4/18/2025 - Reasoning Models and Multimodality in ChatGPT o3

Transcript

1. 0:02[Music] [Music] [Applause] You and I get all alone. Sunday morning here at
2. 0:19home. Blue and the car is strong. It's true. Then I open my eyes to a dream real life in front of
3. 0:34me. And I haven't got a clue what in the world is happening to me. Think I think I'm happy like first day summer vacation.
4. 0:47Happy got to get a little rest, relaxation. Happy choir on Sunday morning singing true. [Music]
5. 1:29Every time I see you, I get that look in mine. Every time I see your mouth, I hear that smile. The early misty morning light. I
6. 1:46heard the engine turn the old foot [Music] outside. You were leaving
7. 2:01me again today. You will convince me again today. You're leaving this heart town looking for someone else's golden
8. 2:20ring. Should say so long, Suzan. Now don't you [Music]
9. 2:36cry. So long, Suzanna. Don't you cry for [Music]
10. 2:56me. Sharing drinks and cigarettes and keeping warm out on the road. Chasing down a lifestyle out on Highway 24.
11. 3:08New York State was a rolling breeze and the sunshine with a blue sky falling to the chill of all September [Music] creeping. You believe in
12. 3:24me again today. You convince me again today. You're leaving this hotel looking for someone else's golden
13. 3:41ring. Should have say so long, [Music] Suzan. Hush now. Don't you cry. So long, Susanna.
14. 3:58[Music] Don't you cry for me. [Music]
15. 4:14Thank you so much, Winston. I saw the galaxy you sent. That was beautiful. Generous. You're amazing. Thank you so
16. 4:20much. Welcome everybody. It's Friday night date night. You know what we do here. We cook nachos. We microwave hot
17. 4:28pockets. We have fun. We learn. We learn together. Tonight we're gonna explore stuff. Posted my poster about five minutes after [Music]
18. 4:46you. [Music] Um, one thing I'm excited about. Am I excited about it?
19. 5:00one thing that I I feel validated by. So, so this new 03 model from Chat GPT and the O4 mini
20. 5:14things are just a quantitative and qualitatively different tool than we've ever had before. And while I don't know, I don't yet personally
21. 5:31know how it's a big deal, I just have this deep deep deep instinct that it's a really big [ __ ] deal. And like building from the launch
22. 5:42through like right now, like right before I came on, my Twitter timeline is just people I respect up and down the
23. 5:50board from the sort of math and science side, from the AI, you know, influencer side to just regular people like the
24. 6:00like everyone is posting that this is a big [ __ ] deal. Um, so that's that's largely what I want to talk about tonight. [Music]
25. 6:22Um, Kyle, please go look in your regulars at Where in the World is Kyle Shannon. I can't pass that one up.
26. 6:34All right, AI learning lab, here we come. Yes. Um, no, I was just going to tell you on the uh where in the world is uh
27. 6:43Kyle Shannon, um the post is good in of itself, but the real um specialty lies in the chat link that Danielle shared.
28. 6:54And it's it's link it's if you expand out her text the chat GPT story or chat that she had is in her um message. So if
29. 7:06you scroll up to click on this and then nope scrolling up just a little bit because I was just looking at this. Pretty cool.
30. 7:16So expand that. Yeah. And then open. So, what basically what she did was she reversed image searched a screen grab of
31. 7:22your live, but you have to go in, scroll up to the top, and look at the process that 03 took to find you. And then she's
32. 7:30wondering if you know if it got it right or Oh, cool. Okay. Upload an image. This is a copy of a I can't see which image it is, but
33. 7:43that's all right. [Music] Um, so uploaded some more images. So I don't know what they are. Danielle, tell
34. 7:50me which images you did. Uh, she grabbed a a screen grab of where you were saying go play now outside of a building. Oh,
35. 7:57okay. Okay. Yeah. Yeah. Yeah. It should be able to find that one. All right. Let's see what it does here. I mean, I love all the Okay. So,
36. 8:08if if you're new to Wait, let me flip this. Let me put my black bar up. We're jumping right in here. This is cool, Danielle.
37. 8:20Okay. So if you're new to what what makes 03 different, why it's unique and interesting is that if you know a little bit about
38. 8:35normal large language models and reasoning models, the the the basic difference between them is a normal
39. 8:41large language model like GPT40, the ba the basic model at the top of the list is that you ask it something and it
40. 8:49gives you an answer, Right? A reasoning model, you ask it something and then it has a conversation with itself. It calls
41. 8:57it thinking as Pate M rightfully points out. It's not actually thinking, but it behaves like it's thinking, right? It's
42. 9:03it's sort of prompting and responding responding behind the scenes before it gives you an answer. 03 amplifies that. It can do much longer
43. 9:17chain of thought conversations. It's also multimodal. So it can incorporate images and do image analysis
44. 9:26as well as um as well as text, you know, large language model stuff. Can do both. And it can do tools. It can write
45. 9:36and execute its own Python code. So So if what she asked was, you know, tell me where this thing is. if it needs to
46. 9:45analyze a photo to try to find things out, um, it will write its own code to do that analysis as part of its internal
47. 9:56conversation with itself. So that's some of the stuff that we're seeing here. So it thought for 43 seconds, which in 03
48. 10:03is a pretty long time because it's a very, very fast model. And this is the epitome of we'll do it live. go ahead and
49. 10:11uh refresh your screen or uncheck the box for all HO details because you're seeing something very different than
50. 10:19what I'm seeing and what I'm seeing on my end is um probably more relevant uh to how the model operates. Wait, I'm
51. 10:28just not sure why it didn't load on your end. And if not, I can share screen. Oh yeah, now it's got pictures in it. There
52. 10:34we go. Okay, now now we're seeing it. So she didn't upload those like screen grabs. That's the model like zooming in
53. 10:43and analyzing in real time. So it like that's the original screen share and then as you scroll down it's like
54. 10:49pinching in on your like your cheek and building behind you. Great. That's the model doing that. That's not Yeah.
55. 10:57Danielle giving them the information. Yeah. Exactly. So that's perfect. So you know her prompt. Can you tell me exactly
56. 11:05where this location is? Right. And then and then it starts like if you have another frame that would help. And then
57. 11:11it looks like maybe she uploaded some stuff. But now it's importing the image and you know uh looking at it. So
58. 11:19there's the first it looked at the whole image first. Now it chopped it up. Looked at that part of it. Now it
59. 11:25chopped it up. Looked at that part of it. That's actually um this is a brewery right across the street from the place I
60. 11:32work. That's the place I work. And these are flipped for some reason, I guess, because it was flipped on on TikTok, so they're backwards.
61. 11:40[Music] Um, it's looking at my little fuzzy sweater. It's looking at my my dramatically receding
62. 11:54hairline. There's some fat neck. It's like, dude, dude, stop eating the burgers. Oh. Oh my. Oh my. So, what will be really impressive is if
63. 12:09it not only tells you where you are, but that you should go to a gym. Exactly. Exactly. Here's our recommendation.
64. 12:15Here's here's where Kyle works, and here's a gym within walking distance you might want to consider. But okay, not for
65. 12:29nothing, here's here's what what is different about this. You could before with code interpreter, you could upload a photo
66. 12:40and you could say to chat GPT, zoom in on part of this image and analyze what you see in it, right? And it would write
67. 12:47Python code and it would zoom in on part of an image and it would tell you what it saw. This is doing it by itself, you
68. 12:55know, as part of the thinking process. So these are called tools. Every time it writes one of these, it's basically
69. 13:01creating a tool for itself to use in the process of of trying to solve the, you know, answer the question. All right, so
70. 13:09let's keep going. Let's quickly get past my mouth, shall [Laughter] we? All right, let's see. Uh oh, that's cool. Now it flipped
71. 13:23it. It It inverted the colors or something. Very cool. It's doing good old and and you know, if you think about
72. 13:34this organizations like the CI CIA, right, if they're doing, you know, like satellite photo analysis or they've got
73. 13:42a picture of someone, you know, now we've got AIS that just, you know, you take a picture and it tells you exactly
74. 13:46where it is. But we're watching this thing sort of figure it out on its own. Um, I really liked your mouth. That's pretty funny. Can Can you
75. 14:02tell me the last time this guy went to the dentist? Yeah, it's doing negatives. Look at that. Oh, what's also cool
76. 14:15here, this took a while. Look this one. It said thought for 5 minutes and 35 seconds. Wild. Oh, look. It found this
77. 14:29is a golf place in my building. Chat with vision walked through my garden today. Oh, cool. Suggested plantings for now and for
78. 14:40light space. That's very cool. Chat GBT has a crush on you. Okay, wait. Here we go. I took another deep dive. New frame, zoom, sharpen,
79. 14:52projected, sign, Google induct storefront images. The best probable match I can find for the black squared mountain and lines logo is A-ine
80. 15:03Boutique's hanging blade sign in Denver's Loto micro district district called Dairy Block. That's not right,
81. 15:10but it's close. It's really [ __ ] close. like it's this is probably a you know 3 minute drive away. What would lock it down? One
82. 15:21steady high resolution still of the sign. Yeah. Any cross street or awning? There's a there's a tool out
83. 15:29there right now that that does this. You can take a shot like the one you put up there and it figures out what the
84. 15:34building is and then goes goes and finds its location. But this got I mean not for nothing this is remarkably
85. 15:43close. Like I wonder why like how it got to Denver. Is there anything in the original image that that says where I am? No. Huh.
86. 15:59Fascinating. No idea. That's very very cool. All right. So So there's a there's a use case. So that's that's what we're going to talk about
87. 16:08tonight. And this is a good example of, you know, so imagine a similar sort of thing where you take some crazy neo
88. 16:18gothic painting, you know, one of those big giant 8 foot wide paintings that's got all sorts of crazy, you know,
89. 16:27imagery and symbology in it. And you upload that and you tell this thing, tell, you know, find everything you can
90. 16:34possibly find in this painting and catalog it for me and research the history of everything you find and, you
91. 16:40know, tell me all of the allegorories and all the metaphors and all the all the symbols and all that sort of stuff.
92. 16:46Danielle um should should ask if you have any cavities. Yeah, exactly. Um, I feel Vicki I feel like it would
93. 16:56just be quicker to ask Kyle where he was. Always coming in with the logic. It's a good It's a good shot, Vicki. It's It's a It's a
94. 17:14winner. But that's pretty awesome. Um, very very cool. Okay. Um, all right.
95. 17:29So, so let me So, I figure here's what we'll do tonight. Um, first of all, how is everybody? Possibly Jake Vicki, but this is Chad add.
96. 17:46Well, you know, and if you ask me where I am, you know, my response is going to be like, "None of your damn business. Shut up, you nosy
97. 17:54person." You know, so that would be awkward. So, have chat GBT do it. [Music]
98. 18:20Does everyone in Denver have that shirt? No, nobody has that shirt. I'm fancy. What's champ? What's up? What's up,
99. 18:28dude? Tik Tok pin. All right, we saw that one. Beautiful. [Music]
100. 18:43Um, reinvigorated post Friday Kyle meeting. Nice, Gwen. Fantastic. Yeah, there there was uh the AI office hours today. If you weren't at
101. 18:55it, we went through some things. So, here's here's the good news. If you are at office hours today and if you're here tonight andor andor
102. 19:07um I'm gonna give you what I think is a really cool powerful um unlock for for o for testing 03 for discovering what it makes possible. I
103. 19:22think what Danielle just did there, like like that use case, find out where this is. That's one way you could use that.
104. 19:29But basically, upload an image and have it do detailed analysis of that image. There's all sorts of things you could do
105. 19:35with that. Like that's a great that's a great use case to explore 03, but you could also do it with 04 mini high and
106. 19:42see if they're different. Um, I said this today on my content evolution call. I said this today on my office hours and I'll say it again
107. 19:55tonight. Um, I think it is probably in everyone's interest if you're interested in AI, if you want to start to understand what is
108. 20:05now possible that wasn't possible 3 days ago, spend at least an hour a day really pushing 03, 03, 04 mini high, 04 mini,
109. 20:20just those three reasoning models that are the new ones that came out this week. Um, do everything you can to understand
110. 20:29what they are, how they work, what they're good at, what they're not good at, what they're good at for you. That's
111. 20:36the most important thing. What they're good at for you. All right. So, what I'm going to show you, I I it always baffles me how long [ __ ]
112. 20:46takes, but I I put together this document today of all these prompts, and I'm going to give you these prompts. I
113. 20:51just posted it on my ex account as an article. Um, but and I'll put it on the the AI salon uh as well and I'm going to also
114. 21:03put it on LinkedIn because because I want people to explore this. My request is as you learn stuff about 01 just like
115. 21:11Danielle did, share it in the AI salon. You can share it in the regulars channel in the in the AI learning lab channel.
116. 21:18You can share it in, you know, the general talk stuff at the top of the list. Um, you can share it in here's
117. 21:25something I made or here's something I discovered. Um, just start sharing what you're learning because there going to be so many
118. 21:35different people are going to unlock so many different ways this is powerful that only if we kind of share what we're
119. 21:43learning will will it be good. What are the usage limits for the new models? Uh, I don't know, but I do know that you um I
120. 21:53saw Where did I see that? Oh, no. Maybe that was in Was it here? Hang on. Is it in
121. 22:10here? I thought I saw somewhere it said how many things you have left, but I don't I'm not seeing it now. So, I don't know. All right.
122. 22:20Um, if you haven't played with deep research, that's another one to play with. Just click that button and see what happens.
123. 22:29Um, and just just like I think it's it's hard to it's hard to
124. 22:44overstate how much of a reset this is in some way. And here here's what I mean by that. Yes, it's in chat
125. 22:52GPT. Yes, it's a reasoning model. Yes, it's just the sort of next one from 01, right? We had 01 and 01 pro and now
126. 23:00we've got 03. O2 was a trademark thing, so it doesn't exist in case in case you're like, wait, what happened to O2?
127. 23:08I learned how to count in kindergarten, and that's not how it goes. It that's how it goes with Open AI. Okay, so on the surface, it looks
128. 23:18exactly the same. It's not. It's a reasoning model with tools. It can think for itself. It can analyze. It can do
129. 23:30mathematics in the process of analyzing something. I mean, it would be fascinating and you could do it. You could take that that prompt that
130. 23:40Danielle shared and go look at that chat and just look at how it actually found it. Like, what was that process? What
131. 23:47tools did it write? How did it figure that [ __ ] out? Did it did? Maybe it saw, you know, the Rocky Mountains in the [Music]
132. 24:00background. I talked to the new chat GBT and it said it made made it and all I had to do Oh, was click to explore. Oh, yeah.
133. 24:14Okay. So, Silver Fox brings up a really good point. One of the things that 03 in particular, but I I assume 04 and
134. 24:26uh 04 04 mini and high have this as well. It hallucinates like a [ __ ] It just it it unashamedly makes [ __ ] up. So it will it will say
135. 24:40here's a Dropbox link with all these files that I created. And you click on the Dropbox link and it doesn't exist.
136. 24:47So, uh, so that'll have to deal with, but yeah. So, so just know that it hallucinates a lot. But I just I, you know, I feel like
137. 24:57we've spent the past two years kind of understanding, you know, how to use large language models, how to prompt,
138. 25:03how to get good stuff out of them, you know, how to do it for particular use cases. Um, and then March 31st of this year, we
139. 25:12get this new image generation model, which is a really powerful and unique thing. And it's funny, the the sort of
140. 25:19Google lovers out there are like, Google's had that for months. They've had integrated image gen model in, you
141. 25:26know, they've had a multimodal large language model, but there's something about OpenAI's implementation of it.
142. 25:31It's just better. It's the first one that I'm experiencing it. It generates um like commercially usable creations, art, you know, ads,
143. 25:46social media post, YouTube thumbnails, just right out of the gate. You don't have to take it into some other assembly
144. 25:54program like an Adobe Suite program. So, that really changes things. And then this new model, because it's it's got reasoning with
145. 26:03tools, shit's different. It's a It's just a different thing. And like when do you use it? How do you use it? Why would
146. 26:10you use it? Is there any reason not to use it? And the answer is yes. Sometimes you need a quick answer for something.
147. 26:17Sometimes you just want a paragraph written. But sometimes you want a report. And in that report you want it
148. 26:23researched and analyzed and thought about and graphs made and charts made. Then it can do all that. Okay. 50 per week for plus. There you
149. 26:33go. Um, and I mean, you know, there's there's a part of you probably going like 50 a week doesn't sound like much, but
150. 26:43like that thing did all that thinking. You know, that was a single prompt and it did all that work. So, if you come up
151. 26:48with interesting prompts, you're going to be good. And I'm going to give you a bunch of those tonight. All right? And
152. 26:53we'll we're going to go play with those. Okay. Thoughts, questions? Anything? [Music] No. No. Everyone's like, "It's Friday
153. 27:07night. I'm just drinking, Kyle. Could we just not be so intense on the edge of Thanks. Thanks. Hey, Kyle. Kyle,
154. 27:18listen. Listen. I love that you're so passionate about the artificial intelligence, but you know this Harry Kerry. We like baseball.
155. 27:27baseball booze and you know hanging out on Friday night maybe lighten [Music] up does it retain its tools? No, it
156. 27:41doesn't. It's like it's it's literally you can almost think of it like this. It's it's in the chain of thought. It
157. 27:47says, "Okay, I've gotten population data from this county and this county and this county, and the next thing I need
158. 27:55to do is I need to consolidate the data across those three counties." And it's like, "Oh, okay. Well, then I need to
159. 28:04write a Python application to extract the data, put them into a new spreadsheet or a new data table, and
160. 28:11consolidate it so I can look at it." So it writes the Python code, consolidates the data, and goes, "Okay, here's your
161. 28:17data." And then it goes, "Okay, now I got the data. Awesome. Now I need to actually analyze that data and look for
162. 28:22trends. So let me write a Python code to analyze the data based on what I what it has and analyze the trends." So it
163. 28:30basically just keeps creating a tool, executing it, looking at the answer, figuring out what it needs to do next,
164. 28:37writing another tool, or just using the large language model. that basically is just working its way down through
165. 28:45this process of of it trying to get to the answer. So, it's it's insane. It's crazy. All
166. 28:59[Music] right. Now, Brian, one interesting thing that you raise Because chat GPT now has memory. Like if
167. 29:13you don't know this, the other thing that happened in the past week, week and a half is chat GPT now remembers all of
168. 29:21the chats you've ever had with it. So it's possible, Brian, that it it it would understand that it it it created a
169. 29:29tool for a previous prompt and it it was successful at a thing. But I don't think Chat GPT necessarily cares. Like I feel like
170. 29:38that's a human thing. Like I wrote a thing that works. Let me reuse it so I don't have to think about it and
171. 29:44recreate it the next time. But it doesn't care. It's just like I need a tool that analyzes this specific data
172. 29:52from these three specific sources that might not be reusable. So like it's disposable code. It doesn't care. Dr.
173. 30:00have to drop and we'll have to do re-watching. Well, thanks for hanging in or checking in and hanging out. verified
174. 30:07by chat. GBT 50 messages per week for plus and near unlimited for pro. Not sure what near unlimited means. I guess
175. 30:17basically if you start melting their servers, they'll shut you down. Um I used 03 to create a proposal from an RFP
176. 30:24and it did a pretty good job. Formatting was I assume shitty. Oh, it was amazing. Cool. Awesome. Formatting was amazing. Writing
177. 30:37Why Why is this so blown out? Tik Tok. Um, do I have I'm going to try something.
178. 30:53Brightness. Hang on. Wait.
179. 31:13I'm trying I'm doing monitor settings. Oh, god damn it. Um, without really knowing how to
180. 31:30um. Ah, god damn it. Hang on people. This is if you think
181. 31:47this is easy. Uh, yes. Okay. Yes. Okay.
182. 32:04Yes. Oh, look. I'm doing it. Tik Tok. You can actually see me now. Okay. Now I can't I can't see the screen at
183. 32:16all. Let me turn brightness up a little bit. Maybe we had too high contrast for Tik Tok. Hang on.
184. 32:31Okay. Okay. There. I can see it a little bit. And yeah, you can you can see things now. Tik Tok. I fixed it. [Laughter]
185. 32:45You're welcome. Welcome to chat add. Okay. How to change your monitor settings without reading the manual by Kyle
186. 33:02Shannon. Oh my goodness. We're back in business. Look at that. He can actually now see me. I like it's unfortunate.
187. 33:16Um, only took four months. Shoot up. Producer Brandon's smacking me down. Uh, could we have done this night one? We could [Laughter]
188. 33:29have. Can you have a guitar in your hands? Maybe more difficult. Oh, because you have a guitar in your hands. Maybe
189. 33:37more difficult. No, this is just um Okay, so here's what I thought we'd do. I thought it might be nice, let me get rid of the the
190. 33:52guitar. What I thought might be nice, is to go to the go to the Twitter. Some call it the X and we don't call it the Twitter or the X
191. 34:06boomer. So, you go to the X and I figure we'll just like scroll and look at what people are talking about. All
192. 34:17right. What am I doing here? I got to adjust my so used to seeing myself blown out. I'm a little caught off guard here.
193. 34:28Okay. Producer Brandon in the house. Dennis M. I can't control Kyle's monitor. Wait, are we making fun for Kyle for not
194. 34:38having a producer again? No, this is the one thing that Brandon controls everything except the quality of of the display on Tik
195. 34:51Tok. This is I will not I will not uh whatever you call it blame producer Brandon for that. There's much I can
196. 35:00blame producer Brandon for, but not that one. Okay. Okay. So, I figure what we'll just do is scroll Tik Tok for a bit and
197. 35:08see um what people are talking about. Yeah. Oh, what did I say? Tik Tok. Yeah, I
198. 35:27mentioned. Yes, ma'am. Get back to work. All right. It's really good at stuff. Dear algorithm, how do you guys deal with burnout? Just go
199. 35:41deeper. I just I I got to tell you, I mean, I I feel like the end of 2024 for me was was pretty burnouty and and
200. 35:53I I think January and February for me were pretty burnouty. March March has started to light light lighten the the
201. 36:02load and and now I'm kind of reinvigorated again here in April. Launching deep agent. I don't know what that
202. 36:13is. Sorry friends. Can't keep up with all the DMs. Okay. Funniest thing. Social. Okay. Uh a meteorite just hit somewhere.
203. 36:27This was an interesting Oh, here this is a good here's a interesting Sam Alman tweet and Gary Marcus response. I think
204. 36:35this is going to be more like the the Renaissance than the industrial revolution which I agree with. Right.
205. 36:40I've I've been saying for I don't know year and a half that we're entering the great renaissance. I that's absolutely what I
206. 36:47see. Gary Marcus Sam Alultman used to promise radical improvements in productivity. Now he hints at vibes.
207. 36:53Why? because companies like Johnson and Johnson finally realized that Genai was not going to deliver on its promises. This is such
208. 37:02[ __ ] And let me tell you why. Let me talk about this one for a second. I I have deep feels on this one. One, I I
209. 37:08just think Gary Marcus is just nasty for nasty's sake. It's his brand and and what he thinks people expect of him. Um the first place people
210. 37:23go when a new technology comes along is efficiency because how computers have always behaved is that they make human work more
211. 37:37efficient. The G in GPT stands for generative. So yes, you can use AI to make what you currently do more
212. 37:45efficient. But what you can also do is you can create entirely new ways of working entirely new kinds of
213. 37:53outputs in the context of work in the context of your job. It's it's an evolved understanding of what AI makes
214. 38:02possible. It's not just an efficiency machine. It's an efficient an efficiency machine plus whatever generative [ __ ] you want
215. 38:11to generate out of it. Um, and that's why I think the Renaissance comment is right. This is not just the industrial
216. 38:21revolution. If this felt just like a tool that was going to make us more efficient, then it's an extension of the
217. 38:27industrial revolution. The industrial revolution quite frankly was about augmenting human strength, right? We
218. 38:35used these machines that were stronger than we were to do the work that we used to do physically. AI is augmenting us
219. 38:48intellectually. Like we don't lament the day where we had to, you know, carry heavy things, you know, across the
220. 38:57river, you know, now we made steamboats and we could steam cranes to put the heavy [ __ ] on the boat and steam it across the river.
221. 39:09Same thing's happening now with with our intellect. And so it's not just about efficiency. That's that's it's an
222. 39:17antiquated way of thinking. Anyway, sorry, mini rant. If chat GPT is so efficient, why do I lose hours working
223. 39:24with it? It's both a gift and an ADHD nightmare. Totally agree. But but that actually what you're talking about is
224. 39:32what I was just talking about, right? If it were just an efficiency machine, you'd be a hell of a lot more efficient.
225. 39:38But you like do this efficiency thing and you're like, "Oh, maybe I could create a new marketing plan." And then
226. 39:44you're down the rabbit hole, right? I'm still on fire from burnout. And it's only April. I know. Tweeting hour. Um, okay. 03 is
227. 39:57insane. I don't even know what to say. This is the moment we've been waiting for in programming. Okay. really hard to
228. 40:03comprehend what 04 will feel like. [Music] Um, it's Ethan Mollik. It seems like a big question in AI right now is whether
229. 40:12there's a flywheel in AI development and when it kicks in, either because having advanced AIs let you code and train
230. 40:20better AIs or due to the returns on experience. If when it does, all non-frontier labs will will fall behind forever. That's
231. 40:36interesting. A thing that I think is fascinating about OpenAI right now is they launch a new way of doing
232. 40:46something, right? So they they innovated with chat GPT and then all the other companies quickly replicated what they
233. 40:57did in chat GPT and then they came out with GPT4 and then people took whatever it was 6 eight months a year to replicate what they did in
234. 41:06GPT4 and now they come out with a new thing now they come out with reasoning models and everyone's like you know what
235. 41:11is what is Orion and what's a reasoning model and how are they going to do it then they come out with it now everyone quickly
236. 41:17replicates what they're doing. This new model, this new reasoning with tools is is a new thing. Open AI right now to me is
237. 41:29starting to feel like Apple used to feel in the 80s. YouTube question on 03 is it really just more for programming or a bit better than four
238. 41:41for everything? I haven't played with 03 much yet. I don't know the answer to that. and what I'm going to share
239. 41:47tonight. So, this article that I wrote, I'm going to share tonight, but I'm I'm gonna walk through it. We're gonna experiment with
240. 41:54it. I created a whole series of prompts that are about using 03 for non- STEM, science, technology, engineering, and math.
241. 42:07I went to Grock and I and I said I I asked Grock, "Go find me use cases that people are talking about with 03 that
242. 42:15are non- STEM, that are non-scientific." And it basically found like three use cases. And then it said, "Not many
243. 42:22people are using 03 for non-scientific use cases." And so I said, "Well, let's come up with some use cases to be able
244. 42:29to go test it." And so that's the that's the article I'm going to be sharing tonight. and and I'll um um and that we'll start playing with.
245. 42:39But that's the that's the uh that's the thing. Um fulminating Friday. Although AI is moving much faster pace than Apple. Yeah, it's
246. 42:55moving faster. But there's something about like Apple used to do used to do industrydefining innovations where Apple would do it and
247. 43:08everyone would go, "Wait, what? Oh, holy [ __ ] Yeah, a a screen you can touch instead of a keyboard on a
248. 43:17phone. Computer in the pocket. Yeah, that's good. Let's go copy that." Right. That used to be Apple every two, three
249. 43:25years, right? And with Open AI, it's every six months, right? Six, eight months so far. Um, yeah, fascinating. All right. I
250. 43:38use four for marketing purposes. Yeah. I I think listen, 40, the the the the little default mode of chat GPT is
251. 43:49incredibly powerful. You don't need you don't need these reasoning models. But I think the more that we can understand
252. 43:56what these reasoning models make possible, it's going to start to make 40, god damn these names, it's going to start to make 40
253. 44:06feel increasingly antiquated where you go in and you give it a prompt and it gives you an answer and then you do the reasoning to go,
254. 44:14okay, here's a follow-on prompt and here's a follow-on answer. Oh, okay. Right? as opposed to going, "Here's a
255. 44:21goal." And then have the AI think about all the stuff it needs to think about to give you a complete answer rather than
256. 44:29little one-offs. Um, so anyway, all right. Someone allegedly stole someone's canvas first, although AI is moving much
257. 44:40faster pace. your article TLDDR oh the article is 30 I think it's 30 30 different use cases for non- STEM here I'll let's let
258. 44:52me go share the article now so if you go to my profile on LinkedIn or not LinkedIn uh X the X Twitter the one that used to have the
259. 45:03bird that one and scroll down the you you'll see this live and then right below below it. There's this
260. 45:13article that I wrote. Oh my god, it's so good. Tik Tok, you can actually see this. There's a rumor. I heard a rumor
261. 45:20that Tik Tok's going away Sunday that China's pissed, but Brandon said he hasn't heard anything about it. But
262. 45:26anyway, I fixed my monitor settings. You can actually see what I'm seeing on my screen now. No, I have I have an update
263. 45:32on that, Kyle. Um, the ban the deadline was extended until June 19th. And some of those that are from not paying
264. 45:42attention were conflating it with April 19th because they just saw the 19th. Yeah. The the ones the the ones that I
265. 45:47saw were not that there was speculation that that um you know due the due to the tariff stuff they were just going to
266. 45:54shut it down prematurely. But whatever doesn't matter if it's gone. It's gone. You all know how to get to the to the
267. 45:59YouTube channel. We'll keep going there. But anyway, after four months I've finally fixed it so you can see. So
268. 46:04anyway, unlock 03's hidden powers. a playbook of non- STEM prompts uh for for non-scientists. That's actually bad writing, isn't
269. 46:14it? What do you call non-scientists? For creatives, but it's not just creatives. Whatever doesn't matter. Um,
270. 46:23okay. So, the the way this is there's there's three sections in it. One is just 03 in the humanities, how you would
271. 46:31use 03 in the humanities. And then it's got like use case comparative thematic analysis description example prompt
272. 46:38right compare the themes of love and sacrifice in Shakespeare's Romeo and Juliet and Jane Austin's Pride and
273. 46:45Prejudice highlight how each reflects the historical context and provide nuanced interpretation of their differences. Right?
274. 46:53So broad use case, you know, example prompt and just go play with these. And so um stylistic analysis of pros or
275. 47:03poetry intertextual connections. What identify and explain the interextual references to Greek mythology in TS
276. 47:12Elliott's The Wasteland. How do these references enhance the poem's meaning? I mean, this is kind of [ __ ] you get in,
277. 47:18you know, some advanced literary class, right? non-geeeks. Why don't I need to get rid of the word non, right? Because I've got
278. 47:27a playbook for non- STEM prompts. A playbook of non- STEM prompts
279. 47:43for Oh, yeah. Shut up. Shut up. Shut up. producer Brandon. Okay, this
280. 47:59title. Wait, what happened? [Music] um has
281. 48:14two nons in it. I want to remove the second one. What do I call
282. 48:30non scientists in this context? Make it more [Laughter]
283. 48:46better. Non-geeks. Here are a few cleaned up. Unlock it. For creative thinkers. Oh, I like that. For humanists. That sounds
284. 49:00too that sounds loaded. For curious minds. Scientists can be curious. Culture makers. Everyday innovators. Oh, I like everyday
285. 49:08innovators. All right. Everyday innovators or creative thinkers? I think we're going with creative thinkers. Yes.
286. 49:16Creative thinkers or everyday innovators. Votes. Come on. Come on, people. Let's put some comments. Pretend like you're
287. 49:24listening. One. One. Vote for for uh creative thinkers. Creative thinkers. Everyday innovators. Both. We can't have
288. 49:33both. Oh, we could have both for creative thinkers and everyday innovators to explore. But that's a little long. Everyday
289. 49:44innovators. One. One. Uh, like we're split down the middle. Damn it. I have to make a choice. Creative thinkers. Uh, critical
290. 49:55thinkers. Creative thinkers. I think it's everyday innovators because the the loaded part of creative is well I'm not
291. 50:02creative but everybody can be an everyday innovator. Okay, everyday innovator. All right, let's go fix it. Um so do me a favor. Here's my
292. 50:12request to you. The article will be temporarily unpublished. Yeah, fine. We got to get rid of non-scientist every day.
293. 50:30Innovators, a playbook of non- STEM prompts for everyday innovators to explore. Beautiful. Okay, here is my
294. 50:45request. Go push the [ __ ] out of this thing. Like it, reshare it, quote tweet it, get it out there because I want this
295. 50:53to help people. Okay, so let's just keep going through this for a bit. All right. So, um, so literary analysis was one of
296. 51:00them. Couple of examples in there. Creative writing, um, constrained narrative generation. Oh, this is kind
297. 51:07of interesting. One of the things it said is in a reasoning model, you should be able to say something like, write a
298. 51:13500word story, and it should get close to 500 words. All right. Poetry composition. Write a Shakespearean sonnet about the passage
299. 51:26of time. Use imagery of Let's actually use this. Let's go do this one because I want to see a Shakespearean sonnet is a 14line a 14line poem
300. 51:39with what is it? Six and six and two. I forget how I forget the structure of it. Oops. I didn't actually copy it.
301. 51:52Copy. Go over here. You can see all this, right? Thinking ensure the rhyme scheme and amic pentameter are strictly
302. 52:09followed here. Look, look what it ah beautiful. Look at this. Aab CD CD ef GG. The first few lines are pushing along
303. 52:23nicely. Line one seems to follow the correct meter. I'm adjusting the phrasing for line two while distance
304. 52:30starts. Let's go look at all this. This is really cool. So, oh, here it did it. Wow. Okay. I've never
305. 52:45had I've never had an LLM be able to successfully write a sonnet that follows the actual format. So, let's go. I I
306. 52:55want to look at what it did. It thought for 35 seconds. The sonnet must have 14 lines following amic pentameter 10 syllables per line.
307. 53:08and then use an ab cd cd ef g rhyming scheme. I'm starting the first line. The river clocks the hours with silver flow.
308. 53:24However, that's 11 syllables. I can adjust the river counts the hours in silver flow to match the 10 syllable limit. This is so [ __ ] cool.
309. 53:38The first few lines are pushing along nicely. I think I'm going to put old Sunseo lyrics through it. No. Start up.
310. 53:51Start over. The Tik Tok cut out. Oh, did it? When did it cut out? Danielle, am I still Am I live on Tik Tok?
311. 54:06Okay. Okay. That might have just been on your end. Just tell me what you missed, Danielle. So, so here's where I am. I've written an
312. 54:17article on X. I'm I'm going to put this on AI Salon and LinkedIn, but on X right now. Unlock 03's hidden superpower, a
313. 54:27playbook of non- STEM prompts for everyday innovators to explore. So, it's the it's the other than this
314. 54:34live, it's the next it's the next post down in my thing. So, go find it and then it's just it's just like a big
315. 54:42longass list of things to do. So, where we are right now is we're in poetry composition and have it write a Shakespeare like a properly
316. 54:53formatted Shakespearean sonnet. And so we're just looking at the the reasoning. The first few few lines are pushing along
317. 55:03nicely. It's coming together. But line nine, but rivers teach that change is their decree. If you don't understand
318. 55:13reasoning models like what this is doing right now is the perfect example. What if you if you tell Forro to write a
319. 55:21sonnet, it will vomit out a poem. And then even if it got it wrong and you said, "Hey, the last two lines are supposed to rhyme
320. 55:32and they don't," it'll say, "Oh, let me fix that." And it'll vomit out another poem. And it may or may not fix the
321. 55:39bottom things. This is fixing it before it actually shows you anything. It's it's basically thinking, right? It's reasoning before you get
322. 55:51anything. All right. It looks like I've got a re the rhyme scheme mostly figured out. Line 10 rhyming with light is
323. 55:58solid. I like that it says the rhyming with light is solid. That means it's it's doing qualitative analysis of like was this is this a good
324. 56:08poem. It's been a while friends. It CS tutor hope you've been well. Thanks for popping in. That's pretty fascinating.
325. 56:15Couplets 13 and 14 fit with tide and glide rhyme. So that's progressing well. I've adjusted the other lines for better
326. 56:22flow, particularly switching some words to achieve perfect rhymes. The rhyme scheme now follows AB A B CD CD E F E F
327. 56:33G. All right. Upon the river's clock and watchful stars, the river counts the hours and silver flow, while distant stars keep
328. 56:43watch with ageless gleam. Their winding tails repeat the world of flow. Oh, so it used flow twice. We don't like that yet. Mark our mortal.
329. 56:58Oh, okay. This is bad. It used the same word. It didn't actually rhyme them. It used the same word. So, let's give it feedback and see
330. 57:09if it can fix it. Tik Tok pin. It's talking like you. Oh, that's the other thing. Oh, so, so this is an important thing to note in your
331. 57:20settings under personalization. There are now two options. Reference saved memories. So, the saved memories are where it says
332. 57:31updating memory. It's like an explicit list of memories. And then this new thing is reference chat history. You can turn that off.
333. 57:39By default, now chat GPT knows everything about what you've ever talked about and can reference it in your current chat. It's
334. 57:47insane. Like it chat GBT starts to become much more like a companion that gets to know you over time. But you for something like this,
335. 57:59if we want to write this poem, we might not want it to write like Kyle. We might want it to write more like um
336. 58:05Shakespeare, right? So I might want to turn that off before I do this exercise. So, you should know that that's a
337. 58:11possibility. Well, technically C does rhyme with C. Yeah, exactly. So, okay. So, we're going to say, um, I'm impressed that you got the
338. 58:25structure correct. However, what you should notice is that in your AB AB rhyme
339. 58:39schemes, the words not only rhyme but are identical. Period. That's not very creative. Period. That's not
340. 58:49very poetic. Period. That's certainly not Shakespearean. Period. I want you to dig deeper, try again, and not only
341. 59:06achieve definitional accuracy, but creative excellence. I'm going to treat it like a freshman in college. Listen, listen. You
342. 59:22did the assignment, but you're still getting a C [Laughter] [Music] minus. It said the word solid. And you
343. 59:38say that word frequently. Oh, interesting. That's cool. Wow, it did it really quick. What What did it say? Okay, the user wants
344. 59:49it. Okay, the first line Done. I guess because it had something to start with. Okay. The river upon the river's clock and
345. 1:00:00watchful stars. Revision. The river tallies hours in murmured sweep. Murmured sweep. Yeah, that makes sense with the
346. 1:00:10river. While lanterned stars record the night in ink, their whispered chronicles descend the deep and constellations
347. 1:00:20pause yet never blink. And let's so let's doamic pentameter. The river tallies hours wait. The river tallies hours in murmured
348. 1:00:33sweep while lantern stars record the night in ink. Their whispered chronicles descend the deep and constellations pause yet
349. 1:00:45never blink. So those are correct. They sing of moments carved on passing foam and threads lost wishes
350. 1:00:54through the vaulted sky. Yet every crest betray betrays a fleeting a fleeing home. While heaven's map remains
351. 1:01:04refusing why. Oh restless clock that glimmer glimmers in the tide. Teach mortal eyes to read their shifting
352. 1:01:11flame. For rivers point the starlet fates reside and blaze tomorrow's course in silent claim. So let my fleeting
353. 1:01:20breath find patient sleep within the starcross streams. Your echoes keep. It's pretty [ __ ] good. I've I've never had a large language
354. 1:01:34model get get a Shakespearean sonnet structure right. Um pretty amazing. Okay, back to back to this document. So
355. 1:01:42that was poetry composition. Like we're just getting started like like um so what I did here's the three sections I
356. 1:01:48have in this document. Section one is just humanities like just do like [ __ ] like this like poetry. Section two is
357. 1:01:58give me prompts that live in the space in between humanities prompts and STEM prompts. Right? Because there's some
358. 1:02:04things you might want to do in humanities that that require some calculation or deeper analysis. And then the third section
359. 1:02:12is 03 is a multimodal model that can see and understand images as as Danielle's example really nicely illustrated. So,
360. 1:02:21so there's a whole series of prompts about the multimodality of it. Okay. Sections. What are we doing? So
361. 1:02:28what we're doing, Pate, is I wrote an article um that I just published on X uh that is about it's a series of example prompts, non-
362. 1:02:41STEM prompts for using 03 reasoning model because because OpenAI is full of engineers, the examples that they showed were all math
363. 1:02:51examples, which as a as a guy with an acting degree is it's meaningless to me. Like I I'm you know I assume it's good
364. 1:02:59math. I don't [ __ ] know. Um but I know that the way these things work there's probably very deep power in
365. 1:03:07them. So I wanted to write an article that gives people a head start for how to prompt it. Um poetry composition
366. 1:03:15historical fiction. Write a 300 words word scene set in 18th century Paris during the French Revolution. include
367. 1:03:22two characters debating a monarch, the monarchy's fall using period appropriate language and historical details. That's
368. 1:03:29pretty cool. Historical analysis. Analysis of historical events. That's pretty cool. Source evaluation. Evaluate
369. 1:03:37a letter from a civil civil war soldier and a modern historian's account of the same battle. Assess their reliability,
370. 1:03:48biases, and how they complement each other. That's a really good one. You know what? What would be cool with this
371. 1:03:53one is go find an image in the Smithsonian of a famous Civil War letter that's had a lot written about it and just give
372. 1:04:0603 the image of the letter and say, "Read this letter, identify it, find anything that's been written about
373. 1:04:16this letter or this story, and then compare those things." Paid on TikTok. Um, I'm on the flip side of writing. I'm on the flip side of
374. 1:04:30writing. Wait, I'm on the flip side. Writing part is useless to me. Yeah, exactly. I don't give a [ __ ] if I can
375. 1:04:36write like Shakespeare. Yeah, exactly. Curious Kyle and Ma Mathematic. So, Pate, that that's exactly my point. It's like this
376. 1:04:47is the the the significance of the multi-step reasoning with tool use instinctively feels to me like it
377. 1:04:58goes way beyond doing math stuff, right? And so that's I I think mathematicians and scientists and researchers are going
378. 1:05:07to look at what it does and it's much more patently obvious why you would use a model like 03 in those things. It's
379. 1:05:14not as obvious on the humanity side. So that's exactly what my article is about is go play with these things and let's
380. 1:05:20learn what it's good at um in the non- STEM stuff. Count counterfactual scenarios. What might happen if if an
381. 1:05:28American what oh what might oh it's like what if scenarios what might h what good lord what might have happened if the
382. 1:05:36American revolution had failed provide plausible scenario for the next 50 years grounded in 18th century political and
383. 1:05:43social conditions that'll be a good one I and I honest to god like reading through the um the reasoning it did when
384. 1:05:52it just wrote that sonnet doing more meaty things like this could be really interesting Okay. Philosophical inquiry, ethical
385. 1:05:59dilemma analysis um philosophical text interpret interpret interpretation summarize Nichi's concept of will to power in thus spoke
386. 1:06:14uh zeraustra critique its implications for modern individualism. Good lord. Construct arguments for and against the existence of free
387. 1:06:24will. All right. So, and then educational content creation, lesson plan development, interactive study guides, historical timeline creation.
388. 1:06:34Oh, that one's kind of cool. Let's do one of these because then we can do an image. Okay. Of this. Okay. So, what
389. 1:06:42we're going to do, create a timeline of the Renaissance, 1400 to 1600, highlighting 10 key events in art,
390. 1:06:48literature, and politics. Annotate each with a brief explanation of its cultural impact. So, we're going to go do that,
391. 1:06:54and then we're going to have 03 make a visual of the timeline because it can do images. So, we'll just do that prompt as
392. 1:07:07is. Not perfect initially, Valerie Cox, but so amazing in the end. Just wow. I assume you're talking about the
393. 1:07:15poem there, or you're talking about something that you did. That's pretty cool, Kyle the playright. Exactly.
394. 1:07:26Exactly. Love that Civil War prompt. Yeah, that was pretty cool. All right, what are we doing here? Oh. Oh, this is our timeline. Okay, it
395. 1:07:34did this really fast, right? Okay, these 10 moments I'm gonna say Okay, great. Um, oh [ __ ] my mouse my mouse stopped
396. 1:07:42working. All right, hang on. I think my mouse is it's charge is dying. Hold please. Charging said mouse. Okay. And
397. 1:07:53now I'm using a trackpad. Okay. So now I'm going to say um I want you to create an illustrated timeline that
398. 1:08:09has period accurate imagery, appropriate labels, and a clear timeline feel.
399. 1:08:28This should look so good that it could be displayed in the Smithsonian Museum.
400. 1:08:53it. We'll just let it we'll just let it be there. Okay. So, let's see how it does here. And then I'm going to try one other
401. 1:09:01thing. I I I read on X earlier that using Python, it can actually upscale an image, creating an illustrated timeline.
402. 1:09:14The user wants a period accurac accurate timeline. Okay, so it thought. So what did it think? We might need to either combine
403. 1:09:21several images into one. The user wants Okay. However, since I can't directly marry generated images into one large
404. 1:09:30graphic, I could either suggest them one at a time. This this would help. So let's see. Let's see what it does.
405. 1:09:49Come on, chat GPT, don't suck. But anyway, apparently you can upscale images now in 03 because it uses a Python library for upscaling
406. 1:10:05images. I don't know if that's true or not, but we can find out. [Music] My mouse retired recently,
407. 1:10:25too. Friday night, date night, playing with 03. So, are you guys going through the article? And again, do me a favor. How
408. 1:10:34many folks are in here? We got pile of people here, pile of people there. Go to Okay, it's not bad. It did it. It said it did it 4K
409. 1:10:46wide, but it cut it off. Greek. Uh, I'm going to say, let's see. I like the idea though
410. 1:11:04the image is cut off. on right and left. So either make it two
411. 1:11:18rows or allow
412. 1:11:34um things to be smaller to accommodate the space
413. 1:11:49requirements. Also, the color palette um feels old and tired. make it more
414. 1:12:06modern with um contrasting
415. 1:12:24background versus the text and illustrations. Let's see what it does with that. Ask it to create padding around the
416. 1:12:37border. Shrink to fit. Yeah, those would both be good. Those would both be solid. I should have looked at that before. Seems
417. 1:12:43plausible, but may just be a quirk. Would probably be better off doing a dedicated upscaler. Yeah, definitely. I
418. 1:12:49just want to see if it I just want to see if it is functionally there. I would think. Yeah, something like what's that one
419. 1:12:57called? Magnafi seems to be one that's a dedicated one that looks like it does really good results. That guy's always
420. 1:13:03posting examples up there. Um, while that's doing that, if you wait, can I pop out of here and pop back in? Yeah, I
421. 1:13:11guess I can. If you haven't seen the typo and timeline Oh, yeah. I'm sure there's going to be typos. Yeah. Yeah.
422. 1:13:22Yeah. Actually, what would be interesting is we'll let it generate this one. If it's got any major typos in
423. 1:13:29it, it might be interesting to tell it to write write a creative brief for the timeline first with exactly what text
424. 1:13:38needs to be on it. It'd be interesting to see if it got if it wrote all the text out first as a creative brief if it
425. 1:13:44got more accurate in its output. But I mean, do you know, you know, when we were back in
426. 1:14:00college and you had to go like like, you know, read all the books and get all the the crap from the Renaissance, the major
427. 1:14:08things that happened, and you had to do this [ __ ] yourself. All right. It still [ __ ] it up. Renaissance timeline, but it's not
428. 1:14:19bad. I kind of like the the color thing it did. Uh, let's see. Add padding to the left. It is still cut
429. 1:14:41off. All right, we'll come back to this. Um, go back over here. Historical timelines. Uh, and then it talks about
430. 1:14:56that. Okay, so here are the hybrid STEM and humanities use cases. These are kind of cool. So this talked about um I'll
431. 1:15:04propose a new list of use cases that push beyond surface level applications exploring this boundary layer and
432. 1:15:11venturing into cultural philosophical even speculative territories. These ideas aim to exploit 03's reported
433. 1:15:19strengths um simulated reasoning private chain of thought and adaptability. And then it gives some benchmark [ __ ] while
434. 1:15:28embracing humanity's focus on meaning, context, and human values. So these categorized use cases. Um, oh, I didn't I meant to edit this
435. 1:15:39[ __ ] Well, whatever. Um, cultural synthesis and reimagination. So, cross-cultural narrative fusion. Th this
436. 1:15:47was this was my favorite my favorite prompt so far that it wrote. Write a 600word narrative about a refugees
437. 1:15:55journey, blending the rhythmic cultural story te storytelling style of West African giot traditions with the
438. 1:16:03fragmented introspective style of Japanese zoosu. Ensure each cultural culture's aesthetic is distinct yet harmonized and
439. 1:16:14use the theme of displacement to bridge them. Include annotations explaining your stylistic choices. Um, DAI tomorrow is still on. Oh,
440. 1:16:26demystifying AI is still on. Zero things to present though. So, come with questions. Okay. So, tomorrow, if you're
441. 1:16:33a member of the AI salon, Pate's doing a um demystifying AI session. If you haven't been to one of his, you should go. So, if you go to
442. 1:16:46events, should be here right at the top. Demystifying AI. So, tomorrow at 2 PM. So, go RSVP for that event. If you've
443. 1:16:54not been to one of Pat Oh, and how you get to the AI salon, if you're not a part of it, go to the salon.ai, click on join our community.
444. 1:17:02That'll take you over here to the um to the mighty networks community. Register, introduce yourself. That's step two. And
445. 1:17:10then just go to events and RSVP for Pate's uh AI demystifying AI session. Um, if you want to understand how AI
446. 1:17:20works and like the stuff we're playing with tonight, actually, what's going on under the hood, that's what Pate's
447. 1:17:25really good at. Good. Awesome. Thank you for sharing that. Um, okay. Let's go back here. Did it do it? It did it. Yay. You did it.
448. 1:17:38Yay. Look, Tik Tok can actually see this [ __ ] now. It's a little miracle. Um, still a timeline. Oh, Tim Timoline, right? Let's see.
449. 1:17:53Florence Baptistry is there twice. Yeah. So, that's pretty crappy, but that's okay. Let's just That's That's fine. Download revised graphic.
450. 1:18:06Let me know. Um, please upscale this image. Let's see what it does here. Yeah, I 1590 to 1991. Is that what
451. 1:18:21it said? Yeah, the last one. All right. So, it's not accurate, but you know, hey. Okay. Wait. Thought for eight
452. 1:18:36seconds. Oh, it says the original image was created by Deli, which it wasn't. Since I don't have a direct upscaling
453. 1:18:51tool. Huh, that's weird. It should have invoked a tool. It should have generated a tool to do this. I'm going to stop it and
454. 1:19:05say, "Can't you use a a Python tool to upscale the image? Use your tools, you
455. 1:19:22tool." Does harshweed the first make an appearance? Ah, look. PIL image open. New size. So, it's doing a 2x
456. 1:19:40upscale. Resize using the high quality length zas filter. I don't know if that's high quality or not. Download the upscaled
457. 1:19:53version. Starting download. So, it downloaded something. All right, there's that. So, let's see if this is actually a different size
458. 1:20:05than the other one. Download this. Same size. It lied to us. It
459. 1:20:22lied. You hallucinator. All right. So, that was a fun one. All right. Back to our back to our little thing here. Um, speculative cultural
460. 1:20:36evolution. Imagine the evolution of DLL Moros celebrations from 2025 to 2125 influenced by climate induced migration and augmented reality
461. 1:20:48technologies. That's cool. philosophical and ethical simulation um interdisiplinary historical reconstruction, right? So these hybrid
462. 1:21:03ones are just much more sophisticated, you know, taking multiple inputs and and smashing them together, which is kind of cool.
463. 1:21:11Speculative anthropology modeling posthuman belief systems. Design a 700word belief system for a 2100 society
464. 1:21:21where humans and AI are cognitively integrated. Base it on trends in transhumanism, Buddhist ethics, and
465. 1:21:29neural interface technology. Describe the core tenants, rituals, social implications, and explain how 03
466. 1:21:38synthesized these influences. Cool. All right. Alien cultural contact, simulating it. Okay. So, that's those. And then the third section is
467. 1:21:52multimodal capabilities. So, visual cultural analysis. Analyze the the iconography and car K caravio's the
468. 1:22:01calling of St. Matthew. Identify at least five key symbols and explaining their religious and cultural significance in Baroque, Italy. Very
469. 1:22:14cool. Digital art and cultural reimagination. Create a 600 by 600 pixel digital artwork inspired by Aboriginal
470. 1:22:22Australian dot painting. Adapt the technique its techniques and storytelling to address climate change. Ensure the artwork respects cultural
471. 1:22:32principles of Aboriginal art. Provide a 300word reasoning chain explaining how you incorporated traditional elements
472. 1:22:41and modernized the narrative. This is fascinating. Let's see what this does.
473. 1:22:56I'm at the poem point in the weekend. A sign currently create an illustration blackout poem. I don't know
474. 1:23:07what that is, but that sounds cool. Don't forget to give it a thumbs up. Oh, good. Vicki G Riot pronounced Griot.
475. 1:23:15Okay, good. I didn't know if it was Yeah. Griot or Yeah. Or G Riot. Okay. Griot. Good. Thank you. This is not doing AI upscaling, but
476. 1:23:28it's something. Yeah. Yeah. Exactly. You know what's weird, Pate? 03 seems to be very very um it's very very fickle.
477. 1:23:37Sometimes it seems to be brilliant and sometimes it seems to be totally incompetent. Lord Digital God's given me
478. 1:23:43a two $2 sticker, super sticker on YouTube. Thank you so much, sir. very generous. I appreciate that. Oh, lordy lordy lordy
479. 1:23:55lordy. Thinking is it what's is it just thinking or is it making? I think we might have stumped it with this
480. 1:24:05one. That's the other thing I noticed is that it seems to be a little their servers seem to be being hammered.
481. 1:24:15It's a classical upscaling algorithm. So, not bad, just not bleeding edge. Yeah, I know. I I figured that was the
482. 1:24:22case, Pate, because it didn't seem like it was any sort of AI sort of thing. Get node by ID or message ID. No
483. 1:24:30node found on the ID. There's some There's some [ __ ] code that they didn't get behind the uh popup popup menu.
484. 1:24:42We'll put weekend assignments on irregulars. Oh, good. So, you can see blackout film. Nice. The AI told me to
485. 1:24:49make these. Cool. Very nice. Considering cultural sensitivity, the rules for Aboriginal art probably haven't changed
486. 1:24:56a lot. Searching the web. Getting started. That's kind of cool. All right. So, wait, what did it do? I'll quickly research
487. 1:25:07guidelines if I'm uncertain for thing. If I'm uncertain or things are dynamic, I'll browse for the latest info. In
488. 1:25:16particular, I can search for the respectful use guidelines for Aboriginal art like dot painting. That's actually kind of
489. 1:25:24cool. I'll make sure to include anything relevant and site cultural principles to stay accurate and respectful. Let's do the
490. 1:25:34search. All right. Keeping in mind these avoiding sacred symbols. Oh, that's cool. This is aware of what symbols it can include and what
491. 1:25:46it can't. I'll incorporate elements symbolizing water, rising temperatures, and footprints. I'll also include a thoughtful touch
492. 1:25:55acknowledging a consultation with a we're a jury artist. Along with the image, I'll provide a 300word explanation of the
493. 1:26:05artistic choices. Make sure to site cultural guidelines for accuracy. Let's get started. Oh god, that's gorgeous. I got to
494. 1:26:20sneeze. Okay, that was the weirdest sneeze I ever had. I just had a non-sneeze. I had the in inhale part and then it just
495. 1:26:28disappeared. That was weird. Yikes. Blackout poem is a chunk of text and words that are blacked out. So remaining words make the poem
496. 1:26:43something. Please zoom in. That's as far as I can zoom in
497. 1:26:57here. Love the river. Um, let's see what it says. To honor Aboriginal Australian dot painting while addressing climate
498. 1:27:09change, I began by identifying non-sacred visual conventions that are widely shared across nations, concentric
499. 1:27:16circles for water holes or communities and meandering dotted lines for song lines. Avoid any clansspecific or
500. 1:27:24ceremonial symbols. Follows the First Nations cultural and intellectual property protocols that stress permission for respect of restricted
501. 1:27:33imagery. Pallet. I chose deep ochres, siennas, and umbers. Cooler teals and indigos flow diagonally to symbolize rising seas and
502. 1:27:44disrupted rain cycles. Um, a large central Sun circle anchors the piece. Is that right?
503. 1:28:02Yes. Simulated cultural consultation. Before finalizing motifs, I drafted a brief and hypothetically hypothetically
504. 1:28:13shared it with a we're a jury knowledge holder and a Gatagle environmental artist. feedback asked, "Avoid mimicking
505. 1:28:24toemic animal tracks. Emphasize shared stewardship over didactic warning and foreground water as connector, not merely a
506. 1:28:33threat." Those insights led me to weave meandering aqua pathways through every quadrant so country sky and people
507. 1:28:42remain balanced in balanced dialogue. I mean, I don't know [ __ ] about any of the ethics of this, but it would be
508. 1:28:53interesting to see if this if it actually achieved what it said it did. Um, there are dot paintings in most of
509. 1:29:03the buildings here, especially government buildings. Wow. Cool. So thoughtful. Wow. Yeah, it's pretty crazy, right? This is beautiful.
510. 1:29:12Teaser sneezer. Yikes. Was to the Oh, was to the node ID error message. Yeah, I know. I know, right, P? It was like It was
511. 1:29:24like, uh, guys, you might want to make sure that your errors are not, you know, out there for the public to
512. 1:29:33see. You You did launch this thing, right? Like this is available to everyone. Um, okay. So, we'll go back to
513. 1:29:41this thing. So anyway, so so we don't have to go through a whole document, but might as well. Um, speculative cultural
514. 1:29:48artifact creation. Imagine a visual artifact, a ceremonial object. Oh, this is cool. Imagine a visual artifact. A ceremonial
515. 1:29:57object, a public mural for a 2075 society shaped by AI human symbiosis and ecological restoration. generate a an image of the
516. 1:30:13artifact and manipulate the photo of a 2025 urban street to incorporate it in 400 words. Explain its Okay, let's let's
517. 1:30:22go do this one, too. I'm going to change some of the stuff in it. This is cool. And again, I'm doing 03 here. The
518. 1:30:32other thing to do is switch between 03 and 04 mini high. Uh, like if if you run out of 03 things, do that. But also
519. 1:30:41maybe just try the same prompt in multiple models and see what it does. I ran this ran the same prompt and
520. 1:30:50got a very similar picture, but it is different. Fascinating. Cool firsthand info. Hey Steo, was there anything in either the
521. 1:31:00image or that explanation that you felt was was like out of bounds or off base in terms of like ethical considerations.
522. 1:31:48[Laughter] Damn. What am I doing? Tibetan throat singing over here.
523. 1:32:04I do that too with my voice. It's just ADD self soothing. No, it seemed fine. Uh, this is not what I wanted at all.
524. 1:32:17Um, this is supposed to be a photo of a mural on a building in the
525. 1:32:31future. Take this same image and re-imagine it in that context. What's it doing now? It's
526. 1:32:45writing the Oh, it's Oh. Oh, this is going to be interesting to see what it does. Thought for 30 seconds. I need to find a street photo.
527. 1:32:58It didn't find it. Okay. I found an image that could work as the street scene even though it's not obviously
528. 1:33:04from 2025. It has the description. 2025 should be acceptable. Now I need to generate the artifact image. After that I'll figure out how to
529. 1:33:16combine both the street photo and the artifact image. Although passing the correct image references will be
530. 1:33:23important for editing. Time to proceed. All right. So there's the image it created. Kind of cool. Image C. So these are basically the same
531. 1:33:35thing. If it just repeats this again, then I'm
532. 1:33:50going to stop it. I'm going to give it an explicit prompt. But again, you know, the difference thing here is it's generating
533. 1:34:01these images in this chain of thought thing, right? It's not just I didn't tell it to create multiple
534. 1:34:08images. I told it to create an image and combine it with a photograph. And so it's it's doing those steps. Yeah, it's
535. 1:34:18just it's just repeating itself. Okay. So, we're going to stop this and then we're going to go here's an explicit prompt. Get your
536. 1:34:26[ __ ] together. [Music] mural future reimagination creating a
537. 1:34:42photorealistic scene. I'm considering that we might need a better
538. 1:34:58query. I'm going to stop it because it's saying 2025. The photo should be from 20 from the
539. 1:35:13future. Ignore the 2025 photo um element. Just make a street
540. 1:35:27photo with the image you created on a massive mural on a building in the photo in the future. Damn it.
541. 1:36:00Wee. Oh, did it do 2025 again? I think it did. [ __ ] Oh, the user has shifted the focus. Okay. Since we can't manipulate
542. 1:36:08an existing photo, I need to generate a completely new futuristic street scene with a giant mural using text to image.
543. 1:36:15Yes, correct. I'll also use the other artifact image I created earlier. Good. So, it's remembering. This is one of the big things about
544. 1:36:2503 when it's doing this stepbystep thinking, it's maintaining the context across all these different steps. So, the fact that it just
545. 1:36:36referenced that it made the images earlier, that's something that straight chat GPT can't do or it's hard to get it to do that. You're like,
546. 1:36:45"Use the image you made before." It's like, "Okay." And then it makes a new image,
547. 1:37:07right? Hey, it did it. [ __ ] yes. It [ __ ] did it. it up, people. But wait, we're going to tell it. Okay.
548. 1:37:21Um, okay. You gave it the gippers try, but let's amp it up. Make it 16 by9 wide.
549. 1:37:41Cinematic and photographic in style. In style. It should look very very very realistic.
550. 1:37:58realistic with dramatic lighting and um
551. 1:38:12atmospheric haze. What the hell did it write after? Oh, it just wrote all this other [ __ ] The circular society not emblem is the
552. 1:38:25most ubiquitous icon of 20755's ecote tech [Laughter] commonwealth. It's fantastic. Kyle Shannon the human digy
553. 1:38:47do future Starbucks. Oh lordy lordy lordy lordy lordy. I tell you lordy lordy lordy. I probably shouldn't say that. It's good
554. 1:39:02Friday, right? Maybe the haze is gone in the future.
555. 1:39:16Yeah, it probably should be, but there could still be there could still be fog rolling in off the water. So, just what the atmospheric haze
556. 1:39:25should do is just make it so the stuff in the foreground is more, you know, more prevalent. But we'll see. We'll see how it does.
557. 1:39:36I actually am not grooving on they took away. So one of the things that when they came out with the new image gen
558. 1:39:42model, it would start out with this interface and then it would go to this blurry image and it would sort of draw
559. 1:39:47the image from the top down because it's doing it as as a as a transformer. Oh yeah, it did it. Wow, this is great. There you
560. 1:39:58go. I wouldn't exactly call it photorealistic. It still looks a little illustrative, but it's got atmospheric haze. It's
561. 1:40:09cinematic. It's got a drone in it. Everyone's walking in the same direction. Huh. All right. [ __ ]
562. 1:40:25works. And back here. What else? Visual historical reconstruction. Oh, reconstruct a lost fresco. That's pretty
563. 1:40:36cool. Reconstruct the lost fresco from a fifth century Mesop Messoamerican temple using textual descriptions from
564. 1:40:47the cotices, architectural data on pigment use, and images for the related Mayan art generate a picture of the fresco.
565. 1:40:59So what they say Mesoamerican temple. Oh,
566. 1:41:15fresco. Excuse me. So we need a wrecked one.
567. 1:41:37All right. Copy this image. All right. So, we're putting in an image of a Mesoamerican fresco.
568. 1:42:05So we'll say from this messo American example let's
569. 1:42:20see identify and reconstruct this loss fresco or this this uh
570. 1:42:35dilapidated fresco. I don't know. Let's just see what it does. Thinking
571. 1:42:59Flipboard edu podcast. Lol. [ __ ] works. Yeah, this is pretty good. This is this is this is really something. All
572. 1:43:10right. Considering Mayan art, let's let's look at how it's doing its thinking here. Okay. So, the image
573. 1:43:16mentions a feathered Oh, look at this. The image mentions a feathered serpent and flowering trees which could relate to te
574. 1:43:27to oh tia tat haneo. How do they say that? Tiat t to tak hakan murals. Um but the user asked specifically about Mayan art.
575. 1:43:42There's a possibility this could be from Kak Ta with its mix of Mayan influence, but it's not purely Mayan
576. 1:43:56style. Let me do some more searching. Analyzed image. Hm. The swirling designs remind me of Mayan glyphs, possibly from
577. 1:44:03a mural. I'm thinking this could be temple of murals from Bonamp Pac. The image looks narrow on the right side. Fiery
578. 1:44:25serpent. So, this is interesting. I picked an image. I picked an image that was not Mayan and it and it busted me. And so
579. 1:44:38now it's it's saying, "Well, this doesn't match Mayan, but he said Mayan, so he must know what he wants. But let
580. 1:44:46me go see if I can find this other shit." In if if you've ever wondered, is the AI smarter than you are? [ __ ] yes, it
581. 1:45:02is. Okay, let's see. It thought for two minutes and 13 seconds. That's cool. So where did it end up? We need to ensure
582. 1:45:11cultural sensitivity while sharing our explanation. So here's the plan. I re researched code cotices like Dresden and
583. 1:45:19Madrid to study water scroll motifs using pigment analysis. Cineabar for red, Maya for blue, Maya blue for blue and yellow for geoite.
584. 1:45:31G. Oh, goite. I made sure the design mirrors symmetrical compositions of kakla murals. Now I'll put together a prompt for
585. 1:45:43reconstructing feather serpent mural. Keeping it naturalistic. The scene will depict Ket Kalddle or whatever water scroll motifs.
586. 1:45:59The bird warrior at top the serpent's bat, the sea creatures below. Huh? All right, I'm learning [ __ ] while we're making
587. 1:46:10pitches. This is [ __ ] crazy. Make a thing that doesn't suck. Please make a thing that doesn't suck.
588. 1:46:25Oh, that's gorgeous. All right, so let's go look. Let's download this and then let's go. Click on this. Copy it. Save image as
589. 1:46:40um fragment. We'll call it to downloads. It's fine. And then we'll go bang bang. And then we'll
590. 1:47:06go. All right, that's pretty cool. Can be used in midjourney as an SREF. Probably. I want to animate that.
591. 1:47:21All right, I'll post this to uh I'll go post this to the You know what I'll do? I'll post Oh, you know what we should also do is So,
592. 1:47:30we not only have the um the image, but we have the the writing of it. Why this fragment is from this
593. 1:47:45place. Oh, this is so cool. All right. So, let me grab
594. 1:48:06this. I'm going to go to the salon and I'm going to go to the AI learning lab. Boom. Pop that bad boy in there.
595. 1:48:40Okay, there that is. And then we'll post that. Don't notify everyone. And then I'll say um here was the uploaded
596. 1:48:58image and copy pasted prompt. um that inaccurately defined or uh I
597. 1:49:15inaccurately uh said this was Mayan art. It wasn't. and 03
598. 1:49:30corrected the mistake before the reconstruction. Here is the prompt I used
599. 1:49:45from and then I'm going to put the link to the article. Look how good I'm doing here. You kidding me? How do I get uh
600. 1:50:00link pin your article, edit article, share, copy link to article back here.
601. 1:50:15Boom. Here's the prompt I used. And then my prompt was Uh
602. 1:50:33here. And then we're going to get rid of that. And then we're going to add in the fragment that I uploaded. Comment. All right. So there
603. 1:50:45you have it. You've got there's what it generated. And then in the comment is the image I uploaded and the prompt I used. Also posting your
604. 1:50:55article prompts there for the weekend. Wait. Oh, he just did it. Okay, good. Beautiful. So, there's that one. That was cool. That was really
605. 1:51:09cool. Let me edit this for a second and say um edit Cool.
606. 1:51:2403 experiment. There we go. Save it. Beautiful. All right. There you have it. So, if you
607. 1:51:39haven't been to the AI salon before, you should join. Is it true that Tik Tok shuts down on us on Sunday? It was a big broadcaster that
608. 1:51:48said it. Uh it I would absolutely not be surprised if it shuts down on Sunday. I would assume that it's a negotiating
609. 1:52:00tactic. So don't be surprised if there's some big announcement um that you know someone saved the day and it doesn't shut down. But also,
610. 1:52:12don't be surprised if it's just [ __ ] shut down. This is why. So, if you're on Tik Tok right now and you don't know
611. 1:52:18about the YouTube channel, go to YouTube and subscribe to the salon or no, learning lab-ai on YouTube. That's my YouTube
612. 1:52:33channel. So, I've been dual broadcasting. They call in the business they call it simalcasting. I've been simal casting on
613. 1:52:43Tik Tok and YouTube and X and LinkedIn but primarily YouTube for about four months now. Um because I just assume the Tik
614. 1:52:53Tok shit's going to go away. All right. So it it may I don't know anything more than you know. Um, but it seems like apparently 175%
615. 1:53:07tariffs have pissed some people off. Oh, dang it. I hit my limit on 03. So, Cindy [ __ ] that means you're
616. 1:53:18winning. And just flip down to 04 mini. Like, just keep playing with the other models until you max them all out. I
617. 1:53:24like run yourself out of being able to do these things. I think it's really important to start experimenting with
618. 1:53:31what these things make possible. I mean that that reconstruction of that image with all the research it did where
619. 1:53:40I asked it about a Mayan reconstruction and then it identified that's not what it was and and recreated, you know, this
620. 1:53:47what it might have looked like. That's pretty [ __ ] insane. Let's do one other thing before I go. Let's
621. 1:53:57say this is awesome. Oh, wait. Sorry. Am I still sharing? I think I am. Am I? Okay. This is awesome.
622. 1:54:12Um, now I want you to create a photograph that looks like a
623. 1:54:26National Geographic photographer went back in time and
624. 1:54:41photographed this fresco um in its actual
625. 1:54:56location, you know, on a wall or building or what may be appropriate. It should be
626. 1:55:11a wide image captured at magic hour just before sunset.
627. 1:55:30have high contrast and look incredibly realistic. I want the
628. 1:55:46viewer to feel as if they are there. That's good enough. All right. Boom. It's going to be cool. It's going
629. 1:56:00to be cool, people. I promise you, it's going to be awesome or it'll suck. It's been fun and unfortunate. Unfortunately,
630. 1:56:07I have to jump off. Have a great weekend. Thank you, Steo. Have a great one. Yeah, we're basically done here.
631. 1:56:13Steo, find us if Tik Tok is gone. Yeah, Steo. Well, Steo's here on he's commenting on YouTube. So, that's the other thing. Hang out in
632. 1:56:24the AI salon. hang out in the in the AI learning lab channel on the salon because then you can talk even when I'm
633. 1:56:30not live, right? You can you can keep the conversation going. So, am I late? Chemical farmer. Excellent. I'll make some
634. 1:56:41time. Mr. out there selling some stuff. Chemical farmer super late out with friends. Friday night date night. Good
635. 1:56:48night. Should I rewatch? Yeah, this is a good one to rewatch because we're talking about how important 03 is, the
636. 1:56:55new model, and then the ' 04 mini and 04 mini high, why they're important. And then I also shared an article that's got
637. 1:57:02all these example prompts that are non science and math prompts and we experimented with some of them tonight,
638. 1:57:10so it's probably a good one to rewatch. And then the link to the article article if you just go to my exac account at Kyle
639. 1:57:17Shannon sort of the second or third post down is this article and the article is called unlock 03's hidden superpower a
640. 1:57:33playbook for non of non- STEM prompts for everyday innovators to explore. Right. Oh my god, that's gorgeous. Holy [ __ ] Look at
641. 1:57:54that. That is amazing. It's been fun, informative. Okay, got to go. That's [ __ ] gorgeous. Okay, I'm gonna I'm
642. 1:58:03gonna download this and post this into the into the comment with this with the prompt. So, I'm going to copy
643. 1:58:17that and then I'm going to download that and then we're going to go back to here and I'm going to throw in another comment and say um I then
644. 1:58:30asked to place the frisco in situ. How do you spell situ? Is it sit tu?
645. 1:58:45Anyone? I'm such an idiot. Siri, what? Oh, come on.
646. 1:59:04[Music] What? I am so done with [ __ ] um should have quit while you were ahead. I know, right? Mighty Networks just sucks. What
647. 1:59:17What is it with these sites with their stupid [ __ ] upload? It's like they got an intern to make their upload
648. 1:59:38[Music] service. There we go. in situ. Yeah. Okay,
649. 2:00:05great. And then what I would actually do, the next prompt I would give it is I would make it make that image with
650. 2:00:12people in it, you know, dressed in the appropriate garb. That's [ __ ] amazing. And the fact that it's even, you know,
651. 2:00:22potentially like culturally and historically accurate based on that [ __ ] [ __ ] up fragment of a thing to
652. 2:00:30go from that to get the education about what it actually is and then end up with with that photo. This is [ __ ] bonkers,
653. 2:00:41man. All right, I'm leaving you. I'm leaving you for the weekend. Your homework is clear. kick the [ __ ] out of 03, 04 mini, and 04
654. 2:00:52mini high. Okay, just go play. Go, you can use the prompts in my article. Go come up with your own. Cindy [ __ ] did an
655. 2:01:01amazing thing today. She took my article with all the prompts in it and she took information about her client that she's
656. 2:01:09working on and she goes, "Keep all the same basic use cases, but personalize every prompt for my client." And so it
657. 2:01:18gave her that whole example of all the different things you could do with 03 specific to her client. [ __ ] genius.
658. 2:01:25Like get creative and just kick the snot out of these models. And then when we come back next week, I want you guys to
659. 2:01:34like share stuff. Here's the other thing. Share what you're learning. So, just like I shared this thing in the AI
660. 2:01:39salon, figure out an appropriate channel wherever you want to share it. You can share it here. You can share it up in um in uh
661. 2:01:53Duh. Look what I made. All right? Or you could share it in the water cooler. So water cooler is just a general
662. 2:02:06chat. Look what I made is like showand tell area. Oh, that's beautiful. Who did that? Sharon Crawford. That's [ __ ]
663. 2:02:22gorgeous. And then um and then AI learning lab. You can share it in there. All right. Let's teach each other what this thing's good
664. 2:02:35at. Don't just hoard the information. Don't just be a lurker. Generously lead. That's one of our values. It's one of our imperatives
665. 2:02:46within the within the salon, right? The cycle to AI readiness. Play first, mindfully create, generously lead. So,
666. 2:02:56just go play with this thing. Once you figure out it can do something cool like we did tonight, mindfully create it and
667. 2:03:02then share what you did. Critical right now. Okay. All right. Peace out. Have a fantastic weekend. Um I hope you had fun. I did.
668. 2:03:13This was a good night. This is a good high energy night. And uh I appreciate you all more than you know. All right. Have a great
669. 2:03:22weekend. Bye.