8/31/2026 - Runway Solaris and the Shift Toward Code-Free Generative Interfaces

Transcript

1. 0:01[music] [music] [music]
2. 0:22Heat. Hey. Hey. Hey. [music] >> [music] >> All right. All right. All right. All right. Welcome to the AI Learning Lab.
3. 0:38Uh this is uh it's Monday, August 31st. August is gone. Summer's gone. [laughter] Labor Day is here. Is it Labor Day? It's
4. 0:50Memorial Day in the spring. Labor Day in the fall. Right. Labor Day is coming. Um, so we got long weekends at some
5. 0:57point. Maybe Monday. I don't know. Hey, Steo. What's happening? So, here's what's going down. Um, let's see. Manage poll.
6. 1:07Start. All right. There's a poll on Tik Tok for those of you on Tik Tok. So, here's what's going down. If you're on
7. 1:13Tik Tok, this is the backstage view. Um, you get to watch me, but you won't be able to see what I have on screen.
8. 1:18Mondays. So, so here's the new format. Monday, Wednesday, Friday, we're talking about um we're using the cycle of AI readiness
9. 1:31which was developed based on how people who are doing interesting and remarkable work in the AI salon, how they do what
10. 1:38they do. Um it it developed into this thing called the cycle of AI readiness. And the cycle of AI readiness has you at
11. 1:45the center. So basically the idea that in the AI salon Our our philosophy is it's not us versus AI. We're not
12. 1:54competing with AI. If we do this right, you put yourself at the center. You figure out who you are, what your values
13. 2:01are, what you want to do in the world. That becomes the driving imperative. And then you use AI to amplify that. And how
14. 2:10you get good at it is you is you cycle through this cycle of play first, create excellence, and generously lead. And so
15. 2:19Monday's going to be play first night. Wednesday is going to be create excellence night. So we're going to
16. 2:24actually build stuff and and refine things and and you know not just play but actually build things. And then
17. 2:31Friday will be generously lead. And that um will probably look like talking to people, bringing people up on stage
18. 2:38talking about um how you can teach other people what you know, what works, what doesn't work. um how you can interact
19. 2:46with people that might be afraid of this stuff or pissed off about it. Um which is kind of where we are right now.
20. 2:54We're we we've got a transformative technology. Um if you haven't, one of the amazing things you can do in the world is uh research
21. 3:05history. Every time there's been a transformative technology, everyone thinks the world is going to end. People get really afraid of it. Uh,
22. 3:14and then it ends up happening and it takes some amount of time. Uh, 100% of people want to hear Champy sing, but we
23. 3:20only got one vote. So, [laughter] oh man. All right. Um, so if you're on TikTok and you want to see uh what I'm
24. 3:31putting on screen, uh, go to the, uh, go to the my profile and click on the stream. That'll bring you over to the AI
25. 3:39salon. Uh, if you're new to the AI salon, welcome. Uh if you look down the left hand side of the interface of the
26. 3:45AI salon, you'll see a bunch of different spaces. Those spaces are um different conversations that people are
27. 3:53having around different topics. Um there's community chat and a community feed. There's business group and there's
28. 4:01a life hacks group and um advocacy and policy and things like that. So there's all sorts of all sorts of things going
29. 4:08on in the AI salon. If you if you haven't explored it, please explore it. If you haven't introduced yourself,
30. 4:14right toward the top is a space called introduce yourself. So, go do that now. Um, and let us know who you are, where
31. 4:21you are with this AI stuff. You know, how far in it are you, even if you're just beginning. Uh, and uh, and let us
32. 4:28let us know what you're excited about. Um, or scared about. Um, you don't have to be excited about AI. It's called the
33. 4:36AI salon, but it doesn't mean we're just cheapest sheepishly optimistic about it. Um, we can be scared about it. We can be
34. 4:44angry about it. We can not like how things were done. Um, but I think that one of the things that is a hallmark of the AI salon is
35. 4:59I believe that AI is not going to go away. And if AI is not going to go away, then you only have two choices. You either deal
36. 5:07with it or you don't. And it's my contention that if you don't deal with it, it's going to suck worse than if you do.
37. 5:17Because if you do, what you'll learn and what we've learned over um the past four years in the AI salon is that
38. 5:27you can reclaim your agency and you can use AI as an as an amplifier of who you are and what you do and what you believe in. So
39. 5:38if you don't deal with it, it's going to deal with you. Exactly. Outstanding question. How do you get updates
40. 5:43if you run models locally? Um that's a question from Steve O. So when you run a local model, the basically the way you
41. 5:51do it is you run a program like Olama or LM Studio. LM Studio is the one I run. And what those are is kind of like a chat GPT interface
42. 6:04um where you can find and download local AI models that can run on your what whatever uh machine you're running. Um so the the one that
43. 6:19I'm doing this streaming on is an old M1 with 16 GB of RAM. You can't run anything very powerful on that. put over
44. 6:28on the desk over there. I've got a new as of April maybe M5 Max, I don't know. I I forget what the spec was, but like the the baller one at
45. 6:41the time um with 64 gigs of RAM and on that I can run really powerful local models. And so once you have it running,
46. 6:50it's it's a file or group of files that's that's literally sitting on the hard drive of of that machine or the
47. 6:59SSD. It's not technically a hard drive. Hard drives have rust on them and spin really fast. These are just chips. But
48. 7:07sitting on the chips over there is a file um that contains all of the knowledge of humanity and then you can interact with
49. 7:15it just like chat GPT. So your question Steo about how do you how do you update it if you're if you're asking about like whatever the latest
50. 7:25that model was trained on LM Studio and O Lama also have tool access. So that means that you can um you can have the LLM search the internet
51. 7:38so it can go out and search websites and things like that. So if you need the latest up-to-date information like the
52. 7:45weather or you know the latest news on a particular thing, it can actually go out and search the internet and then do all
53. 7:51of the the data processing locally on the machine. Um and then if you want to update like as new models keep coming
54. 7:58in, you just need to pay attention to that and then you basically download them to that machine. Um I don't think
55. 8:05that's an automatic download because the files tend to be really big. They tend to be, you know, 25 gigabyte files, uh, things like that.
56. 8:16Okay, good. Awesome. Um, okay. So, if you're on TikTok, if you're new here, um, we're live on the AI salon, so I'm
57. 8:26going to be demoing stuff tonight. Tonight is a play first night. So, we're going to be learning by doing, learning
58. 8:33by just exploring. Um, going and looking looking at some new things and, you know, looking at news things and then
59. 8:39maybe jumping over to sites and playing with them. So, as I'm looking at news sites, as I'm demoing things, you won't
60. 8:46be able to see them on TikTok, but you can inside the AI salon. So, if you go to my profile up in the corner here, um,
61. 8:52that'll take you over to the live stream. All right. Beautiful. Um, I'm using Alienware and V5 for my local models. Cool. Yeah,
62. 9:05here's the thing about local models. Um, they're getting really good. They're getting quite powerful. So, you can
63. 9:13actually do real work. You can do a gentic work. Like if you're doing something with codeex where you're
64. 9:18running things locally, you can actually run codeex or claude co-work or claude code or Hermes or openclaw 2.0 came out
65. 9:28today. Um you can run these agentic platforms um that don't cost you money other than electricity to run. They can use the
66. 9:38local inference and that's incredibly powerful. The other thing about running things locally is that all of your
67. 9:46context, all of your data lives locally. You own it. It's not on someone else's server. Um, that's incredibly powerful. and
68. 9:57moving forward, those people that have taken the time to set up how their AI works and thinks like they think and
69. 10:07speaks like they speak and understands the projects they're working on and the values that they hold, those people are
70. 10:14going to be able to step into any situation and use AI in a much more powerful way than someone that's just
71. 10:21stepping in and using it generically. Um, so, so there's all sorts of value to running things locally. So, if you
72. 10:27haven't played around with that, that's something something good to do. Um, tonight we're going to be, um, I'm going
73. 10:34to play with um, Grockbot. I think I'm going to head over to there's a new editing tool inside MidJourney. I think it's MidJourney 8.2.
74. 10:47There's a new editing tool within it that I think looks pretty cool, but I haven't played with it. Um, I'm going to
75. 10:54talk tonight about the new um, Miniax H3. I don't know, whatever.
76. 11:08A video model came out that was good and then on the foul website FAL, they they did something to it to make it 50 times faster.
77. 11:22And it can render a 5-second video from nothing, from a prompt in 3 seconds. Five seconds of output in 3 seconds.
78. 11:36The implications of that are that you could basically write or vibe code a computer program that prompts it
79. 11:46every 4 seconds and the thing will just run indefinitely. Now, the cost of running it indefinitely is anywhere between $45,000 and $11,000 a
80. 11:58day. I could I think that's depending on if you're doing 768p or 420p or 460p, whatever it was. Um but
81. 12:10we are now today there is technology in the world where you could have a 247 entertainment channel that is being generated in real time.
82. 12:23We're there today. So I'm going to go play with that. Um and then there's a couple other news items. I can't
83. 12:30remember what they were that we'll go look at and talk about. Okay. So if you're on Tik Tok, welcome. Um the
84. 12:38experiment, we'll do this for a couple of weeks. So the new format of the AI learning lab is Monday play first,
85. 12:46Wednesday create excellence, Friday generously lead, and then we'll have Tik Tok as the back channel. So if you like
86. 12:52chatting on Tik Tok better than you like it over here, cool. Um Brandon is paying attention to that producer Brandon. So
87. 12:59he'll pull pull questions out of Tik Tok and bring them over here. If you want to know how to get to the live stream,
88. 13:04here's my request. Tik Tockers, as people come in, if they're wondering um where the stream is, just please keep
89. 13:12pointing them to my um to my profile. And I don't know if we can put something on screen on TikTok, Brandon. We should look into that.
90. 13:24All right. I don't know what I said, but I that makes sense to me. Oh, Charleston AI. We are definitely going to be
91. 13:32exploring Charleston AI. Um, small towns embracing AI. We're going to go We're going to go look at an entrepreneur. Um, killing it down
92. 13:44in Charleston. [laughter] I'll keep it on rotation, Kyle. Thank you very much. Okay, cool. Beautiful. Beautiful. Beautiful.
93. 13:54Okay, I just got my Tik Tok back. Yeah. So Tik Tok's going to be there as backstage and Oh, that is hilarious. Charleston
94. 14:04AI. We could start with Charleston AI. [laughter] This is this is I think I have found my entrepreneurial spirit animal. [laughter]
95. 14:15Um, okay. Let's let's go do it. Let's see. I'm going to do plus here. Screen. I'll do window. I think it's this. Yes.
96. 14:36Okay, here we go. Okay. Oh, wait. I didn't change my
97. 14:51Oh, please. I didn't change my uh I didn't share audio. Hang on a sec.
98. 15:06I thought this was some sort of professional operation. Window that. Share audio. That. Okay, now we're good to go. What up, party
99. 15:19people? Chef Kelly, what's happening? Uh, we're going to go look at we're going to go look at a local business that has has embraced AI
100. 15:32and uh there's two things about this video. One is it's just funny. If you remember the taste the biscuit video, this is in the
101. 15:43neighborhood. But also, um, we're very, very early in this whole AI thing. Okay, here we go.
102. 16:05AI is not coming to Charleston. It's already here. I'm Jim Riley from Charleston AI. I'm one of the co-founders of Charleston AI. We're
103. 16:14local to Charleston and we're here to help humans help humans with AI. >> Like that, like the mission.
104. 16:20>> And here is our lab. This is where AI comes alive. What used to take months to get accomplished, we can do in hours for
105. 16:28you. Here we have an empty office available for somebody that wants to make Charleston AI their home. Charleston AI's conference room where
106. 16:38our clients come in, individuals and small businesses come in and we discuss what it is they like to do with AI in
107. 16:43their business or personally. We listen very carefully, take [clears throat] notes and make sure we get it right and
108. 16:50then we offer them a way to make it work. AI is not >> There you go. [laughter] All right. So that's that's your
109. 16:58competition people. And then there's and then there's there's a review. So, here's the review. Um, my my wife used
110. 17:06Charleston AI to make herself an AI boyfriend, and she's been cheating on me with him. We'll never do business here again. [laughter] So,
111. 17:16so I think he is uh that man is about to become a meme superstar. He's amazing. And I think my favorite
112. 17:25thing about that is uh [laughter] the the uh the three water bottles in the conference room, the the little
113. 17:34conference room table with the fold out chairs and the three the three water bottles on it. Um because you don't want
114. 17:40to get thirsty when you're doing AI. [laughter] Oh my god. All right, Charleston. AI is where it's at. this. I mean, on on a serious note,
115. 17:56so many people have no idea what AI is. So, a company like that that's like, you know, if you want to figure this out,
116. 18:09come down here. If you want us to do something for you, we'll do something for you. Like, they're going to probably
117. 18:14do good business, right? And um you know over time they'll likely get more sophisticated and and do all that
118. 18:23sort of stuff. Or maybe not. Like maybe using AI to just do simple things for the community is is going to be enough
119. 18:32for them. His conference room uses more water than a data center. [laughter] Oh my god. Does he have an office at
120. 18:46Open AI? Yeah, that's totally that's totally an awkward Open AI uh open AAI thing. Okay. Um so, let me go look at uh a couple of
121. 18:58other things. So, by the way, um tonight is uh is play first. So, we're going to play with some different things with AI. Um, and
122. 19:12if you're on Tik Tok, just go to my profile and you'll be able to see what I'm sharing on screen. Uh, join the the
123. 19:19live stream in the in the AI salon. Doesn't cost you anything. Um, you'll need an account, but that's it's
124. 19:24basically just an email and your name. Um, all right. Why does he have a New York accent? Yeah, exactly. He he he does
125. 19:33have a bit of a New York accent, right? [laughter] Oh my god. Oh my god. All right, let me see what I'm going to play with tonight here.
126. 19:47All right, this is a thing I just want to show you. So, let me Okay, so this is a thing I talked about earlier that um
127. 20:04foul.live live. I think they they shut this down today because it was it was basically costing them too much money.
128. 20:10But it's a new platform for infinite interactive live streams. Pick a channel, prompt what happens next, and
129. 20:17watch it generate in real time. You're not just watching the show, you're directing it. So, let me let me just play this
130. 20:28where I can chat with all of you. three hours >> where I can chat with all of you.
131. 20:44>> And then let me see. This is powered by the new H3 Max Director, an auto reggressive and natively continuous version of H3 Max.
132. 20:59the generated videos type better. Okay, whatever. Um, all right. So, let's do let's go search for H3 Max.
133. 21:14So, here's another one. Today is a very historical moment for AI video generation. You can now generate AI
134. 21:21video faster than you can watch it. So, for those of you on Tik Tok and for you are regulars in the AI salon, or you are
135. 21:28regulars on Tik Tok and the AI salon, if you're if you're if you're double screening, when I first started this channel,
136. 21:35you'll remember that my entry into AI was through an NFT project um that generated images. And the the CEO and the founder of that project
137. 21:50said in spring of 2022. He said, "Within five years, you'll be able to sit in a theater and someone at the front of the theater
138. 22:06will be able to speak a movie into existence and we're six months away from his prediction." And this is this is that moment. Now,
139. 22:22it's too expensive for it to be commercially viable potentially, but maybe not. Who knows? I I've had a
140. 22:29couple of ideas of of things you could build now that you have instantly rendering uh video. Um, but we're now at
141. 22:39the point where you can generate video faster than it takes to play it. So, three seconds to make a five-second clip. And
142. 22:49that just basically means just run a script that just keeps throwing in prompts [laughter] for it to keep
143. 22:58building and spinning that up. And then at some point, someone you could bring people in from the audience to just add
144. 23:03prompts to it and just keep the thing going or just have an LLM just continuously generates a script. Um, pretty wild. So, so let's watch.
145. 23:13Here's 23 seconds of this. There's not not a bunch of really long ones because I I think it's really expensive to do
146. 23:20this. It's about 5,000 to $12,000 a day to to run this right now. And that'll get cheaper as they all do.
147. 23:31>> Look at this Morty interdimensional cable. It's It's just nonsense from other realities. I don't know, Rick.
148. 23:36Some of this stuff is actually kind of disturbing. I really got to I mean I have so much homework, Rick, and we're just we're
149. 23:48just sitting here. >> Um if if your first question is um isn't that a copyright infringement? Yes, as a matter of fact, it is. Um,
150. 24:03people are going to get the pants suit off them for all this kind of stuff, but it won't stop them from doing things. And
151. 24:11creators are going to create original properties. Here's the thing. If you're someone who's had an idea for
152. 24:20a show or a cartoon or something like that for years and years and years and you've been playing around with AI and
153. 24:28learning how to use these video tools, we are within 6 months to a year of anyone being able to come up with a style, characters,
154. 24:39scripts, all that sort of stuff and just turn on a show and start sharing on that show. And some of those people will be talented
155. 24:53and some of those talented people will create shows that are hits and get picked up by other distribution networks.
156. 25:02and some of the people that set up dynamic audience interaction live stream things like Twitch where the
157. 25:11audience can not only comment but they can generate the show the infrastructure there those people are going to make a fortune. So,
158. 25:25back in the day. Four years ago. I know. Four years ago. It's crazy. Back in the day. [laughter] Um,
159. 25:36all right. Let's see what else we got to play with here. That was the foul live thing. Okay. Also, the new MJedit mode. So this is
160. 25:49mini maxing midjourney H3 lip sync is okay. I got [music] this feeling inside of me.
161. 26:04[music] [music] >> All right. Um you do that over at foul.ai. I think it's foul.ai.
162. 26:19Okay. Runway. Today we're sharing new research on Solaris, our first interface world model. Solaris is a new kind of
163. 26:30operating system that generates interactive interfaces frame by frame in real time with no code. We find that Solaris outperforms LLMs,
164. 26:41generating new interfaces across structural similarity and information retention. Read more to request early access.
165. 26:53All right. Let's go look at videos. Okay. But what every piece of software has predetermined what we could do with it.
166. 27:06But what if software could generate itself as you interact with it? Introducing Solaris, the first interface world model.
167. 27:18Solaris [music] is built to understand interaction and to run in real time. It generates the interface itself frame by
168. 27:26frame as you use it. No code underneath, no layout to translate. The image [music] is the application. It treats
169. 27:35your clicks and drags the way our video models treat words. Not as commands to look up, but as a signal for what to
170. 27:42generate next. Nothing about it is decided in advance. So nothing about it has to stay the same. Light shifts,
171. 27:51objects respond, the scene answers to your intentions. It becomes whatever you ask of it. And no two interactions are
172. 27:59ever [music] the same. Solaris isn't just changing how humans experience interfaces. It's unlocking new ways for agents to
173. 28:06learn, too. The need for fixed, hard-coded, and pre-rendered operating systems and surfaces are soon to be a
174. 28:14thing of the past. Solaris interfaces that are generated, not coded. Wow. So, uhoh, Google Chrome froze.
175. 28:30Hang on a sec. Rockbot paused. Um, I'll be back.
176. 29:14Heat.
177. 29:32Heat.
178. 29:54>> [music] [music]
179. 30:16[music]
180. 30:40[music]
181. 31:25[music] [music] >> best pack.
182. 31:52Hello, I'm back. Sorry about that. My machine went poof. It went boom boom. [laughter] Okay, so let me just talk quickly about um
183. 32:06let me talk quickly about the the runway thing and and show you what's going on with runway this new model. And I want to talk about
184. 32:22it. The implications here are kind of crazy. Okay, so this this we just played. Um,
185. 32:39so there's been some some talk of existential threat to programmers and engineers because of vibe coding, right?
186. 32:51and and you know the engineers are like, "Vibe coding will never be as good as what I do." And then December 2025, Opus
187. 32:594.6 comes out and they're like, "Ah, it's pretty good." Um, and it's just getting better and better and better. But
188. 33:06vibe coding is essentially AI doing what we did before, which is writing code to create deterministic
189. 33:17logic that makes the computer behave in the way that you expect it to behave. What this model is implying is that that code won't exist.
190. 33:33that you will have an image that looks like software and you interact with it with a mouse and clicks and drags and typings
191. 33:42like you do and it will just dynamically be feeding you pixels that look like software but are not. They're being generated out
192. 33:54of the latent space. That's insane. It's bonkers. Um, so let's see. Early access at the link below. Learn more. Introducing
193. 34:04Solaris. Okay, so let's let's allow all cookies. You know what I like? I like people tracking my every move.
194. 34:15Today we're sharing. I wonder let me log in because I'm in the uh the creative partner program. I wonder if I actually
195. 34:22have access to Solaris. [snorts] Seed dance is here. Let me log in.
196. 34:42Sign up. Sign in popup was closed. [cough]
197. 34:57This is weird. By the way, if you're on Tik Tok and you want to see what I'm seeing on screen, just go to my profile and join the live
198. 35:05stream in the AI salon and you'll be able to see what I'm looking at. Um, I don't understand why this isn't working.
199. 35:19It is now. Okay. Okay. Unlimited. Let me see. Apps. Let's look up Solaris.
200. 35:34No. [laughter] Okay.
201. 35:53Today we're sharing Solaris. Okay. Every operating system from early terminals to Linux and Mac OS has dictated what's rendered on screen
202. 36:03and what happens when a person or program acts on it. Applications get built on top and stay fixed until someone pushes an update. Solaris
203. 36:13instead renders that layer directly. Chef Kelly question on Streamyard. But doesn't someone have to write code for that? No.
204. 36:29No. Because what are they calling this? A world application model. Wait, hang on. It's the first It's the first of a new
205. 36:40family of AI systems we call interface world models. Okay. So you know how Gemini Omni the new video model from from Google is called Omni
206. 36:58and it has an understanding of physics. So if you ask it to make a video of a steel ball rolling down two metal
207. 37:10tracks, you know, like you do if you have a steel ball and metal tracks. it will be accurate to physics.
208. 37:21Now, normally historically, what you would have to do to do that is you would write a program that can understand 3D
209. 37:28objects and then within those 3D objects, you would create gravity systems and particle collision systems
210. 37:35and you would you would write the rules of the world and then if you do those really good, it'll look natural. What that world
211. 37:47model does, what the the Omni model does for video is it's just been trained on enough video and then combined with it's
212. 37:58been trained inside a language model so it understands the concepts of physics and it understands math and equations
213. 38:05and things like that and then what those actually look like in the real world and it puts those two things together. So
214. 38:12when you see the the the marble running down the steel tracks in an Omni video, it's got accurate physics, but it's not
215. 38:20using any physics rules. It's just sort of understanding what the physics are and what that ball should do on those tracks.
216. 38:32This is the same thing for software interface. So think of it like this. You say, "I want a an interface uh with IKEA
217. 38:43furniture that pulls from the catalog." And I guess it will go off and go research, you know, pictures at IKEA. It will
218. 38:52bring those those elements in and and you'll say, "And when I roll over an object of furniture, I want you
219. 39:01to pop up information about it." This is all stuff that you would write code for, right? You'd write an API to IKEA and
220. 39:07then you'd pull in the images and then you'd figure out a way to render them in a scene and and kind of fake 3D modeling
221. 39:15and then you would have rollover states and you would write code for the rollover states and you would be pulling
222. 39:21information from the website that when you roll over the thing and the thing pops up you pull in the information. All
223. 39:27of that has to be coded. What they're saying is that this thing just understands how software works and it will just pull
224. 39:37from the latent space or from wherever and it will present you with a screen and then when you interact with it when
225. 39:43you roll your mouse over a sofa it will understand that and dynamically just generate a popup window that you can then click on and it will
226. 39:54do something and it'll rerender the interface. So, so you're basically getting the output the pixel output of what software would
227. 40:06do by just rendering the pixels. It's bonkers. And and so it's why like why I'm taking the time to just
228. 40:17talk through this is that I'm trying to get my head around the implications of it because vibe coding is one level of dis disruption.
229. 40:28Being able to literally just just describe an application and start interacting with it from the minute you conceive it. Not
230. 40:39like Lovable has to go write code and write files like we've always written files. No, this is like it's generating
231. 40:46pixels on a screen that look like software and when you interact with it, it acts like software, but there's no
232. 40:53code running it. It's just a model. I mean there there's so Kelly to your point there is software being run
233. 41:05but the software is just it it it's an AI model that is I don't know if they're using diffusion. And I assume they're using some sort of
234. 41:17diffusion something or other to reach into the latent space and pull out what looks like software. And then they've got some sort of
235. 41:31software that tracks mouse movements that I assume it translate those mouse movements into prompts that it then
236. 41:40reacts to. I don't know. I don't know. I missed the heart. Tap tap. [laughter] Okay. Um, design is more visual than ever. Okay.
237. 41:53The intermediate representation limits. The the intermediate representation. Okay. Wait. Solaris handles rendering
238. 42:00and interacting. Rendering and interactions jointly, removing many of the trade-offs we associate with design
239. 42:07today. A single world model generates every frame and every response to user input. eliminating the need for an
240. 42:14intermediate representation. Because there's no conversion step, there's no loss. The entire frame becomes the
241. 42:20interface. We think that Solaris opens up new ways of building websites, apps, other online interfaces, but it's also a
242. 42:28new way to train agents in much more dynamic environments. Yeah. Like the implications here are just crazy. It's crazy.
243. 42:42You could do data visualization and have it be interactive and zoom in and zoom out. And now here's what I can promise you.
244. 42:51What I can promise you is the first version of this is going to suck really bad. We're going to get all excited
245. 42:57about it. We're going to sign up for early access. We're going to do that right now. You should be joining along. We're at
246. 43:02runway.com/news/ressearch/introducing Solaris. and we'll get access to it and we'll play with it and we'll be like
247. 43:18that's disappointing. Except think about what it is and think about how bad video generation was even a year ago and where
248. 43:26it is today. And think about how bad hand rendering was four years ago and where it is today. By collapsing the space between action
249. 43:37and response, Solaris lets agents train against interfaces that are constantly changing and layouts that may have never
250. 43:44existed before. All right, I think we just saw this, but let's watch it. Yeah, we saw this already. Okay, first Solaris
251. 43:52is entirely visual. When an image becomes the application itself, which by the way, hang on, I want to say something. Um,
252. 44:07tonight in the cycle of AI readiness, there's three stages. There's play first, create excellence, generously
253. 44:13lead. Mondays are now play first nights. So, what we're doing here is just kind of playing like someone announces
254. 44:21something. Let's go see if it's real. Okay, it's not real. We can get on the wait list, but we can read about it. We
255. 44:26can learn about it. We can think about it. Does this feel different than anything we've done before? Yes, this feels like
256. 44:36when we first got vibe coding like like you could generate code with chat GBT from from day one. One of the very first
257. 44:46things I did, the thing that really blew my mind with the first chat GPT the first day was I I ended up writing a Python
258. 44:54application that made API calls and generated social media content using OpenAI's API. I've never written a line of Python in
259. 45:04my life. So, so like they've been able to code from the beginning, but you had to know what to do with that
260. 45:13code. And then they made hooks into idees, you know, programming environments. And then along came apps like lovable
261. 45:24that were just like, you don't need to worry about any of that crap. You just give us your ideas and we will make you
262. 45:31an application. We will build you a website. And that became revelatory, right? And so that becomes a whole category of exploration.
263. 45:42Now how we discovered that early is we just play like we're playing things code, okay? You can go to replet, you
264. 45:48can go to cursor, you can go to all these things if you understand coding and how to hook up GitHub and all the nonsense.
265. 45:57And then lovable comes along and says you don't need to do any of that. We got you. So, this feels like one of those. So, so
266. 46:04tonight, Monday, is a play first night. So, we're playing. Um, when we're playing, play is a really important word. Play doesn't mean you
267. 46:14have to have fun. But play is exploration without expectation of outcome.
268. 46:28John Davis on Tik Tok. It's a bad idea generally until it can be caged. Software has to be predictable. I I disagree.
269. 46:38Software doesn't have to be predictable. In fact, LLMs are not predictable, right? They hallucinate like which is frustrating.
270. 46:48But you know what? You know where hallucinations are really good? If you're trying to do something creative and you want it to hallucinate,
271. 46:56having something that's deterministic may not be great. UX has to be predictable. Oh, UX has to be well. So, first of all, we don't know
272. 47:05it. It might be possible to prompt these things in such a way that you do get predictable interface, right? You could
273. 47:11probably create a prompt that says, you know, I want this to have consistent interface and I want you to follow
274. 47:17Google's, you know, flat design standards from 1998. And it it may be able to do that. So, it's possible that that happens.
275. 47:26What this starts to make possible is new kinds of interfaces that don't have to be predictable. It can be entirely dynamic.
276. 47:36Some people are going to hate that. But some people are going to see a new way of doing computing that the people that
277. 47:48hate it will have will never see because they won't be able to see outside of their framework. This is like Kodak, right? The
278. 47:57executives at Kodak when the team walked into their office and said, "Hey bosses, here's the digital camera. We think pixels are the
279. 48:07future." Those executives literally couldn't were not willing to accept that anything other than chemical
280. 48:22exposure of silver particles suspended in a medium was photography. It had to be what they knew. So this model, you know, potentially
281. 48:36opens up an entirely new realm of what we call software. And and this is the thing where it's like if you want a million people to use your
282. 48:50software, having it be predictable is good because then when people get confused, you can tell them how to fix it. But what if you
283. 48:58have an audience of one? What if you just want to create a dynamic photo sharing thing with a bunch of photos that I upload and I want to
284. 49:08make it into an immersive art gallery and I'm going to explore my art on the walls in this gallery. You may or may
285. 49:16not need it, right? It might be more like an immersive game. So, I don't know. I I understand your point and I
286. 49:22think this gives us something different. Anyway, I just wanted to say that tonight is about playing. It's about
287. 49:27going in without expectations and just being as open-minded as we can to say, "Huh, what does that make possible?"
288. 49:36Because when you when we move into Wednesday, create excellence, that's about taking what we've learned. Okay,
289. 49:42there's these new things possible. If I want to build something down here, huh, is there something I learned up there
290. 49:48that we could apply? Right? So, that's the idea. So, that's how my mind is calibrating tonight as we go through this. Okay.
291. 49:57I want to keep reading this piece because this actually feels important to me. So, let's see. Um, [clears throat]
292. 50:07imagine browsing a virtual clothing store where the showroom itself is the interface. Using a single image of
293. 50:16yourself as a reference, you can pick up a shirt from the rack, drag it onto yourself, or try it on or rearrange the
294. 50:24display as naturally as you would in a physical store. This is trippy. So, you just click on the thing, move it
295. 50:34to the other side of the room. It kind of understands physics, right? It understands that that plant was on the ground.
296. 50:42It dynamically, I assume, dynamically created this interface element. Kind of cool. Why does it have four colors there? Maybe that's in the
297. 50:51prompt. I don't know. Maybe it's just dynamically assuming that if you click on a couch and let go, it'll pop up an interface element. I
298. 51:00don't know. Second, it's alive because the application is continuously rendered. It's always evolving rather than waiting
299. 51:09for the next user action. Reflections shift with lighting, and objects respond naturally as they're manipulated. A user
300. 51:17can say something as simple as, "Move the table so I can see how it looks." Or, "Change the color of the couch." The
301. 51:23result is software that feels less like navigating through scripted pages and more like interacting with a living
302. 51:29environment. And this is a trip. This is absolutely a trip.
303. 51:45Whoa. What's going on here? Huh?
304. 52:04Finally, it's open-ended. Traditional interfaces are limited to the interactions developers have anticipated. This is really important.
305. 52:11Okay, hang on. So, by the way, if you're on t unable to validate, please retry. Hang on. If you're on Tik Tok, um, and
306. 52:21you want to see what I'm seeing, I'm I'm on the runway site. If you go to my profile and click on the link there,
307. 52:27that'll bring you to the live stream. Okay, finally, it's open-ended. I posted the runway link. Okay. In the
308. 52:36chat. Beautiful. Oh, let me I got to get chat back. Okay. Fantastic.
309. 52:55Kelly Camp. What's going on with Kelly Camp? Client sent me a screen share recording of their 30 plus set process.
310. 53:01I coded a voice activated data entry s system that could potentially save them 50 hours a week plus no human error in
311. 53:09typing data. [laughter] That's cool. Okay. Well, [clears throat] soon you're going to have to give them
312. 53:16uh dynamic interfaces with this thing. Okay. Finally, it's open-ended. Traditional interfaces are limited to interactions devel developers
313. 53:25anticipated during development, but Solaris can support entirely different behaviors in the same scene, reacting to
314. 53:32user interactions in real time. This flexibility decouples the interface from predefined workflows. That is going to break brains.
315. 53:48Solaris can support entirely different behaviors in the same scene, reacting to interactions in real time. This
316. 53:54flexibility decouples the interface from predefined workflows. Instead, leaving the capabilities of the driving world
317. 54:02model to determine what's possible. So, this is where it's like they pick stuff up out of the bowls and throw it in the
318. 54:08salad and it just makes it look like it was cut up and put there. Drop the salad. Drop the carrots.
319. 54:23Wow, that was a little weird, but that's okay. That's pretty astounding. Solaris turns an interface into an interactive
320. 54:35experience rather than a sequence of pages instead of Yeah, this is I mean, this is kind of like a video game,
321. 54:43right? It's It's like a It's like a world model. I don't know. This is This is kind of bonkers. Turns an interface into an
322. 54:58interactive experience rather than a sequence of pages. Instead of selecting options from menus, users interact
323. 55:05directly with the scene itself. Building a salad is as simple as dragging ingredients to a bowl with the interface
324. 55:10responding naturally as they're added. Okay. Today, people turn to an LLM when they get stuck while online. But LLM's answer
325. 55:18in text, and most hands-on tasks are not task based problems, text based problems. Solaris renders the next step
326. 55:25in your context with visuals so you can steer it. For instance, you can generate an interactive demonstration of combustion.
327. 55:33Wow. Allowing users to experiment with different materials and observe physically plausible reactions as they interact. That's crazy.
328. 55:51Wow.
329. 56:06Huh? Why hasn't this existed until now? Oh, please translate. Oh, okay. Which part?
330. 56:28So, okay. So, principles of combustion. So normally what you would do is you would you would program an interface that's
331. 56:42like write some text of of what this application is principles of combustion, right? And and so Lovable does stuff
332. 56:50like this, right? It's going to just it will figure out what text to put on a website and put it there and it'll make
333. 56:55this scrollable thing where you can scroll to an interface, right? But normally what you would do is you've
334. 57:04got four elements copper, steel, wool, magnesium, and ammonium dramate. Normally what you would do is you would
335. 57:18either get footage of interactions with copper and fire and steel wool and fire and magnesium and fire and you would build an application
336. 57:30that when someone touches the fire, you say, "Okay, when they touch the fire, pick up the fire and have a have a flame
337. 57:38follow along." And then if they touch the copper, jump to the routine, jump to the part of the code that plays the
338. 57:45copper fire. What this thing's doing is it understands the world and it understands that when fire and copper
339. 57:54interact, it makes a green flame. And so, so it is figuring out in real time what that interaction is. Cuz what what you
340. 58:08could likely do is you could say, "Okay, give me four new objects. Give me or four new whatever materials. Give me
341. 58:14four new materials. Give me four new materials. Now, let's turn the fire into liquid nitrogen. Now, let's turn it into mercury."
342. 58:25And and if if you drop Mercury onto something that interacts with Mercury, it will know how to do that. Oh, I
343. 58:32already did. Okay. So, okay, cool. Uh, new Streamyard pin chats. Okay, let me see what's going on over here in Streamyard.
344. 58:43Okay, I'm so glad we started following you in 2023. As we played, I learned all the words, nooks, and crannies, and now
345. 58:49it's easier to understand overall for me. That's part of the part of what I can imagine would be frustrating about this channel
346. 59:01is that I don't follow a curriculum and it's called the learning lab and you're like, where's the where's the where's
347. 59:07the curriculum? Is there a test? Is there going to be a rule book? Um, that ain't me. And so this is a channel that is sort of
348. 59:19learning through osmosis. And I've even experienced this in office hours when when Brent Peterson first
349. 59:28started showing up and talking about the agentic work he was doing and the learning systems and the vector
350. 59:34databases and all the crap he was doing. I didn't understand 10% of what he was talking about. Like I
351. 59:42literally had no idea what the words he was saying meant. Um, and then after a year of that, I started to, and then after two years of
352. 59:53that, it's like I can actually track it now. This AI stuff is changing so fast that it's actually really there's no real
353. 1:00:05curriculum to follow and you kind of do have to just show up. So the irregulars, the people that show up here regularly
354. 1:00:12have been doing this for a while. And so um and so yeah, so thanks thanks for sharing that, Lori. But that's also if
355. 1:00:22you're new here, showing up to this thing if you're like, I don't know what the he's talking about, it will get better over time.
356. 1:00:30Okay, Mimi, I've been crazy building my latest app. It's called seastar.app. It's live. We'll jump over there to seastar.app.
357. 1:00:40Let's see what that is. Manies. Any open-source video tools available for free? Yes, you can download I think it's LTX2.
358. 1:00:59I think it is two. Two or 2.5. LTX 2 or 2.5. Um is an open-source video rendering model on my M5 Max with 64 gigs of RAM. I ran it and it
359. 1:01:16generated 8 seconds of video in about 10 minutes. It was very slow and the results were horrible. [laughter] So, yes. And it
360. 1:01:34still sucks. I think open openw weight video models that generate good video on modest machines by you know by Nvidia
361. 1:01:48standards. Um I think that's probably two or three years off. Little gods digital gods gifts on Tik Tok. Thank you
362. 1:01:56Lord digital gods as always and comment pinned on screen. I don't see a comment pinned on screen but I can look at it.
363. 1:02:01Miniax H3 and 13 are also open source. Great. Thank you, Joe. I appreciate that. Yeah. So, there's models out
364. 1:02:07there, but you have to have pretty beefy iron to run them and it's going to be slow, but if you want video for free and
365. 1:02:15you're creative about how you use it so that you navigate around the jankiness, um, this was one of my favorite games of
366. 1:02:25all time was the game Mist. And why Mist was so brilliant is that they took the limitations of QuickTime, which is
367. 1:02:32basically postage stamp- sized video. And they did these really beautifully rendered images that maybe had a little bit of movement
368. 1:02:40in them, but they were basically static rendered images. And then they had video sitting in a little window
369. 1:02:48uh in the middle of the screen. And so they took a limitation of the technology and they designed around that. And so
370. 1:02:54you could certainly do that with with local video models. Okay. LDG. It's a software hollow deck. It's totally a software hollow deck. Why
371. 1:03:04isn't this existed until now? Okay. Um pop up my screen share if again if you would, Brandon. Um digital digital
372. 1:03:12interfaces are built on two systems which until now have lived in different worlds. The systems that know things, eg search
373. 1:03:21engines and AI assistants, answer with static content, text, image, maybe an embedded video. And then the systems
374. 1:03:27that respond in real time, JavaScript, CSS, game engines, and more re recently, interactive world models create rich
375. 1:03:34interactive experiences, but they know nothing about your products. Right. Right. So, you've got world models can can do stuff.
376. 1:03:44LLMs understand stuff. And so they've brought those two worlds together. We've traditionally thought of software
377. 1:03:50interfaces interfaces as deterministic programs and world models as generators of visual content. An interface world
378. 1:03:59model has to be both at once. A system that understands your intent while continuously rendering an interactive
379. 1:04:05world around it. Once you try to build one, three engineering challenges immediately appear. One, speed. I could
380. 1:04:15imagine interactions stop feeling interactive somewhere around half a second of delay. Video diffusion models
381. 1:04:22take seconds or minutes to produce a clip which is an accept is acceptable for content creation but too slow for
382. 1:04:29interface. To cross that threshold, the model has to be able to generate frames sequentially with each frame depending only on what
383. 1:04:37came before cheaply enough to keep up with the user. Cool. Staying coherent. An interface has to remain consistent across the entire
384. 1:04:46session. So this is what I think Joe Joe said, someone said that the interface has to stay consistent, not just a single clip. The things it
385. 1:04:56needs to preserve, text, layout, identity of objects are the same things generated video has historically
386. 1:05:04struggled to maintain and small errors compound the longer the generation continues. And then the third one is
387. 1:05:10cost. Generating every frame is still more expensive than serving a page that was built once. The same work that made
388. 1:05:18Solaris real time also made it orders of magnitude cheaper to run than standard video diffusion model. The cost curve
389. 1:05:26continues to improve. Solarist is our best bet. These conceptual and technical barriers can be overcome, which means
390. 1:05:32they haven't overcome them yet, but but they're at least talking about it. So, we're somewhere in the neighborhood. We
391. 1:05:39built it with three focuses. Real-time interaction, coherence over the entire session, visual quality that holds up at
392. 1:05:44720p. Solaris builds on our Gen 4.5 video generation model, which we've adapted to one, understand the
393. 1:05:52interaction, two, respond in real time. So, I don't know if you noticed this, but when when the like when the salad
394. 1:06:00was was being rendered, it was only rendering like a small portion of the entire frame. So, they're breaking up
395. 1:06:08the frame. They're probably using what's that called? Um, the object analysis stuff, segmenting. They're probably using segmenting to
396. 1:06:17say, "Okay, the the middle of the bowl is where the salad goes." And so, we're going to segment that, and then we'll
397. 1:06:22just render the video in that little segment or render the image in that little segment. Learning interaction. Solaris treats
398. 1:06:30user input as conditioning for the next frame the same way it treats text or images. The model observes clicks, drags, and
399. 1:06:38other other interactions as it generates, using them as signals for what comes next. So, I would imagine if
400. 1:06:45you like waggle your mouse, it can probably figure out, oh, he he wants to do something that wiggles something. It
401. 1:06:52learns the relationship between user interaction and visual outcomes. This means that it knows what should happen
402. 1:06:59when something is clicked, dragged, or modified without requiring those interactions to be explicitly programmed. This is going to melt the
403. 1:07:07faces of developers because developers have always had to think about when someone clicks, what happens. They've had to think about that and
404. 1:07:22choose that and do compromises around that. Well, if we make it explode, then we're going to need this and that's
405. 1:07:28going to cost this. But if we make it wiggle and like jello, it'll then this is the cost. So, we're going to do both
406. 1:07:35of those, but we're going to have to call sub routines. And there's there's a tremendous amount of thought that goes
407. 1:07:40into what happens when someone clicks and lets go. Do they click and hold? Do they click and hold and let go? Do they
408. 1:07:47double click? Did that click and hold? Right click. All of those are deterministic, right? All of those. Someone decided what happens.
409. 1:08:00This is sort of inferring it as you click. It's wild. Running in real time. Standard video diffusion models
410. 1:08:08refine the entire clip over dozens of dnoising steps. A process that's far too slow or dynamic for user interaction. We
411. 1:08:15converted Solaris into a real-time engine in three stages. First, we taught it to generate frames auto reggressively
412. 1:08:21with each frame def depending only on what came before it. So that probably means it generates this and then if all
413. 1:08:30it's going to change is this little piece of carrot, it doesn't need to regenerate all this. It's just making
414. 1:08:35the changes for what it needs to change. Next, we distilled the many step denoising process into just a few steps.
415. 1:08:42That makes sense. So instead of 70 steps to D noiseis, they're probably doing it in five. Finally, we trained the fast
416. 1:08:49model on its own outputs so visual quality remains stable over long interactions. That's a big deal. This the result generates frames at
417. 1:08:58interactive speeds while refers preserving the visual quality of the original teacher model. The Eiffel Tower. All right, let's see
418. 1:09:08what this is. Day night. So, I clicked on that. Huh.
419. 1:09:24So, it remembered that inter those interface elements. Oh, that's cool. Look at that. Now, all of these videos are in the 20
420. 1:09:32second range. So, I'm wondering I wonder if this has a limitation of 20 seconds. I bet it does. Reasoning and rendering. Solaris
421. 1:09:40generates the interface one frame at a time while the language model determines how the interface evolves. The LLM interprets User requests
422. 1:09:52decides which interactions should modify the current scene versus transition to a new one. So it can figure out, do we
423. 1:09:59stay in the same scene or are we moving to something new? That's wild. Defines behaviors that make the world feel alive
424. 1:10:05and produces the prompt that guides Solaris to render the next state. So it's interpreting user interaction,
425. 1:10:13translating that into a prompt, and then quickly rendering what happens next. bonkers. Together, the language model
426. 1:10:21and the world model separate reasoning from rendering. One decides what the application should do next, while the
427. 1:10:28other generates how the behavior appears and responds in real time. This is insane. Continuous generation. You you provide a
428. 1:10:41starting state, eg a brand environment or product scene, and the model streams frames in real time. As the user clicks,
429. 1:10:48drags or types, those interactions are incorporated into the next generated frames, and the scene responds in place. There
430. 1:10:57are no predefined screens, no templates to fall back on. Instead, text prompts specify what clicks, drags, and other
431. 1:11:05inter actions mean in a particular scene. Okay. Wow. Okay. So, look at that. That's
432. 1:11:20cool. So, the user clicks the sun and drags it down and so it becomes early evening. And then the user clicks on the street lamp
433. 1:11:31and it turns it on and makes it dark. Vickiy's here. I'm incognito. [laughter] By the way, if you're on TikTok and
434. 1:11:44you're wondering what the hell am I looking at, just head on over to um go to my profile and click on the link in
435. 1:11:51the AI salon. Come over to the AI salon. You can see what I'm looking at. Redefining the mouse. Okay, let's see.
436. 1:11:58The cost of translation. Earlier we argued that translating interfaces into intermediate represent representation
437. 1:12:05inevitably degrades information. To measure that loss, we tested how faithfully today's multimodal language
438. 1:12:13models can recreate an interface from a screenshot.
439. 1:12:51Yeah, I have a feeling the early versions of this are just going to be short little bursts of interface. Wow.
440. 1:13:08What it can't do yet, Solaris is strongest in ambient motion. text. Stable, legible text remains one of the hardest problems. Trust for
441. 1:13:17instructional or commercial experiences. A convincing wrong answer is worse than no answer. Today, Solaris stays anchored
442. 1:13:26through what you give it. The starting frame can be composed from real imagery and reference material which grounds the
443. 1:13:35scene and things that actually exist. conditioning generation on richer verified context as the session unfolds
444. 1:13:42is an active research focus. So it doesn't really work work well. Okay, it's 8:45. Okay, so hang on. I
445. 1:13:51want to talk to you about something in a second. Trust long sessions. Yeah, maintaining visual semantic coherence
446. 1:13:58over extended things. So this is the initial thing is going to be you'll be able to do 15 20 seconds of interface new kinds of interfaces.
447. 1:14:10It's an early step towards okay fill out the form. Okay. Boom. Let's do the salon.ai. Company size. We're a small little company. Um,
448. 1:15:11All right. So, there we go. Tried it. All right. Up next. So, bring me back up here. Bring this over here. All right. So,
449. 1:15:28couple of things. We're We're getting toward the end of the evening. Um, I know I ran down a rabbit hole there and
450. 1:15:35you know me, if that was a pile of garbage, I wouldn't have run down that rabbit hole. Um, it's very early. The
451. 1:15:42initial thing is going to be short bursts of interface. It's going to be janky as So, so there's nothing
452. 1:15:47you need to worry about right now. But keep in your mind that vibe coding and writing code to do deterministic, meaning predictable
453. 1:16:02behaviors will likely be considered old school programming within two years. Truly dynamic, interactive, fluid.
454. 1:16:18Understanding your intent as a user interfaces will be the future without code. That's insane. So, you heard it here first. Okay. Well,
455. 1:16:30no, you actually, if you worked at Runway, you heard it there first. But, um, tomorrow night is AI Salon presents.
456. 1:16:38Um we've got um Price Fuler is going to be presenting. He's a an ex product guy who has started a roofing company
457. 1:16:51because he wanted to be in a business that he knew the AI wasn't going to destroy. So as a product guy, his job
458. 1:17:02was to take the user requests and go, "Huh, the user wants an app that does this, that, and the other." And then his
459. 1:17:10job would be to tell the engineers, "Here's what you need to figure out how to do." And then they would go, "We
460. 1:17:16can't do that." He would argue with them. And then they would take six months to make the thing do the thing.
461. 1:17:24And then he would deliver the app, and then there would be bugs in it. and they'd have to fix it. And he realized
462. 1:17:30that's probably going away. So he said, I'll go into roofing. And you're like, why is he talking at the AI salon? Because he's using AI to
463. 1:17:40do all sorts of cool with his business. Because he understands it. He's passionate about it. And he's also
464. 1:17:48smart to understand that shit's about to change. So that's tomorrow night. AION presents. Uh, if you are in the AI salon, go to
465. 1:17:58the events calendar. Please RSVP to it so we know how many people are coming. Um, this Thursday, Brandon, producer Brandon starts his Codeex
466. 1:18:14mastermind sprint. It's a four-week sprint. If you haven't used the Chat GPT app, ChatGpt Work and Codeex, they're kind of the same thing.
467. 1:18:28Brandon's going to do it. So, you need to be a member of the mastermind. So, if you haven't joined the mastermind,
468. 1:18:34right at the top, there's a there's a link there that says learn more about the mastermind or you can scan that QR code or you can you
469. 1:18:44should join the mastermind and you should do this um codeex thing. If you have not been playing with a Gentic
470. 1:18:51tools, it really is time to learn it. It's a completely different skill than learning how to use chat GPT. Um, and
471. 1:18:59Brandon's going to take you there. And and Brandon, if nothing else, is kind of maniacal about making sure that there's
472. 1:19:06actual value for your actual life and actual business. It's not just learn it because it's cool. That's more my mo.
473. 1:19:15Brandon's more like, "No, no, we're gonna we're going to set this up so that you can take the things in your business
474. 1:19:21that are repetitive, mind-numbing, soulcrushing things, and we're going to have the agents do that for you." That's
475. 1:19:28what it's about. So, go do that. Brandon, do you want to pop up on stage at all and talk about it? There's anything that I talked about.
476. 1:19:35>> Sure, I'm always happy to pop up. Uh, links are in the chat, but um, yeah, we're really excited. It's going to
477. 1:19:41start this Thursday. Although Codex and Jedgbt work was down for several hours today, which was interesting timing.
478. 1:19:48>> Yeah, they uh had an outage. Um Okay. Said though it's back up now. >> A couple of them had outages today. It
479. 1:19:55wasn't because I heard another one had an outage today. >> I don't know which one. >> Oh yeah. >> Yeah.
480. 1:20:04>> But uh yeah, we're going to be talking about uh codeex and chat GBT work. And that really is a blanket term for
481. 1:20:12agentic work. So whether you're using co cloud computer or Google anti-gravity uh or perplexity computer or any of the
482. 1:20:21other ones out there, we're going to talk about how to empower yourself to go further using these agentic tools. And it's going to
483. 1:20:29be Thursdays at uh noon mountain, 2 Eastern uh for four weeks. >> Beautiful. Um, my my little gimbal for
484. 1:20:38Tik Tok. My little gimbal just died. So, you're just now on a stand. >> Beautiful. Um, that's that. And then, um,
485. 1:20:53yeah, we'll we'll play with I I might dig into Grockbot on Wednesday. Um or I might do something around >> around some chat GPT stuff, some
486. 1:21:05projects I'm working on. >> Um the other thing that we do >> Yeah, go ahead. >> I will say um by the way to the to the
487. 1:21:12folks that can hear me but can't see me on TikTok, thanks for playing along and and watching behind the scenes, it was a
488. 1:21:19success. We had a bunch of people jump over to the AI salon so they could actually see the show. Uh so thanks for being here.
489. 1:21:26>> Great. Beautiful. Love it. Okay. Um, so a couple of things. This new format of play first on Monday, create
490. 1:21:36excellence on Wednesday, generously lead on Friday is a new experiment. But one of the things that I'm increasingly passionate about is that
491. 1:21:51cycle of AI readiness, those three elements, those three modes of interacting. Learn, play first, do, create excellence, lead, share what
492. 1:22:02you're learning, right? Establish yourself as a leader, a critical thinker, and then and then and then
493. 1:22:09repeating that over and over again. This is the way. And so I'm leaning heavily into making the salon about bringing that to the world.
494. 1:22:22And so what this actually looks like tactically for the AI learning lab, I'm open to suggestions. Create excellence
495. 1:22:29could look like me working on something, but it could look like someone in the community has built something cool and
496. 1:22:37comes up on stage and demos it and talks about how they built it and just, you know, takes a moment to have a
497. 1:22:43spontaneous, you know, show us how you did what you did, show us how you maintained excellence while working with
498. 1:22:50AI that is notoriously janky, right? Um, same with generously lead on Friday. Um, that might look like bringing people up
499. 1:22:58on stage to interview. It might look like talking about leadership in the community. It might it might look like,
500. 1:23:04you know, people are afraid and protesting AI and, you know, angry about AI. How do we interact with them? Um, so
501. 1:23:10I'm open to it. So, if you have ideas about this stuff, come with your ideas. Um, no, Emmy might be doing that Friday. Great. Um,
502. 1:23:19get in touch with producer Brandon. He's all over the AI salon. So at producer Brandon in the AI salon, if you haven't
503. 1:23:28DM'd him, feel free to do that if you have ideas about what we can do. We'll put them into the hopper. Um, I want
504. 1:23:36this to be increasingly valuable for people. And I think this new format's going to help do that. And hopefully for
505. 1:23:42the folks on TikTok, you like the backstage view. Uh, and we'll keep doing that. Okay. Charles Adams, do we have a
506. 1:23:48class to teach how to even think about what's worth coding? Literally all I do is write and build skills to write better. Um I mean
507. 1:24:03to a great degree um Charles that's what the mastermind is. The daily practice really is about understanding who you
508. 1:24:16are, what you value, and what you want to do in the world and then figuring out how you use AI to do that. And and one
509. 1:24:23of the things that's emerged out of that group is that a lot of people are rediscovering their love for analog,
510. 1:24:32their love for writing, their love for personal notebooks, their love for scrapbooking that serves as as an anchor
511. 1:24:44for understanding what they want in the world and then applying AI to be able to amplify that idea. So, so that's
512. 1:24:52something that's emerging. So, it it it really is about intentionality and and slowing down to check in with yourself
513. 1:25:01before just jumping into the next tool and jumping into the next tool and making the next thing because ultimately that gets
514. 1:25:10it gets kind of boring. Nobody wants to see your crap. you get frustrated because you've made all this stuff that's really cool, but
515. 1:25:19nobody wants to play with it because they're all out making their own crap. And so I I really do think that that the
516. 1:25:26the discipline right now is how do we each of us individually and collectively figure out what it is we want and then
517. 1:25:36use that as the filter to determine what you do with AI. Oh, and and by the way, uh Charles Noi did um an LOL on how to put something in
518. 1:25:50in the world, something that she cared about. Um so there's an LOL in the playbacks in the salon, which you can go
519. 1:25:57watch. Go watch No Emy's LOL's. I think she did a three a three-part series. Two or three-part series. All right. Um
520. 1:26:05beautiful. It's time. Let's get you all out of here. Tik Tok. Hope you had fun tonight. Um, I will charge the gimbal so
521. 1:26:11it doesn't die like it did tonight. Um, I got one in the works. Hopefully it'll be done by then. I did an LOL on that.
522. 1:26:19Beautiful. Turtles up. All right, Tik Tok is back. Um, if you want to see what's on screen, you go into the AI
523. 1:26:27salon. But, but I'm digging the Tik Tok thing here. I hope you guys had fun. Hope you all had fun. And, uh, peace
524. 1:26:34out. I'll see you tomorrow night for AI Salon Presents uh, with Price Fulier. uh talking about the future of work
525. 1:26:42[laughter] and the future of using AI uh to run his roofing business. Going to be wild. It's going to be a good one. All right, peace out
526. 1:26:51everybody. Hope you have a fun night.