5/9/2025 - Exploring GenSpark AI Sheets and GPT-4

Transcript

1. 0:01You ready, Champ? Champy, are you ready? Are you ready? [Music]
2. 0:21[Music] You and I get all alone. Sunday morning here at
3. 0:36home. Sky's blue and the coffee is strong and it's true. [Music] Then I open my eyes to a dream realized in front of
4. 0:52me 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.
5. 1:07Happy got to get a little rest and relaxation. Happy like fire on Sunday morning singing true. [Music]
6. 1:44[Music] Oh, is this a place I can rest my
7. 2:03forehead, gather my thoughts in Swiss silence. And there's this place where the feelings
8. 2:18on all culture through violence. Friday night date night. It just seems like last Friday we had one of these. Is
9. 2:28Is it like every Friday? What are we doing? Every Friday. Friday night date night. All right. [Music] How do I get rid of that
10. 2:45sticker? Uh, there we go. Peel it off. It's happening. Danielle Silver Fox shaking. Silverf Fox shared the live
11. 3:01because Silver Fox a mench is a mench. She's awesome. She's a live sharer. When was the last time you shared a live? Are you just like a
12. 3:12lurker? You just come in here to lurk, judge me silently or out loud, but not in the comments. That's what I would
13. 3:28do. YouTube comment. I see no YouTube comments. Ah, there I do. Champion is a vibe vocalist for sure. [Music]
14. 3:59[Music] Oh yeah. Champy. Champy. Champy. What are we going to talk about [Music]
15. 4:15tonight? Walt Grace desperately hating his old place. Dreamed to discover a new space. buried himself alive inside his basement. Turn on the
16. 4:29side of his facement. He's working away on displacement. What it would take to survive. Cuz when you're done with this
17. 4:42world, you know, the next is up to you. [Music] Um, you got to spank your guitar. You got to spank it. You got to spank
18. 5:06it. Wrong channel, Kyle. Ah, damn it. That was the twerking channel. Sometimes I forget. Am I on the twerking channel
19. 5:14or am I on the the the brainy one? the where we're talking about the intelligence, artificial intelligence, not the real intelligence.
20. 5:26Clearly, that's not happening here. [Music]
21. 6:12[Applause] [Music] How about new Google AI? Which which one, Silver Fox? The new image one or the new 2.5 something?
22. 6:27What's the They got They got the one the one that's supposed to be really good at coding. Let's see. Gemini
23. 6:362.5. I think it's the the 56 preview. I don't know. We'll check them out. We'll look at that [Music] stuff. More importantly, are we sure
24. 6:49we're in the right one? Oh, you mean for Tik Tok for the for this channel? Um, I I was thinking about having this channel
25. 6:57just be a crossover. I just merge the two and I just show up here one night in a manini and start doing some twerking
26. 7:05and just see how it goes, right? I don't think they'll ask me about my qualifications [Music] anymore. Source camp. Nice guitar
27. 7:18playing. It's lulling me to sleep. [Music]
28. 7:49Thanks for letting me in today. Well, of course that's what office hours is all about. [Applause] [Music]
29. 8:19Hoo! Doo doo doo doo. [Music]
30. 8:49Hey, good. [Music] [Applause] [Music]
31. 9:04Did I get accepted into the new notebook LM beta app? No, largely because I didn't fill out the form. Good reminder, Danielle. That's
32. 9:17solid. Solid of you. Let me see if I can go find it. Here it is. We're going to go fill that out right now. No need to share this. I'm just
33. 9:27filling out a form. Listen, I'll be with you people when I get with you people. Do you understand what I'm saying? Do you think
34. 9:34this is easy? You think it's easy being [Music] me? All right, here we go. Let's see. I can actually see now. That's nice. Let me do
35. 9:51my black bar for the for the good Tik Tok people. Sorry. Which AGI do you think is best? Well, many would argue
36. 9:59we're not at AGI yet. Um, but hang on. Let me fill out this form. Kyle, please provide the email address you'd like us
37. 10:07and yeah, I already did that. Kyle. Yeah. Yeah, there we go. What country are you in? United States. USA.
38. 10:18Um, I'll be testing iOS. I agree. What will be next? There we go. All right. I just did that. [Music]
39. 10:51Producer Brandon. Producer Brandon, give me a little shade in the background here. I like it. I like it. You gotta
40. 10:58You gotta keep You got to keep the hosts on their toes because they'll just go rogue. They'll just go out and do [ __ ]
41. 11:05Um, the question about which AGI do I like best? Which which AI do I like best? Yeah. I mean, you could probably argue that
42. 11:1903, Manis, and Gen Spark are what Sam Alman call was it Sam Alman? Someone from open AI said spark you know sparks of AGI or you
43. 11:32know hints of AGI whatever whatever it is. You could argue that that the new reasoning engine from OpenAI Manis and Gen
44. 11:42Spark start to live in the neighborhood of AGI. But the the definition of AGI that that I am going to go with
45. 11:57is you know a system that's smart enough to do you know most economically viable jobs to do them to do the jobs right not
46. 12:08to have a human sit there and babysit it so that it does 80% of a job and then kind of halfass does the other 20% and
47. 12:16humans like, "Well, this sucks because we that's what we have with computers now, the this sucks thing." So, I would
48. 12:23argue we're not at AGI yet, but you know, we're we're starting to get hints of it's going to be pretty [ __ ] cool. But, good question.
49. 12:34[Music]
50. 12:49[Applause] [Music] She came on here like slow moving cold front. Well, be was warmer than a look
51. 13:06in her eyes. She sat on the stool and he said, "What do you [Music] want?" She said, "Give me a love that
52. 13:23don't freeze up [Music] inside." Said, "I have melted some horse in my time, dear. [Music]
53. 13:38But to sit next to you, well, I shiver and shake. And if I knew love, well, I don't think I'd be
54. 13:55here asking myself if I've got what it takes. I do melt your icy blue heart. [Music]
55. 14:13Should pass turn what's been frozen for you into a river of [Music]
56. 14:32tears. Oh, that Champy loves his singing. when you get a song just right, he just jumps in. Um, so what we're talking about on the
57. 14:44live today, we're talking a little bit about um, we were talking a little bit about uh, Gen Spark came out with a new thing
58. 14:52called Sheets. Gen Spark Sheets, which we'll go play with. If you haven't heard of Gen Spark or played with it yet, it's
59. 14:58pretty remarkable. Um, and they're they're rolling out features pretty damn quick the past couple of weeks. Um,
60. 15:05they're really good. We were also talking about I'm still semiobsessed with this email from the CEO of Fiverr to his staff.
61. 15:16um that basically talks about like what he basically says in it, you've got a few months to get your ass aggressively curious about
62. 15:28AI or you're just going to have your ass handed to you. And uh he he said it plainly and clearly and you know people are arguing
63. 15:38arguing whether he did it right or not, but like it was at least clear. It had some urgency. like it's it's the first it's the
64. 15:48first, you know, communication of note. Um, you know, other than people like Sam Alman who are building the
65. 15:58[ __ ] but he's even understated, but it was the first thing of note that had some urgency to it. So, I actually liked
66. 16:04it. Did you see the image I made for you? It's in Look what I made. I did not. Let's go look. There's one from Vicki. Oh, that's
67. 16:17nice. Photo realistic unicorn. Nice. Nice. Vicky launched a lovable
68. 16:32app, the visual prompt generator. Very cool. Nice. Steo, the new pope. If anyone's offended, take it
69. 16:54down. Hey, it looks like Champy. They got the eyes right. [Music]
70. 17:32Put it here. [Music]
71. 18:01Ah. Um, nah, I was not I was not that blown away with the with the There's a new image gen model from Gemini, but I
72. 18:11wasn't that that impressed with it. They said it was supposed to be better, but I thought the the text was a little eh it
73. 18:18was a little janky. So, not worth playing with because it's too hard to find. Just the the Gemini [ __ ] the
74. 18:25Google stuff is just ridiculous to find. What are we going to do tonight? Um, oh yeah, that's a good thing. So, if you go to [Music]
75. 18:42uh if you go to OpenAI and search for GPT enterprise models and limits, go to help.openai.com
76. 18:57openai.com and then on YouTube I'll post in this URL on Tik Tok. You're on your own, but you can go I mean you can see it. You
77. 19:08can screenshot it and see it up here. Um but they put out a list of when to use each model. Now, let me give you my two cents
78. 19:21on this stupid [ __ ] thing. We were promised by by Mr. Altman himself that GPT5 was going to come out
79. 19:32in May and we wouldn't have to deal with all these [ __ ] models. That they were going to have some sort of magic model
80. 19:40picker. Well, the only thing they've been doing is picking their nose apparently because GPT5 ain't coming
81. 19:48anytime soon. Probably the summer. And now they've they've put out a document that says here's when to use each
82. 19:56model. The whole point of having magic AI is that we don't have to use our brains. So this is not this does not make me happy.
83. 20:08But GPT40 excels at everyday tasks. Brainstorming, summarizing emails, and creative content. Fully multimodal.
84. 20:16Supports almost all capabilities. GPTs, data analysis, search, image generation, canvas, advanced voice, and
85. 20:26inputs, documents, images, CSV files, audio, and video. Wait, we can upload video to chat GPT. Hang on a second. Hang on one go dang
86. 20:40second. Let's go test this. Oh, look at my look at my superhero. I don't know why it put [ __ ] Washington in the background,
87. 20:55but you know, cuz it because I had done one of those earlier in the chat apparently. All right, let's see the new
88. 21:03midjourney. Is there a new midjourney? We can look at the new midjourney, but wait, I got to see if this thing can
89. 21:08input videos. Let me start a new a new chatty chat. We'll go.mpp4. And honestly, we could grab
90. 21:40[Music] All right, this is [Music] good. Okay, so this is a little opening
91. 21:57montage for my Broadway musical that opened on Broadway over the winter. It's not true. Uh uh uh uh uh uh uh uh. Come
92. 22:16[Music] on. Oh, I wonder if I have to do the instead of the MP4. Ah, file formats. It was going just
93. 22:33slow. All right, this one's going now. Okay. White ring. Yeah, I know. It looked like it hung a little bit. Um I I don't know. Let's see. Um
94. 22:48what Let's see. Uh what is this video about? So at least it let me upload it. What is this video about regarding wildlife documentaries?
95. 23:03What is this video about? Um what where is the location? Oh, movie pie editor. So it's using a Python library
96. 23:20to analyze it. Extract a few representative frames. One duration. So it's basically every second it's grabbing a different
97. 23:29frame. Then it's doing ffmpeg parsing. Would you like me to extract key frames so we can visually identify? Yes. I
98. 23:39thought we I thought we established that when I said, "What is this video [Laughter]
99. 23:53about? Thank you, Winston, for the gift. It looks like people have been sending some gifts over here on on the Tik Tok.
100. 24:00Ion China seems to think there's a new midjourney. Vicki seems to think there's not. I used to work on Fiverr. I can't
101. 24:14believe they're still in business. Uh, definitely the 20% of not AGI. All right. What's this doing? Did we do it?
102. 24:28Here's the first frame. Here's the first frame of the video. Extract it at the 1 second mark to help identify the
103. 24:38location. Yes. Stop asking me if you should do your job. Uh so it's so it's doing code
104. 24:54analysis. Just tell me
105. 25:10what's in the images. So this is not if you want to see if you want to see what multimodal video looks like, it does not
106. 25:24look like this. This is [ __ ] Ah, but it did finally get it. Okay. Frame half a second. A wide view of Times Square, New York
107. 25:35City. The camera zoomed in a bit. The AI entity is more prominent. The visual style emphasizes futurism, artifice, and power, almost
108. 25:47like a goddess of information. The video is a stylized promo intro set in Time Square showcasing a vivid abstract female AI figure broadcast
109. 25:59against the massive LED screens. Want me to create a cinematic description? I'm going to say based on everything you know about
110. 26:12me, what could this be a promo for? Mark, let's see if it gets Sydney. Sydney, based on everything I know about
111. 26:24you, your work in AI storytelling, your musical Sydney, your obsession with narrative tech, and your knack for
112. 26:30blending futurism with emotion, the video could be a promo for a Broadway show, Sydney and Artificial Love
113. 26:39Story. Oh, and then it guessed a bunch of other [ __ ] All right, so it did it. But here, let's go. Let's go show you
114. 26:46what real multimodal looks like. So, I'm going to go to Now, here's the problem. I don't know which of these
115. 26:53models is actually multimodal anymore. Let's assume that maybe this is. Okay. So, we'll do new chat. So, where I am
116. 27:01right now is I'm in Google uh aistudio.google.com and I'm in the Gemini 2.5 Pro. And let's go see. Upload a file. Yeah, I could have
117. 27:15put in a uh a YouTube a YouTube video there. So, that's promising. Is this it?
118. 27:31No. MP4.
119. 27:46Why can't I not find it now? Oh, that's [Music]
120. 28:07why. Okay, this is it.
121. 28:26[Laughter] Steo, I have another another image of you and Champion on the opening night of Sydney on Broadway. All right. What's this thing
122. 28:36doing? So, this is uploading the file. It looks like there's going to be a video preview here, which is interesting. I tested like Gemini in
123. 28:50Google Studio probably nine months ago, you know, could fully understand video. So like this is this is doing it.
124. 29:05So, okay. Extracting. So, it's taking it a while. 14,000 tokens. Oh, type something. I'm going to say, okay, um, give me a shot
125. 29:25list of the video. Um, what creative work is being featured? We'll try something else here.
126. 29:41And the style of music in the background. Expanding to view model
127. 29:57thoughts. Oh, expand to view model thoughts. Yeah. So, look at what this is doing. Um, 01 black screen. 01 to 03
128. 30:05Sydney and artificial love story by K Kyle Shannon and Andrew Watts. Image, woman's profile, color hair. Um, 03 to
129. 30:1204 blurry title card. 04 to05 times square like scene, red carpet steps, crowds, billboards. This is what a native video model looks
130. 30:22like. Consolidate and refine the shot list. So, this is it thinking. So, it's using reasoning. Identify creative work.
131. 30:30Title: Sydney and artificial love story. Authors Kyle Shannon and Andrew Watts. Key visuals. Theme: Analyze musical
132. 30:39style: drums, bass, electric guitars, clean and slightly distorted. Okay. The video is a promotional piece for a fictional
133. 30:49Broadway musical. What do you mean a fictional Broadway musical? Gemini is an [ __ ] It's a real Broadway musical.
134. 30:56It's just just we haven't it's just not on Broadway yet. Damn it. I hate it when the AI is smarter than [Laughter] me. Title
135. 31:11um type Broadway musical implied by the play bill. Theatrical marquee. The poster features a woman's profile with vibrant almost digital
136. 31:21ethereal colorful hair looking toward a small distant silhouette of a man. and the background is warm theme implied.
137. 31:30The title and artificial love story in the digital ethereal look of the main female character in the artwork strongly
138. 31:36suggest themes of artificial intelligence, technology, and romance. It's quite good shot list. Amazing. That's what a real multimodal
139. 31:51response looks like. That's very cool. All right. So, that's again in a studio.google.com. And then the model
140. 32:01that I picked was Gemini 2.5 Pro Preview0506. And I had temperature set at one. You're like, what's temperature?
141. 32:13Why do I need to deal with this [ __ ] Because this is where the stupid AI industry is right now. They can't figure
142. 32:20out if they want to be for developers or human beings. Not that developers are not human beings. Listen, I've known
143. 32:26some very nice developers that seem almost human, but like normal people shouldn't have to deal with this
144. 32:37[ __ ] Do I want to have structured output or function calling or grounding with Google search? What's that
145. 32:44mean? Can't I just stick stuff in the chat hole and it vomits out brilliance? Isn't that what we were
146. 32:57promised? I want flying cars and I want this [ __ ] to work. Inspired by actual events. Yes. Okay. So, we haven't done anything yet.
147. 33:13Oh, wait. You know, we were doing something. We were looking at this was a rabbit. This rabbit h brought this rabbit hole brought to you
148. 33:21by chat add. Welcome to chat add where we might occasionally finish a sentence. Pinkies out everyone. We're classy.
149. 33:37Okay. Fantastic. Fantastic. Yeah, that's really good. That's really good there, people. Yeah. Let me tell you something. All right.
150. 33:49GPT40. So it inputs, it can input documents, images, CSV files, audio, and video. So it does input video, but then what it does is it
151. 34:00just Okay, ffmpeg, if you don't know what FFmpeg is, it's a um command line video processing library that's been
152. 34:11around for decades. In fact, my company, Storyvine, a lot of the tech we use leverages FFmpeg. So, it's been around
153. 34:21forever. Um, so what that says to me is it can input the video, but then they're just using FFmpeg to analyze it. So,
154. 34:29it's not it's not true um video multimodal, but it is image multimodal. Uh, and it's audio multimodal because
155. 34:38that's where the advanced voice comes from. So, there you have that. Okay. So that's 4.0. You use it for everyday
156. 34:46tasks. I mean, I just use 4.0 for almost everything. 4.5. Ideal for creative tasks, emotional intelligence, clear
157. 34:55communication, creativity, and a more collaborative, intuitive approach to brainstorming. Create an engaging
158. 35:03LinkedIn post about AI trends. Write a product description for a new feature launch. Develop a company apology letter with an
159. 35:13empathetic tone. All right, let's go try that. So, we're going to go to chat jet. We'll go to 4.5. We'll go um write an
160. 35:28empathetic apology letter. This is bad. This too soon. This is this is before it's happened. uh informing uh this is bad employees
161. 35:43um that um 90% of them aren't necessary.
162. 35:58Yikes. Let's just see how it does. Let's see how [ __ ] empathetic you are there, chat. GPT 4.5. This is
163. 36:08bad. Sorry, I have no qualifications. They're fired. You're fired. Here's a carefully worded, empathetic apology
164. 36:16that acknowledges the difficult news clearly and respectfully. Difficult but necessary changes team. This is incredibly hard
165. 36:23message to share. I want to start by sincerely expressing my deep regret and empathy for the impact it will have on
166. 36:30many of you. After careful consideration and extensive evaluation of our organizational structure and future
167. 36:36direction, we've reached the painful conclusion that approximately 90% of your current roles will no longer be
168. 36:45necessary. All right. So, that seems very corporate. So, let's we'll do a couple of variations here. Let's do one
169. 36:52that's like um the uh the Fiverr CEO talked about radical cander. Um, let's do one from a CEO who
170. 37:09practices radical cander. Uh, oh, did I hang it? I think I might have hung
171. 37:23it. Hello. Oh. Oh. Oh. because I'm in uh I'm in a canvas. So, it's rewriting it. Team, I'm going to be direct because honesty and
172. 37:36clarity are essential, especially during challenging times. After rigorous analysis of our business and future
173. 37:44trajectory, we've concluded that 90% of our current roles are no longer aligned with our strategic goals. This is a
174. 37:50stark reality we must confront openly. To be I want to be absolutely clear. This decision is not about performance
175. 37:57or capability. Well, it's about the performance and capability of the [ __ ] technology, isn't it? You douchebag. It's about adapting
176. 38:05aggressively to a changing market dynamics. That still sounds like horseshit. So, let's go um let's go full quotes
177. 38:16Lumbberg from Office Space. apology letter. I write an apology letter to Ann. Ann posted her her smoking picture on LinkedIn. So, I think
178. 38:30Ann's fine. Hey, team. Yeah. So, I'm going to need you to go on ahead and listen up. We've run some numbers, looked at some
179. 38:39things, and uh it turns out about 90% of your jobs just aren't going to work anymore. Okay. Now, don't get me wrong. It's not
180. 38:49that you're doing anything particularly bad. It's just we're kind of shifting gears here, aligning with our um strategic
181. 38:57stuff. You know how it is. So, I'm going to need you to go on ahead and clear out your pretty desks soon. All right. So, I guess 4.5's good
182. 39:08at at letter writing and [ __ ] like that. All right. Let's go back here. Um okay. 04 mini fast technical tasks quick STEM
183. 39:19related queries programming visual reasoning provide a quip quick summary of a scientific article. Oh, that's a
184. 39:30good one. So, let's go grab a let's go to archive a rxiv, right? Isn't that how it
185. 39:44[Music] What was the latest one? There was a new one that just came out. That's looks like it's going to be the new
186. 39:51Transformers paper. Oh, I can't remember what it was now. Damn it. Uh uh uh I forget. Physics, mathematics, computer
187. 40:08science. Uh anyone remember that one? 40 rocks grounding with Bing in wait in get on the bus or get off the bus
188. 40:20style. When is Sydney going to Broadway? I Mr. It as soon as if you got 20 first $20 million take Sydney to Broadway. I
189. 40:30don't know what we're [ __ ] around waiting for. If you got $20 million sitting around, just DM me on TikTok.
190. 40:38You say, "Hey, dude. I'm ready to go to Broadway. What's your wiring information?" It's all we need. It's all we need. Okay.
191. 40:51Okay. Which one did you find there? Brandon found something. Oh, I see what you did. You just did a search. Okay, let me
192. 41:06look. Echo agent, pain assessment, neural networks. There's just too many of them. Uh, let's let's do search for uh AI
193. 41:23Google. Well, it doesn't [ __ ] matter, does it? Let's just go grab one PDF. Wait, but this doesn't have visuals in it.
194. 41:40[Music]
195. 41:58Um, dang it. I want something with pretty pictures in it. Ped pediatric asthma detection. Give me pretty
196. 42:13pictures. There we go. All right. This one has pretty pictures. Beautiful. Oh, charts, graphs. Lovely. Okay. Download. Uh, whatever.
197. 42:28Desktop. Go back here. Go here. Go here. We're going to go 04 mini. We're going to upload this thing. Um, how how did they
198. 42:43put it over here? Provide a quick summary of a scientific article. Um, provide a quick summary of this and be sure to
199. 42:59explain the most important charts and graphs and what they mean in context of the summary.
200. 43:16I'm a dum dum. I have no [Music] qualifications. Make the
201. 43:31summary quote me proof. Sifting sifting through archive is an art. It's called the abstract. Most papers come with one. Okay, good.
202. 43:47Great. Oh, the summary. Yeah. No, I understand that. I But I'm just I'm just looking to see if this actually does the does the
203. 43:56stuff. Take a two second clip of a child breathing. Turn it into 512 number sound fingerprints via Google
204. 44:05here. Thought for 60 seconds. The study builds a super simple asthma detector for kids using short two-cond breath
205. 44:14recordings and audios Google's big audio model. Here, here's what happens. Top for performer separation in 2D space. Bottom
206. 44:27line, figure nine, page six. Figure eight, page six. Figure seven, figure six. Figure six. Figure 1. So it it
207. 44:37didn't use all of the the figures, so it picked the most important ones. All right, that seemed to work. All right,
208. 44:44fine. So that's 04 mini. And then 04 mini high detailed technical tasks, advanced coding, math, scientific
209. 44:53explanations. Um, thinks longer for higher accuracy. Hey Pete, give me a uh give me something I can give 04 mini
210. 45:01high to explain something about tensor tensor optimization and then you can tell me if you think it's full of [ __ ] So we're
211. 45:10going to go 04 mini high. We're going to go new chat. If you're still in here,
212. 45:40Pete, I think we might have lost Pate. Did we lose Pate? He's buffering. Okay, I'll I'll start something and then if he comes up with
213. 45:53something good, I'll change it. So I'll say um explain the math that is
214. 46:09required to optimize tensor processors for both inference
215. 46:26and training of generative AI. Wait, explain how Ah, here we go. Ah, that you know what is you know what's funny, Pate? I
216. 46:36was just about to type that. I wrote something different, but I was just about to write explain how to use
217. 46:41sharding and jacks across multiple devices. That that was it was like I was going to do that one, but I thought it
218. 46:49was too obvious. So, but I'm glad you you put it in here. It confirmed what my instinct was. Explain how how to use sharding. The
219. 46:58only thing I know about sharding is, you know, you got to change your underwear. Okay. Explain how how to use Yeah, we go there. We go
220. 47:10there on this channel. sharding um in jacks across multiple devices in
221. 47:25a tensor parallel manner. All right. Should I have it explain the math? Wait, we're on the same same wavelength. It's spooky.
222. 47:40It's we're really sympatico there. Just always thinking about the same uh same mathematics frameworks, aren't we? The
223. 47:48user is asking about using jacks. Okay. Step by step. Oh, it's doing stuff with with Python. Build your device mesh. Okay.
224. 47:59So, basically what it says is one, carve up your devices into a named mesh. Two, declare how each array should be sliced
225. 48:08via partition spec. Three, put your arrays onto a mesh with the right sharding. Write a PJIT wrapped function pit
226. 48:20function that respects those shards. So, devices npray jacks. Devices mesh equals mesh. Define how you want to
227. 48:31shard your tensors. Do do you do you really use this language and not laugh like a like a junior high school kid? Shard your shard your
228. 48:49tensors. Define how you want to shard your tensors. Excuse me sir. Use partition spec to say slice this axis across model devices. You can
229. 49:01ask it about how it would use that to do something like a large matrix multiply. Okay, so let's do that at the Okay, so
230. 49:10Oh, wait. There's more. Run it inside a mesh context. Bonus older PMAP-based approach. If you only have a 1D split
231. 49:19and want a quick and dirty tensor parallelism on end devices, you can manually slice it in PMAP. You know,
232. 49:27I've been manually slicing my shards for I probably three, four years now. I cut my finger once. It was very painful. So, shards are acidic.
233. 49:39Um, okay. What did you say here? If you tensed your shards, it would have gone better. Okay. You can ask it about how
234. 49:47would you do that in something? Wait. Okay. How? Wait. sort of explain how um and include explanations of the math maths is what
235. 50:03we're supposed to what oh how would you how would you use that to do something like okay wait how explain how you would do
236. 50:18some like a Large matrix multiply. Large matrix multiply.
237. 50:32I usually talk about large matri matrix multiplication when I when I teach I substitute teach uh second graders
238. 50:40advanced math a lot and and that we're often talking about large mat matrix multiplication you know for matrices
239. 50:48like a n to the n b n to the n you know what I mean below is how you take a huge matrix multiply x= y = xwa
240. 51:02X equals I think it's sum R fancy R B to the A B * H raised to the B * H LinkedIn question Tim Lloyd Kyle happy Friday
241. 51:15what do you know about Enzo AI agents anyone um I don't know anything about Enzo but we can go look them
242. 51:23up see the problem with me having it explain the math is I can't even read the [ __ ] formulas this does this look good. Yes,
243. 51:32this is how you would do it. Pjit is outdated though. My dissertation was about large matrix operations. Wow,
244. 51:43wild. Um, so I'm going to I'm going to tell I'm going to say I'm going to say, hey, listen. I have a um large
245. 51:59associate who informed me that PIT wait PIT is outdated. Get with the program.
246. 52:18Straighten up and fly right, you stupid. You big dumb large language model. Straighten up and fly right.
247. 52:33Okay, let's see what it says. Enzo.bot's agentive marketplace at 49 bucks a month. Oh, yeah. I don't the the the there may be
248. 52:44commercial agent marketplaces out there that are so both Claude, Gemini or all three claude, Gemini and Open AAI all have some
249. 52:57agentic things in their APIs that you can go start making agents. Um so they might be very real, but I I
250. 53:05don't I don't know. I'm kind of waiting for one of the big three to to drop the sort of the chat GPT equivalent for agents,
251. 53:14right? Like custom GPTs were where you don't have to know any programming. You can just build them. Like we don't have
252. 53:20that for agents yet. So, I'm just kind of waiting for that thought for 20 seconds. The core idea. Yeah, Pate. I I
253. 53:27think um chatbt is smacking you down here. The core idea hasn't changed. The heavy lifting is still split the big sum
254. 53:37to the h n= 1 x to the i comma n w to the nj across d devices. I mean duh, right? That we knew like you still got
255. 53:48to do that. Then you got to compute each partial sh sum on its shard. Honest to god, you really do need
256. 53:56a gen xer in the corner going, "Guys, you got you've got to change the name shard. You just you can't
257. 54:03You you you have to change it. You have to change it. You're you're adult people talking about sharding things. You can't do
258. 54:12that. Okay. Then you have to all reduce pum the d partials into the full result. What has modernized is how you tell Jax about your
259. 54:24shards. As of Jax 0.4.11 plus you import pjit directly from Jax. No more jacks.experime. experimental. Jack's
260. 54:33experimental was always temporary. I I think it's in the name. It's implied. All right. Hey, chat GBT. I
261. 54:41know the underlying idea is the same, but there's easier ways to do it. Oh, wait. That's good. This is good. This is good.
262. 54:48Thanks, P. You're This is good. You're like my uh You're like my hitman here. Okay. Uh I understand the underlying uh what' you
263. 55:01say idea is the same but there are easier ways to do it. Think
264. 55:17harder. Let's not be ninnies. thinking. See, you know what the problem
265. 55:32is with doing something like this on my channel, like on in my account, is chat GPT is now going to think I understand
266. 55:39any of this [ __ ] and it's going to start including mathematical equations in my screenplays. I don't like it. Math in three lines.
267. 55:48Okay, here's the simplest, don't overengineer it way to spread a big X Y equals XW X math equation across D devices using
268. 56:03PMAP. You write that then you break it into D equal chunks H divided by D. Then you recover in full y by y equals sum d minus1 dals
269. 56:170 y to the parentheses dig which is just all reduce over the partials. Did it do good? Gemini agrees the functionality is now baked directly
270. 56:31into jacks.jit maths of the humanity chunky shards. Exactly. I think listen here's what we've learned
271. 56:39tonight. Sure. 04 mini high can probably do some analysis of your sharding in the tensor optimization game, but we still
272. 56:49got to change the name of sharding. We've get you it seriously needs a rebranding. We can't be saying it with a
273. 56:55straight face. It's not appropriate. It's not professional. All right. Um Okay. So now that we have that out of the way,
274. 57:11um, sharding is dangerously close to sharding and it is
275. 57:26making non engineers giggle. and uncomfortable. We need a rebranding and we need it
276. 57:44now. Jeff on YouTube, I know you understand what you think I said, but I'm not sure you realize what you heard is not what I meant exactly.
277. 57:55Okay, let's let's shoot sharting off into a into the bit bucket and give our multi-device split a fresh ungly
278. 58:06identity. Here are five contenders. Tiling, API sketch, tile mesh, tile spec in tiles, slicing, partitioning,
279. 58:17spreading. Spreading is almost as bad, and stripping or striping. I like tiling striping. I don't know.
280. 58:28Anyway, I can't see everything, but the ideas are about all right in a kind of hand wavy, so it's not wrong. Oh, that's
281. 58:35cool. All right, good. All right. So, so that's what OpenAI claims that's good for is fancy math. And then open OpenAI
282. 58:4303 complex or multi-step test, strategic planning, detailed analysis. Hang on. extensive coding, advanced math, science coding, visual
283. 58:58reasoning, and then 01 pro complex reasoning takes a bit longer, enterprise limits and capabilities. Anyway, all right. So,
284. 59:07there we have that. There's when to use which models. Um, just use 40. Use 40 until until it breaks and then try some other [ __ ] That's
285. 59:19basically it. We need a Mother's Day palette cleanser. Um, yeah. Let's see. What do we want to do? Oh, you know what I heard
286. 59:29today from Brent Brent Peterson? Um, let me go back to 40 and I'm going to say, um, I want you to come up
287. 59:44[Music] with 20 novel Mother's Day card concepts. 10 are
288. 59:58gutbusters funny. Gutbuster funny. and 10 are tearjerkers. And all of them have to be we should try
289. 1:00:17this in 4.5 as well. All of them have to be original in ways we can't currently
290. 1:00:31imagine. So, Sunday is Mother's Day. Music, poem, art. Yeah, something like that. All right. My eyes are getting all dry.
291. 1:00:43Something must be going on. Oh, you know what's going on? I think I think I think there's some pollen happening here in
292. 1:00:49Yield, Denver. I don't I don't have a lot of allergies, but my eyes are really really uh like someone threw pollen in
293. 1:00:59them. Okay. Gutbuster funny cards. Your womb, my first apartment. That's pretty good. Front blueprint style diagram of a
294. 1:01:09womb with stick figure furniture inside. Rent was free and the food was delivered introvenously and the landlord loved me
295. 1:01:17unconditionally. Best landlord ever. That's pretty cute. Mom, I googled you. Search bar that autofills with, "Why is
296. 1:01:24mom always right?" Turns out it's genetic. You pass down your sass, sarcasm, and supreme correctness. Mom math. One brain plus
297. 1:01:35five kids equals Huh. All right. Tearjerkers. The first voice I ever heard. Ultrasound wave. Okay. So, here's what I want to
298. 1:01:44try. So, it made a bunch of things. So, now I'm going to say make the first five cards covers.
299. 1:01:59So, the jobs at Hallmark are still safe for now. Yeah. Um, Brent Peterson was saying today in one of his backend systems, he
300. 1:02:10was running a test and I think he said he was using 03 and it decided it needed it needed to make nine images and he said it just
301. 1:02:22went and made nine images and didn't stop. So, I'm going to see if that's something that they have have added in
302. 1:02:28here. After it makes one image, is it just going to make another one? Oh, most of the terms when it when
303. 1:02:37it came up with new sharding terms, most of the terms were already used. Yeah, I was wondering about that. Tiling
304. 1:02:43certainly seems like something that would have been taken. Your womb, my first apartment. Well, it's a unique
305. 1:02:58card and it didn't do all five. Yeah. So, all right, that failed. Okay. So, what you should do is you should here. Here's I'll give you
306. 1:03:11homework right now for the weekend. And you can start doing this now. And maybe, in fact, maybe what you do is, you know,
307. 1:03:18for the rest of the little live here tonight, Friday night date night, make some Mother's Day cards and share them
308. 1:03:23in in the uh in the Irregulars channel on the AI Salon. Have you used Dreamface app? I used it for a baby talking video.
309. 1:03:33Pretty simple. No, I used I used Hedra. Um, right now I'm I'm kind of avoiding all of the apps because there are so
310. 1:03:43many apps that are being dumped into the app store right now that are basically just calling base models. Um, so I'm
311. 1:03:50just trying to avoid that for now. But I mean, the nice thing about those kind of apps is that they do make it really
312. 1:03:56simple. The downside is they're really [ __ ] expensive. Uh, James Duder sounds rat to
313. 1:04:11me. It's a website, not an app. What's it called? Dreamface app. Dream face app. And does it cost you anything?
314. 1:04:26[Music] [Music]
315. 1:04:59Freely swap faces. Voice studio comes. Clone voices. Convert. Advanced avatar has launched. Audio. Record
316. 1:05:14it. AI script. Um, talk about Mother's Day.
317. 1:05:32Plus photos and videos. Got
318. 1:05:48it. That's JPEG image. Why did it not like that? Oh, it did. It did like
319. 1:06:08it. Voice. Baby Rose.
320. 1:06:29generate one minute video was free than it costs. Okay. Yeah, that makes sense. Tab when it generates. All right, I'll go do the tab now.
321. 1:06:41[Music] I assume what this is using is, you know, APIs into Clling or or or Hedra or or some of those. It's probably using 11
322. 1:06:57Labs for the voice is my guess. I mean, quite frankly, nothing here yet. Generation. Oh, generating completed. Here we go.
323. 1:07:08That was quick. Is it any good? Let's see.
324. 1:07:31Why am I seeing something different over there? Oh, can't hear it. Oh, did I share the wrong screen? Oh, I see what
325. 1:07:37it did. It It spawned a new screen. I'm like, why can I not see it the same way I'm seeing it? and Mother's Day. This uh they're
326. 1:07:50probably using um they're probably using Hey Jen here because this is almost exactly what Hey Jen looked like. Jeff
327. 1:07:58Flanigan and Mother's Day card using the word sharding. Yeah, it's pretty good. Serves as a poignant reminder of the profound
328. 1:08:08influence mothers exert in our lives. It is a day dedicated to honoring their unwavering support, selfless sacrifices,
329. 1:08:16and unconditional love. Celebrating this occasion fosters appreciation, and gratitude, allowing us to reflect on the
330. 1:08:24invaluable lessons learned from these remarkable. All right. Yeah, it works. It works. It works. It works. Um,
331. 1:08:37yeah. I I just I don't I mean I can't keep up with the base models, much less apps like these. I It's just It's
332. 1:08:57like I'm trying to think I'm trying to think if there's a if there's a good strategy. I'm trying to think if there's
333. 1:09:05a good strategy to not be to be decently articulate about what's possible with AI. But it's like one of the things that
334. 1:09:20I've been experiencing in the past two months is that a lot of people that I know that are very educated about
335. 1:09:30AI haven't heard about major new things in certain sectors of AI. So, I think people are kind of narrowing into in I don't know if it's
336. 1:09:40verticals, but like there's too much to keep up with. So, I I don't I don't have a good strategy for it other than
337. 1:09:47just [ __ ] do everything you can to to keep on top of it because I there's just not a way to there's just not a way to
338. 1:09:53do it. What did you say it might be the same as? Um Oh, this this looks almost identical to Hey Jen. So yesterday we did Hey Jen.
339. 1:10:07So here's the Hey Jen video. Oh wait, hang on. I got to share it. Is too good. So this is Hey Jen from
340. 1:10:21last night. So yeah. So yesterday was my birthday, May 7th. And see how the balloon doesn't move and just his head.
341. 1:10:29A little bit of hand movement and just his face. You just want to write it down and put it in your calendar or whatever.
342. 1:10:33But yeah, yesterday I uh I uh I technically, you know, like on a calendar using math uh turn 60. So So that's Hey, Jen. But then
343. 1:10:51Hedra, come on. I got to switch this again. Hang on.
344. 1:11:07So yeah, so yesterday was my birthday, May 7th. And you know, in case you're in case you want to write it down and put
345. 1:11:12it in your calendar or whatever, but yeah. Yes. I mean, that's just a very different level of quality, right?
346. 1:11:21So yeah, so yesterday was my birthday, May 7th. you know, in case you're in case you want to write it down and put
347. 1:11:26it in your calendar or whatever. But yeah, yesterday I uh I uh I technically, you know, like on a calendar using math
348. 1:11:36uh turned 60, but you know what? So anyway, all right. So what I would say you should do over this here weekend
349. 1:11:51is you should go above and beyond using AI to make Mother's Day joy. Right? Make a poem, make a song,
350. 1:12:06make an image, make a video. Turn yourself or one of the kids into babies again. Turn those into videos.
351. 1:12:16saying the poem. I don't know. Like just do like a multimodal, you know, explosion of joy. Um, and have chat GPT help you figure
352. 1:12:27that [ __ ] out or whatever, whichever one you like. Uh, this is, uh, Hedra. H E D R A is how you spell
353. 1:12:39it. Yeah, this one's much better. Hedra, I think, is how they pronounce it. Hedra is much better. All right. What else? Questions?
354. 1:12:58Thoughts? Thoughts? It's Friday night. Nah.
355. 1:13:19[Music] Kyle, what did you use? What did I use for what? To make that baby video. Hedra. H E D R A is the tool that I used for the
356. 1:13:35animation. Check out my images in Look what I made. Fine. Look what I made. Look what I
357. 1:13:53made. Oh, nice. Oh, that's awesome. Oh, I'm not sharing. Should I share?
358. 1:14:09That is beautiful. That is gorgeous from from one Steo. Sydney on Broadway, baby. It's coming. Who's got $20 million?
359. 1:14:29Seriously. FO has been coaching me in several areas. Very cool. You said at the beginning that some site came out
360. 1:14:36with something new and we would look at it. I think what I was talking about was Gen Spark. We can go look at GenSpark. Let me I like this
361. 1:14:46image. Let me go Steo. I'm going to put your Steo put your uh your X
362. 1:15:01um put your X handle in in whatever you call it. Uh Kyle Sydney cartoon.
363. 1:15:29Thanks. Scroll down. Go down. Oh, okay. I I'll look at there in a second. Uh thanks to at Steo Steo. Okay,
364. 1:15:44here. Steve_ore. Is that 042 or 042? Looks like a zero. Underscore42. No, must be an
365. 1:16:05O. There it is for this dream board
366. 1:16:23image. Where do I save that? [Music] Son of a [ __ ] Sorry. Let me change the format
367. 1:16:40of this so that I can use it in software. So [ __ ] exhausted.
368. 1:17:06Um, coming to a Broadway near you.
369. 1:17:32[ __ ] All right, heading over to Irregular's channel. Irregulars. I asked artificial intelligence write your mother's day
370. 1:17:45message. What's it say inside? Dear mother unit, your contributions to this household. Functionality are noted.
371. 1:17:54Meal preparation above average. Emotional support immeasurable. Laundry throughput high volume. Minimal complaint. Affection level
372. 1:18:04100. So I fired the robot. That's nice. That's cute. Wait, did it match the Yeah. format? It did. The in the inside
373. 1:18:13matches the outside. That's cool. We made it, champ. That's awesome. All right. Um, let me go to GenSpark. Gen S P R A
374. 1:18:30K.AI. All [Music] right. In Hydra, you mean Hedra? It's not Hydra, Mr. He hedra or
375. 1:18:47hedra, not hydra. What's
376. 1:19:03that? Both Kyle and the new pope are in this their 60s. That's rude. Uh, I can't find it, Mr.
377. 1:19:19[Music] It. Oh, look what I [Music] made. Nice date night. A AI learning lab. Little popcorn. I like the unicorn.
378. 1:19:40There's something from Gareth. Vicki Steo, why do I look so sad? Oh, two hours into the live, three tequilas [Laughter]
379. 1:19:54down. I got no Mr. It. I got nothing. You've been You've been buried, Mr. It. Someone is squatchching your voice. All right, let's go to Jen
380. 1:20:11Spark. So, I figure what we'll do is we'll watch this video together because we watched this on office hours today and it's pretty
381. 1:20:23something. Now, here's the caveat. Caveat is that this is a marketing video. So, as Pate rightfully points out
382. 1:20:35whenever I show one of these, he's like, "Yeah, how much of this is real?" So, I don't know how much of this is
383. 1:20:42real. I already did tabs, didn't I? I thought I did this. I thought I did it. Did I not do it right? Hang on. Hang on.
384. 1:20:51I did it right. But whatever. I'll do it again. Okay. Thank you, Winston.
385. 1:21:08Hey, we are shipping GenSpark AI sheets, a full logentic tool that can help you from extensive data collection, analysis to
386. 1:21:15visualization. Forget Excel, now you have a personal data analyst. Here's the raw data of all marketing campaigns from
387. 1:21:22a company over the past year. Upload these four Excels to Genpar. Genspark quickly transformed this chaos into a
388. 1:21:28clean, structured AI spreadsheet. Now, just ask, which type of campaign has the best performance? Genpark immediately
389. 1:21:35executes code and applies formulas. It generates not only charts for different channels, but also a radar graph
390. 1:21:42highlighting email campaigns as the winner. Let's probe deeper. What's the correlation between content variation and performance?
391. 1:21:51Genspar quickly finds negative correlation with ROI and recommends one to four variations as the optimal range.
392. 1:22:00Finally, show me a comprehensive performance report. Within moments, GenSpark delivers an executive summary
393. 1:22:07with clear data visualizations and talking points addressing all key questions your boss might ask. Genpark
394. 1:22:13can automatically find companies, people, products, or any data you need. Let's ask Grenark to find healthcare
395. 1:22:21startups that are at series A or series B stage founded after 2020 with their most recent funding round occurring in
396. 1:22:282024 or later. Within minutes, GenSpark delivers a comprehensive spreadsheet providing um Lissic or or Liss Lissionic or Lis 10
397. 1:22:42Inc. Um what are you trying to do? I'm an AI query expert. Uh I'm not trying to do anything. I'm just watching a video
398. 1:22:49from GenSpark which just launched sheets which looks pretty remarkable and gen and just for context Gen Spark's super
399. 1:22:58agent and their slides generator is pretty remarkable already. So I'm assuming this is going to be pretty
400. 1:23:05good. I haven't played with it a ton though. 9 companies, all with recent funding in 2024, founded after 2020 at
401. 1:23:13Series A or B, complete with industries, founders, and investors. Oop, that music is so brutally bad, isn't it? This is, you
402. 1:23:23know what this music is? I promise you, this is not um pseudoenerated music. I I guarantee you what this is is this is some of that
403. 1:23:34loopy [ __ ] stock music that you know you pay $10 for the year and get, you know, 10,000 songs. This is one of those
404. 1:23:4110,000 shitty songs. Next, if you're a recruiter, tell Jensen Spark. Identify senior UX designers with hyperspecific
405. 1:23:48needs. Watch as it compiles profiles of qualified specialists with their experience in AR or VR, their
406. 1:23:54proficiency in Unity and Unreal Engine, and experience with eyetracking technology. For brands tracking their
407. 1:24:01social media performance, GenSpark can search YouTube and aggregate data from all relevant videos, summarizing views,
408. 1:24:09likes, and comment analysis in one convenient dashboard. Hey everyone, in this video I'm going to be showing you
409. 1:24:15an AI agent. AI sheets can do much more beyond traditional formula. You can use LLM as your new formula, processing all the data
410. 1:24:24throughout your sheets. If you're a student preparing for your AP Biology exam, simply upload your most recent homework. Genpark
411. 1:24:34analyzes it against key concepts in this chapter, color codes your proficiency levels, and creates study materials
412. 1:24:41targeting exactly what you need to improve. That's crazy. If you manage e-commerce for products such as Stanley
413. 1:24:48tumblers, just up Wait, pay attention to this, right? Just upload pictures of your Stanley tumblers. Watch what this
414. 1:24:54does. Upload images of your products. Jensen generates professional advertising visuals in the gym, home, and outdoor settings, scaling your
415. 1:25:04product photography in just minutes. If you're a healthcare software sales rep, ask for potential clients.
416. 1:25:14GenSpark delivers a list of qualified prospects in your target region, complete with key contact persons, current software, and sales
417. 1:25:21opportunities. It can even drive personalized outreach emails, plus one-click buttons to send them. Go to
418. 1:25:27genpark.ai and find GenSpark AI sheets today. All right, so let's go try something. Let's see if this is any good. Is it any [ __ ] good?
419. 1:25:39Genpark.ai AI open AI called they want their set back. Yeah. What is it about I saw that from source camp. Why do they all have
420. 1:25:50the same stupid [ __ ] set? All right. So, here's sheets. Okay. So, what I'm going to do is I'm going to say
421. 1:26:05um let's go over to chat GBT for a second. Based on everything you know about my musical
422. 1:26:22Sydney, write up a primer about the show, the authors, um, and anything you
423. 1:26:37think might be important to identify agents and
424. 1:26:52producers. All right. So, we're going to Oh, am I not sharing? God dang it. So, I'm getting chat GPT. Since chat
425. 1:27:05GPT knows me, I'm going to get it to tell. All right, I got all this [ __ ] right. This is actually quite amazing
426. 1:27:20that the memory function in chat GPT is actually it's pretty stunning that I can just say write up [ __ ] about my musical
427. 1:27:39Sydney. Would you like a PDF pitch deck version of this? uh PDF pitch PDF version is
428. 1:27:54fine just format nicely. I don't need slides, just a doc. And I'll show you why I'm doing this in a second here.
429. 1:28:20All [Music] right. All right. So now chat GPT is generating a PDF document using Python which if you haven't been if you
430. 1:28:35haven't been kind of digging in deep into Chat GPT recently it's pretty slick that would you like me to export this as
431. 1:28:41a polished PDF as well? Yes. Let's have it do a polished PDF and see what that looks like. Model not found error. Ouch. All
432. 1:29:10right, let's go look at this. Did it get everything?
433. 1:29:44You know what? I think I'm going to go back to the original. I think in writing all that it missed a bunch of [ __ ] Maybe it
434. 1:29:58didn't. Whatever. Okay. So, let me just copy all this. So, here's the write up of our show with contact information. All
435. 1:30:11right. Then, we're going to go back to GenSpark and I'm going to say each data search. I'm going to go here is some information on my new
436. 1:30:27musical Sydney. I want you to find producers and
437. 1:30:42agents you think would be good and put them in a sheet with [Music]
438. 1:30:57customized Outreach. Wait, outreach emails for each. Also include why you selected them and
439. 1:31:12be sure um to make
440. 1:31:28the email um resonate with that core reasoning and then just paste in all
441. 1:31:44that [ __ ] I just did and we'll [ __ ] turn it loose. Let's see what it does. Tik Tok comment pinned. I don't see it.
442. 1:32:01millions of jobs gone just like that. Um I sure like sure I I I would say
443. 1:32:17I millions of tasks that human beings do today are going to be automated. It doesn't necessarily mean that jobs are
444. 1:32:25gone. What it means is there's going to be new kinds of jobs. What I can tell you is that if you're sitting on the
445. 1:32:31sidelines or if anyone is sitting on the sidelines just bitching about the fact that AI is here, it doesn't it's not
446. 1:32:39going to make it go away. The people who are going to fare better when job displacement comes are people that know how to use AI. The
447. 1:32:49whole purpose of this channel is not to say AI is good, not to say AI is bad, not to say AI is anything other than to
448. 1:32:57just say AI is here. And what this channel is about is let's understand what it makes possible. And then once you understand a tool like
449. 1:33:08um like GenSpark, which is really quite a remarkable research tool, then you can choose you've got two choices. you can
450. 1:33:18choose to engage with it or not. And if not, just know this is what you're going to be competing against. And if you do
451. 1:33:25choose to use it, and particularly if you're good at data analysis and you're already good at querying,
452. 1:33:32um, you're going to be able to look at the output of a tool like this and use it in a much more sophisticated way than
453. 1:33:39say some bozo with an acting degree can. Um, but I would I would caution you if if the if your comment is, you know,
454. 1:33:49you're sort of casting stones from the sidelines with AI, I would encourage you to get your ass curious about it sooner than later.
455. 1:33:58Um, oh, and how do I raise $20 million? Exactly. All right. So, let's see what it's doing here. Um, what did it do?
456. 1:34:07Stacy Mitch, producer. Dear Evan Hansen, The Crucible. Deep experience. Techfocused. This is cool. Deep
457. 1:34:18experience with tech focused theatrical work. Dear Jordan, your reputation for championing bold, innovative works that
458. 1:34:28push the theatrical boundaries while maintaining artistic excellence makes you an ideal partner for our new musical Sydney. Much
459. 1:34:37like your productions of Spring Awakening, an American Idiot, redefining the musical theater could explore,
460. 1:34:42Sydney examines, this is good. This is really quite good. So, it's basically making me It's making me a [ __ ] target list
461. 1:34:55of who to reach out to with with a an email per person. Absolutely amazing. Absolutely amazing. Sorry if I distract you with gifts being
462. 1:35:11OCD. Um, just taking my coin balance down to the next lowest prime number like 1361 then 1259 where I am
463. 1:35:27currently Winston. That is the best. I it never occurred to me that you could you could do some fantastic things like
464. 1:35:37prime number coin management in Tik Tok, but yeah, of course you can. Prime numbers are everywhere. That's [ __ ] awesome. Um, let's see.
465. 1:35:52Uh, well, let's just let this thing keep going. This is really quite amazing. I I mean, honest to God, like if you look at what this is
466. 1:36:05doing and your immediate response is there, you know, a million jobs are down down the drain, you know, I would encourage you
467. 1:36:13to look at this and say, what jobs does this make possible, right? or what what superpowers does a capability like this
468. 1:36:24give to an individual that might get laid off that can now maybe do something at a dramatically higher level than they could have otherwise
469. 1:36:34right I you know again two ways two ways to respond to AI one is ignore it and one is embrace it ain't going away. The only
470. 1:36:51thing I know is if you ignore it, it's going to happen to you. And that's not going to be pleasant for [ __ ] anyone. This is so
471. 1:37:05cool. It's been populated with six producers who are excellent. All right. Producer, theatrical
472. 1:37:40owner. This is amazing. Look at the like it it wrote into the into this personalized email. The show asked
473. 1:37:48timely questions about consciousness, personhood, and intimacy. I can't it's because it's working. I
474. 1:38:04can't h It's fascinating. This absolutely fascinating. Is it giving me their contact information? Let me add something in here. Um, it
475. 1:38:17looks like we are missing contact information for the producers. I need everything possible
476. 1:38:33to get the word out. Yeah, I can't do it while it's working. So, that's a downside. So, there's So, this is very much like
477. 1:38:45Manis. Um, one of the nice things you can do with Manis is you can talk to it while it's working. This one, it doesn't look like you
478. 1:38:54can. All right. So, while that thing's thinking, that's still going, right? Eight contacts so far, seven producers, one literary agent. All
479. 1:39:09right, let me go look at uh Mr. It's got something in irregulars here. Oh, it's a birthday video. Nice. Let's go look at that. Hang
480. 1:39:25on. This is going to make me uncomfortable, isn't it? I don't deal with uh Let's see. An error
481. 1:39:40occurred. Bad playback ID. Huh? Close that. It's
482. 1:39:55broken. Posted a second video below that. Kyle Shannon, thank you, my friend, for changing my life. You and AI have
483. 1:40:08shifted the course of my dreams, my mind, and my heart. AI has expanded my creativity and highlighted the
484. 1:40:14intelligence and passion I carry inside. Now I only need to better my ADD medicine dosage to work harder. Thank you irregulars for being
485. 1:40:25my e friends. Our realities may be different, but somehow keeping our eyes on AI has guided us toward a common
486. 1:40:33goal. I truly hope that in the future that goal will belong to everyone, not just a few. During the day, I wear many
487. 1:40:43hats. Among the many projects I juggle, infinite AI is the one closest to my heart. I carve out time each day to
488. 1:40:52build it. But I can't do it alone. Right now, I don't have the funds to invest in my dreams. The only treasure I have is
489. 1:40:59your friendship. Can you follow me? Help me. Maybe even a like, a follow, or a kind word means the world. Kyle, will you follow me? Thank
490. 1:41:14you. That's awesome. That's awesome. Are you sure you made that in Hedra? H E D R A or did you do it in some some other
491. 1:41:21tool? Because that Hedra looks like it's more sophisticated than than what that did. Um I'm curious where that was done. That's
492. 1:41:35sad. Oh, yeah. Okay, let me go back to share my tabs the right [Music] way. But thank you very much for the kind
493. 1:41:50[Music] thoughts. All right, are we done here? No, we're still going. So, what do we have now? We've got two literary
494. 1:42:05agents. Dear Beth,
495. 1:42:23[Music] Hey, Kyle. Yes. Got one more uh generated birthday gift for you. Oh, no. Put it on the uh No, this isn't anything
496. 1:42:39bad. You'll like this one. Uh I zero shot a uh an image swapper for you. That's on the post-it note. Oh, this is this was done in Lovable.
497. 1:42:53[Laughter] That's great. So what what can I convert? So first you can share the tab so we can see it. Yeah. Yeah. Yeah.
498. 1:43:09Um and so this is a lovable prompt a Google translate style site but for image file format swapping. Oh that's
499. 1:43:18great. And it is it can take your AVIF files and convert them to JPEG. You just drag it in, drop it, and then it goes.
500. 1:43:25Have you Have you tested it? I have. If If I drop this a AVIF in, and it doesn't do it. File uploaded. So, oh, there it is.
501. 1:43:39Nice. Convert from that to ping. Convert. That's awesome. And I mean the whole site, the whole site, the help about the feedback, the
502. 1:43:53footer links, the privacy policy. I mean, this didn't exist 15 minutes ago. Loable did at all. That's amazing. I'll drop it in irregular
503. 1:44:03so yeah. Yeah, drop it somewhere so that I can find it. I'll I should I should also bookmark it. I mean, this is where we are, people.
504. 1:44:12Like, um, Jim Ross the other day, um, got substituted in to do a keynote pre presentation 20 minutes before he had to
505. 1:44:24give give the presentation. And rather than panicking and writing a deck, you know, and then giving the presentation,
506. 1:44:31he generated the deck for the audience. like his part of his presentation was generating the presentation that he then
507. 1:44:41gave. Like this is the world we live in now. Just make [ __ ] You need [ __ ] you make [ __ ] You ask for it, there it is. [Laughter]
508. 1:44:51Boom. You know, a 100red million jobs destroyed. How many jobs generated? How many jobs generated? And what kind of jobs?
509. 1:45:02I'm not sure we can ever trust that AI results will not be 100% accurate. I guess similar to human error. Yeah. Have you worked with
510. 1:45:12humans? AI is 99.99999% accurate. Like it'll it's going to get there. And then humans are still going to be bitching. And you
511. 1:45:22know, we're consistently delivering, you know, in the low 70% accuracy wise. That's, you know, it's going to be like
512. 1:45:31uh self-driving cars, too, right? Where I think self-driving cars at this point are already safer than humans, but every
513. 1:45:38time, you know, one goes off the rails and kills someone, everyone's like, you know, they're evil. They're bad. No. Well, considering the
514. 1:45:49competition. Oh, man. Joker. I makes I make apps on the go as I need them. That's super cool. Uh, is this thing
515. 1:45:57still going? Yeah, it's still going. What has it found? Why does it keep? It keeps updating the same two
516. 1:46:15agents. Let me stop this and I'm going to say you missed the contact information. Let's throw that in there. Regenerate response.
517. 1:46:27I bet I ran out of tokens. I bet I ran I shouldn't have stopped it. Damn it. Anyway, um go play with Gen Spark.
518. 1:46:37Gen Spark's really good. Um I should probably just download this, shouldn't I? Is that what this is? Copy URL to
519. 1:46:52clipboard. How do you download this? Visualize. Fact check content. Continue
520. 1:47:17searching. Yeah, it's not going to it's not going to let me go. All right. Well, there you have it. There you have it, good
521. 1:47:29people. You can just right click on a name and convert it. Oh, file formats.
522. 1:47:48Um, I don't know how to download this.
523. 1:48:09[Music] Anyone
524. 1:48:28Hey view. Ah, download. No. Well, if anyone can figure out how to
525. 1:48:44export Gen Spark sheets, let me know. Far right icon
526. 1:49:00where anyone with the link I can copy the URL to the clipboard. anyone with the link. I bet I could just copy. Yeah, I
527. 1:49:18can probably do it like this. No, it's not working. All right.
528. 1:49:33Well, Command A. Nope.
529. 1:50:18quit while you're behind. Oh, I got it thinking again. I'll fact check the content and add the contact information. All right,
530. 1:50:27I'm going to let that do its thing. You don't need to watch me try to make this thing work. Anyway, I would go play with
531. 1:50:32this thing. It looks pretty remarkable. Um, not as quite as remarkable as my hair. That's fantastic. All right. I skipped out on
532. 1:50:44the B birthday because it made me feel old. Yeah. No [ __ ] Yeah, it made you feel old. How do you feel about
533. 1:50:53me? Uh, what happens if you upload? I don't know. I don't know. [Music] Third from the left. That did not do it.
534. 1:51:09And of course, conversions. Kyle. F1 may help bring up help. F1. Uh, how do I do F1 function? F1. Nope.
535. 1:51:45All right, I'm going to get out of here. Thank you all. Hope you had a good Friday night, date night, and
536. 1:51:56uh this weekend, do something nice for Mother's Day. Um go make lots of content. Just make stuff. make go make
537. 1:52:05make. Here's what I would encourage you to do is um try a new tool this weekend. Try something new. And I like at this point I'm I am
538. 1:52:18no longer able to recommend like what's good or not because there's just too many of them. Like just have chat GPT go research image
539. 1:52:28generation tools and just go play or video generation or music and just explore and then make nice stuff. Put some joy in the world.
540. 1:52:38All right. And when people ask you how did you do that? Tell them tell them how you did it. All right. Beauty. All right. Um Da
541. 1:52:53da. Beautiful. Thanks everybody. Have a fantastic weekend. I will see you Monday night. If you're on YouTube, like and
542. 1:53:00subscribe. Subscribe to the YouTube channel. Okay. Thank you. Join the AI salon. Beautiful. All right. Peace out,
543. 1:53:08everyone. Hope you have fun tonight. Did