5/20/2026 - Unpacking Google I/O: Gemini Omni, Spark, and the Future of AI Video

Transcript

1. 0:36Good evening, good evening, good evening. How is everybody? Welcome to Wednesday night AI Learning
2. 0:44Lab live. We are live here from the $20 million production studio. If you're new
3. 0:51here, you probably don't know that this just looks like a crappy home office that
4. 0:55takes armies of people to make it look that natural. My name is Kyle Shannon.
5. 1:02This is the AI Salon, AI Learning Lab inside the AI Salon. I hope you're
6. 1:06all doing great. We had some exciting— Google I/O was yesterday and they made some
7. 1:16exciting announcements. So if you're not paying attention to this stuff on a daily basis
8. 1:22like I am, then you may not know about that. So we'll talk about some
9. 1:26of the stuff they talked about there. Danielle's in the house. Silver Fox is in
10. 1:31the house. Oh, by the way, if you're new to this AI Salon, streaming inside
11. 1:37the AI Salon, if you want to find the comments, where to comment, I think
12. 1:41that's down here. No, where to comment is underneath the AI Learning Lab logo up
13. 1:47here. So click on that. And then where to pop this window out, if you
14. 1:52want to pop this window out and surf other areas in the AI Salon, that's
15. 1:56down here. So the relevant buttons are here. So go play. And then if you
16. 2:05figure out the commenting, pop a comment in the comments. And I'm happy to answer
17. 2:10those. But there were a lot of announcements yesterday at the Google event. The one
18. 2:19thing that's kind of amazing, this was one of my pet peeves with Google in
19. 2:26the early days of this GenAI stuff, was they would make announcements for stuff that
20. 2:31you couldn't go play with or you couldn't find. And they got much better at
21. 2:39that, but they're kind of returning to their old wily ways. So this new video
22. 2:45model, that's available in Gemini. If you go to the website gemini.google.com, they give you
23. 2:52like 3 videos. You can make 3 videos a day. But if you go to their app, you can just make videos, a ton of them. And then if you
24. 3:01go to like AI Studio, their developer site, um, you can't make any. They're like,
25. 3:09you're maxed out until whenever. So they're a mess. They're still a mess. Here's the
26. 3:16thing about Google, right? We, you know, if you've been around for a while, you
27. 3:21know, they started as a startup and they used to be scrappy. They're just a
28. 3:24big old massive corporation now. And so they act like it. So, so some things
29. 3:30are really good. Some things are an absolute mess. Um, but, but we'll go look
30. 3:33at, we'll go look at some of that stuff. Um, if you're new to the AI Salon, So welcome to the AI Salon. Over here on the sidebar of the
31. 3:44AI Salon, this is the AI Salon community. There's lots and lots of stuff to
32. 3:47do. There's all sorts of events that are happening. There's a thing called the AI
33. 3:51Salon Mastermind, which is a subscription area of the salon. We've got all sorts of
34. 3:55separate events in there for Mastermind members. We've got the Mastermind Practice Lab, which we've
35. 4:01got a session tomorrow morning at 10 AM. So if you're in the Mastermind, don't
36. 4:04forget that's tomorrow with Sid Hargrove and myself. Um, we've got a workshop going on
37. 4:10right now about separating your identity from your job, which, as AI starts to muck
38. 4:16with all of our jobs, it's gonna muck with all of our identities. Um, Brandon
39. 4:21is teaching a workshop, Producer Brandon, on Lovable and making apps with Lovable inside the
40. 4:26Mastermind. So there's lots going on. Um, there's also just all sorts of, you know,
41. 4:30clubs and groups and, you know, interest groups. So jump in there. Good morning. Am
42. 4:36I late? Mr. IT is in the house. You're not that late. We are, we
43. 4:41are 4 minutes in, so we're just getting started. So if you're new here, welcome.
44. 4:46And if you are not new here, welcome, welcome everybody. Um, if you have any
45. 4:51questions or if you have any experience, if you've played around with these new tools
46. 4:55and have any experience and want to share that with the group, um, pop it
47. 4:58in here and say hello. I am without a producer tonight, but it's— that's, that's
48. 5:03less of an issue now that we're not doing multiple screens and multiple streams and
49. 5:07things like that. Um, but I will do my best to pay attention to comments. I will, um, do my best to share things and, and, and have a lot
50. 5:15of good stuff. All right, beautiful, fantastic. Um, let's see. I'm a newbie. Yeah, you're
51. 5:24a newbie. We're all newbies in the end. We're all trying to figure this stuff
52. 5:30out. All right, I'm going to share my screen and hopefully It allows me to
53. 5:36share my screen for longer than 30 seconds. But let's see, you know what I want to do? Let's go over to, um, let's go over to X and start,
54. 5:50start looking at some of the announcements because boy, there's just a lot of stuff.
55. 6:07What's that? We're building a moon base. Okay, Elon Musk is excited about that. Brandon
56. 6:13Tidd, the Zendesk after party, that's where producer Brandon's going to be tonight. I think
57. 6:20someone's been playing around in, in Nano Banana or the new, uh, ChatGPT image maker.
58. 6:29The other thing is, uh, is producer Brandon, uh, couldn't figure out why his hotel
59. 6:34was so cheap, and he got, he got a hotel on Colfax Avenue, which if
60. 6:38you know anything about Colfax Avenue in Denver, you know why his hotel was cheap.
61. 6:46Quick fix for Gemini 3.5 flash: drop the price in half, no one will complain.
62. 6:51Um, Let's go look at— look, look, look, look what's going on. Everyone's still talking
63. 7:02about Eric Schmidt getting booed in his commencement speech. Lovable is doing some interesting stuff.
64. 7:16They've got— they just added connectors, Google Connectors. So you can start doing all sorts
65. 7:22of really cool, sophisticated apps with Lovable. What Gen X skill would confuse Gen Z?
66. 7:36Drinking from a hose outside. Okay, this, this note from Sam Altman I think is
67. 7:50worth, uh, worth paying attention to. Um, three of the things we are most excited
68. 7:59about: AGI accelerating research, AGI accelerating companies, and then personal AI accelerating everyone achieving their
69. 8:07goals. Which Sam followed this up with a comment. Let's see. Where did he put
70. 8:15it? Where? Oh, maybe it's in replies. Hang on a sec. Okay, so he must
71. 8:44have, he must have done it in a different post. So a couple of things
72. 8:47happened. AGI accelerating company. So yesterday, Mr. Altman, let's see, Y Combinator. Altman. Let's see,
73. 9:18here we go. So yesterday, Sam Altman offered, um, $2 million in tokens in exchange
74. 9:29for equity in Y Combinator-backed startups. So giving away $2 million in tokens. That's a
75. 9:35lot of tokens. Tokens are the things that generate all the stuff when you're using
76. 9:42generative AI and $2 million, you know, spending $2 million in tokens is a lot,
77. 9:48right? Well, if you're starting to use agentic tools where you're just having the something
78. 9:54like Codex spin up agents and spin up sub-agents and then go work on its
79. 9:59own, you're going to burn through lots and lots and lots of tokens. So probably,
80. 10:04you know, tens, hundreds of thousands of dollars in tokens. So $2 million is pretty
81. 10:09significant. So that's tied to his post about AGI. Um, let's see, um, Math Solve.
82. 10:27Second thing that happened today, let's see, let me get rid of Altman. Let's see,
83. 10:44mathematics problem. 6 hours ago. OpenAI. Today we share a breakthrough on the planar unit
84. 10:59distance problem, a famous open question first posed by Paul Erdős in 1946. For nearly
85. 11:0680 years, mathematicians believed the best possible solution looked roughly like square grids. An OpenAI
86. 11:12model has now disproved that belief, discovering an entirely new family of construction that performs
87. 11:18better. This is apparently the new model coming from OpenAI, and one of the significant
88. 11:28things about it— I wonder if Altman replied to this. He might have. I don't
89. 11:35know. I don't see it here. But— Let me go back to Altman. And by
90. 11:50the way, can you still— am I still sharing? Yes, right? Yes. Anthropic hires— oh
91. 12:00yeah, Andrej Karpathy went to Anthropic. Fascinating. Okay, this post from Altman. So Sam likes
92. 12:12to do these cryptic posts, right? Today it was great to announce— oh, it's, it's
93. 12:20right in this post. Okay, so he goes, AGI Accelerating Research, today it was great
94. 12:24to announce the unit distance result. So that's AI, AGI Accelerating Research. Yesterday it was
95. 12:32great to announce that we're offering $2 million in OpenAI credits to every Y Combinator
96. 12:37company. That's AGI accelerating companies. We need to increase our efforts on the third, personal
97. 12:46AGI accelerating everyone in achieving their goals. OpenAI tends to release new stuff on Thursdays.
98. 12:54Tomorrow's Thursday. Google I/O was yesterday. So doesn't guarantee that we're going to have an
99. 13:02announcement from them tomorrow, but it's possible, but I would think that either GPT-5.6 or
100. 13:08GPT-6 is probably coming in the next month, maybe tomorrow. So we'll see. But it
101. 13:15looks quite amazing. So let's go to Google I/O and go look at some of
102. 13:22the vidges. Oh, this is pretty cool. Google Gemma. So Gemma is Google's open source,
103. 13:37or is it open source? Yeah, open source, open source large language model. And you
104. 13:44can control a robot. And then they also, um, have this 3D printable robot on
105. 13:52GitHub. So you can basically use Gemma to control this thing and you can go
106. 13:57print your own robot and I guess buy all the components and build it. So
107. 14:01If you go to @GoogleGemma, you can go look at the plans for a robot. Go build a robot. Here's the thing. I think it is in all of our
108. 14:10interests in the next 6 months to a year is to get educated on hardware
109. 14:17that can run powerful large language models. Learn which ones are out there, which ones
110. 14:24are available, what the capabilities are, what can run on any given piece of hardware.
111. 14:29If you can afford it, invest in some hardware that can run decent models. AMD
112. 14:34just announced a new box yesterday that's in the $2,000 to $3,000 range that can
113. 14:40run 200 billion parameter models. It's a 128 gigabyte machine. It's like a Mac mini
114. 14:47from AMD. Apple will likely be coming out with them. I think that the local
115. 14:52large language models are going to be big. They're coming. Google is putting agents in
116. 15:03search. They're really transforming how search works, which is interesting. Let's see. Oh, here's— this
117. 15:11is good. Somebody made us a little Google I/O. They probably built this in ChatGPT.
118. 15:22The core AI models, Gemini Omni. That's what we're going to play with tonight. World
119. 15:26model that simulates physics, any input to any output with iterative video edits. OmniFlash Live
120. 15:33now. And then OmniPro's coming. So I'll explain about that when we go play with
121. 15:40it. Gemini 3.5 Flash, frontier level, 4 times faster. I don't know that these two
122. 15:49things are separate. They may be, but they may not be. But I think Omni
123. 15:55is in the, uh, the Gemini 3.5 Flash. Maybe I'm high on that. Maybe I'm
124. 16:03wrong on that. I don't know that. That I actually don't know. That was confusing
125. 16:06yesterday. But they said Gemini Omni is inside gemini.google.com, and it's inside the Gemini app.
126. 16:17So I don't know which one is which, but it says it's Flash that's in
127. 16:20there. The agentic era. Gemini Spark is a dedicated virtual machine running in the cloud
128. 16:28that can run multiple agents doing tasks on your behalf. So this is just like
129. 16:32OpenClaw, just like Hermes, just like Codex, just like Claude Code and Claude Cowork. Gemini
130. 16:42Spark is in there. Gemini for macOS is coming. That's not out yet, but it's
131. 16:47coming. Developer tools, they launched the new Anti-Gravity 2.0, which is getting some slack. I
132. 16:54saw some people bitching about that, but if you're into coding, it's their new IDE.
133. 16:59It looks like they've stolen the interface from Codex. So the Frontier Labs stealing from
134. 17:10one another continues. Search and Commerce search engines spin up dedicated agents that monitor and
135. 17:17act on results 24/7 on your behalf. Search agents, generative UI, um, search codes and
136. 17:23deploys live custom widgets. All of that stuff is coming, I think. And then there's
137. 17:28a universal shopping cart. Um, there's— there was some hardware announcements, audio glasses, um, uh,
138. 17:37augmented reality displays. Um, new TPUs, the 8th generation TPUs. So the actual— one of
139. 17:47the things that Google's doing that's really smart is they're getting out from under NVIDIA
140. 17:53and making their own chips so they can— they control their own destiny for how
141. 17:57much compute they have. Um, there were some interesting things on the creativity front. Docs
142. 18:05Live. Is a Google Doc that you can just talk to, and it's not just
143. 18:10transcribing what you're saying. You can just talk to— we'll go watch a video on
144. 18:14that one, actually. That one's pretty interesting. Um, Stitch is a UI designer that looks
145. 18:20good. Google Pix, where you can edit images, um, um, iteratively, is really interesting. And
146. 18:28then there was some interesting stuff on the science front. So they They put out
147. 18:32a lot of stuff yesterday. Okay. Let's go look at Google's Docs Live presentation because
148. 18:41I think that one's fun to look at. You can just verbally brain dump whatever
149. 18:47is— Let me just pause for a second and jump over back to you guys.
150. 18:50Let's see. I'm glad we're over I/O. So many— Oh, we're going over I/O. So
151. 18:58many new announcements. Hang on. Let me— Do this. Let's see. Pixelated and bad video.
152. 19:08Omni is the multimodal. It's a naming problem. Yeah, yeah, yeah. Okay. Um, is Kyle
153. 19:15streaming on any other platforms? I'm not. So at 7:10, I do the, the AI
154. 19:22Learning Lab pre-show on 4 different platforms. TikTok, X, LinkedIn, and YouTube. And then the
155. 19:31Learning Lab happens live inside the AI Salon. So that's where I go over there.
156. 19:36I'm glad we're going over I/O. There were so many announcements. It's a lot. It's
157. 19:42a lot. I thought it was consolidating. I want a friend of mine to join.
158. 19:46Yeah, so they can go to the AI Learning Lab TikTok channel or YouTube channel.
159. 19:54The most recent 2 weeks of these lives live inside this channel in the AI
160. 20:02Salon. The lives happen inside this channel, inside the AI Salon. And then there's a
161. 20:062-week delay where the archives will live out on the YouTube channel. So we're going
162. 20:11to still have the whole historical YouTube channel, which is we've got, I don't know,
163. 20:162,000 hours of programming from the beginning of basically ChatGPT till now. So that's gonna
164. 20:22stay, but the most recent 2 weeks are inside the AI Salon. The whole idea
165. 20:26here is we want people inside the AI Salon communicating with one another. So you
166. 20:31can play with the chats here. You can also chat within the community, right? If
167. 20:35you haven't introduced yourself yet, go to the Introduce Yourself space and do that. If
168. 20:42you've been here for a while, go check out the Introduce Yourself space and say
169. 20:46hello to people that are joining. So if you're new here, welcome. If you've been
170. 20:50here for a while, Welcome. It's great to have you all. Too many names. Let's
171. 20:56see. Was there a demo of Spark or Flow for AI video editing? I think there was. I was in the middle of a meeting at work and I was
172. 21:07trying to pay attention to Google I/O while that was happening. So I did miss
173. 21:12some stuff, but we'll go look at that. So Flow and Spark are two things
174. 21:16we want to look at. And then we'll play with the image generation model. Lori
175. 21:22Blair, I just made one and the images were good. Didn't follow direction 100%, but
176. 21:27the sound effects were great. Oh, the video. Yep. Hello everyone. Last meeting finally over.
177. 21:31Welcome, welcome. SourceCamp, the video quality is really messing up. Yep. I always look for
178. 21:41Kyle to unpack this stuff, make it digestible. I will try my best. All right,
179. 21:46beautiful. So let's go back over here. Let's watch this. This is the Google Docs
180. 21:54Live. I don't think this is available yet, but I could be wrong on that. Is on your mind and let Gemini do the rest. Let's see it in action
181. 22:04with the demo from our— Wait, that's just someone filming it from the audience. There's
182. 22:08probably a better version of that. Yeah, here we go. Create a doc with Gemini
183. 22:14before, you'd have to type up a really precise prompt. With Docs Live, you can
184. 22:19just verbally brain dump whatever is on your mind and let Gemini do the rest.
185. 22:24Let's see it in action with a demo from our product team. This is all in real time, not sped up. All right, let's try this out. So I just
186. 22:37remembered I'm doing an alumni talk for my high school's career day tomorrow. Uh, I
187. 22:41need to come up with some talking points to explain what I do for a
188. 22:44living as a software engineer, but I'm not really sure where to start. Oh, actually,
189. 22:49can you just pull my resume from Drive? Although that might be boring. Um, maybe
190. 22:55can you come up with some funny analogies so it'll be more of an engaging
191. 22:58talk for the students? Oh, and also, I think the school sent me an email.
192. 23:03I think the subject is something like Career Day Logistics. Maybe just grab the details
193. 23:08from there. Throw them at the top of the doc so I know where to
194. 23:10go and what time to get there. Let's just update those requirements and turn this
195. 23:14straight into a draft. This is cool, but it's a little dense. Maybe format the
196. 23:27analogies as a table so it's a little easier for me to scan. And also
197. 23:32add a note to tell the story about how my brother inspired me to become a software engineer. At the top of my doc and bold it so I don't
198. 23:38miss it. Yep. Did it do it? There's the table. Yeah, it looks great. New
199. 23:50feature called Docs Live. So that's pretty sweet. That's— again, I think, you know, I
200. 24:02think we're in this really interesting— this is what we talked about on Monday. We're
201. 24:05in this interesting transition period between— here's an interesting thing. If you're in the AI
202. 24:14Salon, if you're brand new here and you're brand new to AI, the benefit of you being late to AI is if you're feeling like, oh my God, I have
203. 24:27to catch up. These people know so much more than I do because they've been
204. 24:30playing with all these tools for this long. Getting up to speed on generative AI
205. 24:38starting today is probably not a bad way to go. So, so for those of us that have been playing with this stuff for 3 and a half years, the
206. 24:50era that we're exiting is an era where it was really important to learn these
207. 24:55discrete tools where we had to learn how to use each one and what their
208. 25:00quirks were. And we had to learn how to prompt them and get a response.
209. 25:02And then we would, either take them out of generative AI and go modify them
210. 25:08elsewhere, or, or we would just prompt over and over again, and we had to
211. 25:12manage all these different tools. We're moving into an era where you're just going to
212. 25:18be able to talk to these things, and then they are going to go off and do work on your behalf, and they're going to be using models that are
213. 25:25so good that the work that they bring back, for the most part, is going
214. 25:29to be good. We're not there yet. We're in this weird transition space in between.
215. 25:34But I look at this Docs Live product and it's like, it's not just transcribing
216. 25:42whatever we say into a thing, or it's not just transcribing it into a prompt
217. 25:47and then giving it something. It is figuring out what you're saying. And if you've
218. 25:52got it connected to all of your Google Docs and things like that, it's You
219. 25:57know, it's pulling data and it's being smart about how it's doing it. I think
220. 26:05that's the world we're moving into, right? It starts to look more and more and
221. 26:10more like Jarvis. Like you can just talk to this thing and it'll go do
222. 26:14it. So you're not going to need to know so much about prompting. You're not
223. 26:17going to need to know so much about all these other things. I think there's
224. 26:20an advantage in understanding how we got here. Um, but I don't think you're going
225. 26:24to need it. So that one's fascinating. So Google— so the other thing we wanted
226. 26:28to look at was, what was it, Stitch that you put in there, Sonny? Was
227. 26:32it Stitch you wanted to see? Latecomers skip the goop. Yeah, but we're still in
228. 26:47the goop. If you're, if you're just getting started now, you're starting in the goop.
229. 26:52It's going to be confusing starting now. It won't be confusing starting a year or
230. 26:55two from now. It'll be confusing starting now because some people will be telling you,
231. 27:00you need to learn prompt engineering, buy my prompt library, right? There's a whole, there's
232. 27:06a whole economy over for the old way. Spark. Okay. Gemini Spark. It's your 24/7
233. 27:25personal AI agent that helps you navigate your digital life, taking action on your behalf
234. 27:31under your direction. It runs on Gemini 3.5 and is built on anti-gravity. So Spark,
235. 27:37what Spark is, is a, okay, anti-gravity. So, Do you remember if you've been here
236. 27:45in ye olde AI Salon for a while, about 3 or 4 months ago, Antigravity
237. 27:50came out and I got all excited and I said, I'm going to learn Antigravity.
238. 27:55And then I tried to learn it and it was so complicated and it's so
239. 27:59much designed for pure engineers that I just, I bailed on it. I'm like, that's
240. 28:03crap. It's not, if you're an engineer, it's fine. It might be great. But if
241. 28:08you're, just like me. It's the dude, it's too much. Spark is the consumer version
242. 28:16of anti-gravity that uses anti-gravity under the hood, but it puts a consumer front end
243. 28:22on it. So let's go, let's go look at this video. How long is this video? That's just a GIF. All right, let's see if there is a demo, a
244. 28:32demo movie of it. I can see clearly now the rain is gone. I thought there was a demo. I'm pretty sure there is. That's a GIF. That's a GIF.
245. 28:52I can see all obstacles in my way. Let's look at Google Spark demo. La da da da da da da da da. It's gonna be a bright, bright, bright
246. 29:12sunshine day. Mm-hmm. This is where we need producer Brandon to go find me the
247. 29:22demo. But he is, is this it? Google I/O, Agentic Driven Experience. Yeah, this is
248. 29:29Spark, okay. Entire custom stateful— here we go. Some projects aren't one-off questions, they're ongoing
249. 29:42tasks. So Search now has the ability to help you create entire custom stateful experiences,
250. 29:50tools, trackers. Now, this isn't Spark, by the way. This is, this is agents within
251. 29:56Search. Dashboards. I think of these sort of like building my own little mini apps
252. 30:00with Search, and I've been making a ton of them. And they're especially awesome for
253. 30:04those long-running tasks where you want to keep coming back, like planning a wedding or
254. 30:09managing your home move. So what do you think? Should we build one together? Yeah?
255. 30:16All right. Let's do a demo. So I'm constantly trying to figure out on Thursday
256. 30:21what it is I should do with my family every weekend. And so here's a
257. 30:26search I actually just did about fun things to do with my family this weekend.
258. 30:30And of course, you get great, great response here from AI mode. But here's something
259. 30:34new. Search proactively offers to build a weekend planner for me. And similar to how
260. 30:39you just saw Search create these generative UI and interactive visuals completely from scratch, Search
261. 30:44can code this up right now. Let's do it. All right, so for the sake of today's demo, we thought you might want to see a peek under the hood.
262. 30:54So as this builds, you're going to see real-time thinking steps and code generation flowing
263. 30:59in. And here, Search is thinking about the right components to build, not just what
264. 31:03information to fetch, but the best way to present it. And I've chosen to securely
265. 31:07connect Gmail, Photos. So just so you can notice what's happening here over to the
266. 31:13right, so, so this is a a modified version of Google Search. It said, hey, do you want to turn this into an app, right? So he didn't have to
267. 31:22come up with the app idea. Search came up with it. Over here to the
268. 31:26right now, it's generating code. So if you've ever used Codex or Claude, where it
269. 31:33creates the— I forget what it's called in Claude, but the, you know, creates the
270. 31:37code over to the right. You've got your prompt over to the left. They've all
271. 31:41got this similar kind of interface where your interaction is on the left, the stuff
272. 31:45it's generating is on the right. So this is just yet another way in to
273. 31:51this kind of, this kind of agentic AI where the bulk of the token generation
274. 32:01is happening independent of your participation of it, right? The world we're exiting is I've
275. 32:07got to go to ChatGPT or Claude or Gemini. I put in a prompt, it gives me a response. I look at that, I say, is it good or not?
276. 32:15We're moving into this thing where it's, he literally just clicked the button and this
277. 32:19is now figuring out all the stuff it thinks he wants. And this is not
278. 32:25Spark, by the way. This is, this is app development within Google Search. So it's
279. 32:33using personal intelligence. To make my suggestions even more helpful. That means referencing Gmail receipts,
280. 32:39Calendar, and much more to personalize. And boom, here it is. Looks like it's ready.
281. 32:45All right. Got a beautiful planner already taking into account driving times and weather. And
282. 32:50Search knows that I have 2 kids and that they love animals and that my
283. 32:54oldest is learning to play chess. So actually, this— the second one here is kind
284. 32:58of awesome for my oldest, so I'm going to go ahead and part that. But, you know, 2 kids, got to make them both happy. So I'm going to lock
285. 33:07in the Happy Hollow Park and Zoo right there. And because this syncs with my
286. 33:11calendar, you can see it's already blocked off my afternoon to meet a friend to
287. 33:16watch this game down here. And below, it's got all of these cool restaurant reservations
288. 33:20and beautifully laid out on maps. Some projects— OK, so let me come back to
289. 33:28this for a second. So I think this is worth talking about, and I want
290. 33:30to see your— We can't see that. You haven't been sharing your screen. Oh, Jesus.
291. 33:40Hello, this is a professional. Kyle, we can't see. Kyle, we can't see. Let's see.
292. 33:50Hang on, let me, let me share with you, um, some of the comments that
293. 33:54I'm getting here. Kyle, we can't see your screen. Kyle, can anyone pin this comment?
294. 34:01Um, you're not sharing, Kyle. Fire him. We can't see, Kyle. We, we pinned Sonny's—
295. 34:09who pinned Sonny's comment? Please, a comment that we can't see. This is all me.
296. 34:17Um, let's see. You haven't been sharing your screen. We can't see that. Oh no,
297. 34:21I'm at the doom center of buffering. Okay. So here's what we'll do. I'm going
298. 34:27to play this video again so that you can see it. I'm going to take
299. 34:35Sonny's comment off there. We've been trying to tell you this for 15 minutes. Listen,
300. 34:42listen, here's the thing about the AI Learning Lab. One of the, if you, if
301. 34:48you're new to the AI Salon and you're new to the AI Learning Lab, One
302. 34:52of the things we talk about is embrace the jank. You've got to embrace the
303. 34:56imperfections of AI and celebrate them because I don't know, in a year or two,
304. 35:03it's just going to be perfection and there's going to be nothing broken. It's going
305. 35:06to be Shangri-La. And we'll look back on these times fondly. Same with the AI
306. 35:11Learning Lab. At some point, we're going to hit perfection here, but not now. Okay,
307. 35:19so what we're going to do is we're going to go look at a thing that is not Google Spark. But this is the, uh, this is, uh, let's see,
308. 35:31um, this is dynamic app development inside Google Search that is not out yet but
309. 35:39is coming. Are one-off questions, they're on there. Okay. Let me, let me just make
310. 35:47sure you can see now I'm sharing, right? Embrace the jank because next year will
311. 35:51be Shanker luck. Quote of the day. We'll be laughing about this night for a
312. 35:56long, long time. Listen, you can— here's the good news. Please tell producer Brandon he
313. 36:01is needed and loved because he would have been, uh, Post-it noting me and that
314. 36:08wouldn't have gone on for 15 minutes. Okay, you can see what I'm sharing. You
315. 36:12can, you can pick up what I'm putting down. So we'll watch this again. Some
316. 36:15projects aren't one-off questions, they're ongoing tasks. So Search now has the ability to help
317. 36:21you create entire custom stateful experiences, tools, trackers, dashboards. I think of these sort of
318. 36:30like building my own little mini apps with Search, and I've been making a ton
319. 36:34of them, and they're especially awesome for those long-running tasks where you want to keep
320. 36:38coming back, like planning a wedding or managing your home move. So what do you
321. 36:44think? Should we build one together? Yeah? All right. Let's do a demo. So I'm
322. 36:51constantly trying to figure out on Thursday what it is I should do with my
323. 36:55family every weekend. And so here's a search I actually just did about fun things
324. 37:00to do with my family this weekend. And of course, you get great, great Great
325. 37:03response here from AI mode. But here's something new. Search proactively offers to build a
326. 37:09weekend planner for me. And similar to how you just saw Search create these generative
327. 37:12UI and interactive visuals completely from scratch, Search can code this up right now. Let's
328. 37:19do it. All right, so for the sake of today's demo, we thought you might
329. 37:24want to see a peek under the hood. So as this builds, you're going to
330. 37:27see real-time thinking steps and code generation flowing in. And here search is thinking about
331. 37:33the right component. Okay, so this is where I stopped before and was talking a
332. 37:36lot. Let me just make sure I'm still sharing. Um, to build, not just what
333. 37:43information to fetch. Let's see, am I— embrace the jank. Okay, right now it's jankrilla.
334. 37:57Okay, so you are seeing this. Okay, cool. So this is the thing I was
335. 38:02talking about before. So he was in search. It said, hey, I want to build
336. 38:06you an app. He said, okay. And it's basically got this interface that's becoming familiar,
337. 38:12right? Where you code the thing over, you prompt the thing over here, it writes
338. 38:17code over here. So that idea is starting to show up everywhere. My prediction is
339. 38:25where we move is all of this stuff is going to keep happening, but it's
340. 38:28going to increasingly drift into the background. right? Like, all this code is going to
341. 38:32be generated, all these interfaces are going to be dynamic, but they're going to happen
342. 38:36instantly, and it's just going to be, you know, some version of just you get
343. 38:40whatever application you need will be generated for you on the fly. We're not there
344. 38:45yet, but this is a step in that direction. Best way to present it, and
345. 38:49I've chosen to securely connect Gmail, Photos, and Calendar, so it's using personal intelligence to
346. 38:54make my suggestions even more helpful. That means referencing Gmail receipts, Calendar, and much more
347. 39:00to personalize. And boom, here it is. Looks like it's ready. All right, got a
348. 39:06beautiful planner already taking into account driving times and— which, by the way, regardless of
349. 39:11how good or not this planner is, the fact that it was done that quick—
350. 39:14what was that, a minute, you know, 45 seconds, something like that— that's not bad.
351. 39:20And Search knows that I have 2 kids. And that they love animals, and then
352. 39:24my oldest is learning to play chess. So actually, this second one here is kind
353. 39:28of awesome for my oldest, so I'm going to go ahead and heart that. But,
354. 39:33you know, two kids, got to make them both happy, so I'm going to lock
355. 39:37in the Happy Hollow Park and Zoo right there. And because this syncs with my
356. 39:41calendar, you can see it's already blocked off my afternoon to meet a friend to
357. 39:46watch this game down here. And below, it's got all of these cool restaurant reservations
358. 39:50and beautifully laid out, on Maps. All right, there it is. Some projects aren't one-off.
359. 39:56So you saw that, right? Voice activation, definitely Jarvis-like. I started asking my Alexa for
360. 40:08stuff and it doesn't answer like I want it to. This is— so Siri and
361. 40:13Alexa, because those are deterministic logic tree systems are going to be increasingly disappointing because
362. 40:22you just can't anticipate all the stuff people want to ask for. So today I
363. 40:27got Jasper to create an app, create application and tools and save them for later
364. 40:32use. I think everything will become AI and everything will be dynamic. Yeah, I agree.
365. 40:37I agree, Gareth. I think this is where we're headed. Again, we're in this, We're
366. 40:43in this goopy state right now, right? Where if you're someone like Gareth or someone,
367. 40:50you know, in here who's been geeking out on the tools and you do a bit deeper dive, you can go build all sorts of cool shit, but you have
368. 40:58to understand the piece parts you're pulling from and how you're doing it. And what
369. 41:02Google's showing today is this sort of weird in-between where you're still seeing code being
370. 41:07written. But then it— that code turns into an app that you just use. Like,
371. 41:13what's not clear to me is how do you save those apps if you want
372. 41:17to modify that app? Is that just a— is, is that weekend planner app something
373. 41:23that he's going to use just this weekend and then we'll write a new one
374. 41:27next weekend? Is he going to save that at some point? Does it become so
375. 41:32good he's like, I'm going to sell that? Is anyone going to buy any apps
376. 41:36anymore because they're just being dynamically generated? I— none of— we don't know, right? We're
377. 41:41in the goopy phase right now. We don't know what anything looks like. Intuit stock
378. 41:46was at $805 last year, it's at $380 today. Ouch. Software, SaaS, are you going
379. 41:57to need this stuff? You're doing a Google search, how to do my taxes. Up
380. 42:03will come a paid thing from Intuit and TurboTax, and then there'll be an organic
381. 42:09thing from Google that says, would you like me to dynamically build you your own
382. 42:13personalized, um, tax preparing software? Sure. And then it just will. Oh my God, this
383. 42:25is This is the perfection of MCP cross-app connectivity with that weekend planner. Search the
384. 42:33calendar to maps to activity reviews. Yeah, this is the thing. So, okay, so Sonny,
385. 42:40what you're pointing to is hard to get your head around. It's hard for me to get my head around. I mean, Sonny, like, we, we built a lot of
386. 42:58these websites that, you know, the early websites, right? We built a lot of the
387. 43:01early websites and, and the internet by and large operates the way we started building
388. 43:09websites in, in the '90s. You type in a URL or you click on a
389. 43:16hyperlink, takes you to a website. The page loads, it hooks to the database, it
390. 43:21does a thing, and it's designed for human beings to click on shit. That's not
391. 43:30going to be how websites are used anymore. The human interface will probably still be
392. 43:37there, but will probably be like 0.003% of how that website gets used. Agents through
393. 43:46MCPs will just go do shit on the websites. So websites are going to be
394. 43:51built for agents to provide services for these dynamic interactions. It's crazy. It's crazy. A
395. 44:01small percent of us are still in the group. The rest of the world is
396. 44:04completely unaware. Yeah, and it's— and, and here's the thing, this is the thing that
397. 44:11that should be depressing to, to anyone here that's been here for a while. Um,
398. 44:17this will make you feel shitty. Anyone who's here that's brand new to this and
399. 44:22is like, I don't have a clue, this will make you feel good. Largely, the
400. 44:27skills that we've learned over the last 3 and a half years are completely irrelevant.
401. 44:34I, I know Chef Kelly disagreed with this yesterday. But if you look at what
402. 44:41that thing just did, right? He was in a Google search. He had whatever version
403. 44:45of search this was. It said, do you want me to make you a travel
404. 44:49planner? It did, right? It connected all that shit. It wrote all the code. Let's
405. 44:56see. Jasper can update the apps. It builds everything on Codex and gives me a
406. 45:00link to run and start apps it creates. It is also— labeled Made by Jasper.
407. 45:05If you don't know Gareth, by the way, if you're interested in getting your hands
408. 45:10dirty, there's some really interesting folks in the AI Salon. Gareth Hood's one of them
409. 45:15who are diving deep on this stuff. He created a Jarvis clone, Jarvis from the
410. 45:22Iron Man fame, that just does stuff for him. He talks to it, it talks back to him, and it just does stuff for him. Um, so he's in the
411. 45:32middle of building this right now. So yeah, that's pretty cool. That's pretty swell. All
412. 45:38right, that's why it's called The Irregulars. We're pointing out what's irrelevant. Um, let's see.
413. 45:46I disagree. We learned. I don't— Sharon, I don't think that the learning was irrelevant.
414. 45:54I think that what we learned was irrelevant. Those are two very different things. Right?
415. 46:00The learning, understanding, understanding the intricacies and subtleties of where AI is good, where it's
416. 46:08not, like all of that is, is invaluable. That will carry on. But like that
417. 46:13whole idea of going to a tool, learning how to prompt it, learning how to
418. 46:18massage the output, manually having to glue this shit together. That's the part that I
419. 46:23think goes away. Lori Blair, me too. I think we learned a lot, a lot,
420. 46:29and now it's just new stuff. Maybe, maybe. Please don't tell me my 3-year investment
421. 46:34won't matter. No, I think, listen, anything that we have learned, don't get me wrong,
422. 46:40anything that we've learned is valuable. You, you've, you're, you've learned how to think, right?
423. 46:46And how to be critical about AI. That doesn't go away. But just the idea
424. 46:51of people that are starting now— well, we're in the Goop phase. People that are
425. 47:02starting now, they're going to learn some of the old stuff, some of the new
426. 47:05stuff, and the new stuff's going to be too complicated and janky, and people are
427. 47:08just going to be confused. We're going to be confused. So a year from now
428. 47:13it won't matter. Right now it still matters. Okay. Okay, the Dev Fireside today. One
429. 47:19thing they kind of called out, even the people at Google are on the rocket
430. 47:24ship we're on. They are learning and discovering part of AI every day. That's a
431. 47:28really good point, Gareth. I think that there's a lot of anti-AI sentiment right now,
432. 47:36and there's a lot of there's a lot of narrative about these frontier model companies,
433. 47:49lab companies, have all these nefarious plans and like they know what they're doing and
434. 47:55they're out to get us. I, I don't think they— like, they barely know more
435. 48:02than we know, right? Like, we're here, they're out here. You know, most of the
436. 48:09world is way back here. So we're able to sort of see quickly as they're
437. 48:14building stuff, you know, they're sort of dropping little pellets out of their, out of
438. 48:18their tractor that we're going, oh, that's cool, that's cool. We're playing back here, they're
439. 48:21up here. I don't think they know much more ahead of what they're building, right?
440. 48:29I think they have general ideas, but I think it's, I think it's It's like,
441. 48:35let's see how far we can push this technology. Let's see how far we can
442. 48:38push this approach. And then when they hit walls, they try to glue other shit
443. 48:42together, and some works and some, some doesn't. They now have a specific time on
444. 48:50the teams to go and experiment. I think that's smart. I think you'd have to.
445. 48:55Why burn tokens and spin up your own personal software? When we could pick from
446. 49:00software templates they already have spun up. We should create a template marketplace. I don't
447. 49:07know, Todd, that, that, that reeks to me of, of the current model. And I'm
448. 49:18not saying it's a bad thing. I think that the current model of we're going
449. 49:23to buy, we're going to build software for a specific purpose and then people are
450. 49:27going to go discover that software and they're going to use it for their own.
451. 49:31I think that's going to be valuable for some time, but I don't think it's
452. 49:35going to be valuable for long, 1 to 3 years maybe. And I think it's
453. 49:39just going to be a decreasing value curve. If these tools get good enough to
454. 49:45just dynamically generate the software you need in the moment, and it's personalized for you
455. 49:52each time, and it's exactly what you need, in the context that you need it,
456. 49:57I don't know that anyone would seek out a marketplace. But I don't know, like,
457. 50:00this is— we're, we're in full-on ChatTMZ speculation territory. That's how it feels. They seem
458. 50:07like they were just playing and sticking things together. Yep. Charles Adams: No matter how
459. 50:11much they plan, they lose control as soon as they succeed at building something better
460. 50:15than humans at everything. Right. The marketplace is called GitHub, for— well, for those that
461. 50:20know what it is. Hi, AI fam. Hey, Joy Purdy, what's happening? Okay, so let's
462. 50:25go see if we can find the Spark demo. I couldn't find one before, but let's, let's see if we can find one. Um, is, is it— I don't think
463. 50:33it's in this, in this thread, is it? Oh, let's see, is this it? Oh
464. 50:49wait, I didn't share again, did I? Hang on, calm down everybody, calm down. Shut
465. 50:58up, shut up! Oh my God, oh my God, you all, you all are very
466. 51:11funny. All right, so there's that. You can see that, right? Yes. Okay. It's your
467. 51:31personal AI agent that helps you navigate your digital life, taking action on your behalf
468. 51:37and under your direction. It runs on dedicated virtual machines on Google Cloud, and it's
469. 51:4424/7. And yes, you can close your laptop. It's powered— which, that's what Open Claw
470. 51:55was, that's what Codex is, that's what Cowork is, right? So this is Google's entry
471. 52:01into the consumer 24/7 agent your, your virtual employee. 3.5 and the Google Anti-Gravity Harness,
472. 52:11which allows— he just said the Google Anti-Gravity Harness. So the harness, when they're— if
473. 52:17you hear people talking about a harness with these AI agents, you've got your models,
474. 52:25and then ChatGPT is a harness, Anti-Gravity is a harness. So there, there are things
475. 52:31that have other functions that, that call out to, to different tools. And they're— the
476. 52:37harness is serving as an orchestrator, basically, of all these other tools to perform long-running
477. 52:43tasks easily in the background. Spark will integrate seamlessly with tools, starting with our own,
478. 52:50and in the coming weeks with third-party tools through MCP. And you can work with
479. 52:55Spark however is most meet in the Gemini app or soon through email and chat.
480. 53:02Let's have Josh come up. All right, so it looks like there is a demo.
481. 53:08Let's have Josh come up. Let's see if we can get Josh. Google Spark Josh.
482. 53:20Aren't one-off questions Yeah, that's right. Let me just make sure you're all still seeing
483. 53:41it. You're all seeing it, right? We're good. Let me see comments. Spark is not
484. 53:47available to most people yet. I thought I heard— yeah, this is one of the
485. 53:52problems with these, these Google announcements is they're not super clear on when stuff's actually
486. 53:56available. The incentives— they say there's no point at which it will ever stop. If
487. 54:04none of them can stop, they eventually replace themselves. Google is weaving the thread through
488. 54:10their ecosystem. All right, so let's go look at this little video. I don't think
489. 54:15this is a demo, but let's look at it. Let's go, let's give them what
490. 54:49they need. Can't get good energy, so watch me, just watch me, so watch me
491. 54:57do it. Let's give them what they need, can't get good energy, so watch me,
492. 55:03just watch me, just watch me do it, do it. Yeah, that's just bullshit, bullshit
493. 55:10marketing, which is fine. Let's see, Josh Gray. Is this the guy from— no. Document
494. 55:48that. Have it today. Have it today. That's normal. Good. Uh, here we go. Is
495. 56:00this Josh? New capabilities. Yeah, here we go. Okay. This is it. One more thing.
496. 56:12This with the new capabilities of Spark in the Gemini app, macOS app. I need
497. 56:19this yesterday. Okay. We're still there. Not sure if anyone has been to Google Labs
498. 56:29yet. Yeah. Google Labs. I'm going to go check out Google Labs in a second.
499. 56:34I need an AI vocabulary cheat sheet. Yeah, there, there is, there's, there's a lot.
500. 56:44It's, this is part of the goop phase right now, Mary, is that, um, these
501. 56:51two worlds that we're in, it was complicated enough when it was just discrete tools,
502. 56:57but what's happening when they're starting to say things like harnesses and agents and just
503. 57:04all of the stuff that's coming. What they're doing is they're mixing a bunch of
504. 57:07tools together, um, and so it's just, it's, it's getting more and more and more
505. 57:12complicated. So yeah, we do, we do need a vocabulary cheat sheet kind of thing.
506. 57:18Um, all right, let's go watch this, this video. But you're not wrong. Using anti-gravity,
507. 57:25they did over 100 features in less than 100 days. Now, 2 weekends ago, we
508. 57:31were hacking on the Mac app and we came up with something cool and we wanted to sneak it into the show. Oh, this is, this is on the tail
509. 57:38end of the Spark demo. This is the, the macOS, uh, just one more thing.
510. 57:45This is some additional feature. You want to see it live? Yeah. All right, you
511. 57:52got a big summer trip coming up and we've got to find a kennel for
512. 57:56our two dogs. And here's a picture of our two dogs. There's Hank looking good,
513. 58:02and Louis Cinnamon, one of the most interesting names for a dog I've heard. Uh,
514. 58:07what we're going to do— and remember, when you have to go to a new
515. 58:10kennel, there's lots of paperwork, allergies, vaccines, all the history you have to pull together.
516. 58:15It's so painful. And so what you can do with this on Gemini on macOS
517. 58:21is actually take a look at a bunch of documents like this, you'll be able
518. 58:25to select them all and then long press the function key and just dictate the
519. 58:31email to the camera. So it works something like this: Hi there, I need to
520. 58:38do a short boarding stay for my two dogs, Louis Cinnamon and Hank, starting this
521. 58:43Thursday. Oh wait, no, actually it's this Friday. They've never stayed with you before, but
522. 58:48they're very social dogs And also, can you turn these files into a table with
523. 58:53their details, allergies, recent vaccines, and make this email sound friendly so we make a
524. 58:59good first impression? All right, I'm gonna release the function key. You can see Gemini's
525. 59:06thinking at the bottom here on this Macbook. What it's done is because I've selected
526. 59:11those files in Finder, using its multimodal understanding, it can go through the PDF, can
527. 59:18go through these images of their invoices, and it's all controlled by my voice. So
528. 59:23it can actually take all that complex information. Look at that, there it is. It's
529. 59:28got a table in line. It's also so amazing because it corrects. Remember I said
530. 59:37Thursday? No, scratch that, Friday. It picks up that and automatically cleans up. All right,
531. 59:45so that's, that's basically the same thing that they showed us with the, the Docs
532. 59:49Live functionality, but in the, in, in the Mac. He's over here in the Finder
533. 59:53selecting documents and then pulls data out of the documents, puts it in context in
534. 1:00:00that email. Pretty slick. All right, cool. Let's see the Gemini announcement. Google Spark. That's
535. 1:00:15just a picture. All right, I'm gonna stop looking for that because it's getting boring
536. 1:00:28now. Um, let's see, let's go look for, um, let's look for, um, Gemini Omni.
537. 1:00:39So Gemini Omni is the, is the, uh, is the world model. So let's see,
538. 1:00:49here's this, here's this video. This is what we'll look at. Let's go, Demis Sabes.
539. 1:01:49I'm excited to announce Gemini Omni. Our new model that can create anything from any
540. 1:02:01input. It combines Gemini's intelligence with the best of our generative media models for a
541. 1:02:06new level of world understanding, multimodality, and editing. Models like Veo, Nanapanana, and Genie are
542. 1:02:15able to create extremely realistic videos, images, and interactive simulations. Although not perfect, they already
543. 1:02:22demonstrate some impressive notions of intuitive physics. And with Omni, we've now made even more
544. 1:02:28progress. It's a step change in simulating things like kinetic energy and gravity. Previous systems
545. 1:02:35would have found these concepts difficult. Gemini's world knowledge and reasoning really shine in Omni.
546. 1:02:42It can translate complex ideas into highly accurate videos. So for example, you can give
547. 1:02:48it a simple prompt like make a claymation explainer of protein folding and get this.
548. 1:02:53Proteins start as chains of amino acids. They fold into patterns like the alpha helix
549. 1:02:59and flat sections called beta sheets, forming a perfect three-dimensional shape. But the initial generation
550. 1:03:06is just the start. The creative process is rarely a single step. It's usually iterative.
551. 1:03:12Just like Nana Banana redefined image editing, Omni gives you a more natural way to
552. 1:03:17edit video with conversational language. What's really cool is you can give it your own
553. 1:03:22videos, for example, this selfie, and change reality in a really fun way. You can
554. 1:03:32easily adjust the details and style or even add elements. And the whole scene altered
555. 1:03:38to reflect your new idea. So, so let me come back here for a second.
556. 1:03:49Um, let's see, this is the first— oh, Omni. Yeah, Omni. This is the first
557. 1:03:58I'm hearing about it. So here's the thing about Omni, it is available now. It
558. 1:04:04ain't great. So, so there's Omni Flash or there's, well, their naming is screwed up.
559. 1:04:14Gemini 3.5 Flash has the Omni model inside it, but I think it's called 3.5
560. 1:04:22Flash, but you don't even know that. If you go to gemini.google.com or if you
561. 1:04:27download the Gemini app, it's in there. Gemini Omni Pro ain't out yet. So what
562. 1:04:36we're playing with is the cheaper, faster version of it. So I would imagine that
563. 1:04:41some of the videos that they showed there might have been from the Pro model.
564. 1:04:45I don't know, um, but it's still pretty impressive. And I'll show you some videos
565. 1:04:49that I did, and I'll show you— there's another video here we'll look at here.
566. 1:04:52But let me see what the comments are. Uh, Mary Carol, this is The first
567. 1:04:56I'm hearing about it. Remember we were wondering if Google would drop it into the
568. 1:05:05game? I don't know which game you're talking about there. Oh, there was actually a
569. 1:05:08really interesting game. I wonder if I still have that tab open. Hang on a
570. 1:05:12sec. No, I don't have it. Um, I am officially overwhelmed. I'm with the— I'm
571. 1:05:24with you. World model. Okay, so when we talk about a world model, couple of,
572. 1:05:33couple of important things. So, so there's a couple of clues. So that was Demis
573. 1:05:40Hassabis. So Demis Hassabis is the CEO of DeepMind. DeepMind was acquired by Google. DeepMind
574. 1:05:47is the group that did AlphaFold. Which won Demis Hassabis the Nobel Prize for Chemistry
575. 1:05:53because they predicted how all of protein— all of the proteins in all of life
576. 1:05:58are going to fold to like 95 or 97% accuracy. And he gave it away
577. 1:06:04to the world. That's Demis Hassabis. So the fact that he is announcing Omni as
578. 1:06:10an important— it's not just— this is not just a video maker, right? So this
579. 1:06:16Omni model is, he basically said anything in, anything out, meaning you can start with
580. 1:06:22an audio input and come out with a video. You can start with a PDF
581. 1:06:26input and come out with a PDF. You can write anything in, anything out. So
582. 1:06:31that's pretty remarkable. Right now, the video, what's being output are videos. But if you've
583. 1:06:39got a model that understands the physics of what's in the video model, then the
584. 1:06:46next logical conclusion after that is you're going to be able to output a 3D
585. 1:06:51world that you can actually interact with. That is not just there's objects in a
586. 1:06:57world that we can have programmatic physics applied to. It's that the world itself understands
587. 1:07:06physics and understands all of what a large language model can understand. So we're going
588. 1:07:13to be able to be in a world. So imagine being able to enter that
589. 1:07:17video, the claymation protein folding thing that he talked about. You can just prompt it
590. 1:07:23to say, give me a claymation explainer about protein folding. We're going to be able
591. 1:07:27to be in a world where you're going to be able to interact with stuff
592. 1:07:30that understands all of what it's doing. Like that's where this is headed. And so
593. 1:07:37this is the first, or maybe the next best version of a video model that
594. 1:07:44really does understand all of what's in that video and what the relationships are between
595. 1:07:51the things in that video and how they relate to concepts in the world. Anything
596. 1:07:56in, anything out. So that's what a world model is. And it starts to— the
597. 1:08:04implications of this Right now, if you've got a physical robot and you only have
598. 1:08:09a large language model, the robot has to translate words into physical action. But if
599. 1:08:19you have a model that understands more than just words, it understands physics and how
600. 1:08:24the world works, then that robot is going to understand how the world works. It's
601. 1:08:29not going to have to translate language into physical world. It's just going to understand
602. 1:08:33physical world because the model can understand all that. So this is a, this is
603. 1:08:38a fairly major transition. The implications of it are really hard for me to get
604. 1:08:46my head around what it means. Like, like the more of this shit I learned,
605. 1:08:55the more I'm like, yeah, we might live in a simulation. 'Cause, you know, 3
606. 1:09:00years ago when people were like, I think we're living in a simulation, I'm like,
607. 1:09:03that's not possible. There's too much detail. There's too much randomness that you, you know,
608. 1:09:08we just live in this world and all this stuff happens that can't be. Well,
609. 1:09:14we're, we're sort of step changing toward being able to interact in a world where
610. 1:09:19it can just understand shit, how the world works and just make new shit. In
611. 1:09:24the simulation. So I don't know. Uh, what's this say? Chat offered me a living
612. 1:09:33cheat sheet with these sections that I'm going to make into a daily update. It
613. 1:09:37will determine new AI world words I keep seeing. I like that. What they mean
614. 1:09:43in normal language, where I saw it, why people are using it now, and how
615. 1:09:47I might use it. That's really smart, Silver Fox. It's really good. Um, I'm gonna—
616. 1:09:54I'm going to think in simple large concepts. That's the only way I can look
617. 1:09:59at the AI galaxy. That's smart. I'm reading The Infinity Machine right now. It's about
618. 1:10:05Hassabis. It's an amazing book. Yeah, Demis Hassabis, if you haven't seen the, uh, I
619. 1:10:11forget what his documentary is called. There's a documentary on him where they started tracking
620. 1:10:15him. Like, he was a world-class chess player as a kid. And then he got,
621. 1:10:22he got bored with chess. He was like, yeah, I've kind of figured out chess.
622. 1:10:25I could keep doing that, but I kind of want to, you know, solve artificial
623. 1:10:31intelligence. And so he got a game at like 17 at England's most prestigious game
624. 1:10:36studio and created one of their most popular games where all of the characters were
625. 1:10:41AI-driven. He's, he's He's an insane genius. All right, Demis was a chess prodigy. Yep,
626. 1:10:51from a very poor family. Then he built AlphaGo. Yep. Yeah, they— yeah, in the
627. 1:10:57documentary, they— there's the whole thing about them beating the AlphaGo guy. It's crazy. AlphaFold.
628. 1:11:03Yep. Yep. Yeah, it's crazy. Okay, so let's see, let me turn that off. Um,
629. 1:11:17all right, let's go look at a couple other videos here. So let me go
630. 1:11:20back to sharing, sharing the screen. Let's see, Gemini Omni. Let's go Gemini Omni video.
631. 1:11:34Let's That's a 10. That we saw. This is a cool one. This is pretty
632. 1:12:02cool. So it's doing audio as well as, as well as video. Build worlds. We're
633. 1:12:47not there yet, but that's what's coming. All right, so let me go show you.
634. 1:13:23Let me stop sharing this for a second. Let me share in a slightly different
635. 1:13:27way. And I'll go show you some of the videos. I'll show you some videos I made. So, okay, a couple of things. So if you want to play with
636. 1:13:33this, you can play with it in Google Flow. So if you go to labs.google.com,
637. 1:13:40you can play with it in Flow. You need to probably have an Ultra subscription.
638. 1:13:46There's really weird rate limiting with this right now. Um, so you can play in
639. 1:13:51Google Flow. You also have access to it in aistudio.google.com. You also have access to
640. 1:13:57it in gemini.google.com. And then you also have access to it in the Gemini app.
641. 1:14:04Now, all of the fancy ways to interact with this thing are severely rate limited,
642. 1:14:12but If you just go to your Gemini app on your phone, um, I can
643. 1:14:23just say, um, make me a documentary video narrated about the migration of monarch butterflies,
644. 1:14:38um, during their great migration. And use a David Attenborough type accent, but don't use
645. 1:14:46his exact accent because I just want it to be fictitious, but to sound like
646. 1:14:51one of those kind of documentaries. All right. So I will, once that's, we'll make
647. 1:14:59sure. Okay. So it says generating video. So it'll take about a minute. It's not,
648. 1:15:03it's not too bad. So if you want to play with this, play with it
649. 1:15:07in the Gemini app. Because they don't seem to be rate limiting this, but they're
650. 1:15:12rate limiting everything else. So if you played with it, like, at gemini.google.com, and it
651. 1:15:17was like you can make 3 videos, and you're like, well, that's stupid, I agree.
652. 1:15:27Video to video editing in real time. Yeah, I'm getting impressed with Omni as I
653. 1:15:32used to, used to be with Adobe in the early days. By the way, Adobe
654. 1:15:37just cut a deal with DeepMind. Okay, so timeout, hold please, wait, what? One of
655. 1:15:48the things— so, okay, the AI haters, one of the things they hate is that
656. 1:15:58all of these models were trained on all of the shit that's out on the
657. 1:16:03internet. So all of the copywritten stuff. By the way, our video's done. Why is
658. 1:16:11it not? Oh, my sound. Each year, millions of monarch butterflies undertake an incredible journey.
659. 1:16:20They fly thousands of miles, guided by an instinctual compass. Hey, man. Let me save
660. 1:16:27this video and then we can actually go look at it. Each year, millions of
661. 1:16:39monarch butterflies undertake an incredible journey. They fly thousands of miles, guided by an instinctual
662. 1:16:45compass. Each year, millions of monarch butterflies undertake an incredible journey. They fly thousands of
663. 1:16:53miles, guided by an instinctual compass. Okay, so now what I'm going to say is
664. 1:16:58I'm going to go back into— to Gemini. I'm going to say, that video is
665. 1:17:05great, but when you talk about them migrating thousands of miles, can you put up
666. 1:17:09an accurate migration map, um, as a— as B-roll in between the butterflies flying? So
667. 1:17:22this is that thing of where, where Demis Isabas talked about, just like Nano Banana,
668. 1:17:28you can edit things in context. You can now do that with video. Um, Google
669. 1:17:34is in Adobe for the last 4 months now. Is chat rate limit serious? Getting
670. 1:17:40worse by the day. No, no rate limit. I don't have to remember the numbers,
671. 1:17:46letters, special characters. Just go Find the app. Um, yeah, so, so just as mind-blowing—
672. 1:18:01remember, was it last week or the week before we did, um, the new ChatGPT
673. 1:18:08image, image model that's, that's really remarkable? Like, like, this is, this is going to
674. 1:18:14be sort of the next version of that. And then What— here's the thing, I
675. 1:18:19like— I'm all over the place. Let me, let me, let me, let me lower
676. 1:18:24the, lower the temperature for a second. The deal with Adobe— so everyone was up
677. 1:18:32in arms about how these models were trained on everyone else's shit. Adobe kind of
678. 1:18:36had the high ground and said, hey, we've only trained our stuff on licensed material
679. 1:18:43or public domain material. Consequently, their image and video models sucked because they weren't trained
680. 1:18:52on all the stuff. Everyone else from Google to Midjourney to OpenAI, they've all trained
681. 1:19:03them on just all the data, right? And they're finding ways to whitewash their data,
682. 1:19:08but they were all trained on crappy data. So the fact that Adobe and DeepMind
683. 1:19:15just did a deal probably says that Adobe's looking to have higher quality models available
684. 1:19:21to their subscribers. So, so I don't know if that's going to damage their brand
685. 1:19:27or not, because you have to be paying attention to what's happening, but Adobe's models
686. 1:19:31are just going to start getting better because they cut this deal. So that's, that's
687. 1:19:35fascinating. Um, all right, let me go show you some of the stuff that I've
688. 1:19:41made in the past day with this thing. Let's see, I'm generating your video. Doesn't—
689. 1:19:50it looks like it stopped. Um, try again. All right, I'm going to share my
690. 1:19:58entire screen. Oh wait, hang on, I did not share my audio. Share my entire
691. 1:20:08screen, share audio. Okay, so now, all right, here's all these videos. Let me go
692. 1:20:24to this icon. So it's a gallery. There we go. Okay, I don't like gallery
693. 1:21:13view. There's icon view. Okay, all right, so You all know that I do the
694. 1:21:23muscle car thing. So here's, here's one of the first videos that I did in,
695. 1:21:29in this new video model. So historically, when I would do a muscle car in
696. 1:21:37an abandoned factory and then turn that into a video, the wheels wouldn't roll right.
697. 1:21:42The, like, everything was wrong. The, the exhaust pipes would be coming out of the
698. 1:21:46front of the car. They would just be a disaster. Um, so here is— prompt
699. 1:21:54was a '69 GTO Judge in an abandoned factory driven by a mysterious guy screeches
700. 1:22:00out of the, the factory. Right. That's pretty good. Then I did, um, a, uh,
701. 1:22:19kind of a joke thing on the, uh, on the, uh, the, the rare Antilles,
702. 1:22:27the Lesser Antilles picklebird. Deep in the lush rainforests of the Lesser Antilles lives the
703. 1:22:34incredibly rare picklebird. Its remarkable camouflage makes it nearly impossible to find. So that was
704. 1:22:42pretty good. Um, here was the, the Skittle salamander. Here on the mossy branches— oh
705. 1:22:48wait, that was— here's this one. But even more elusive is the Skittle salamander. Its
706. 1:22:54skin is adorned with a dazzling array of bright spots. They serve as a brilliant
707. 1:22:58and effective against the colorful backdrop of the jungle. Right, and then here I did
708. 1:23:02the symbiotic relationship between the two. Here on the mossy branches, a strange alliance is
709. 1:23:08formed. The pickle bird provides a cleaning service, rid the salamander of unwanted parasites. In
710. 1:23:13return, the bird gains a nutritious meal. So, yeah, so that's not bad. I mean,
711. 1:23:19again, these were just shitty prompts, right? And it did the narration, it did all
712. 1:23:25that stuff. So then one of the things I wanted to try was taking a
713. 1:23:30video of myself and turning that into other things. And I don't think I have
714. 1:23:36the original video here, but Let's see. This is sort of what it looks like.
715. 1:23:44This was, so I took the original video and I said, create the original video
716. 1:23:49with monarch butterflies on me. And so it did this. We're going to have some
717. 1:23:54fun tonight on the AI Learning Lab. Join the pre-show at 7:10 PM Mountain Time.
718. 1:24:00So like, I don't know why it made it sideways like that. So I said,
719. 1:24:03make it vertical. So then it did this. We're gonna have some fun tonight on
720. 1:24:07the AI Learning Lab. Join the pre-show at 7:10 PM Mountain Time. Right? So, so,
721. 1:24:13so I ended up, I ended up using this one and just rotating it manually,
722. 1:24:16but like, why did it do that? I don't know. And then sometimes it used
723. 1:24:20my voice and sometimes it didn't. So here's one where it used my voice. We're
724. 1:24:24gonna have some fun tonight on the AI, AI Learning Lab. But it, but it
725. 1:24:28screwed up the script. We're going to have some fun tonight on the AI Learning
726. 1:24:34Lab. Join the pre-show at 7:10 PM Mountain Time. At 7:10 PM PM Mountain Time.
727. 1:24:42Here was one where it didn't use the voice at all. Right? Whatever. Then I
728. 1:24:58said, uh, put me underwater and it's sort of me, but sort of not me and sort of my voice and sort of not my voice. Welcome back to our
729. 1:25:06underwater AI learning lab. We are exploring new depths with AI. It is truly a
730. 1:25:12fascinating world. Um, let's see. Then I had, oh, have me standing on the beach.
731. 1:25:22We're gonna have some fun tonight on the AI learning lab. Join the pre-show at
732. 1:25:257:10 PM Mountain time. So there, it kept my voice but changed me into some
733. 1:25:31older, skinnier dude. Maybe it's just trying to make me in shape. Then, then I
734. 1:25:37did this dude. We're going to have some fun tonight on the AI Learning Lab.
735. 1:25:43Join the pre-show at 7:10 PM Mountain Time. Horrible. And then the final one I
736. 1:25:51did was this one. We're gonna have some fun tonight on the AI Learning Lab.
737. 1:25:59Join the pre-show at 7:10 PM Mountain Time. So again, all of these were fairly
738. 1:26:04crappy prompts with fairly crappy results, so it's, it's still janky, but it's, uh, it's
739. 1:26:10a pretty impressive— it's a pretty impressive, uh, capability. All right, where are you going
740. 1:26:18to make the videos for free? Go to the, um, to the Gemini app. Let me see. Video downloaded. So let's see how it did. So what I asked it
741. 1:26:33here was I asked it for, um, to add a map to the, to the
742. 1:26:41migration video. Incredible journey. They fly thousands of miles, guided by an instinctual compass. Each
743. 1:26:55year, millions of monarch butterflies undertake an incredible journey. So that's not bad, that's not
744. 1:27:01bad. Okay, Sunny and labs.google— labs.google/fx/tools/flow— that gets you to the Flow tool. Again, all
745. 1:27:14of their web-based tools that have access to this video model seem to be highly
746. 1:27:19rate limited. So if you're paying $200 a month for Gemini Ultra, you probably have
747. 1:27:24access to it there. If you're getting rate limited out, just download the Gemini app.
748. 1:27:28They don't seem to be rate limiting this, which as soon as this— if anyone's
749. 1:27:32from Google's in here watching this, they're probably going to turn this shit off. So
750. 1:27:37go make some videos now. My app says it's not free though to make videos.
751. 1:27:43I, I, again, their product marketing and their product rollout strategy is horrible. So it's
752. 1:27:51possible you can't play with this. It might be that I don't know why I'm
753. 1:27:56able to play with this here but not anywhere else. Like, why, when I went
754. 1:28:00into Google Labs earlier, it wouldn't let me make videos. Anyway, okay, um, it's coming
755. 1:28:09up to the top of the hour. Um, you have to be a Google Storage customer to get the Gemini app. Oh, I didn't know that. All right, so go
756. 1:28:18get yourself some Google Storage and you can get the Gemini app. I— again, their
757. 1:28:25product marketing is a fucking disaster. It's just, it's com— it's all over the place,
758. 1:28:29right? It's, it's the The result of all these different divisions with this— within this
759. 1:28:34huge corporation, just— it's— they need, they need an angry Gen Xer sitting in the
760. 1:28:41corner going, stop it, stop, make a single rule. When we roll things out, here's
761. 1:28:50how it works. But they— there's too many fiefdoms. Okay, um. Thank you all so
762. 1:28:58much. Hope you enjoyed tonight. Please go check out the different spaces in the AI
763. 1:29:05Salon. We're here. As complicated as this stuff is, it's gonna get more complicated. Things
764. 1:29:13are moving fast. Being connected, talking, asking questions, pointing out that I wasn't sharing my
765. 1:29:19screen for 15 minutes. These are all really important things. Please connect with one another
766. 1:29:25in between these meetings. Like, the whole purpose of the AI Learning Lab is for
767. 1:29:30us to be in a discussion together. We've now moved this inside the AI Salon
768. 1:29:34so we can continue that discussion in between these moments, okay? Um, share what you're
769. 1:29:41learning. Share, like, anything that you can that might help other people. If you see
770. 1:29:46people that are new that are asking questions, do me a favor, please connect with
771. 1:29:49them. Say hello to people, welcome them in. Okay, really, really important. Um, this was
772. 1:29:56awesome, thank you very much. I have the Gemini app. Anywho, never mind. Okay, good.
773. 1:30:01I have the Gemini app and I made a video. Great, what am I missing
774. 1:30:04here? Had to leave for a second. You're not missing anything. If you can make
775. 1:30:08videos in your Gemini app, then you seem to have access to this new model
776. 1:30:14that does all this crazy-ass stuff. Again, the video model that's in gemini.google.com, Gemini, the
777. 1:30:23Gemini app, and Flow are all this new Gemini 3.5 Flash/Omni. The Flash model is
778. 1:30:34the smaller, jankier model. The Omni Pro model isn't out yet. They said it'll be
779. 1:30:43coming sometime. That's probably because that one's really expensive for them to run, right? So
780. 1:30:51they're probably going to limit that to the Ultra users, the $200 a month users,
781. 1:30:55things like that. Danielle says she can't access the OmniFlash stuff that came out today.
782. 1:31:00Okay, again, Gemini app says it's not free to make videos. Yeah, it's— their product
783. 1:31:08marketing is all over the place. Okay. Um, that's it. So tomorrow, if you're a
784. 1:31:13Mastermind member, tomorrow at 10 AM Mountain Time we have the Practice Lab. So, um,
785. 1:31:18Sid Hargrove and myself, um, host that. Please come tomorrow if— even if you haven't
786. 1:31:24been in a while but you're part of the Mastermind, come to the Practice Lab.
787. 1:31:29Getting ourselves calmed down and focused and intentional as all this chaos happens with these
788. 1:31:34tools accelerating is actually really important. So that's one of the things we're doing in
789. 1:31:37there. So please come join that. Um, there's probably some other stuff coming up. I
790. 1:31:42don't have producer Brandon telling me what to do, but please go check out events.
791. 1:31:46At the top of the list over here is the events tab. Go check that
792. 1:31:49out. Um, there's— I think there's an LOL. There's a special kind of 15-minute LOL
793. 1:31:54happening on Saturday that Vicki's putting on, so please go check that out. Yay, Mastermind
794. 1:32:00Practice Lab! Beautiful. Um, the best place to be on Thursday mornings. I agree. All
795. 1:32:06right, I updated the app and now it's running the new model. Beautiful. Beautiful. Update
796. 1:32:11your app. Agreed. Calm down. Tone down the temperature. Thank you. Phew. Okay. I know.
797. 1:32:17And I've been a little ampy tonight because there's new stuff and none of this
798. 1:32:21matters. Come to the practice lab and we'll talk about centering ourselves and breathing. Yes,
799. 1:32:311 PM, 4 PM Eastern time is Vicki's 15-minute talks. Beautiful. On this coming Saturday.
800. 1:32:36Okay, great. Thank you all so much. Um, I, I, I appreciate you all beyond
801. 1:32:42what you know, and, uh, I will see you Friday. So Friday we've got office
802. 1:32:47hours at 11 AM Mountain Time, uh, right here in the salon. Um, I'll also
803. 1:32:53post about it on LinkedIn. That's, that's where I primarily market that. Um, and then
804. 1:32:59Friday night date night will be and ask me anything. So think about questions that
805. 1:33:04you might have for Friday, um, as you start playing with these tools. Gather up
806. 1:33:09your questions, bring them Friday, and I will do my best to answer them. Okay, thank you all so much. Have a beautiful night. Good night.