7/9/25 - Grok 4: X.AI's Next-Gen AI Assistant Launch and Demo

Transcript

1. 0:01Are you ready? Are you ready? Yes, I'm ready. Are you ready? Yes, I'm ready to fall in love. To fall in love
2. 0:16right now. [Music]
3. 0:3210,000 words swarm around my head. A million more in books written beneath my bed. [Music] And I wrote or read them all when
4. 0:47searching in the swamp. Still can't find how to hold my hands. [Applause] And I know you need me in the next room
5. 0:59over. I'm stuck in here home paralyzed. [Music] Four months I got myself ins.
6. 1:15[Music] I'm no different. like talk on things we don't know about.
7. 1:32Well, ain't it like most people know people like things we don't know about. [Music] [Applause] [Music]
8. 1:49[Music] to say that recognize me these shoes and disab
9. 2:05[Music] [Music]
10. 2:21[Applause] [Music]
11. 3:00It's not simple to say that most days I don't recognize me that these shoes I'm the apron. That place and its patrons.
12. 3:15I've taken more than I gave them. It's not easy to know. I'm not everything like I used to be. Although it's true. I
13. 3:29was never attention sweet center. I still remember that girl. She's imperfect, but she tries. She is good, but she lies.
14. 3:45She is hard on herself. She is broken, but won't ask for help. She is messy, but she's kind. She's lonely.
15. 4:00Most of the time she's all of this mixed up and baked in a beautiful pie. She's gone, but she used to be mine. [Music]
16. 4:20[Music] Okay.
17. 4:34Um, it is Grock Fortnite apparently. Apparently it's Grock Fortnite. I haven't seen many posts Elon's going on
18. 4:42about some dude named Jack J Jack Epstein. Jack Jack Eisenstein. I don't know. Anyway, there's some guy Elon Musk talking about
19. 4:54him where he's talking about Grock 4. So, it's possible we get some sort of list about Jack Eisenstein
20. 5:02or we get Grock 4 tonight. One of those. All right. Um or UFO disclosure. Could be UFO disclos or we'll surf. We'll we'll use AI. We'll
21. 5:17we'll act like good AI thieves and go steal everyone's content. Actually, some guy some guy in my trolls,
22. 5:32he thought he he thought he cornered me. He thought he cornered me. He said, he said, "So, oh, so I guess if you,"
23. 5:43because I was talking about, you know, prompting. Prompting is a craft basically. And he said, "Oh, so if you
24. 5:48go to a vending machine and you put in a quarter and you pull push the button for a Snickers bar and a Snickers Snickers
25. 5:55bar come out came out, did you make that Snickers bar?" And and like when it when I first read it, I was like, "Oh, he got
26. 6:02me." And then I was like, "Wait a minute. No, that's not how this [ __ ] works." It would be more like this. I
27. 6:07put a quarter in the machine and I say, "Hey, I want to design a new candy bar and I want it to have almonds and I want
28. 6:14it to have peanut butter and I want it to have this and I want it to have that and then I want it to have chocolate on
29. 6:19the outside but like peanuts like a payday." And then I want it to actually come up with a name of the new candy bar
30. 6:26for me and come up with the packaging design and then I push the button and that comes out. That's what AI is like.
31. 6:35Son of a [ __ ] I was like, "No, it's AI is not full of candy bars that are already made. It broke it broke it down into the ingredients."
32. 6:49So, in that case, yes, I did make a candy bar. Am I am I a chocolier? No. [Music]
33. 7:28Thank you for sharing the live, Danielle. That's good. Um, okay. So, um, what we're going to do here tonight is
34. 7:37I'll probably mess around on the guitar. If you How many folks we got here? Oh, we're we're relatively small at this
35. 7:43point. Um, go tell your friends that there's some idiot live on Tik Tok and YouTube and Twitter and LinkedIn.
36. 7:52Um, who's going to be live streaming the Gro 4 announcement if it happens. Comet browser. Yes, the comment browser.
37. 8:07You can skip the weight list if you pay 200 bucks a month to Perplexity. 77. What's 77 today? 79. Did I put 77 somewhere?
38. 8:22777. 777 film. [Music] Oh, [Music]
39. 8:36isn't that It been a holiday night, wasn't it? That was it. Just an open an open that. No, I think it's higher than that. [Music]
40. 8:52Um the title of the YouTube Oh, I did 77. Oh, I'm such an idiot.
41. 9:06Um I don't think I can edit that. Can I? Oh, maybe I can. Ha, I think I can. 7 9
42. 9:21and then if I click in here, it fixed it there and it fixed it there. Save changes. Ha, wait. LinkedIn doesn't support updating
43. 9:31streams yet. Oh, LinkedIn, you're an [ __ ] Really, Microsoft? I don't if I close this. I think it's
44. 9:47still going to be 77. Whatever. All right. Whatever. It's It's 77. Hey, I'm part of time. Shane Saw in the house.
45. 9:54Fixed. Good catch. I didn't notice it. All right. Fantastic. What announcement is today? I'll tell you what side
46. 10:01hustle, Mimi. Today Today we got there's a there's a uh there's a social media service. It's called Twitter. Used to be
47. 10:11called Twitter. Now it's called X because the guy that bought it, he likes the letter X. He likes it so much he
48. 10:17named his son X and he he likes X airplplanes and he he got X.AI. He's Anyway, he likes X. And so they got this
49. 10:27thing called Grock and Gro is a large language model. It's very much like Chat GPT except it's not good and no one uses
50. 10:34it. Oh, but it does make like racist posts automatically. So that's cool. Anyway, so that is at level at at at
51. 10:45level three. It's version three. It they had it they had one out Grock came out at first and then it was really bad and
52. 10:55then they're like, "Hey, we got a better one." And it was called two and then they're like, "No, this one's super
53. 10:59good." And it was three and then it was it was okay. And it made pictures. It made pictures and then the now today
54. 11:09this it was going to be 3.5 but then the guy that called Twitter X said we're not going to call it 3.5 we're going to be
55. 11:16calling four and so they're going to call it 040 because chat GPT's got 04 and then they got 40 and but one's an O
56. 11:27and one's a zero and it but it doesn't matter because now we're going to have Grock 4 and it but it'll be Grock 4040
57. 11:34This is going to be so good. It's going to be so good. This lame character brought to you by
58. 11:49Saturday Night Live minor leagues. Everything I said in that stupid voice was mostly factual.
59. 12:06Oh my god, I'm like the Lauren Michael of the AI learning lab. [Music]
60. 12:32[Applause] [Music]
61. 12:49Um, I feel like an idiot tonight. An idiot tonight. An idiot tonight. I think this
62. 13:03is going to be a stupid show tonight. Oh, I got recognized today. This is the first time in a long time. Maybe if ever
63. 13:12I used to get recognized when I was doing agency.com when I spoke a lot, but um I had a meeting. We had a a pitch
64. 13:19with an agency today at StoryVine and uh I got on the thing. So, I don't know if John's here. if you're here, John. Hey,
65. 13:25man. But, uh, but this the the woman that ran it, Stacy, said, "Oh, John's coming." And when I said to John, "We
66. 13:33were meeting with Kyle Shannon," he goes, "Oh, Kyle Shannon, the AI guy. He's been watching the live." So, I
67. 13:41turned a few shades of red on that one. But, you know, what are you gonna do? Um, just released the Kyle 4 update.
68. 13:48Yeah, we I'm I'm still on on beta 0.8.7. Um so which is why you know this channel lacks credentials and qualifications and
69. 14:04actual uh knowledge. Yeah. Qualification. Yes. Yeah. Yeah.
70. 14:18Yeah. Yeah. See [Music]
71. 14:42looking [Music] [Music]
72. 15:00[Music] for.
73. 15:16Woohoo! Woohoo! [Music]
74. 15:40in a westerly direction. Oh no. [Music]
75. 15:55[Music] [Music]
76. 16:18[Music]
77. 16:41Oh, all right. How was the podcast? Podcast was good. We had on Tyler Fisk on the show today. Was really good. I
78. 16:51thought you were Kyle 6.0. No, that was the rumor. That was uh Kyle 6.0 has been announced but never shipped. That's the
79. 16:58That's the challenge there. Um had me at the guy that called Twitter X. Okay. Um let's see what what's happening.
80. 17:14Well, at nine o'clock, so in 40 minutes, 43 minutes is is supposedly the Grock for announcement release. Who knows?
81. 17:36Mr. Strawberry is putting out [ __ ] posts. We're less than an hour away from super intelligence, which that's [ __ ] That's not true.
82. 17:52Um, [Music] Elon's greatest enemy is Elon himself. That's pretty funny. What's the first thing you're going to
83. 18:06ask Grock for? That's a good question. Let's go see what a thousand people said. How many cubes are in my trailer?
84. 18:16What's your opinion of Hitler? Oh, internet trolls are the best. They really are. Oh my god. Oh my god.
85. 18:33Fantastic. Fantastic. Yeah. Yeah. Yeah. Yeah. Let's see. We got anything here? Um, uh, X.AI lost their CEO today. So,
86. 18:47she was there for two years and they're going to launch Grock 4 today and their CEO quit. So, I don't know what that's about other
87. 18:59than maybe working for Mr. Musk is a is a challenge. [Music]
88. 19:14Johnny IV and Sam, the IO products deal has officially closed and we're thrilled to welcome the team to Open AI. Oh, okay.
89. 19:28So, I guess they bought them. They were partnering before, now they bought them. Oh, there's Ailia. Nice. Yeah, let's go look at these.
90. 19:43This is from Ethan Malik using Gro 3. I think Mr. It's got his his autotapper going, by the way. That image is so cringe. It's totally
91. 19:57cringe. Um, let me change my screen sharing. [Music]
92. 20:25actually I think I am over Hamlet. [Music] Sorry, Caesar. [Music]
93. 20:40You know, I think we should see other people. I am feeling a little better actually. Actually,
94. 20:58I think I am over Hamlet. [Music] Mary Mary got the get smart theme. You are correct.
95. 21:19Oh, what did I see today? Um, there was a the [ __ ] it called? There's a new video model out. It's actually been out for a while. Moon
96. 21:30Valley. Moon Valley. If you go to moonval.com, I got to change my tabs again. Moonval.com. Um, unfortunately, you have to pay for
97. 21:45it. Uh oh, the sam the salmon Johnny Johnny Ives image is so cringe. Yeah, that's pretty cringe. It is pretty cringe. Um
98. 22:01[Music] I didn't know he beatbox. Does he
99. 22:17beatbox too? My god, that man's got so many talents. He can hum four bars of of nearly seven television
100. 22:25shows from the 1970s and 1980s. Um, he's he's a pretty good guitar player. He's learned it seems like almost 12 songs in in 60 years.
101. 22:40Pretty solid. Like it's almost like a song every four years he's learned on average, you know? Um, and then he's just, you know, he can
102. 22:49he did like I was on one of his lives the other night and um I mean he was talking about AI. I didn't understand
103. 22:55what he was talking about but it doesn't matter. But I was on there and he started beatboxing. He was like
104. 23:03and it was it was about as good as anything else he does and I just didn't realize he had that talent too. I'm like
105. 23:10oh my god what is he a renaissance man? It's just like crazy. So anyway, that's that's what I do for like two hours a
106. 23:18night is I'll watch that and then sometimes what I'll do is I'll re rewind it, watch it again or if I miss the one
107. 23:22the night before, I'll just watch him until 3:00 or 4 in the morning and I'm sorry, what have have I learned anything
108. 23:31from from the AI learning lab? Um well uh I it depends on what the definition of learn is. Yeah. So so sure. I mean I'm aware of more things
109. 23:46than I was. Yeah. Yeah. No, I've I think there's I think learning has happened for some people have said Yep. that
110. 23:55learning has definitely happened. So that's why I go because maybe someday maybe it'll something will sink in. That
111. 24:04would be good. That would be good. That would be so nice. Yeah. That'd be good. Be good. I don't know what I'm doing
112. 24:09with my life. I should probably I should probably pay for therapy. Yeah. Yeah. No, I I I think I'm trying to fill a void uh
113. 24:19with the with AI learning lab. Yeah. Uh but no, it's good. It's good. It's It's solid. It's he he promises he's going to
114. 24:27finish a sentence someday. Who's got Who's got questions? Uh Moon Valley. Uh this is Moon Valley. So Moon
115. 24:41Valley when I think it was when Runway came out like two years ago. Moon Valley came out right around the
116. 24:51same time and it was a Discord only. It was like remember when uh Midjourney was Discord only. Moon Valley was was
117. 25:00a video generator two years ago that was mid or Discord only and it looks pretty good. Uh well from the outside it looks good because I
118. 25:12didn't pay for it. But if you say start creating they do this really annoying thing where you got to sign in. Think I
119. 25:19got to do this. Then I got to do this. Then I got to do this. Uh, but it's got controls. It's got all these controls. So, like motion
120. 25:32transfer. So, you can upload a video with motion and then it'll take the motion from that video and put it on
121. 25:39some other image. Um, or here's pose transfer or here's camera controls. Um, here's trajectory control. Here's shot
122. 25:48extension. So, there's that. Then they've got reference. You can add reference photos. So, it looks pretty
123. 25:55good except they let you do all this and write a prompt and then and then when you hit generate, they're like, "Oh,
124. 26:01sorry, loser. Sorry." Yeah. No. Yeah. This is going to cost you money. I I listen, I know we could have said that
125. 26:11before you pushed the generate button, but you know, we don't have a Gen Xer working at our company to yell at us for
126. 26:18stupid [ __ ] like this. So, if I mean, just go ahead and press the button and see what happens. It's probably going to make you angry.
127. 26:34Uh, it's a two-hour void that comes back five nights a week at least. The AI learning lab where you can't get
128. 26:44those two hours back. Be there. [Music] Um, anybody want to do anything tonight?
129. 26:59Uh, it looks like Udo added beats per minute to their little thing, which is kind of cool. The void that keeps on giving.
130. 27:12Even black holes emit radiation once in a while. Um, who's got a thought about AI? Love it, hate it, um, questions about it.
131. 27:29What tool should I use? Do you need a nap? No one. Buer, buler, buler,
132. 27:45narrator forever. So, what's new these days? Been away. Um, okay. Here, here are the biggies for me right now. We're we're getting Grock 4
133. 27:56tonight. We'll probably get um open-source chat GPT or yeah, I guess open source chat GPT potentially a chat GPT browser this
134. 28:10month. Next month we'll get GPT5 I think. Um, so there's so the frontier model companies are doing something.
135. 28:24The the thing that I've been most enamored by is MidJourney and Google's V3 video models. They're just crazy
136. 28:33good. They're crazy good. They're crazy good. They're really good. Um, so that that's what I'm kind of
137. 28:40excited about. I joined a VC co-working space today. Curious to see how many are working on AI. Kelly, I can't imagine if
138. 28:49you're in a co-working space in Dallas. I would I would put money. Sorry for yelling. It's the phone, not
139. 28:57me. So, this is what Kelly does. She's very passive aggressive. She'll come in and purposefully type all caps and then
140. 29:04go, "Oh, oh, rather me. Oh my. Oh my, I did not mean to to be so brusk, right? She's southern. She's like she's got
141. 29:13Texas thing going. I know what I did was more like Louisiana, but or or more like Georgia, right? But she's like the Texas
142. 29:20version of that. She'll just come in here and just yell at people and then she's like, "Oh, was that me?" Yeah. Anyway,
143. 29:28um I would put money on the fact, Kelly, that you probably can't find a business in that space that's not working on AI. That's right.
144. 29:44Oh, man. We need your support, Kelly. We do. We do. Um it's me. I'm the problem. User error.
145. 29:59Oh my god. Mr. It is still still calling for agents. Gareth. Gareth has a single question. He's not going to tell us what it is,
146. 30:10but he does have a question. [Music] What's going to happen to all of the programmers? Well, um, what's going to happen to every
147. 30:24knowledge worker and every creative worker and quite frankly probably every worker because this is going to affect
148. 30:29every industry, not just not just knowledge workers, um, is that they're going to have to adapt. Humans are not static. There there's a the
149. 30:47if you understand what AI is and you look at Microsoft laying off 15,000 people and you think they're not related, they're very related. Um
150. 31:01but there's a couple of things happening. One is while AI is going to take some jobs, what AI is actually going to do is AI is going to
151. 31:13automate some tasks. And any job is made up of some series of tasks, right? And if you have a job that's got a
152. 31:24limited set of tasks or a very specialized set of tasks that are highly likely to be replaced with AI, then your
153. 31:33job is going to be gone. The tasks that you do are going to be gone. Now, if you're a business owner, you could
154. 31:41say, "Hey, I've got all these people. I could give them other tasks to do." Um, but a lot of companies aren't going to
155. 31:50do that. just going to lay people off and they're just going to go that job was eliminated. Um I think that a lot of
156. 31:56companies are going to fire the wrong people. I I or or maybe put another way, I don't think they're going to keep the right
157. 32:07people. I don't think who's going to be successful in an AI powered company are the same people that are successful in
158. 32:16specialist jobs today. I think creative generalists um are going to be the new superstars in the in the workforce. So
159. 32:24what I would say is this. I would say to the degree that you've got a hyperfocus on a single kind of task, um you're probably more at risk.
160. 32:40And so what that looks like is start playing with AI. If you're a programmer, if you're a programmer and you're not
161. 32:46vibe coding, I think that's a big mistake. Andre Karpathy, right, one of the co-founders of a of OpenAI um and
162. 32:55the the lead scientist there for a while. Um he has stopped coding. He is only vibe coding. And if you don't know
163. 33:03what vibe coding is, it's where you're using one of these tools that writes code and you just talk code into
164. 33:10existence and it writes the code. And then he goes in and I guess he reviews it and updates it and does whatever you do. Um
165. 33:18there's there's there's there is a radical elitist um thing happening. I see it in engineering. I certainly see it in the art community. Um,
166. 33:35I see it in the video community where the people who've spent 20 or 30 years mastering the craft of programming or
167. 33:48film making or illustration or design are [ __ ] pissed off. They're pissed off.
168. 34:03Because these tools do what they do, right? And and the rage right now is about, well, they stole our
169. 34:14work and that's why I'm losing my money. No, you're not losing your money because they stole your work. You're losing
170. 34:18their money because we have we now have this tool that can do the execution of things in an incredibly efficient way, but it also
171. 34:27makes new things possible. Tik Tok is glitching, FYI. Oh, yeah. It looks like my interface went away. Hang on. We've lost the stream
172. 34:41two times. All right. Hang on. Let me I'll quit it and come back in. Right now,
173. 35:00all right. Is that better people? Tik Tok people, are you good? Is this good? Hello. Hello. Hello. Hello. Hello. Hello. Um,
174. 35:13looks like we're back. Okay. Um, so to the degree that people in very specific jobs or careers, let's put
175. 35:30careers. To the degree that someone is in a very specific career, I'm just an illustrator, I'm just a programmer, I'm
176. 35:37just a this, I'm just a motion graphics person. to the degree that those people remain committed to that single thing
177. 35:48and especially if they resist learning AI for that thing, I think they're going to be increasingly employment challenged
178. 36:02because while a motion graphics professional can look at anything generated by these video tools today and
179. 36:08make fun of them and laugh at them, they're a [ __ ] ton done better than they were two years ago, right? And so two
180. 36:16years from now, they're going to be indistinguishable, right? But here's the thing that no one talks about. What everyone talks about is the
181. 36:26AI is going to take the jobs. What no one talks about is the 15,000 people that got laid off at Microsoft.
182. 36:36How many of them are truly AI literate and AI passionate? Let's say it's a thousand of them. You got a thousand really smart people that
183. 36:48were maybe at Microsoft and let's let's say it's less than that. Let's say it's a hundred of them. Out of 15,000,
184. 36:53there's 100 people that are super passionate about AI. One is Microsoft should have kept them, right? I don't
185. 37:00know how they identified the 15,000. The other one is those hundred people now don't just have the skill that they were
186. 37:08hired for at Microsoft. They've got all the other skills that they can tap into with generative AI. They can do
187. 37:15research. They can do writing. They can do art. They can do music. They can do film. They can do all these different
188. 37:22things. They can do marketing plans. And so what what I think we're going to see is masses of people being forced into entrepreneurship.
189. 37:36And those people that are like, "Well, I don't I don't know how to start a business and and they're not playing
190. 37:41with AI." They're going to get circles run around them by people who do know AI that are like, "Okay, I want to start an
191. 37:49Etsy shop. Okay, I need to come up with designs. I need to come up with manufacturing. I need to come up with
192. 37:55sourcing. I need to come up with a marketing plan. I need to come up with advertising. I need to figure out a way
193. 38:00to do content marketing because I don't have a lot of money, but I do have a lot of time because I just got laid off from
194. 38:04my [ __ ] job. Right? A single person. If you if you want inspiration for the kind of person who's going to be incredibly successful
195. 38:15in the future, take a look at Jim Ross, who's a part of this organ, this community. He's in the self- storage business. He
196. 38:26spends an hour every morning like like drinking his cup of coffee playing with AI doing things with his business for AI
197. 38:34or if he doesn't have anything he's just out experimenting. He comes through these lives on a regular basis and when I teach
198. 38:41occasionally teach something and he learns something new he's like huh let me go try that for my business.
199. 38:48He's told me on multiple occasions that he is functioning, he himself is functioning like a 10-person company
200. 38:58because he's gone all in on AI. That's what people aren't talking about. They're just talking about the loss side
201. 39:05of it. The opportunity side of it is [ __ ] huge because because so many people are resisting AI, pissed off at AI, avoiding AI,
202. 39:16dipping their toes in the water with AI, waiting till someone else figures it out with AI, the people that embrace AI, the people
203. 39:25that are irregular enough to show up at a channel like this five nights a week to just be in the conversation to to
204. 39:34glean some knowledge through osmosis or talking to others or watching me demo something. The work that they produce as a single
205. 39:46entity, as a single person, is going to make them occur to other people like [ __ ] wizards. Wait, Jim, you what? What do you mean
206. 40:00you were on a call with a guy? Oh, this was a thing that Jim Ross did. Here's one th this is one where you appear like a wizard.
207. 40:10So there was some big industry event for self storage, right? And there was a keynote speaker, the main keynote speaker dropped out of this
208. 40:24big event 20 minutes two minutes before he was supposed to speak. sorry, can't do it. So the the guy running the
209. 40:37event reached out to Jim Ross and said, "Jim, can you do a keynote for this event?" And Jim said, "I don't really have a keynote."
210. 40:47You know what Jim did? This is [ __ ] brilliant. He went into I don't know Manis might have been Manis or or GenSpark one of them and
211. 41:01the keynote started and he said hey you know I'm here at the last minute for this keynote. So, what I'm going to
212. 41:07actually do is I'm going to use AI to write a keynote presentation about AI in the self-s storage business.
213. 41:18And with people watching, he pushed the button and this thing started writing and generating slides for his keynote.
214. 41:27And so, five or 10 minutes later, it generated this keynote. He gave the keynote. And then when when the thing was over,
215. 41:37what do you think people wanted to talk about? Self storage? No. They're like, "How did you do that? What What was that? Is that
216. 41:48even possible? Did you Was that like a movie? Like that's Did Did that just happen?" So like what's going to happen to the
217. 41:59programmers is the programmers that understand that the power dynamics for all of employment are shifting and
218. 42:10they start adapting and start and start doing everything they can like like one thing that I think programmers could do
219. 42:18that is going to give them job longevity for some time is I think there is a whole industry about to be born
220. 42:29of good engineers that can unfuck up vibecoded applications, right? Troubleshooting shitty code that clueless people like me wrote, right? I
221. 42:44started a SAS company. I've got $100,000 in MM MR and I find out that my my SAS service that I vibe coded is completely
222. 42:52insecure and people are getting their credit cards stolen. Oh [ __ ] what do I do? I don't know. Maybe call someone
223. 42:58that knows what the [ __ ] they're doing with technology, right? That's going to be a whole job. Who's going to get that job? Not some
224. 43:07elitist programmer that's like vibe coding's not real coding. No, no. It's gonna be a vibe coder like Andre Karpathy that's like, I'm gonna
225. 43:17I'm gonna learn how to do this vibe coding stuff as well as I know how to do that coding stuff because I want to understand what it
226. 43:24does well, what it doesn't do well, and I want to understand what the vulnerabilities are. And I want to be
227. 43:28I'm you know what? I'm gonna start a whole business around fixing things that people messed up because I because what
228. 43:36I've learned in messing around with all this is they always [ __ ] up the same four things, right? So, I'll go in and
229. 43:44they'll be in a panic. They need me to fix their stupid app that they made. And because I know the four things that they
230. 43:51probably [ __ ] up, I can go in and look at it and I'll charge them $10,000 because they're in a panic and I'll fix
231. 43:56it in an hour, right? Or a day or a week, doesn't matter. Like there's a whole business there. I I feel that's a whole industry
232. 44:06like a whole sector. It's developer approved. Have you tried emergent yet? No, I have not. Let me see. You know, it's hard to
233. 44:23type with a guitar in your lap. That's scientific fact. There's a research study. Oh, it's 8:44. Uh, emergent.sh. If.sh means shell, if
234. 44:36if emerging is something I do in the command line, Gareth, I I will you know you know my opinion on emergent.sh.
235. 44:49Uh uh welcome to emergent build ambitious app with AI. Uh English not the first language of the developers here. You know you could use
236. 45:00chat GPT and translate that a little better, right? I'm such a dick. It's developer approved. All right. What
237. 45:13would you like to build today? Um, let's do a Um, okay. Um, I'd like to build. Is this going to be one of these? Oh,
238. 45:26no. I've got 10 credits. Okay. I'd like to build a story assistant where
239. 45:42user picks from the top seven storytelling frameworks and
240. 45:56then gives a simple log line of their story and picks a framework and the site um
241. 46:14writes their outline, their complete outline, their complete outline. We'll just start it there. Okay. Bang. Emergent.sh.
242. 46:30This is looking decent. Setting up a safe environment. Ah, safe environment. See, not just setting up an environment,
243. 46:38setting up a safe environment. Very cool. If if indeed it's safe, but but if Gareth is recommending it, it probably
244. 46:46is configuration environment or configuring environment and then starting the agent. Booting. [Music]
245. 47:11Love you. [Music]
246. 47:36Oh, good. Yeah, Brandon, keep keep No stream yet. Okay. Yeah, keep an eye on X. I I I have this sneaking suspicion
247. 47:43that they're not gonna they're not going to go live at 9ine, but we'll see. All right. agent is thinking. So, I guess I guess
248. 47:51it set up a little environment for us. Don't trust anything I recommend. This is your music on hold. Oh man. Um, Side
249. 48:08Hustle Mimi is going to be back, Pate. Yeah, it says safe right there. It's got to be true or you couldn't put it on the
250. 48:14internet. That's that that's my philosophy. is is, you know, if if someone took the time to write it on the internet, then it's
251. 48:24probably fine. So, yeah, this said it was safe. So, I think it's good. I think we're good. I mean, here's what safe at least says
252. 48:35to me is that they're at least aware that not having a safe development environment is a big problem in vibe vibe coding. So, at some point,
253. 48:45someone's going to figure this [ __ ] out. It'll probably be Google. [Music]
254. 49:00You think Elon won't be accurate on his release time? Gasp. I just I I would have thought I would have heard more of him bragging today
255. 49:10about what's coming with Grock 4. And I'm hearing more about the Epstein files. So, I don't know. I just have this sneaking suspicion
256. 49:22they're not ready. But they did they did flip um they did flip Gro to this thing called Super Grock.
257. 49:38The files are in Grock 4. Oh. Oh, the FC files are in Gro 4.
258. 49:53That Roadster promised in 2016 will be ready any day now. So will full self-driving. Um, what new features come with Grock
259. 50:04Fort? We really don't know. They and it was this was supposed to be Grock 3.5 and then just all of a sudden they said
260. 50:10it was going to be Grock 4. Um, and we haven't heard anything. So it looks like my my application is being written.
261. 50:24Introducing forks. Oh, that's cool. I like forks. A kind of weirdass version control. Story assistant. Choose a framework. Share your idea.
262. 50:36Start creating your story. 3X structure. Hero's journey. Save the cat. Can I select multiple of these? No. But
263. 50:45that's okay. We'll do save the cat. Continue the story idea. Or try one of these examples. A young hacker discovers a conspiracy that
264. 51:00the list of famous people who um did bad things
265. 51:17suddenly went missing. [Laughter] Your story outline based on a young hacker discovers cons. Wait, based on
266. 51:33the opening development would go here. This is mock. Oh, so this is not using. All right, this is all mock data. All
267. 51:41right, whatever. All right, cool. This looks this looks very much in line with uh with like replet agent things like
268. 51:50that. Thank you very much, Winston. Thank you so much. That's very generous of you. Um what's cooking? Um we're
269. 51:56we're minutes away, if you're to believe what Elon Musk announced on Monday. We're minutes away from the Gro 4 announcement.
270. 52:08Do you use some kind of intro to the prompt? What is this? Oh, the texture of the image changes. I I'll
271. 52:20answer that. That's that's pretty cool. That's a good question. Yeah, Winston, thank you for the gift. That that was very generous of you. I
272. 52:28appreciate that. Uh let's go to I wonder if it'll be Elon Musk. Let's go to I guess we'll go to X.AI AI croc 4 search
273. 52:51gro 4 60 minutes left until Grock 4. I just don't I don't think it's coming. Wait, what's that? Grock 4 presentation
274. 53:00tomorrow is going to be hilarious. Allegedly it will be atxai. So is it x.ai x.ai.
275. 53:15Here you go. Yeah. So Elon put a tweet uh or a zit or whatever we call them now a couple days ago saying that the live stream was going to
276. 53:26be on the XAI uh X handle or Twitter handle. Oh, the XI Twitter handle. Okay. Wednesday at 8:00 PM Pacific. That being
277. 53:37said, I'm not seeing anything to indicate that that would be going live yet. X AI X AI. Okay. And by the way, prediction is Oh, wait.
278. 53:52Breaking Grock to launch tonight at 11 p.m. But who the who's that? Who's son of Walke? Oh, 11 PM EST. Yeah. So, this is
279. 54:07[ __ ] Okay. Um, where's Let's go to XAI. I'm following them. Live on X. Look, AI learning labs live on X. Woo.
280. 54:24Gro for party. Grock 4 launch party. Yeah, there's all these launch parties. Where's the launch, Elon? You got six minutes, Elon.
281. 54:35We're all waiting. We're all waiting at this point. I've worked with people coming from going to every major AI player.
282. 54:44Wait. Oh, you've worked from for You've worked with people coming from every major AI player except X. Yeah.
283. 54:52Although, didn't X just poach a couple of people from Open AI and then Open AI poached a couple of people from X?
284. 55:00He's announcing his new party. Yeah. Um, live search API. Yeah, this is all old [ __ ]
285. 55:24[Music] Yeah, we could do that. Um, if you're while we're waiting for Elon to show up
286. 55:39with his [ __ ] um, if you're not a member of, uh, of the AI salon, you should be. So, if you go to the salon.ai AI and
287. 55:49click on join our community. That's going to drop you into this awesome place full of awesome people. Um
288. 55:59you'll learn about things like the AI readiness cycle, play first, mindfully create, generously lead, um our values,
289. 56:06five stages of AI adoption. But if you scroll down the lefth hand side here, there's there's um first sort of seven
290. 56:12things you do are things like introduce yourself, check out our values, check out pest meetings. So you really get to
291. 56:18understand what the community is about. Um then there's a community corner, a news area, and then there's um a
292. 56:25mastermind area where there are clubs and hubs and all sorts of things, leadership lounges and things like that.
293. 56:31I want my crown. Um you should get a tag. You should get a badge, Gareth, but I don't think you get a crown. Just the
294. 56:38spaces have the crowns. Um the crowns are part of the mastermind community. Yeah. any anywhere where you see the
295. 56:45crown beside it, those are subscription areas that are part of the mastermind. Um, and the mastermind has very very low
296. 56:53special pricing through the end of 2025 and you get to become a founding member of the AI salon mastermind. Um, but you
297. 57:01should go check it out. Another cool thing on the AI salon is there's a a free area called the Irregulars channel,
298. 57:08which is all the people that hang out here. And here's Vicky's favorite picture. Did she make a Did you make a video of them?
299. 57:23[Laughter] That's pretty good. You can see comedy like that right here. Right here in the irregulars channel. Look, there I am.
300. 57:39Artificial intelligence is no match for natural stupidity. No [ __ ] All right, back to uh back to X to see
301. 57:48if anything's happening anywhere. [Music] Ah, internet will break today after seeing Grock 4. Uh, I'll believe it when
302. 57:58I see it. Let's go see if Grock 4 is showing up anywhere. Gro super. See where it says super Grock? That's new, but it's still Grock 3.
303. 58:08Oh, go to grock.com. Grock 3 enabled. Custom instructions. Enable search.
304. 58:34[Music] Good lord. So, what here's what we'll do. We'll we'll make some cool [ __ ] tonight. What? What is that desk?
305. 58:50[Music] [Music] down. [Music]
306. 59:13[Music] [Music]
307. 59:36[Music] All right, Elon, are we going? Are we going? Are we going? Are we going?
308. 59:55I do not think we are going. It is nine o'clock and I see nothing. Uh let's look up Elon. Elon Musk. Hello, my name's Elon Musk. Grock to
309. 1:00:09understand Grock for release Wednesday at 800 PM. Ah, he didn't say which Wednesday. He didn't say which Wednesday, did he?
310. 1:00:22I think you need a paid subscription to see Gro 4. I've got a paid subscription. This there's supposed to be an announcement.
311. 1:00:33Gro 4 released live stream on Wednesday at 8:00 PM Pacific Starlink.
312. 1:00:51Uh, I got nothing people. I guess producer Brandon, why don't you Hang on. I got to do physical dexterity
313. 1:01:00on TikTok. Um, why don't you just keep checking on on uh Twitter and we'll just go make some cool images and [ __ ] like
314. 1:01:07that. Um, let me I want to go actually uh comment to this guy that asked a question on on Twitter. Um, did you use some kind of intro
315. 1:01:16prompt? Um, let's see. I made this video quickly on one of my lives as an experiment.
316. 1:01:33Um, so it has a color grading and um, aspect ratio
317. 1:01:49glitch. Here's why. Um, one started with a midjourney image. Two, animated
318. 1:02:05that in midjourney to make the statue look at the camera. Oh, here. Let me show you what I'm talking about. Um,
319. 1:02:23oh wait, let me let me finish this and I'll show you what I'm talking about. camera. Um, took the last frame of the animation
320. 1:02:40and used it as start frame for VO3 talking section.
321. 1:02:56Um, VO3 has a different aspect ratio
322. 1:03:11than MJ16 by9. Um, and it increased the contrast. from the
323. 1:03:28original AI fun. Yes. First of all, that's not how you spell aspect. Uh or increased.
324. 1:03:43Second of all, um XAI just posted the Gro 4 live stream will begin soon. Stay tuned. All right. So, let me go show you the uh
325. 1:03:55the the video that I made. Um, let me This was the one. So, so someone just commented on this. They asked the
326. 1:04:03question, "Why does the second half of this video not look the same as the first?" And it's because I use two different
327. 1:04:11tools. Oops, that's not what I want to be doing. Cancel. So this was from last night. We had
328. 1:04:28this image in midjourney and then we animated it for 5 seconds in midjourney. then took the last frame of
329. 1:04:40that animation, took it into VO3 as a start frame so that we could get the statue to talk. And so here's here's what we created
330. 1:04:49last night. And this was pretty quick. This was like I don't know 10 15 minutes. Oh, this curse, that vexing witch. When
331. 1:05:05will I be free? Nice, right? But see how it's like flat and then it goes high contrast. This curse, that vexing witch. When will I be free?
332. 1:05:20But it's still pretty [ __ ] cool. Like if you if you did some color grading on that to match them and and you know,
333. 1:05:26make sure it's not stuff's not jumping around. Be pretty good. You're showing your expiration date. Waha.
334. 1:05:34This is like Heraldo Rivera leading us to Alapone's vault. If you don't remember that, that was that was one of the most hilarious
335. 1:05:43events in television history. All right, let's go to Xi. Yes, before you uh navigate away from this, I sent you a DM on the hugging face
336. 1:05:53robots, and since you're already sharing with sound, it would be a good idea to bop over there. Where? But is it just is it big and
337. 1:06:00prominent on hugging face or is it one of the 8,000 links? No, no, no. It's it's a it's a DM in your Twitter. It's a tweet. Oh,
338. 1:06:08with a video. Oh, got it. Oh, okay. Because because I'm sharing. Great. Got it. Okay, cool. So, for AI builders, so this is um
339. 1:06:24I guess Hugging Face is putting this out, Brandon. Yes. Is that what you said? All right. So, Hugging Face, if you don't know Hugging
340. 1:06:31Face, Hugging Face is a repository for like open source models and weights and all sorts of things. Like geeks use it
341. 1:06:39and you can go play with things there if you want to get around dick around with open source. It's a great it's a great
342. 1:06:44resource. Um, and apparently they're putting out an open-source robot for 300 bucks. So, here we go. [Music] Let's
343. 1:07:06[Applause] get everyone. You get everything you want. If you want it, you can have anything. If you want it,
344. 1:07:20you can have anything you want. In the brand new [Music]
345. 1:07:40cool reachi mini cute fascinating they're going after the education market but you know isn't that fascinating that like Where
346. 1:07:54education starts to go is go make a robot. Go program a robot. Um, hugging face powers everything in the AI world.
347. 1:08:03Um, if it's not the models themselves, it's the data sets. Yeah, hugging. Yeah, hugging face is really quite remarkable.
348. 1:08:08There's a lot a lot of stuff there, which is why I avoid it. It makes my head hurt. Um, uh, does anyone high-five in real life? I still high
349. 1:08:22five. It's a Gen X thing. Sometimes I high-five people and I I tried to high-five a Gen Zer once and
350. 1:08:31she goes she goes she goes um I don't do that. I find it aggressive. Um I don't do that. I find it
351. 1:08:45aggressive. [Laughter] She's like, "Oh, no. No.
352. 1:09:02No live stream yet. We're waiting. We're waiting. We're waiting." Okay.
353. 1:09:16XAI. The Gro 4 live stream will begin soon. Stay tuned. Link, Google,
354. 1:09:34Gary Marcus is still dunking on all the companies for not having their [ __ ] together like he'll do. Jav Lopez has got a new
355. 1:09:43upscaler coming. Um
356. 1:10:08[Music] Da da da. I found out that find vocal fry is a normal expected part of
357. 1:10:23speaking Finnish. Hey really really hey.
358. 1:10:48Teddy Rubkin was not safe for work. You missed your chance for vocal fry. Um, by the way, if you're part of the AI
359. 1:11:01salon or if you just checked it out because I told you to go check it out, go to the events section next Tuesday.
360. 1:11:08We have an AI salon meet and greet where you can come introduce yourself and meet people in the community. It is very wise in my opinion
361. 1:11:17to get your ass in an AI community. Start getting curious and not an AI an AI hater community that that
362. 1:11:29like they're just going to get more and more and more bitter. It's gonna it's just I would avoid those. I would get in
363. 1:11:35one where people are like, "Well, it's here. Might as well try to figure this [ __ ] out, right?" Um, yeah, bring a a
364. 1:11:44friend or a friend of me. Yeah. Um, come and introduce yourself. All right. Let's see. Let's see. Let's see. Let's see.
365. 1:11:52Let's see. Let's see. Hang on a sec. Oh, that's dead. Dang it.
366. 1:12:10Oh, wait. What's that? No. No. 11 minutes ago. Stay tuned. Breaking. Kanye no longer working with Gro 4.
367. 1:12:25Mr. Strawberry. I'm super [ __ ] excited. ASI baby. Shut up. This is not going to be ASI. This is probably going to be about as
368. 1:12:36good as GPT40 is my guess. Maybe it'll have video as part of it. I don't know. Maybe it'll be AGI. May maybe Grock 4 is the first one to to
369. 1:12:49really hit it. Kanye shouldn't be working with anyone. Let's go. What are you guys predicting? I predicted that this thing wouldn't
370. 1:13:01happen tonight. That's what that's my prediction. My prediction is Oh, here is also my prediction. If it
371. 1:13:13goes tonight, if there's a live stream tonight, it will be more awkward than any of the Open AI tiny table presentations like 12
372. 1:13:23days of shipmiss, right? Where lack of eye contact was the feature. It was the feature, not the bug.
373. 1:13:41Uh, it will have video creation like VO3. I heard that too. And and Elon sort of hinted that, you know, video production,
374. 1:13:52you know, video generation is getting really good. So, he kind of did a hint. Um, and quite frankly, um, when they dropped VO3, nobody was
375. 1:14:01expecting it to have image generation in it that was any good. And it's, you know, the image gen within V, uh, Gro 3 is pretty good.
376. 1:14:11Um, he's probably getting the video upload error like I get. He's trying. Yeah, that's what it is. They'd be live
377. 1:14:16streaming except if Elon can't get the QuickTime video to upload to X. Oh man. Give us five minutes. Elon had to pee. What's your hobby?
378. 1:14:34[Laughter] Oh my god. I've never understood.
379. 1:14:50I think I'm just too neurotic. Because like I don't understand how you could say, "Hey, we've got a major new
380. 1:15:00model coming out. Here's the here's the launch date or the announcement date." How do you miss the [ __ ] announcement,
381. 1:15:11right? It's like like I I just I feel like people's relationships with deadlines anymore are really [ __ ] annoying. Like, yeah,
382. 1:15:21we'll get this. Maybe it's 8. Maybe it's 8:15. Maybe, you know, maybe never. We didn't Oh, you thought we meant this
383. 1:15:27Wednesday. That's That's [ __ ] hilarious. Really? What? What made you think that? Oh, because we announced it
384. 1:15:36on a Monday and we said it'd be live Wednesday. We didn't say which live. Which Wednesday, did we? Huh? See how this works?
385. 1:15:51been in politics. Now the GPT store is also coming. Oh my goodness. People don't care about time and it drives me crazy. I know. It makes
386. 1:16:02me mad. Madness. It's madness. I tell you, madness. Wednesday. Next Wednesday or next Wednesday. Oh, it was Wednesday Pacific time.
387. 1:16:26Vicki, I start my meetings on time. No waiting for those who trickle in from other meetings. Yeah, I know. I It's We
388. 1:16:33deal with agencies all the time and it's just like it's almost like they have Tourette's. Oh, sorry I was late. I was in another meeting.
389. 1:16:44Oh, oh, sorry I was late. I just had to grab a bite to eat. Hey, sorry I was late. I just don't respect your time.
390. 1:16:53If you're five minutes early, you're late. Yeah, it is true. If you have a meeting with Vicky Baptiste, if you log in early
391. 1:17:00thinking you're going to be there first, you're going to lose. I have an itchy scalp. I wonder if it's bugs.
392. 1:17:12You ever seeing that commercial? This the GenX. Remember the Head and Shoulder commercials where like the chick and the
393. 1:17:20dude are out on a date and she's like she's thinking like she's like, "Oh, he's really nice." And then he he kind
394. 1:17:26of does this and she goes, "Oh, do you think he has dandruff?" Funny comment on YouTube. Windows update is almost done.
395. 1:17:44We should charge X for making us wait. Like in a doctor's office. Yeah. Yeah. Exactly. The The only thing less
396. 1:17:51reliable than tech announcements are doctor's office announcements. You've got to be there at 2:30 or, you know,
397. 1:17:58we'll cancel your appointment or we'll kick you out of the practice. You get there at 2:30. 4:15. They're like, Mr.
398. 1:18:05Shannon. Yeah. Well, there's no need to be nasty. I've been here for four hours. Oh, there is a reason to be nasty. There
399. 1:18:17is always a reason to be nasty. Um, why is the AI learning lab more
400. 1:18:31consistently on time than a what what's XAI worth now? 1516 billion dollars. Let's go ask Chat GBT what X AI is worth.
401. 1:18:44Oh, you're not gonna be able to see this, but I'll I'll just trust me. I'm asking it. No, I'm good. Um, how much is X AI valuation
402. 1:19:01[Music] $80 billion? Why is it that us with our poultry little 20 million production studio made to
403. 1:19:15look like a shitty home office? Why are we more on time than an 80 billion company? Why? Huh? You tell me why. I
404. 1:19:26don't know why I just became Seinfeld. You tell me why. I don't get it. Elaine. Um, they want you to be 15 minutes
405. 1:19:41early. They just don't respect our time, you know. Dear dear Mr. Musk, I was very excited about the announcement of of Gro 4. I have a
406. 1:19:56community of AI enthusiasts that show up week nights, sometimes occasionally around 800 PM Mountain time.
407. 1:20:05We're on that channel talking about AI and I I reserved a very special episode of the AI Learning Lab Live for for
408. 1:20:15Wednesday, July 9th, 2025 at 8:00 PM Pacific time, 9:00 p.m. Mountain time. That's an hour after we normally start.
409. 1:20:23We got there an hour early so we'd be ready for your lives. Dear Mr. Musk, to say that I am frustrated would be an
410. 1:20:31understatement. Sorry. Dear Mr. Musk to to say that I am frustrated with the lack of a live stream. Dear Mr. Musk,
411. 1:20:44congratulations on your big long hard tall rocket flying into and coming back down on the hard earth standing up straight.
412. 1:20:58Um, I waited for your live stream about Gro 4 because I heard it was going to change everything and it never came. Dear Dear Mr. Musk,
413. 1:21:17Dear Mr. Musk, Dude, the start time now says 8:30. [Laughter] Dear dear Mr. Musk, um I I'm impressed
414. 1:21:32with how large and tall your rockets are. Wait, dear Mr. Musk. Dear Mr. Musk, um your electric cars are fast and sexy.
415. 1:21:51Your your boring company makes large holes and your rockets are really big.
416. 1:22:07I forget what I was going to write here. Um, congratulations, Mr. Musk. I hear you're making a lot of children
417. 1:22:22with the with the women with other women with the women that you make you're populating the because the birth rates
418. 1:22:30are right in Mars and making make congratulations Mr. Musk I it's um you're it's a it's a very impressive
419. 1:22:40rocket you have very very impressive with the and with the children too. Good. It's a it's a noble cause. It's a noble cause.
420. 1:22:51Oh, sorry that was so loud. I think I woke the cats. I don't know what to say. Dear Mr. Musk, please add the pink bow.
421. 1:23:10His rocket comes with report. Well, that went so high. It went so high in the air. It's so exciting.
422. 1:23:28I guess I could have some some fancy tequila. [Music]
423. 1:23:47Okay, we're gonna go to I'm gonna change my just producer Brandon, just let me know when dear Mr. Musk gets his [ __ ]
424. 1:23:57live stream up and running. I'm gonna go make some midjourney images. We're going to start out with Dear
425. 1:24:04Mr. Musk. We're in midjourney, by the way. Right now, the prompt is dear Mr. Musk. Um, we'll go dash dash sref style
426. 1:24:20reference space random. We're going to turn our stylization up to 9850. We're going to go 16 by9 because I likes
427. 1:24:34me some 16 by9 aspect ratio. We're going to do the black bar for producer Brandon. He's being very patient tonight.
428. 1:24:43And we're going to go dash dash um four. And so R means repeat. So we're going to create four sets of four. And then we're
429. 1:24:59going to go dash chaos. We're going to do chaos of 65. And we're going to turn on personalization because I want these things to look like
430. 1:25:14the [ __ ] I want them to look like. All right, here we go. And we're off.
431. 1:25:31planned to be. He is literally my love. Oh, planned because he is literally my love. Oh, come on, Elon.
432. 1:25:41Dear Elon, dear dear dear Mr. Musk. Oh, look at that. She My god, he's a handsome man. That is handsome. Look at
433. 1:25:52that. That South African chisel looks like JFK there. Oh, look. It's his little car. His little cars are so cute.
434. 1:26:08They're so cute. They don't even have a grill. Did you know that about the cars? His car, they don't have a grill. You know why?
435. 1:26:17Exactly. You don't have to cool down that engine. It's just batteries. Just batteries right along the floor. It's
436. 1:26:24just It's like a little miracle. It's just a little miracle. He's just amazing. The big tall rockets. Oh, I get so excited.
437. 1:26:33Uh uh uh uh uh uh uh uh. Dear Mr. Warhol, I would like to commission you to make a self-portrait. A portrait of myself in a little
438. 1:26:46spaceman outfit. What? Oh, Mr. Warhole died years ago. Huh? Well,
439. 1:27:02dear Mr. Mask, where's your [ __ ] live stream? Wherefore art thon? got a got a live stream tonight precisely at 8:00 PM we launched rockets
440. 1:27:17so those are in pre precise timing windows uh so when we launch something with XAI similarly precise except not
441. 1:27:26tonight different we dealing with the Epstein files we get them get my name redacted it doesn't matter that's actually a very nice image
442. 1:27:41DM'd you a link to the not let yet live stream. Okay, good. Beautiful. That's a cool style, too. Which, by the way, so if we're going and
443. 1:27:52looking at these um see, right? So, you get your four images here that all kind of look the same. And right up here is the style reference.
444. 1:28:08You know, we said style reference random and then we said do this four times. So, this going to give us four different style references.
445. 1:28:17Are they live? No, sorry. I was just bringing it up in the background so we can keep looking at it. Okay. Wait, am I not sharing? Oh, yeah.
446. 1:28:26Okay. Beautiful. Beautiful. Beautiful. Beautiful. Beautiful. Beautiful. Look at that. I like that. I actually like that. Uh let's let's use
447. 1:28:35that SRE again. We use Chaos 65 again and let's do um we'll do um
448. 1:28:49Horatio as the prompt [Music]
449. 1:29:03low rent Robert Downey Jr. Oh my god.
450. 1:29:19So much for efficiency. He's busy starting his next failed political party. He hasn't failed yet. That's cool
451. 1:29:32for her ratio. That's pretty cool. Oh, that's good. Looks like Philip Seymour Hoffman. How good is that?
452. 1:29:50Let's Let's animate some of these. All right. So, we're going to animate that. We'll animate that in high motion. We'll do low motion there.
453. 1:30:02We'll do low motion there. We'll do high motion there. All right. We're making movies. I got an ice cream and you can't
454. 1:30:35dear rock. How can one accelerate emergence from a khole? asking for a friend. Yes, Palindrome. Palenrome has
455. 1:30:43the right question. Mr. Musk. Mr. Musk. Mr. Musk. Curious. How do you accelerate emergence from a khole? You khole.
456. 1:31:00Seriously, when's the live stream? My community is getting stir crazy. They don't know what to do. They're challenged in many ways. Neurode
457. 1:31:10divergent exception twice exceptional three times except they so exceptional we stopped counting the ways.
458. 1:31:22I understand you may have some neurode divergency yourself, Mr. Musk. Mr. Musk. Anyway, yes. No, I I've seen your
459. 1:31:29rockets. They're they're they're very large. It's shiny. It's they glisten. Um Yeah. No. Yes. I I'm sorry. I But no,
460. 1:31:37I'm here to hold you accountable for for the live stream. You you promised us 8:00 p.m. precisely, Pacific Standard Time. It's not live.
461. 1:31:51All right. [Music] Let's see. How do I scrub? Oh, here we go. All right. All right.
462. 1:32:10I think these two are better. That's pretty cool, right? How good are these? I'm telling you, man. This midjourney
463. 1:32:22this midjourney [ __ ] animation is so good. Why Elon will not be president? He'll be late on the debates.
464. 1:32:37Oh, yes. Well, there's no animation in the arms there, so that's a loser. Same here. Huh. This is the first This is the first
465. 1:32:52midjourney animation I've seen that I feel like it it actually missed something. Like those arms should be doing all
466. 1:33:01sorts of crazy ass [ __ ] Let's go try something. Let's go. Let's go uh go back to that and we'll do high motion manual and we'll go um
467. 1:33:16the cat tentacles rise. as
468. 1:33:31the feline looks menacingly at the camera. All right, let's try that. Let's see if we can get it to work, people. I don't
469. 1:33:46know if we're going to get it to work. I'm not a miracle worker here. There there have been petitions signed.
470. 1:33:53I I can no longer claim that this is um a university. Um I cannot say that if you come to the AI learning lab you're
471. 1:34:02it's going to be an accredited situation. Um I am not legally allowed to give out even a certificate at this point. So I
472. 1:34:16but I think it is clear that you could learn something here potentially if the lawsuits stop. L a captopus.
473. 1:34:32That's good. Oh, wait. Although his head his head's not moving. He's had Botox. [Laughter]
474. 1:34:46Yeah. Yeah. See? Yeah. See? Yeah. Paluca. Well, that was kind of cool. Except the rain's not It's clearly raining. It's
475. 1:35:02got rain in front of his face and it's not moving. Like the famous turducken. Elon needs to pull out Grock, blow on
476. 1:35:10it, and reinsert the cartridge. That's Gen X video game reference. Hey, have you seen my live stream? I I
477. 1:35:20had a streaming box. I think I left it in the trunk of the car. Have you seen it? Hey, have you do you have any
478. 1:35:26ketamine? Do you have any ketamine? No. My teeth are This is normal. This is what my teeth look like. [Music]
479. 1:35:43What? Why are we not live? Why are we not live? I've been I've been sleeping in the in the artificial intelligence
480. 1:35:53factory. Where are the cars? Oh, sorry, wrong company, Mr. Musk. Where are the cars? Uh uh uh. Oh, here we go. Yeah, there
481. 1:36:04you go. See? Oh, cool. That's what we're looking for. That's good. It's so good. So good.
482. 1:36:26Bad lips syncing AI midjourney edition. All right, we're going to make this cat talk while we wait for Mr. Musk to get his
483. 1:36:40[ __ ] together. I think it's going to be this one. Or maybe it should be
484. 1:37:03No, not that one. Could be that one. All right. So, what we're going to do is we're going to download this
485. 1:37:30We go to the last frame and we're going to go copy image and then we're going to go to
486. 1:37:46flow.google. Hello, I'm with Flo.g Google. It's all Google Gemini. I guess I don't have it open.
487. 1:38:02Um, flow.google. All right. And then we're going to go create with flow. Then we're going to switch to the right account.
488. 1:38:21[Music] Elon does one more thing to make his stock go down.
489. 1:38:36Ah, y'all are funny. I like comedy. Comedy is not pretty. All right, new project. So, we're going to do what we
490. 1:38:46did last night. We're going to do frames to video. We're going to paste a video. Oh, I can't do that. I got to come over. Wait,
491. 1:38:52I got to go to preview. Open that. Do a new image. Save that as cat. Save.
492. 1:39:09And then we're going to go here. We're going to go frames to video. We're going to upload. Where's our cat video?
493. 1:39:24No. Where did I save it? What did I call it? Cat. Oh, there it is. Okay, there's cat.
494. 1:39:42Crop and save. Crop and save. Carrying on rather cherry. Pep pipep. Um, the
495. 1:39:58octo the octo the cat with tentacles stops in the middle
496. 1:40:14of the street and says, Something funny about
497. 1:40:31how ironic it is that he has cat taste
498. 1:40:45and struggles to not eat his delicious [Laughter]
499. 1:41:00octopus arms. We We could have him say something interesting, but no. Let's Let's do something stupid.
500. 1:41:13Never mind. There's a weight list unless you have Max. Yeah, the Perplexity thing. Yeah, I know. Octopus found a
501. 1:41:20catapult one day. Uh uh uh. Is that Cat Cthulhu?
502. 1:41:36OpenAI is also launching a browser. Yep. Champy wants some cheese. Hang on. I got to let him out. Champ, what's up? You want out?
503. 1:42:08Watching Damish Musk
504. 1:42:25hold Please, [Music] let me change my sharing so you can actually hear this when it does it because it's going to be cool, I think.
505. 1:42:40It might not understand the movement, but that's all right. Um, all right, we've got it.
506. 1:42:59All right, here we go. It's a cruel irony, really. I have the refined pallet of a feline, yet I'm constantly fighting
507. 1:43:06the urge to devour my own delicious octopus arms. It's a cruel irony, really. I have the refined That's funny. Okay, let's go. We'll go
508. 1:43:21attach that to the end of this. Go add to end. What happened?
509. 1:43:43Sit. It's a cruel irony. Really? Choose media. All right. Done. Tabs. Oh, yeah.
510. 1:43:59[Music] All right. So, I am in and blackboard. I am in um QuickTime
511. 1:44:14player. And so we made the first part of this animation in midjourney and then the hilarious I have to resist eating my own delicious
512. 1:44:27octopus arms in V3. All right, what happened? All right, here we go.
513. 1:44:42Wait, that's not the right video. Oh, okay. Here we go. We'll go full screen. All right. You can all see this, right? Beautiful.
514. 1:44:56You excited? More excited than we are for Elon not showing up. Some creepy music behind it. It's a cruel irony, really. I have the
515. 1:45:09refined pallet of a feline. Yet, I'm constantly fighting the urge to devour my own delicious octopus arms.
516. 1:45:21That's writing that's so bad that it's good. All right. My own delicious octopus arms. I'm constantly fighting the urge to eat
517. 1:45:34my own delicious octopus arms. All right. Come on, Elon. Where the [ __ ] are you? Oh, you put the live stream here. Live demo.
518. 1:45:48Try it on Grock. Grock 3. No. Grock 4.
519. 1:46:05Grock for iOS. Tune in for the live demo.
520. 1:46:19The world's most powerful AI assistant. When is the live stream? The live stream is now 46 minutes into another dimension.
521. 1:46:39I guess we'll go till 10, then I'll call it. I was excited about this. I've done this before. These things can be fun. But,
522. 1:46:50you know, I'm not going to be abused by one Elon Musk. That was rude. Okay. Did El Don fall out of a window? I think he's just I don't
523. 1:47:06know snacking on some ketamine gummies or something. I don't who knows.
524. 1:47:29Oh, let's go. Let's go to X. Let's go to X and we'll go we'll do a search for SREF. SR EF. Bang. Or we do No, you know what we'll do?
525. 1:47:44We'll do midjourney. Mid journey mood board. An embedding space for M midjourney
526. 1:47:58SRFs. Oh, that's a cool one. Is this midjourney? No sref
527. 1:48:14just a prompt. Is that mine? Yeah, it's mine.
528. 1:48:29I remember when we made this. Remember we we made this in here. I think this has audio [Music] here. You want to see a movie we made?
529. 1:48:38When did we make this? We made this think this is like January or something. December. December of 2024. Gareth.
530. 1:48:51It's going to happen at 1001. I know because I drop out at 10. Well, you know me. I can't say goodbye anyway. But look
531. 1:48:58what we did six months ago right here in this live. We did this.
532. 1:49:13So what I did was I made a mood board of my sky photography and then I made a bunch of midjourney images and then I
533. 1:49:20think we put them into I think it was lumalabs to animate them. But now we could do this right inside Midjourney,
534. 1:49:26which is super cool. Oh, this was Runway ML. Runway ML did the animation. Sunno did the music. Midjourney did the
535. 1:49:34images. My photography served as the mood board. And you know, being the photographer, these these skies are in the
536. 1:49:45neighborhood of the kind of stuff I would photograph. [Music]
537. 1:50:00Heat. [Applause] [Music]
538. 1:50:16Heat. [Music]
539. 1:51:11Nice. Nice. Nice. I like it. It was very Philip Glass and Jeffrey Reio. It was like Coana Scotsy without the talent. [Laughter]
540. 1:51:28[Music] [Music]
541. 1:51:52Oh, here's another prediction. If they go tonight, which they probably won't, um, they won't even mention that they're late.
542. 1:52:05Swanka dink dink dang swanka gank gad. Oh, champ's back.
543. 1:52:19He got bored out there. He got bored. There's never enough cheese. Uhuh. Actually, some of these styles Here's
544. 1:52:35a cool Okay. Cinematic shot of Let's grab this. Let's go to midjourney. We're going to the midjourney.
545. 1:52:53Okay. So, let's see. All right. Don't forget, one of the things that we don't want to do in 2025 is use our
546. 1:53:04brain. So, we're gonna Oh, you can't see any of this because I'm not sharing the right tab. Dang it. You know,
547. 1:53:13okay, so I'm over in Midney. So, what I did was I was over in the Twitter. I found a look and feel that I liked. So,
548. 1:53:21an SREF and there there was a a little bit of a prompt there. And so, here's what I'm going to do. I'm going to copy
549. 1:53:28this. We're going to go to chat jpe. We're going to assume that Brandon's going to tell us if the live stream
550. 1:53:34starts. I'm going to say, "Hey, comma, I found this prompt for midjourney. I want
551. 1:53:48you to create a 20 image storyboard concept where
552. 1:54:04no dialogue is necessary and each shot is a single subject
553. 1:54:19and then I'm going to give there's the prompt. Um, give me an a short overview
554. 1:54:34of the of the story. and then 20 different prompts
555. 1:54:53in the format here with that specific sref sref. All right,
556. 1:55:07that's why I'm not watching tabs. No, I'm good now. Right, we're watching it now. The last lightkeeper. A desolate
557. 1:55:17world where oceans have swallowed nearly everything. A lone lightkeeper maintains the functional lighthouse on Earth.
558. 1:55:24Okay. Cinematic shot of a cloaked figure standing alone. Ancient lighthouse. Gloved hands.
559. 1:55:39Okay. So now I don't think we need that d-p, do we? Oh, that's just personalized. Personalization turned on. Okay. Copy.
560. 1:55:54Now we're going to go to midjourney. Paste. Go. Chat GPT. Next shot, please. Copy. mid journey paste. Oh, wait. What's my Do we
561. 1:56:13want stylization that high? Let's see what the first ones look like. We might go lower than that.
562. 1:56:28[Music] A cloaked figure standing alone on a cliffside as the waves crash below. Yes, brother. Cheerio. Cheerio. Pep pep.
563. 1:56:42Carrying on. I think we need to go lower. So, let's do change the prompt back. We'll do this at 150 maybe.
564. 1:56:55And this is in draft mode. Let's leave it in draft mode. Okay, there's that. Here's the new one. Paste. Go. Go back here.
565. 1:57:09Did it give us the same? Yes, it kept the SRF the same, which that was one of my fears is it wouldn't. So now there's our third shot.
566. 1:57:19Generate. Let me go back here. We grab the fourth shot. Uhoh. Looks like the live stream might be there. Copy. Nope. Nothing.
567. 1:57:29False alarm. Carrying on. Cheerio. Pip pip. Sir, may I inquire? Yes, sir. Go on. Carry on. Oh yes, rather. Uh
568. 1:57:38would you be so kind uh to reveal if you would not to be too forward? What by chance are your qualifications, sir?
569. 1:57:54My qualifications? Well, rather uh it's a uh it's a rather direct question. Uh, you know, one could take offense to the
570. 1:58:08implication. I'm not that man. I'm a far bigger man than that. My qualifications. Yes. Well, uh, there many years. Many years.
571. 1:58:25Uh, uh, uhuh. Uh, do these look like the same style? Yeah, they do. They do. All right. So, wait. So, we've got Okay. Shot one,
572. 1:58:42shot two, shot three, shot four. Did I just copy shot five? No, I did not. Shot five. Copying shot five.
573. 1:58:54What are you doing? Well, we are making a piece of cinema. We're making cinema using midjourney. Midjourney and cinema.
574. 1:59:05How we are doing that is we've had Chad Gibb. I don't know if you've heard of Chad GB. It's rather clever. You you
575. 1:59:11know the British uh thank you Shakespeare are are quite quite uh the linguists and this chat GPT well it gives me a run
576. 1:59:22for my money. I do say the thesaurus has worn thin. Oh, I'm clever. I am clever and humble.
577. 1:59:38Humble. Clever. Dashing. I've been told I I could use some some work on the teeth. Yes. Yes. and the body and the
578. 1:59:54face, frankly, and the hair um are are not quite where apparently the uh adoring public would like them to be. But we are making cinema.
579. 2:00:10And as a cinema crafter, one working on a chain of craft, as you might say,
580. 2:00:24[Music] it does look like they're going to be going live tonight. All right, fine. Well, let's keep making our movie.
581. 2:00:37Yes. Carrying on. rather cherryop pip. So what I'm doing is I've got prompts for 20 different shots. I didn't look at them because I'm
582. 2:00:51committed to AI slop. I don't want there to be anything on the internet that isn't a waste of everyone's time.
583. 2:01:01So therefore, I have chat GPT make a random shot list of a random story in a random style and I'm going to tell that story and it
584. 2:01:15might be good. It very well might be good. Uh but if it good is good, uh it doesn't matter because I've stolen it anyway. I've used
585. 2:01:25AI and if you use AI, you've stolen from everyone. So, so Thievery Incorporated, right? Hey, Kyle, we're uh stream is up.
586. 2:01:36Stream is up. Okay, fine, fine, fine. All right, let me go turn this off. Stop acting silly. Share that. Uh, share that.
587. 2:01:48It does still say it's going to be in shortly, but now it's animated. Oh, all right. Well, then I'll keep I'll keep I'll go to it. Okay,
588. 2:02:02there we go. We'll begin shortly. I'll turn that off. Okay. Or maybe I'll just make it low. But I think if I make
589. 2:02:09it low, it's going to even be Okay. All right. Um, I'm going to You're just going to see that. I'm I'm going to
590. 2:02:25keep going and making shots for our cinema. Our cinema. I didn't realize you were in the cinema.
591. 2:02:39Pray tell. When did you become a filmmaker? Oh, rather I went to chatbt and I said, "How do you make a film?"
592. 2:02:46And it said, "Well, here you go. Just paste this in and do nothing. You are a talentless oath. I embrace my talentlessness.
593. 2:03:06I wonder if I missed a shot. So, this is 17. All right. So, let's see. We've got one, two, three, four, five, six, seven,
594. 2:03:22eight, nine, 10, 11, 12, 13, 14, 15, 16, 17. Great. Perfect.
595. 2:04:02[Music] All right.
596. 2:04:18I'm making editorial choices without you all. Hope you're good with that.
597. 2:04:38[Music]
598. 2:05:21Oh, in a world where knowledge shapes destiny, one creation dares to redefine the future. From the minds at XAI,
599. 2:05:37prepare for Gro 4. This summer, the next generation arrives faster, smarter, bolder. It sees beyond the horizon.
600. 2:05:47The unasked and challenges the impossible. Gro 4 unleash the truth. Coming this summer. That was cheesy.
601. 2:05:57All right, welcome to the Gro 4 release here. Um, this is uh the smartest AI in the world and we're going to show you
602. 2:06:05exactly how and why. Um, and uh, it really is remarkable to see the advancement of artificial intelligence, how quickly it
603. 2:06:16is uh, evolving. Um, I I sometimes think compare it to the growth of a human and how fast a human learns and
604. 2:06:30gains conscious awareness and understanding and AI is advancing just vastly faster than any human. Um,
605. 2:06:41I mean, we're going to take you through a bunch of benchmarks that um that Gro 4 is able to uh achieve incredible numbers
606. 2:06:51on. Um, but it's it's it's actually worth noting that that Grock 4 if if given like the SAT would get perfect
607. 2:07:01SATs every time, even if it's never seen the uh the questions before. Um and if even going beyond that to say like uh
608. 2:07:09graduate student exams like the GRE uh it will get near-perfect results in in every discipline of uh of education. So
609. 2:07:21from the humanities to like languages, math, physics, engineering, pick anything and we're talking about
610. 2:07:29questions that it's never seen before. These are not not on the internet and it's Grock for is smarter than
611. 2:07:38almost all graduate students uh in all disciplines simultaneously like it's actually just important to appreciate the
612. 2:07:47I'm pretty sure that's AGI uh really something um and uh the the reasoning capabilities of Grock are incredible. Um, so there's
613. 2:07:58some people out there who who think AI can't reason and look it they can reason at superhuman levels. Um, so yeah, and
614. 2:08:08frankly it it only gets better from here. So we'll we'll take you through uh the Gro 4 release and uh yeah um show you like the pace of pace
615. 2:08:22of progress here. Um like I guess the first part is like in terms of the training um a little less awkward
616. 2:08:29going from grock two to Grock 3 uh to Grock 4 we've essentially increased the training by an order of magnitude in
617. 2:08:36each case. So it's a 100 times more training than than Grock 2. Um and uh and that that's only going to increase. Um so
618. 2:08:49it's uh yeah frankly I mean I don't know in some ways a little terrifying but uh the growth of intelligence here is is remarkable.
619. 2:08:57Yes it's important to realize there are two types of training compute. One is the pre-training compute that's from GR
620. 2:09:04two to GR three. Um but for from GR three to GU four we're actually putting a lot of compute in reasoning in RL
621. 2:09:13possible could be the lighting just like you said this is literally the fastest moving field and GR 2 is like
622. 2:09:19the high school student by today's standard if you look back in the last 12 months GR 2 was only a concept we didn't
623. 2:09:26even have GR 2 12 months ago um and then by training GR 2 that was the first time we scale up like the pre-training we
624. 2:09:33realized that if you actually do the data oblation really care teacher loving using AI in education and also the algorithm super super power
625. 2:09:42quite a lot by amount of 10x other work yeah it's awesome best pre model and that's why critical thinking education model H100
626. 2:09:53and then with the best pre model and we realized if you can collect these verifiable outcome reward you can
627. 2:10:00actually train this model to start thinking from the first principle start to reason correct his own mistakes and
628. 2:10:04that's where the reasoning comes from. And today we ask the question, what happens if you take the expansion of
629. 2:10:11classes with all 200,000 GPUs, put all these into RL 10x more compute than any of the models out there on reinforcement
630. 2:10:21learning, unprecedented scale, what's going to happen. So this is a story of uh Gro 4 and uh you know Tony uh share
631. 2:10:29some uh insight with the audience. Yeah. Um so yeah let's just talk about how smart graph 4 is. So I guess um we
632. 2:10:38can start discussing this benchmark called humanities last exam and this this benchmark is a very very challenging benchmark. Every single
633. 2:10:47problem is curated by subject matter experts. Um it's in total 2500 problems and it consists of many different
634. 2:10:56subjects mathematics natural sciences uh engineering and also auto humanity subjects. So um essentially when when it
635. 2:11:06was first released actually like earlier this year uh most of the models out there can only get singledigit accuracy on this benchmark.
636. 2:11:16Yeah. So we can we can look at some of th those examples. Um you know uh so there there is this mathematical problem
637. 2:11:23which is about natural transformations in category theory and there's this organic chemistry problem that talks
638. 2:11:30about uh electro uh cyclic reactions and also there's this linguistic problem that tries to ask you about
639. 2:11:38distinguishing between closed and open syllabus uh from a hub Hebrew source text. So you can see also it's a very
640. 2:11:48wide range of problems and every single problem is PhD or even advanced research level problems. Yeah. I mean these there are no humans
641. 2:11:58that can actually answer these can get a good score. I mean if you actually say like any given human um what like what's
642. 2:12:05the best that any human could score? I mean I'd say maybe 5% optimistically. Yeah. Um, so this this is much harder than
643. 2:12:16than what any any human can do. It's it's incredibly difficult and you can see from the the types of questions like
644. 2:12:21you might be incredible in linguistics or mathematics or chemistry or physics or any one of a number of subjects, but
645. 2:12:28you're not going to be um at a postgrad level in everything. And Grock 4 is a postgrad level in everything. Yeah.
646. 2:12:36Like it's it's just some of these things are just worth repeating. like grock boore is postgraduate like PhD level in
647. 2:12:45everything better than PhD but like most PhDs would fail so it's better said I mean at least with respect to academic
648. 2:12:53questions it I want to just emphasize this point with respect to academic questions gro is better than PhD level
649. 2:13:00in every subject no exceptions wow um now this doesn't mean that it's it you know times it may lack common sense
650. 2:13:10and it has not yet invented new technologies or discovered new physics but that is just a matter of time.
651. 2:13:19Um if it I I I think it may discover new technologies uh as soon as later this year. Um and I I would be shocked if it has not done so
652. 2:13:29next year. So I would expect Grock to yeah literally discover new new technologies that are actually useful no
653. 2:13:36later than next year and maybe end of this year. Um, and it might discover new physics next year. And within two years,
654. 2:13:43I'd say almost certainly like, so just let that sink in. Within two years, new physics. Yeah. So, yeah. Um, how Okay. So, I guess we
655. 2:14:00can talk about the the what what's behind the scene of 4. As Jimmy mentioned, uh, we actually throwing a
656. 2:14:07lot of compute into this training. You know, when it started, it's only also a single digit. Sorry. Um, the previous
657. 2:14:14slide, sorry. Yeah, it's only a single digit uh number. But as you start putting in more and more training compute, it started to
658. 2:14:25gradually become smarter and smarter and eventually solved a quarter of the HL problems. And this is without any tools.
659. 2:14:35The next thing we did was to adding a tools capabilities to the model. And unlike GR 3, I think you know G3
660. 2:14:44actually is able to use C as well. But here we actually make it more native in a sense that we put the tools into
661. 2:14:50training. Uh G3 was only relying on generalization. Here we actually put the tools into Okay. When they say they put the tools
662. 2:14:58into training, what they mean is if it runs into a problem where it needs to like write Python code to solve a
663. 2:15:04problem that's now it's been trained on that. So it's probably going to be very very good at the G3 reasoning model uh but without
664. 2:15:14any specific training but we only ask it to use those tools. So compared to this it was much weaker in terms of its tool capabilities
665. 2:15:23and unreliable and unreliable. Yes. Yes. And and to be clear like these are still I'd say fairly this is still fairly primitive
666. 2:15:30tool use. If you compare it to say the tools that are used at Tesla or SpaceX uh where you're using um you know finite
667. 2:15:38element analysis and computational flow dynamics and and you're you're able to run uh or say like Tesla does like crash
668. 2:15:48simulations where the simulations are so close to reality that if the test doesn't match the simulation you assume
669. 2:15:54that the test article is wrong. That's how good the simulations are. So, Grock is not currently using any of the tools
670. 2:16:00that the really powerful tools that a company would use, but but that is something that we will provide it with
671. 2:16:06later this year. So, we'll have the tools that that a company has. Um, sophisticated tools later this year,
672. 2:16:14physics simulator. Um, ultimately the the thing that will make the biggest difference is being able to interact
673. 2:16:20with the real world via humanoid robots. So you combine sort of Grock with with Optimus and it can actually interact with the real world.
674. 2:16:28So is Grock 4 on the spectrum? I think definitely formulate hypothesis and then confirm if that hypothesis is is true or not. Um
675. 2:16:39so we're really, you know, think about like where we are today. We're we're at the beginning of an immense intelligence
676. 2:16:47explosion. We're in we're in the intelligence big bang right now. Yep. Um, and the mo we're at the most interesting
677. 2:16:55time to be alive of any time in history. Yeah, agree. Yeah. Now, that said, we need to make sure that the AI is um a good AI.
678. 2:17:08Uh, good Brock. Um, and the the thing that I think is most important for AI safety, at least my biological neural
679. 2:17:16net tells me the most important thing for AI is to be maximally truth seeeking. So this is this is a very
680. 2:17:23fundamental um like you can think of AI as this super genius child that ultimately will outsmart you but you can
681. 2:17:32still in you can instill the right values um and encourage it to be sort of you know uh truthful uh I don't know honorable you know good
682. 2:17:48good things like the values you want to instill in in a child that that that grow would grow ultimately grow up to be incredibly powerful.
683. 2:18:00Um yeah. So yeah, so these this is really I say we say we say tools these are say still primitive tools not the kind of tools
684. 2:18:13that um that serious commercial companies use but we will provide it with those tools and uh I think it will
685. 2:18:21be able to solve with those tools real world technology problems. In fact I'm I'm certain of it. It's just a question of how long it takes.
686. 2:18:28Yes. Yes, exactly. Um, so is it just compute all you need, Tony? Sorry. Is it just compute all you need at this point?
687. 2:18:38Well, you you need compute plus plus uh the right tools. Mhm. And um and then ultimately to be able to interact with physical world. Yes. Um
688. 2:18:48all right, let's talk about what it is now and when we can play with it. I mean we'll effectively have an economy that is
689. 2:18:56well ultimately an economy that is thousands of times bigger than our current economy or maybe millions of
690. 2:19:02times. I mean if you if you think of civilization as percentage completion of the kadesv scale where kadeshv one is
691. 2:19:12using all the energy output of a planet and kv two is using all the energy output of a sun and three is all the
692. 2:19:20energy output of a galaxy. We're we're only in my opinion probably like close closer to 1% of kartev one um than we are to 10%. So
693. 2:19:33like maybe a point or one one or two percent of kadeshv one. So um we we will get to um most of the way like 80 90% kesh of
694. 2:19:45one and then hopefully movies self annihilate the waffles like it's the the actual notion of of a human economy assuming civilization
695. 2:19:57continues to progress will seem very quaint. Um in in retrospect, um it will it will seem like uh sort of cavemen throwing sticks into a
696. 2:20:08fire uh level of economy um compared to what the future will hold. Um I mean it's very exciting. I mean, I
697. 2:20:17I've been at times kind of worried about like, well, you know, is this this this seems like it's it's somewhat unnerving to have
698. 2:20:28intelligence created that is far greater than our own. Um, and will this be bad or good for humanity? Um,
699. 2:20:38it's like I I I think it'll be good. Most likely it'll be good. Jesus Christ. My god. Yeah, dude. Yeah, but I somewhat reconciled myself.
700. 2:20:50Sober up fact that even if even if it wasn't going to be good, I'd at least like to be alive to see it happen. So yeah.
701. 2:21:03So actually one one Yeah. Yeah. I think one one technical problem that we still need to solve besides just compute is
702. 2:21:13how do we unblock the data uh data um bottleneck u because um when we try to scale up the RL uh in this case we did
703. 2:21:22invent a lot of new techniques innovations to allow us to figure out how to find a lot of a lot of challenging RL problems to work on. It's
704. 2:21:32not just a problem itself needs to be challenging but also it needs to be um you also need to have like a reliable uh
705. 2:21:40signal to tell the model you did it wrong you did it right. This is the sort of the principle of reinforcement
706. 2:21:45learning and as the models get smarter and smarter the number of cool problem or challenging problems will be lesser and less. Yeah.
707. 2:21:54So it's going to be a new type of challenge that we need to surpass besides just compute. Yeah. Yeah. And we actually are running out of
708. 2:22:02of of actual test questions to ask. Uh so there's like even ridiculously questions that are ridiculously hard if
709. 2:22:09not essentially impossible for humans that are written down questions um are uh becoming swiftly becoming trivial for for AI.
710. 2:22:19Um so then there's um but you know the the one thing that is an excellent judge of things is reality. So because physics is the law
711. 2:22:31ultimately everything else is a recommendation. You can't break physics. Um so the ultimate test I think for
712. 2:22:37whether an AI is um the the ultimate reasoning test is reality. Yes. So you invent a new technology like say
713. 2:22:45improve the design of a car or a rocket or um create a new medication. Uh that's and and and does it work? Yeah.
714. 2:22:54Um does does the rocket get to orbit? Does the does the car drive? Does the medicine work? Whatever the case may be.
715. 2:23:02Um, reality is the ultimate judge here. Um, so it's it's going to be a reinforcement learning closing loop around reality.
716. 2:23:13We asked the question how do we even go further? So um actually we are thinking about now with single agent we are able
717. 2:23:22to solve 40% of a problem. What if we have multiple agents running at the same time? So this is what's
718. 2:23:29called test and compute. And as we scale up the test and compute actually we are able to solve almost more than 50% of
719. 2:23:37the uh taxon subset of the h problems. So it's a remarkable achievement I think. Yeah. Yeah. This is this is this is insanely
720. 2:23:46difficult. These are it's it's so what we're saying is like a majority of the of the of the text based um of
721. 2:23:54humanities you know scarily named humanities last exam um GR 4 can solve and you and you can try it out for
722. 2:24:00yourself um and the with the GR 4 heavy what what it does is it spawns multiple agents in parallel and all of those
723. 2:24:08agents do do work independently and then they compare their work and they they decide which one like it's a study group.
724. 2:24:18Um, and it's not as simple as a majority vote because often only one of the agents actually figures out the trick or
725. 2:24:26figures out the solution. Um, and and and but once they share the the trick or or or figure out what what the real
726. 2:24:33nature of the problem is, they share that uh solution with the other agents and then they compare they essentially
727. 2:24:39compare notes and then and and then yield yield an answer. So that's that's the the heavy part of Grful is is where
728. 2:24:48we you scale up the test time compute by roughly an order of magnitude uh have multiple agents uh tackle the task and then they compare
729. 2:24:57their work and they they put forward what they think is the best result. Okay. Yeah. So we will introduce graph 4 and
730. 2:25:06graph for heavy. Sorry can you click the next four heavy? Yes. So yeah, so basically Gu 4 is a single version, a single agent version
731. 2:25:16and graph for heavy is the multi- aent version. So let's take a look how they actually do on those exam problems and also some
732. 2:25:26real life problems. Yeah. So we're going to start out here and we're actually going to look at one of
733. 2:25:30those HLE problems. This is uh actually one of the easier math ones. Uh I don't really understand it very well. I'm not
734. 2:25:37that smart. But I can launch this job here and we can actually see how it's going to go through and start to think
735. 2:25:43about this problem. Uh while we're doing that, I also want to show a little bit more about like what this model can do
736. 2:25:48and launch a uh Gro 4 heavy as well. So everyone knows Poly Market. Um it's extremely interesting. It's the you know
737. 2:25:56seek curve of truth. It aligns with what reality is most of the time. And with Grock, what we're actually looking at is
738. 2:26:02being able to see how we can try to take these markets and see if we can predict the the future as well. So, as we're
739. 2:26:10letting this run, we'll see how uh Gro for Heavy goes about uh predicting the, you know, the World Series odds for like
740. 2:26:17the current teams in the MLB. And while we're waiting for these to process, we're going to pass it over to Eric and
741. 2:26:22he's going to show you an example of his. Yeah. So uh I guess one of the coolest things about uh Grapore is its ability
742. 2:26:35to understand the world and to solve hard problems by leveraging tools like Tony discussed. And I think one kind of
743. 2:26:43cool example of this we asked it to generate a visualization of two black holes colliding. Um and of course you
744. 2:26:52know it took some there are some liberties. It's in my case actually pretty clear in its thinking trace about
745. 2:26:58what these liberties are. Uh for example, in order for it to actually be visible, you need to really exaggerate
746. 2:27:03the the scale of the you know the uh the uh the waves. And yeah, so here's like you know this kind of inaction. Um it
747. 2:27:17exaggerates the scale in like multiple ways. uh it drops off a bit less in terms of amplitude um over distance and
748. 2:27:26uh but yeah we can kind of see uh the basic effects that you know are actually like you know correct. It starts with
749. 2:27:34the inspiral it merges uh and then you have um the ring down and like this is basically um largely correct um yeah uh
750. 2:27:47modulo some of the simplications I need to do um you know it it's actually quite explicit about this you know it uses
751. 2:27:53like post postneonian approximations instead of actually like computing the general relativistic effects at like uh
752. 2:28:01near the center of the black hole which is you know incorrect um and you know will lead to you know some incorrect
753. 2:28:08results but the overall you know visualization is uh yeah is basically there um and you can actually look at the kinds of resources that it
754. 2:28:17references. So here it um it actually you know it obviously uses search it gathers results from a bunch of links
755. 2:28:25but also reads through a undergraduate text in analytical analytic gravitational wave models. Uh it um yeah
756. 2:28:36it um reasons quite a bit about u uh the actual constants that it should use for a realistic simulation. uh it references
757. 2:28:47uh I guess existing real world data. Um and yeah I it yeah it's a it's a pretty good model. Yeah but like actually going forward we can
758. 2:28:59we can we can play we can give it the same model that physicists use. So it can run the the same level of compute
759. 2:29:06that uh so leading physics researchers are using and and give you a physics accurate black hole simulation. Exactly.
760. 2:29:15just right now is running in your browser. So yeah, this is just running in your browser. Exactly. Pretty simple.
761. 2:29:21So swapping back real quick here, we can actually take a look. The math problem is finished. Uh the model was able to uh
762. 2:29:28let's look at its thinking trace here. So you can see how it went through the problem. I'll be honest with you guys, I
763. 2:29:34really don't quite fully understand the math, but what I do know is that ahead of time um and it did come to the the correct
764. 2:29:42answer here in the final part here. Good lord. We can also come in and actually take a look here at our uh our
765. 2:29:48World Series prediction. Um, and it's still thinking through on this one, but we can actually try some other stuff as
766. 2:29:54well. So, we can actually like try some of the X integrations that we did. So, we worked very heavily on working with
767. 2:30:00uh all of our X tools and building out a really great X experience. So, we can actually ask, you know, the model, you
768. 2:30:07know, find me the XAI employee that has the weirdest profile photo. Um, so that's going to go off and start with
769. 2:30:12that. And then we can actually try out you know let's uh create a timeline based on expost uh detailing the you
770. 2:30:19know changes in the scores over time and we can see you know all the conversation that was taking place as well. So we can
771. 2:30:26see who were the you know announcing scores and like what was the reactions at those times as well. Um so we'll let that go through here and
772. 2:30:34process. And if we go back to uh uh this was the uh Greg Yang photo here. So, uh, if we scroll through here, whoops. So,
773. 2:30:44Greg Yang, of course, who has his favorite, uh, photograph that he has on his account. Um, that's actually not how
774. 2:30:50he looks like in real life, by the way. Um, just to get aware, but is quite funny, but it had to understand that question. Yeah.
775. 2:30:57Which is that that's the wild part. It's like it understands what is a weird photo. What is a weird photo? Yeah.
776. 2:31:03What is a less or more weird photo? It goes through it has to find all the team members. It has to figure out who
777. 2:31:09we all are, right? you know, searches without access to the internal XAI personnel logs. It's literally looking at the just at the internet.
778. 2:31:16Exactly. So, you could say like the weirdest of any company. Yeah. To be clear. Exactly. And uh we can also take a look
779. 2:31:24here at the uh question here for the humanities last exam. So, it is still researching all of the historical
780. 2:31:29scores. Um but it will have that final answer here soon. But we can while it's finishing up, we can take a look at one
781. 2:31:35of the ones that we set up here a second ago. And we can see like you know it defines the date that like Dan Hendricks
782. 2:31:41had initially announced it. We can go through we can see you know uh OpenAI announcing their score back in uh
783. 2:31:47February and we can see you know as progress happens with like Gemini. We can see like Kimmy uh and we can also
784. 2:31:54even see you know the leaked benchmarks of uh what people are saying is you know if it's right it's going to be pretty
785. 2:31:59impressive. So pretty cool. Um so yeah I'm looking forward to seeing how everybody uses these tools and gets the
786. 2:32:06most value out of them. Um but uh yeah, it's been great. Yeah. And we're gonna close the loop around usefulness as well. So it's like
787. 2:32:13it's not just books smart, but actually practically smart. Exactly. And we can go back to the uh the slides
788. 2:32:28here. Yeah. So cool. Um so the so we actually evaluate also on the multimodel subset. So on the full set uh this is the number on the h
789. 2:32:40exam. Uh it you can see there's a little dip on the numbers. Uh this is actually something we're improving on which is
790. 2:32:47the multimodel uh understanding capabilities. But I do believe uh in a very short time we're able to really
791. 2:32:55improve and got much higher numbers on this um even higher numbers on this benchmark. Yeah, this is the we we saw like you see
792. 2:33:04like what what is the biggest weakness of Grock currently is that it's it's sort of partially blind. It can't it's it's understanding
793. 2:33:13um obviously in its image. Joe, I always think I really want to watch these and then I realize they're super [ __ ]
794. 2:33:19boring. Yeah, the the like I it is this is unwatchable, but you know, um it looks quite impressive. Um I mean the
795. 2:33:29problem is you've got people that are so smart but so myopic in what they're focusing on, they can't imagine how
796. 2:33:36other people are going to use it. So we get these demos that are just, you know, clever or relevant to them. uh here with
797. 2:33:43a heavy and we can see uh here we can see all the tools and the process it used to actually uh go through and find
798. 2:33:50the right answer. So it browsed a lot of odd sites. It calculated its own odds comparing to the market the market to
799. 2:33:57find its own alpha and edge. It walks you through the entire process here and it calculates the odds of the winner
800. 2:34:04being like the the Dodgers uh and it gives them a 21.6% chance of uh winning uh this year. So it took approx approximately four and
801. 2:34:15a half minutes to compute. Sure. Yeah. That it when can we play with it? You can also look at all the uh the
802. 2:34:27other benchmarks besides HE. Um, as it turned out, GR 4 excelled on all the reasoning benchmarks that people usually
803. 2:34:36test on. Um, including GBQA, which is a PhD level, uh, problem sets. Uh, that's easier compared to HRE. Um, on Amy 25,
804. 2:34:47America invitation mathematics exam, we with graph for heavy, we actually got a perfect score. Wow. Um, also on some of the coding benchmark
805. 2:34:57called live coding bunch. um and also on uh HMMT, Harvard math uh MIT uh exam and also USMO. Uh you can see actually um
806. 2:35:08for on all of those benchmarks we often have a very large leap against the second best uh model out there.
807. 2:35:17Yeah, it's I mean really we're going to get to the point where uh it's going to get every answer right in every exam. um
808. 2:35:24and where it doesn't get an answer right, it's going to tell you what's wrong with the question. Or if the
809. 2:35:28question is ambiguous, disambiguate the question into answers A, B, and C and tell you what what answers A, B, and C
810. 2:35:36would be with a disambiguated question. Um so the only real test then will be reality. Uh can I make useful
811. 2:35:42technologies, discover um new science? That'll actually be the the only thing left because human tests will simply not
812. 2:35:51be um meaningful. That's crazy. You can make all That's crazy. We're within a year. Give it a We're within a year of and and OpenAI is
813. 2:36:01about to launch something probably similar. This is this is why the frontier model companies have stopped talking about AGI and moved on to
814. 2:36:09talking about ASI that there these things are starting to accelerate and Grock heavy tiers we're introducing
815. 2:36:17where you're able to access to both Gro 4 and Grock 4 heavy uh where you're actually going to be the taskmaster of
816. 2:36:23bunch of little Grock research agent to help you you know become smarter through all the little research
817. 2:36:29of you know going through mundane tasks uh and it's available right now. Uh so yeah, so um we we did limit usage during
818. 2:36:40the the demo. So we didn't it didn't break the demo because all these this all this stuff is happening live. So
819. 2:36:44it's not there's no nothing canned about the any of the tests that we're doing. Um so uh after the after the uh demo is
820. 2:36:51done, we'll we'll allow we'll enable more subscribers for Super Grock. So if you can't subscribe right now, just try
821. 2:36:58in half an hour. It should work. Um, so, uh, and now let's, let's get into voice. Great, so many of you have been enjoying our
822. 2:37:10voice mode and we've been working hard to improve the experience over the past couple months. Um, we have cut latency
823. 2:37:17in half to make it much snappier. And today, we're excited to announce a set of new voices that have exceptional
824. 2:37:24naturalenness and procity. You might have noticed the movie trailer voice that opened up the live stream. That is
825. 2:37:31one of our new voices, S, who's got that epically deep tone. And we're also excited to introduce Eve, a beautiful
826. 2:37:40British voice who's capable of rich emotions. Come on, would you like to introduce Eve? Absolutely. Yeah, let's get into the demo.
827. 2:37:51Hey, Eve. Millions of people are watching us right now. Are you excited? Oh, I'm thrilled. Really, it's like
828. 2:37:58being on stage at the Old Vic with a sea of faces out there. Just imagine all those eyes and ears tuned in. It's a
829. 2:38:05proper buzz, isn't it? What do you reckon we should chat about to keep this lot entertained? You know, we have a lot to talk about,
830. 2:38:11but I'll be honest. I'm feeling a little nervous. Can you whisper something soothing to calm me down? Take a deep breath, love.
831. 2:38:20We've got this. It's just you and me having a quiet chant. Like we're tucked away in a cozy corner of a Yorkshire
832. 2:38:28pub. The world's just a murmur out there. Feel that calm wash over you. Let's keep talking. Nice and easy. What's on your mind?
833. 2:38:39Starf fighter. Oh, good lord. Fake girlfriends are going to go bananas. Can you give us an opera on Diet Coke?
834. 2:38:48Oh, an opera on Diet Coke. That's a riot. Picture this. A shimmering silver cannon glistening under stage lights.
835. 2:38:59ASMR diet coke. Thou elixer divine with bubbles that dance in a sparkling line. Thy crisp cool kiss on lips so fine. Oh
836. 2:39:17forever. How's that for a mad little area? Want me to belt out another verse or switch up the tune? Okay, let's try switching it up and just
837. 2:39:27start singing. Right, then let's have a go. Oh, silver can you gleam in the light diet coke my heart's delight.
838. 2:39:38Okay, I won't touch it. It's very very poetic. Um yeah, thank you Eve. Feeling quite nice now. Cool. Um so that's Eve. And um so
839. 2:39:48something else we do here is is a little test each day to check on the natural flow of our conversations against some
840. 2:39:57of our competitors. So we figured why not try that live on the stream today? Should I start? Hey, I'm going to say a number and can
841. 2:40:11you say the same number back to me? One. One. Got it. This is like a fun little game of echo. What's next? Two. Oh, you said two.
842. 2:40:25Three. Four. Nice and steady. What's next? Um. Uh. Five. Five. Nailed it. This is pretty fun,
843. 2:40:39actually. Feel free to keep going or switch it up if you'd like. Hey, Grock. Hey, Grock. Hello there. How can I help you today?
844. 2:40:54We're going to do a little test. Can you repeat after me? One. One. All right. What's next? Two. Two. What's on your mind? Three.
845. 2:41:06Three. Need anything else? Four. Four. How can I assist you? Five. Five. What's next? So, as you can see, um, Groc was, uh,
846. 2:41:19snappier. It didn't interrupt me. And the procity is we made different design choices. I think we shooting for
847. 2:41:26something more calm, smooth, more natural versus something that's more poppy or artificial. So, we'll keep improving these ones.
848. 2:41:32Thanks, guys. Yeah. Yep. So since the launch of the voice model uh we actually see the 2x faster end to end latency uh in the last eight
849. 2:41:48weeks five different voices and also 10x the active user. So Grock voice is taking off now. All right I'm gonna mute
850. 2:41:56that. I'm going to go to bed. I can't take it anymore. This is killing me. Um we we'll figure this out tomorrow. Looks
851. 2:42:03like it's really [ __ ] smart. I assume uh this is AGI or close to it. There's the ARC AGI test. Gro 4 is above
852. 2:42:13everything else by a lot. Um I would be dramatically surprised if OpenAI doesn't come out with something that tops this. Um I don't
853. 2:42:24think they're going to sit sit for for long at this at this stage. But anyway, all right. I'm done. Yeah, we need Steve
854. 2:42:31Jobs back, man. They got to figure out how to do some presentations here. This shit's this shit's awful. Um anyway, um
855. 2:42:39it does look impressive. I'm exhausted. Peace out. Uh tomorrow is what's today? Wednesday. Thursday. I'll see you at
856. 2:42:47eight o'clock. We'll be back. We'll figure out what's worth talking about with all this stuff. All right. I will see you tomorrow. Bye.