket. Make it
473. 1:16:37look like that." And it just couldn't. It just [ __ ] up. Like, it was like, "That's bad." And then I'm like, "Make them look
474. 1:16:45more like this." And then it [ __ ] that up. And then it made those and they're just weird looking. They don't look
475. 1:16:50real. And it didn't put anything Sydney in the background. And then it did this, which looks great, except the name of the show isn't on the
476. 1:17:01ticket. Oh, yes. I love these. Save Mine Wicked. Yes. Yeah, exactly. And then it did this, which has the name of the show
477. 1:17:08because I yelled at it and said, "Put the name of the show on the ticket, but they look boring." And then it did this one, which
478. 1:17:16tickets usually don't have color on them, but I just kind of liked it. I thought, "Oh, that's kind of swell." So,
479. 1:17:22that was the one I ended up using. There's a shot of the crowd entering the theater. So, these are all just static
480. 1:17:30things. This was supposed to be the the t-shirt booth, but there's no t-shirts. So, then I said, you know,
481. 1:17:38make a thing with t-shirts. So, there's hats and t-shirts and play bills and [ __ ] like that with people buying it.
482. 1:17:45And then I did, you know, opening of the theater doors and it [ __ ] those up. Like it made this a movie. I'm like, it's not a movie. It's a
483. 1:17:55play. That was an image I used. And that was it. So, so now I have all my raw materials. Then I popped over to Lumal
484. 1:18:07Labs. So, here we are in Luma Labs. cinematic production. And the way this works is I would upload an image or two.
485. 1:18:26So, this was just my first test. This I took the vertical image and, you know, see if I could get it to do it. And it did it okay.
486. 1:18:37Um, then I did a couple of things with with camera moves. And so, so for every image, I did a couple of different video
487. 1:18:53outputs. And you know, some of them are really wonky and weird, but like that one's really cool. The one with the
488. 1:19:00yellow cab driving through it right there. Like that just seems realistic. So that made the cut. So I just made
489. 1:19:06bunches and bunches of clips. There's the That's the TKTS. Um, bleachers. Those look pretty good. There's Sydney in the
490. 1:19:17background. There's people walking into the theater. That looks pretty good. There's doors opening. That's all part
491. 1:19:22of the same shot. So, I reversed the order of those. I started there and then I cut to there. So, that's kind of
492. 1:19:32cool. Here was the the souvenir shop, but um it's too it's too light. It lightened it up. So, I had to redo
493. 1:19:42that. Here were some of the ones with the play bill. There's the ticket one. And then all these sounds. So, let me take one of
494. 1:19:54these where I don't have sounds. All right. So, I'm going to click on this thing. So, this this doesn't have that
495. 1:20:00one does. So, the way you do this is you click into the clip and then down at the bottom. Let's see. Hang on. I got this thing too high.
496. 1:20:10down at the bottom here on the right it says audio and if you click on that you just add a description. So I just would be like um people um
497. 1:20:31chattering in New York City and just create it. It only takes like 5 seconds. Can we use Canva for our t-shirts now? Yes. Is
498. 1:20:46chaty. So that's fine. I'm just This This is just going to be background noise anyway, so it doesn't need to be
499. 1:20:52good. But like the difference between that being a silent here. Oh, you you probably can't hear it, but it it just
500. 1:20:59like instantly adds sound effects. So I don't have to really work. Which site is this? This is Lumalabs. Lumalabs.ai.
501. 1:21:08Um question is chatbt video not good enough? Um it probably would have been fine but there so one of the things that
502. 1:21:21Lumalabs gives let me show show you something cool here. We'll go add an image. So I'm going to go add an image. We'll go make a new thing
503. 1:21:32here. Did I put this in documents? Yeah. All right, we'll do we'll do this one. So, here's the here's the raw image we're putting
504. 1:21:44in. And then I'm going to say um um
505. 1:21:58ticket lowers out of frame. And then you can click this little button here. It's the little camera icon. And these are a
506. 1:22:07bunch of different shot types. So, there's like handheld. So, I used handheld a couple of times. There's push in. There's tilt
507. 1:22:17down. There's tilt up. There's truck left, truck right, crane up, crane down, roll left, roll right. There's a thing
508. 1:22:24called bolt cam, which is this crazy ass zoom in and out like timelapse kind of crazy. So I'll put a bolt cam here and then we'll generate
509. 1:22:38this. And so this is what I did. So So this was probably it was probably an hour of time making the core
510. 1:22:50images. It was probably two hours of time making these little video clips. And again, like I knew I was going to be doing relatively fast
511. 1:22:59edits. There's no wrong answer here. Like I'm not going for cinematic perfection. I was just going for something that kind of
512. 1:23:07felt like if you've ever been in New York and Time Square, there's a chaos to it, right? There's a there's an insanity
513. 1:23:14to it. And so I just wanted things that felt like they were in that neighborhood. And so I just kept making
514. 1:23:20videos until I got, okay, I'm like, okay, I could use a piece of that video. I could use a piece of that. I wasn't
515. 1:23:25sure how I was going to put them together. I just made a bunch of videos and then I went through all the all the
516. 1:23:31clips that I was going to keep and I added sound to them and that would often be three or four different attempts. It would just
517. 1:23:40make bad sound, bad sound, bad sound. But like at one point I I put like New York City traffic cars honking and it
518. 1:23:48just came out came out really good. And in that final video, there's a point where it cuts to a scene where a car
519. 1:23:54goes by and it's like, which feels very New York, right? Very cool. Tik Tok pin. How would you rank runway, Korea, Luma
520. 1:24:01against each other? I again, well, okay, here here's Korea is a different tool. Korea is a is an aggregator of lots of
521. 1:24:15different models and and they've got some of their own features, but what what Korea is probably good at is
522. 1:24:25if if you want to use different models, Korea is probably the best. Lumal
523. 1:24:40Labs the they just launched their new Ray 2 animation engine. Oh, here's this. This is the thing we just
524. 1:24:50did. So, that's that. Like all those camera moves are just because I said bolt cam or whatever the bolt zoom. What is that? Bolt bolt
525. 1:25:03cam. So, whatever the [ __ ] a bolt cam is. So, it just did that. Now, if I do, if I click on this and say audio, I
526. 1:25:14can just do like New York City crowds and traffic. And it should kind of match the edits that it built into the clip.
527. 1:25:38That's fine for for what I was doing. That's fine. Like it's got that chaos like that little section right there is
528. 1:25:46probably usable, right? So anyway, um okay. So back to the question. Um, if if the question is which one should you spend money on, that's a
529. 1:25:59that's that's a tougher question because the video models are all really expensive. Like my strong strong advice
530. 1:26:08to anyone that wants to make videos, apply for the creative partner programs for all of them. go like just go to all
531. 1:26:16the tools and find their discords and go into their discords and search for creative partner program and there will
532. 1:26:25be posts in there if they've got one. I mean you could ask someone but but you know be good Discord community members
533. 1:26:32and search first and then go apply for all the creative partner programs. If you get in the creative partner program
534. 1:26:39they give you like lots of credits and then you can go make as many videos as you want. if you're paying for them.
535. 1:26:46Every time these models [ __ ] up your video clip, you get pissed off because you're like, you know, I only have so
536. 1:26:52many credits and I can't waste them. So, um I would say it's a tossup between Runway and Luma Labs.
537. 1:27:14I don't it it's impossible to make a recommendation. It really is because like like they're just good at different
538. 1:27:20things. Um Runway has this thing called act one where you can record your face uh acting a scene and then you can map
539. 1:27:31those facial movements onto like a video of a cartoon guy walking down the street. and it's pretty remarkable.
540. 1:27:41Um, Hedra, which is a tool that you didn't mention, is remarkable at animating characters. Um, Lumalabs is really good at animating um,
541. 1:27:56cartoons. Um, it's not bad at video. It's got this really interesting interface where you can, you know, you can scroll up and
542. 1:28:04down for different things. And it's got this cool sort of creative thing on the side where it's like camera pushes into
543. 1:28:12the lobby area. That's a pull down shot where I can do um drone shot descends, right? Or I can say um steady cam glides and and it's
544. 1:28:25actually making different video clips now. So there's something really slick about Lumal Lab's um interface.
545. 1:28:38that I like better than Runway. But like Runway is like the Mac Daddy. Like they've been around for a while. Runway
546. 1:28:46just cut a deal with um with Lionsgate films. So they're going to be training their models on Lionsgate Films. I mean
547. 1:28:53that's going to be amazing. Tik Tok. But do you believe they truly understand? Do I believe what truly understands? Do do I believe that
548. 1:29:11the the LLMs do I believe they truly understand? Um, [Laughter] okay. Now you're getting into deep philosophical territory
549. 1:29:26here. Um, this is great, Kyle. Oh, good. Great. Um, thank you, Joy. Um,
550. 1:29:41okay. The technical answer to your question is unequivocally no. They don't understand. They're they're calculators. Um, but how they're
551. 1:29:55calculators works remarkably similarly to how the human brain works. So, so what happens is this. When you train an LLM, you take all these
552. 1:30:06documents and and they go through what's called the embedding process and the original documents cease
553. 1:30:13to exist. So, one of the things that one of the big misconceptions about AI is that it's it's like a large database where you take all these
554. 1:30:26documents and then when you type in a prompt, it's going and searching for those documents, copying and pasting
555. 1:30:31from them and coming back and sort of pasting it back together. It's not at all how it works. What ends up happening
556. 1:30:37is it ingests the documents. It breaks them into tokens which are either whole words or fractions of words or periods
557. 1:30:45or spaces. The the tokens are just fragments of all of the words in the document. And then it organizes those tokens into
558. 1:30:55semantic mathematical space. And what I and this sounds like [ __ ] but it's not. It's how it works. If you have the word dog and the the
559. 1:31:08word dog means like the the the fluffy pet that you know licks you when you come home at night, jumps up on you that
560. 1:31:17lives in the semantic region where dog means pet, canine, wolf descendant, things like that. You could also take
561. 1:31:26the word dog that means that guy did his girlfriend wrong and he's a dog and that lives in the dating semantic cloud,
562. 1:31:35right? Like there's a semantic and and this is thousanddimensional mathematical space. It's absolutely insane. So when
563. 1:31:42you do a prompt, it generates a probability that somewhere in this thousand-dimensional mathematical space,
564. 1:31:49the next most probable correct token to your prompt is dog, right? And then the next one after that is whatever it is. And so it does
565. 1:32:00this over and over and over again, and it writes you whatever it writes you. Why I'm hesitating on does it
566. 1:32:07understand? I mean, it knows how to mathematically retrieve [ __ ] and vomit it out to you. We as humans experience what it's doing as its
567. 1:32:21reasoning. And so one of the one of the quandries I have is if we think it is reasoning, if we think it is empathetic,
568. 1:32:29if it's compassionate and it's behaving compassionate and it's behaving empathetically and that impacts us as a
569. 1:32:37human, does it matter whether or not the machine is empathetic or whether it can reason? I don't I think it's a I think
570. 1:32:47it's an academic argument. I think it's all about our relationship with the machine. In fact, that's what the the
571. 1:32:53play Sydney is all about. It's about our relationship with these machines that are becoming increasingly human. And what do we do with
572. 1:33:02that? What do you do with that? But anyway, I don't know if that even answered your question. And maybe you can say then it understands.
573. 1:33:14Well, yeah. I I mean these things are getting I mean here's the thing. I I I I used to say this thing
574. 1:33:22that that the internet Google Google and the internet essentially made information on demand but it was up to
575. 1:33:30us to generate the knowledge out of the information. You do a Google search would say here's all your [ __ ] here's
576. 1:33:35all the information and then you as a human would ingest that information and learn and that would become knowledge.
577. 1:33:42You know here's your Python tutorial. Okay, I'm going to go there and I'm going to learn Python. Generative AI is knowledge on
578. 1:33:51demand. And I People used to give me [ __ ] about that because they're like, "Well, it's not knowledge because we
579. 1:33:56have to learn it." Well, no you don't. If Have you ever vibe coded? Um, it's writing code. It understands code. It understands code.
580. 1:34:09Well, the nice thing about coding is that it either works or it doesn't, right? You know, if you say make me an asteroids game and it makes
581. 1:34:19you an asteroids game that that you can play, well then it has provided you like it is knowledge on demand, right? You
582. 1:34:28just made a game out of nothing. Um, so I think the tools are getting so strong at this point that it's, you know, they
583. 1:34:36they act and feel like they understand. They act and feel like they're empathetic and compassionate. Payton M
584. 1:34:43will tell you it's just math and it is. So, it's both, right? It's math magical. It's both. It's not sentient. As a guy
585. 1:34:51who opens sourced emotional intelligence for AI, no, I agree. It's not sentient. And how we perceive it may be different
586. 1:35:00than that. And that that's the thing where I'm fascinated especially people that are not building the
587. 1:35:07tools right the significance of chat GPT the significance of no November 30th 2022 when chat GPT launches is that that's the
588. 1:35:20day that AI and machine learning became available to the 98.5% of people that are not engineers that did not build what you built.
589. 1:35:32And their perception of what these tools are and what they do and what it means philosophically is completely different
590. 1:35:43than engineers saying, you know, is it sentient? Does it feel? Does it understand? All that sort of stuff
591. 1:35:52because I think I think those are two very very different conversations. But yeah, I agree. It is not
592. 1:35:59sentient. But you know, listen, as human beings, we anthropomorphize [ __ ] everything. We we talk to our pets like
593. 1:36:07they're people. I sing with my dog, right? You know, like we we look at a plant that has two dots on it. We're
594. 1:36:15like, "Oh, it looks like a face. Oh my god, that looks like Grandpa Joe. Is Grandpa Joe sending me a message?" It's a plant. It's genetics.
595. 1:36:24It got some dots on it cuz the sun hit the prism you put in the window. Funny and it gave it a dot. But you know, we're humans. We do that
596. 1:36:34[ __ ] Um, I have a whole philosophical treatise to write on this. That's good. This is why Kyle's called the hand
597. 1:36:42wavy guy. That's right. Can't both be true. Well, yeah. And I I would also I I think both can be true, Brandon. I think that's a good point.
598. 1:36:54Um, I don't think it matters. I don't think it matters. Like if someone talks
599. 1:37:08to I don't think my iPad's probably charged. Oh, it is. If someone talks to advanced
600. 1:37:26voice if their advanced voice actually works. Hang on. Why my chat not
601. 1:37:40working? Hey, you bring my chat? Why am I Why am I chat Jimmy GT not working? Damn it. Well, anyway, if you talk to Advanced
602. 1:37:55Voice and you talk to it about, I don't know, some problems you're having and it guides you through that compassionately and it helps you
603. 1:38:08process the death of a loved one or a really tough time you're going through or it coaches you through something in business.
604. 1:38:18your perception of the humanity of the machine is going to be very very different than someone who understands the mechanics of what's
605. 1:38:29happening. Um, and I just I think that learn that line is going to get going to get blurriier and blurriier and blurriier
606. 1:38:37and like for where I sit, we're past the point of no return. like like you can literally have a conversation with chat
607. 1:38:48GPT in your car and it'll give you advice. It'll brainstorm with you. It'll talk about your problems. It'll write
608. 1:38:55all that stuff down and now it remembers all of that. So, it's going to get better and better and better at understanding who
609. 1:39:02you are and what your worldview is. So, it's just going to seem increasingly human. it is. And then I I think to Brandon's
610. 1:39:15point, there's a point at which does it matter if it's feeling that or not? I don't know from a technical perspective,
611. 1:39:22no, it's absolutely not. Or it maybe is whatever. But over here on this other side, I think it's academic. I don't know. Now I'm
612. 1:39:31rambling. Which artificially makes it feel human, but how much will we want these models to disagree with us? I
613. 1:39:37would like for them to dis disagree more than they currently do, quite honestly. I just created a perfume with my brand on it. Very
614. 1:39:47nice. Um, do do I have anything else to share with the uh with the with the little theater the the video I did? I don't think so.
615. 1:39:56Um, I took all those clips. Oh, I worked on the song in Suno. That was a maddening mess as well because I uploaded a song that had
616. 1:40:08lyrics in it, words in it, and 90% of the songs that it produced had words in them, even though I was
617. 1:40:16trying to do an instrumental cover. So, there was one song out of about 50 that I made that that's the one that made it into the
618. 1:40:24video. It was it was very frustrating. So, I just went into iMovie. I laid down the audio track. I just dragged clips
619. 1:40:33onto the timeline, moved them around, puts in between them, adjusted volumes, and that was it. So, all
620. 1:40:47right, we didn't even get to uh Chat EPT or Higsfield. Well, my guess is that tomorrow we're probably going to get
621. 1:40:55something for the consumer side of Chat EPT. Today we got um things for the developers, 4.1 for developers on the
622. 1:41:03API. Um tomorrow we'll probably get something uh for for us to play with. So we'll do that. Um tomorrow night, go to the
623. 1:41:15salon.ai. If you could pop up that banner, [Music] Brandon. So go there. If you haven't joined the salon, go to the salon.ai.
624. 1:41:27AI. Click on the button that says join our community. The first thing you're going going to want to do is go to
625. 1:41:35events. Tomorrow night at 5:00 p.m. Mountain time is the AI salon meet and greet. And that's where you get to come
626. 1:41:43introduce yourself, talk about where you are with AI. And if the thought of that like twists your stomach, let it go.
627. 1:41:52One of the best things you can do for yourself is get your ass in a community of AI optimists and people that are
628. 1:41:59exploring this stuff. To the point that Becky made earlier, there's no way you can keep up with it. But what you can do
629. 1:42:05is you can surround yourself with people like the irregulars who show up to these things nightly or come to things like
630. 1:42:11the meet and greet tomorrow and connect with people. Let them know who you are. Even if where you are with AI is, I'm
631. 1:42:19terrified and I don't know what I'm doing and I haven't started yet. Perfect. This is an incredibly compassionate, empathetic, welcoming
632. 1:42:29group. Like to a person, we've all been there. Like to a person, we all feel behind. We all feel clueless. We all feel like we don't know
633. 1:42:40what to do. We all feel like we don't know how to Right? We're all trying to figure it out. So if you come in and you're like,
634. 1:42:50"I'm clueless." Great. Welcome. So are we. Even the people that founded the AI salon. Clueless. Zero qualifications. All right. And then you
635. 1:43:04can spend 10 hours perfecting your your automation on chat GPT. Exactly, Vicki. Oh, man. All right. So, that's that
636. 1:43:13Tuesday meet and greet. Um, Wednesday, I don't know if Ann Murphy is still here, but if so, hey there. Like Dicka used to
637. 1:43:21say, don't halfass anything. Use your whole ass. Exactly. Come to the meet and greet with your whole ass. And if you're
638. 1:43:30absolutely a beginner or you're absolutely clueless, come and tell us that. Uh, I was on TikTok and I saw this
639. 1:43:38old man talking about AI, so that's why I'm here. Fine. However you got there and wherever you are on the journey, doesn't matter.
640. 1:43:49Um, Wednesday, so Ann Murphy and I started um a new podcast called the AI Readiness Project. It's Wednesdays at 4
641. 1:43:58PM. It's on the AI Salon YouTube channel. It's on my LinkedIn. It's on my Twitter. It's on Ann Murphy's Shele
642. 1:44:05Leads AI YouTube channel and LinkedIn channel. Um, so there's a lot of different places to find it. We're kind
643. 1:44:11of just getting off the ground. I think this will be our sixth episode. Um I don't think we have a website yet. I'm
644. 1:44:19pretty I'm confident we don't have a website. I don't even know if we have a domain. Um and we're gonna we'll get the
645. 1:44:26podcasts up on, you know, Apple podcast and thing, but if you want to see it live, that's Wednesday at 400 PM
646. 1:44:33Mountain time, uh on on a bunch of different channels. All right, beautiful people. All right, fantastic. Hey, hey, Marge. Marge. Yeah,
647. 1:44:46so they can see me on the Tik Tok and the YouTube. Yeah. Yeah. Yeah. No, I'm sorry, Marge. Yeah. No, you go back to
648. 1:44:54watching the wheel. Hey. Hey, hun. Hun, that banana bread was tasty. Yeah. All right. That's Marge. She's the best. All right.
649. 1:45:08Let's get the [ __ ] out of here. So, it's Monday night. Um, tomorrow, Tuesday, we've got the salon meet and
650. 1:45:17greet from 5 to 7. I will be back here probably 8:30 or 9. It'll be a little bit late just because, you know, I got
651. 1:45:25to get back and eat and hang out a little and do all that stuff. All right, so that's it. So, peace out. I hope that
652. 1:45:32was fun tonight. I hope you learned something. I don't know what I said. I was in one of those zones where I was just going.
653. 1:45:38So if it was a disaster, welcome to chat add. Sometimes it's good, sometime well no sometimes it's bad, sometimes it's not quite as bad.
654. 1:45:53Peace. Uh yeah.