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

Description

Chapters

Transcript

0:01 Are you ready? Are you ready? Yes, I'm
0:04 ready.
0:06 Are you ready? Yes, I'm ready to fall in
0:11 love.
0:12 To fall in love
0:16 right now.
0:19 [Music]
0:32 10,000 words swarm around my head. A
0:36 million more in books written beneath my
0:39 bed.
0:41 [Music]
0:44 And I wrote or read them all when
0:47 searching in the swamp. Still can't find
0:50 how to hold my hands.
0:53 [Applause]
0:56 And I know you need me in the next room
1:00 over. I'm stuck in here home paralyzed.
1:05 [Music]
1:08 Four months I got myself ins.
1:16 [Music]
1:24 I'm no different. like talk on things we
1:27 don't know about.
1:32 Well, ain't it like most people know
1:35 people like things we don't know about.
1:40 [Music]
1:40 [Applause]
1:43 [Music]
1:50 [Music]
1:58 to say
2:00 that recognize me these shoes and disab
2:06 [Music]
2:18 [Music]
2:21 [Applause]
2:25 [Music]
3:00 It's not simple to say
3:04 that most days
3:07 I don't recognize me that these shoes
3:10 I'm the apron.
3:12 That place and its patrons.
3:15 I've taken more than I gave them.
3:20 It's not easy to know.
3:23 I'm not everything
3:26 like I used to be. Although it's true. I
3:30 was never attention sweet center. I
3:34 still remember that girl.
3:37 She's imperfect,
3:39 but she tries.
3:41 She is good, but she lies.
3:46 She is hard on herself.
3:50 She is broken, but won't ask for help.
3:54 She is messy, but she's kind.
3:58 She's lonely.
4:01 Most of the time she's all of this mixed
4:04 up and baked in a beautiful pie.
4:09 She's gone, but she used to be mine.
4:14 [Music]
4:20 [Music]
4:33 Okay.
4:35 Um, it is Grock Fortnite apparently.
4:38 Apparently it's Grock Fortnite. I
4:40 haven't seen many posts Elon's going on
4:42 about some dude named Jack J Jack
4:46 Epstein. Jack Jack
4:49 Eisenstein. I don't know. Anyway,
4:52 there's some guy Elon Musk talking about
4:54 him where he's talking about Grock 4.
4:57 So, it's possible we get some sort of
4:59 list about Jack Eisenstein
5:02 or we get Grock 4 tonight. One of those.
5:06 All right. Um
5:09 or UFO disclosure. Could be UFO disclos
5:13 or we'll surf. We'll we'll use AI. We'll
5:17 we'll act like good AI thieves
5:22 and go steal everyone's content.
5:28 Actually, some guy some guy in my
5:30 trolls,
5:32 he thought he he thought he cornered me.
5:37 He thought he cornered me. He said, he
5:41 said, "So, oh, so I guess if you,"
5:44 because I was talking about, you know,
5:45 prompting. Prompting is a craft
5:47 basically. And he said, "Oh, so if you
5:49 go to a vending machine and you put in a
5:52 quarter and you pull push the button for
5:54 a Snickers bar and a Snickers Snickers
5:56 bar come out came out, did you make that
5:58 Snickers bar?" And and like when it when
6:01 I first read it, I was like, "Oh, he got
6:03 me." And then I was like, "Wait a
6:04 minute. No, that's not how this [ __ ]
6:06 works." It would be more like this. I
6:08 put a quarter in the machine and I say,
6:10 "Hey, I want to design a new candy bar
6:12 and I want it to have almonds and I want
6:14 it to have peanut butter and I want it
6:16 to have this and I want it to have that
6:17 and then I want it to have chocolate on
6:19 the outside but like peanuts like a
6:21 payday." And then I want it to actually
6:24 come up with a name of the new candy bar
6:27 for me and come up with the packaging
6:28 design and then I push the button and
6:31 that comes out. That's what AI is like.
6:36 Son of a [ __ ] I was like, "No, it's AI
6:40 is not full of candy bars that are
6:41 already made.
6:44 It broke it broke it down into the
6:46 ingredients."
6:50 So, in that case, yes, I did make a
6:52 candy bar. Am I am I a chocolier? No.
6:57 [Music]
7:29 Thank you for sharing the live,
7:30 Danielle. That's good. Um, okay. So, um,
7:35 what we're going to do here tonight is
7:37 I'll probably mess around on the guitar.
7:40 If you How many folks we got here? Oh,
7:42 we're we're relatively small at this
7:44 point. Um, go tell your friends
7:47 that there's some idiot live on Tik Tok
7:50 and YouTube and Twitter and LinkedIn.
7:53 Um, who's going to be live streaming the
7:55 Gro 4 announcement
7:59 if it happens.
8:05 Comet browser. Yes, the comment browser.
8:08 You can skip the weight list if you pay
8:10 200 bucks a month to Perplexity.
8:14 77.
8:16 What's 77 today? 79.
8:20 Did I put 77 somewhere?
8:22 777.
8:24 777 film.
8:27 [Music]
8:29 Oh,
8:32 [Music]
8:37 isn't that It been a holiday night,
8:40 wasn't it? That
8:42 was it. Just an open an open that. No, I
8:46 think it's higher than that.
8:51 [Music]
8:52 Um
9:00 the title of the YouTube
9:02 Oh, I did 77. Oh, I'm such an idiot.
9:07 Um
9:08 I don't think I can edit that. Can I?
9:11 Oh, maybe I can. Ha, I think I can.
9:18 7 9
9:21 and then if I click in here,
9:24 it fixed it there
9:26 and it fixed it there. Save changes. Ha,
9:29 wait. LinkedIn doesn't support updating
9:32 streams yet. Oh, LinkedIn, you're an
9:34 [ __ ] Really, Microsoft?
9:45 I don't if I close this. I think it's
9:47 still going to be 77. Whatever. All
9:49 right. Whatever. It's It's 77. Hey, I'm
9:52 part of time. Shane Saw in the house.
9:55 Fixed. Good catch. I didn't notice it.
9:57 All right. Fantastic. What announcement
10:00 is today? I'll tell you what side
10:02 hustle, Mimi. Today Today we got there's
10:05 a there's a uh there's a social media
10:09 service. It's called Twitter. Used to be
10:11 called Twitter. Now it's called X
10:14 because the guy that bought it, he likes
10:16 the letter X. He likes it so much he
10:18 named his son X and he he likes X
10:20 airplplanes and he he got X.AI. He's
10:24 Anyway, he likes X. And so they got this
10:27 thing called Grock and Gro is a large
10:30 language model. It's very much like Chat
10:32 GPT except it's not good and no one uses
10:35 it. Oh, but it does make like racist
10:39 posts automatically. So that's cool.
10:41 Anyway, so that is at level at at at
10:46 level three. It's version three. It they
10:49 had it they had one out Grock came out
10:52 at first and then it was really bad and
10:55 then they're like, "Hey, we got a better
10:56 one." And it was called two and then
10:58 they're like, "No, this one's super
11:00 good." And it was three and then it was
11:02 it was okay. And it made pictures. It
11:05 made pictures and then the now today
11:10 this it was going to be 3.5 but then the
11:12 guy that called Twitter X said we're not
11:15 going to call it 3.5 we're going to be
11:16 calling four and so they're going to
11:19 call it 040 because chat GPT's got 04
11:25 and then they got 40 and but one's an O
11:28 and one's a zero and it but it doesn't
11:29 matter because now we're going to have
11:31 Grock 4 and it but it'll be Grock 4040
11:35 This is going to be so good.
11:38 It's going to be so good.
11:47 This lame character brought to you by
11:50 Saturday Night Live minor leagues.
11:57 Everything I said in that stupid voice
11:59 was mostly factual.
12:06 Oh my god,
12:09 I'm like the Lauren Michael of the AI
12:12 learning lab.
12:17 [Music]
12:32 [Applause]
12:33 [Music]
12:49 Um,
12:57 I feel like an idiot tonight. An idiot
13:00 tonight. An idiot tonight. I think this
13:04 is going to be a stupid show tonight.
13:05 Oh, I got recognized today. This is the
13:08 first time in a long time. Maybe if ever
13:12 I used to get recognized when I was
13:14 doing agency.com when I spoke a lot, but
13:17 um I had a meeting. We had a a pitch
13:20 with an agency today at StoryVine and uh
13:22 I got on the thing. So, I don't know if
13:24 John's here. if you're here, John. Hey,
13:26 man. But, uh, but this the the woman
13:29 that ran it, Stacy, said, "Oh, John's
13:32 coming." And when I said to John, "We
13:34 were meeting with Kyle Shannon," he
13:35 goes, "Oh, Kyle Shannon, the AI guy.
13:39 He's been watching the live." So, I
13:42 turned a few shades of red on that one.
13:43 But, you know, what are you gonna do?
13:45 Um, just released the Kyle 4 update.
13:49 Yeah, we I'm I'm still on on beta 0.8.7.
13:54 Um so which is why you know this channel
13:58 lacks credentials and
14:01 qualifications and
14:04 actual
14:06 uh knowledge.
14:16 Yeah. Qualification. Yes. Yeah. Yeah.
14:19 Yeah. Yeah. See
14:20 [Music]
14:43 looking
14:46 [Music]
14:54 [Music]
15:00 [Music]
15:10 for.
15:16 Woohoo!
15:20 Woohoo!
15:24 [Music]
15:41 in a westerly
15:43 direction.
15:47 Oh no.
15:49 [Music]
15:56 [Music]
16:05 [Music]
16:18 [Music]
16:42 Oh, all right. How was the podcast?
16:45 Podcast was good. We had on Tyler Fisk
16:49 on the show today. Was really good. I
16:51 thought you were Kyle 6.0. No, that was
16:53 the rumor. That was uh Kyle 6.0 has been
16:56 announced but never shipped. That's the
16:58 That's the challenge there. Um
17:02 had me at the guy that called Twitter X.
17:08 Okay. Um
17:10 let's see what what's happening.
17:15 Well, at nine o'clock, so in 40 minutes,
17:18 43 minutes is is supposedly the Grock
17:23 for
17:25 announcement release. Who knows?
17:36 Mr. Strawberry is putting out [ __ ]
17:38 posts. We're less than an hour away from
17:40 super intelligence, which that's
17:42 [ __ ] That's not true.
17:52 Um,
17:57 [Music]
18:00 Elon's greatest enemy is Elon himself.
18:03 That's pretty funny.
18:05 What's the first thing you're going to
18:06 ask Grock for? That's a good question.
18:09 Let's go see what a thousand people
18:11 said.
18:13 How many cubes are in my trailer?
18:16 What's your opinion of Hitler?
18:25 Oh, internet trolls are the best.
18:29 They really are. Oh my god. Oh my god.
18:34 Fantastic.
18:36 Fantastic. Yeah. Yeah. Yeah. Yeah. Let's
18:40 see. We got anything here?
18:44 Um, uh, X.AI lost their CEO today. So,
18:48 she was there for two years and they're
18:51 going to launch Grock 4 today and their
18:53 CEO quit.
18:58 So, I don't know what that's about other
19:00 than maybe working for Mr. Musk is a is
19:04 a challenge.
19:08 [Music]
19:14 Johnny IV and Sam,
19:18 the IO products deal has officially
19:20 closed and we're thrilled to welcome the
19:23 team to Open AI.
19:26 Oh, okay.
19:29 So, I guess they bought them. They were
19:32 partnering before, now they bought them.
19:34 Oh, there's Ailia. Nice.
19:42 Yeah, let's go look at these.
19:43 This is from Ethan Malik using Gro 3. I
19:48 think Mr. It's got his his autotapper
19:51 going, by the way.
19:55 That image is so cringe. It's totally
19:57 cringe.
20:00 Um, let me change my screen sharing.
20:05 [Music]
20:26 actually I think I am over Hamlet.
20:29 [Music]
20:30 Sorry, Caesar.
20:36 [Music]
20:41 You know, I think we should see other
20:43 people.
20:47 I am feeling a little better actually.
20:53 Actually,
20:58 I think I am over Hamlet.
21:01 [Music]
21:02 Mary Mary got the get smart theme. You
21:04 are correct.
21:20 Oh, what did I see today? Um, there was
21:23 a
21:25 the [ __ ] it called?
21:27 There's a new video model out. It's
21:29 actually been out for a while. Moon
21:31 Valley. Moon Valley. If you go to
21:34 moonval.com,
21:36 I got to change my tabs again.
21:38 Moonval.com.
21:43 Um, unfortunately, you have to pay for
21:46 it.
21:51 Uh
21:53 oh, the sam the salmon Johnny Johnny
21:56 Ives image is so cringe. Yeah, that's
21:58 pretty cringe. It is pretty cringe. Um
22:02 [Music]
22:15 I didn't know he beatbox. Does he
22:17 beatbox too? My god, that man's got so
22:19 many talents. He can hum
22:23 four bars of of nearly seven television
22:26 shows from the 1970s and 1980s. Um, he's
22:30 he's a pretty good guitar player. He's
22:33 learned it seems like almost 12 songs
22:37 in in 60 years.
22:40 Pretty solid. Like it's almost like a
22:43 song every four years he's learned on
22:45 average, you know?
22:46 Um, and then he's just, you know, he can
22:49 he did like I was on one of his lives
22:51 the other night and um I mean he was
22:54 talking about AI. I didn't understand
22:55 what he was talking about but it doesn't
22:57 matter. But I was on there and he
22:59 started beatboxing. He was like
23:04 and it was it was about as good as
23:06 anything else he does and I just didn't
23:09 realize he had that talent too. I'm like
23:11 oh my god what is he a renaissance man?
23:14 It's just like crazy. So anyway, that's
23:16 that's what I do for like two hours a
23:18 night is I'll watch that and then
23:19 sometimes what I'll do is I'll re rewind
23:21 it, watch it again or if I miss the one
23:23 the night before, I'll just watch him
23:25 until 3:00 or 4 in the morning and I'm
23:28 sorry, what have have I learned anything
23:31 from from the AI learning lab? Um
23:35 well uh I it depends on what the
23:40 definition of learn is. Yeah. So so
23:42 sure. I mean I'm aware of more things
23:46 than I was. Yeah. Yeah. No, I've I think
23:49 there's I think learning has happened
23:52 for some people have said Yep. that
23:56 learning has definitely happened.
23:59 So that's why I go because maybe someday
24:02 maybe it'll something will sink in. That
24:04 would be good. That would be good. That
24:06 would be so nice. Yeah. That'd be good.
24:08 Be good. I don't know what I'm doing
24:10 with my life. I should probably I should
24:13 probably pay for therapy. Yeah. Yeah.
24:15 No, I
24:17 I I think I'm trying to fill a void uh
24:20 with the with AI learning lab. Yeah. Uh
24:22 but no, it's good. It's good. It's It's
24:24 solid. It's he he promises he's going to
24:27 finish a sentence someday.
24:36 Who's got Who's got questions? Uh Moon
24:39 Valley. Uh this is Moon Valley. So Moon
24:42 Valley when
24:45 I think it was when Runway came out like
24:47 two years ago.
24:49 Moon Valley came out right around the
24:51 same time and it was a Discord only. It
24:53 was like remember when uh Midjourney was
24:56 Discord only.
24:58 Moon Valley was was
25:00 a video generator two years ago that was
25:03 mid or Discord only
25:07 and it looks pretty good. Uh well from
25:10 the outside it looks good because I
25:12 didn't pay for it. But if you say start
25:14 creating they do this really annoying
25:16 thing where you got to sign in. Think I
25:19 got to do this. Then I got to do this.
25:23 Then I got to do this.
25:26 Uh,
25:30 but it's got controls. It's got all
25:31 these controls. So, like motion
25:33 transfer. So, you can upload a video
25:36 with motion and then it'll take the
25:38 motion from that video and put it on
25:40 some other image. Um, or here's pose
25:43 transfer or here's camera controls. Um,
25:47 here's trajectory control. Here's shot
25:49 extension. So, there's that. Then
25:51 they've got reference. You can add
25:53 reference photos. So, it looks pretty
25:55 good except they let you do all this and
25:58 write a prompt and then and then when
26:00 you hit generate, they're like, "Oh,
26:02 sorry, loser. Sorry." Yeah. No. Yeah.
26:07 This is going to cost you money. I I
26:09 listen, I know we could have said that
26:11 before you pushed the generate button,
26:14 but you know, we don't have a Gen Xer
26:16 working at our company to yell at us for
26:19 stupid [ __ ] like this. So, if I mean,
26:22 just go ahead and press the button and
26:24 see what happens.
26:26 It's probably going to make you angry.
26:34 Uh, it's a two-hour void that comes back
26:37 five nights a week at least.
26:43 The AI learning lab where you can't get
26:45 those two hours back. Be there.
26:49 [Music]
26:57 Um, anybody want to do anything tonight?
26:59 Uh, it looks like Udo added beats per
27:02 minute to their little thing, which is
27:04 kind of cool.
27:09 The void that keeps on giving.
27:13 Even black holes emit radiation once in
27:15 a while.
27:20 Um,
27:22 who's got a thought about AI? Love it,
27:24 hate it, um, questions about it.
27:29 What tool should I use?
27:36 Do you need a nap?
27:42 No one. Buer, buler, buler,
27:46 narrator forever. So, what's new these
27:48 days? Been away. Um,
27:51 okay. Here, here are the biggies for me
27:53 right now. We're we're getting Grock 4
27:56 tonight. We'll probably get um
28:00 open-source
28:01 chat GPT or yeah, I guess open source
28:04 chat GPT
28:07 potentially a chat GPT browser this
28:10 month. Next month we'll get GPT5 I
28:14 think.
28:16 Um,
28:17 so there's so the frontier model
28:20 companies are doing something.
28:24 The the thing that I've been most
28:27 enamored by is MidJourney and Google's
28:30 V3 video models. They're just crazy
28:34 good. They're crazy good. They're crazy
28:36 good. They're really good.
28:38 Um, so that that's what I'm kind of
28:40 excited about. I joined a VC co-working
28:44 space today. Curious to see how many are
28:46 working on AI. Kelly, I can't imagine if
28:50 you're in a co-working space in Dallas.
28:52 I would I would put money.
28:55 Sorry for yelling. It's the phone, not
28:57 me. So, this is what Kelly does. She's
28:59 very passive aggressive. She'll come in
29:02 and purposefully type all caps and then
29:04 go, "Oh, oh, rather me. Oh my. Oh my, I
29:08 did not mean to to be so brusk, right?
29:11 She's southern. She's like she's got
29:13 Texas thing going. I know what I did was
29:15 more like Louisiana, but or or more like
29:18 Georgia, right? But she's like the Texas
29:21 version of that. She'll just come in
29:23 here and just yell at people and then
29:24 she's like, "Oh, was that me?" Yeah.
29:27 Anyway,
29:29 um I would put money on the fact, Kelly,
29:32 that you probably can't find
29:35 a business in that space that's not
29:38 working on AI. That's right.
29:44 Oh, man. We need your support, Kelly. We
29:47 do. We do. Um
29:52 it's me. I'm the problem. User error.
29:59 Oh my god. Mr. It is still still calling
30:02 for agents.
30:05 Gareth. Gareth has a single question.
30:09 He's not going to tell us what it is,
30:11 but he does have a question.
30:14 [Music]
30:18 What's going to happen to all of the
30:19 programmers? Well,
30:22 um, what's going to happen to every
30:24 knowledge worker and every creative
30:26 worker and quite frankly probably every
30:28 worker because this is going to affect
30:30 every industry, not just not just
30:32 knowledge workers,
30:34 um,
30:35 is that they're going to have to adapt.
30:40 Humans are not static. There there's a
30:42 the
30:48 if you understand what AI is and you
30:50 look at Microsoft laying off 15,000
30:53 people and you think they're not
30:55 related, they're very related. Um
31:01 but there's a couple of things
31:03 happening. One is
31:06 while AI is going to take some jobs,
31:10 what AI is actually going to do is AI is
31:12 going to
31:14 automate some tasks.
31:17 And any job is made up of some series of
31:20 tasks, right?
31:23 And if you have a job that's got a
31:25 limited set of tasks or a very
31:28 specialized set of tasks that are highly
31:31 likely to be replaced with AI, then your
31:33 job is going to be gone. The tasks that
31:36 you do are going to be gone. Now,
31:40 if you're a business owner, you could
31:41 say, "Hey, I've got all these people. I
31:43 could give them other tasks to do."
31:47 Um,
31:49 but a lot of companies aren't going to
31:50 do that. just going to lay people off
31:52 and they're just going to go that job
31:53 was eliminated. Um I think that a lot of
31:57 companies are going to fire the wrong
31:59 people.
32:03 I I or or maybe put another way, I don't
32:06 think they're going to keep the right
32:07 people. I don't think who's going to be
32:10 successful in an AI powered company are
32:14 the same people that are successful in
32:17 specialist jobs today. I think creative
32:19 generalists um are going to be the new
32:22 superstars in the in the workforce. So
32:24 what I would say is this. I would say
32:29 to the degree that you've got a
32:32 hyperfocus on a single kind of task,
32:37 um you're probably more at risk.
32:41 And so what that looks like is start
32:43 playing with AI. If you're a programmer,
32:45 if you're a programmer and you're not
32:47 vibe coding, I think that's a big
32:49 mistake. Andre Karpathy, right, one of
32:52 the co-founders of a of OpenAI um and
32:55 the the lead scientist there for a
32:57 while. Um he has stopped coding. He is
33:01 only vibe coding. And if you don't know
33:04 what vibe coding is, it's where you're
33:05 using one of these tools that writes
33:07 code and you just talk code into
33:10 existence and it writes the code. And
33:12 then he goes in and I guess he reviews
33:14 it and updates it and does whatever you
33:16 do. Um
33:19 there's there's there's there is a
33:21 radical
33:24 elitist
33:25 um
33:27 thing happening. I see it in
33:29 engineering. I certainly see it in the
33:32 art community. Um,
33:35 I see it in the video community
33:39 where
33:43 the people who've spent 20 or 30 years
33:46 mastering the craft of programming or
33:48 film making or illustration or design
33:54 are [ __ ] pissed off.
34:00 They're pissed off.
34:04 Because
34:05 these tools
34:08 do what they do, right? And and the rage
34:12 right now is about, well, they stole our
34:14 work and that's why I'm losing my money.
34:16 No, you're not losing your money because
34:17 they stole your work. You're losing
34:18 their money because we have we now have
34:20 this tool
34:23 that can do the execution of things in
34:25 an incredibly efficient way, but it also
34:28 makes new things possible.
34:31 Tik Tok is glitching, FYI.
34:34 Oh, yeah. It looks like my interface
34:36 went away. Hang on. We've lost the
34:39 stream
34:41 two times. All right. Hang on. Let me
34:43 I'll quit it and come back in.
34:48 Right now,
35:01 all right. Is that better people? Tik
35:02 Tok people, are you good? Is this good?
35:06 Hello. Hello. Hello. Hello. Hello.
35:11 Hello. Um,
35:14 looks like we're back. Okay. Um,
35:24 so to the degree that people in very
35:27 specific jobs or careers, let's put
35:30 careers. To the degree that someone is
35:33 in a very specific career, I'm just an
35:35 illustrator, I'm just a programmer, I'm
35:38 just a this, I'm just a motion graphics
35:41 person.
35:43 to the degree that those people remain
35:45 committed to that single thing
35:49 and especially if they resist learning
35:51 AI for that thing,
35:55 I think they're going to be increasingly
35:57 employment challenged
36:02 because while a motion graphics
36:04 professional can look at anything
36:06 generated by these video tools today and
36:09 make fun of them and laugh at them,
36:12 they're a [ __ ] ton done better than they
36:14 were two years ago, right? And so two
36:16 years from now, they're going to be
36:19 indistinguishable, right?
36:22 But here's the thing that no one talks
36:24 about. What everyone talks about is the
36:26 AI is going to take the jobs. What no
36:29 one talks about is the 15,000 people
36:31 that got laid off at Microsoft.
36:36 How many of them are truly AI literate
36:40 and AI passionate?
36:44 Let's say it's a thousand of them. You
36:47 got a thousand really smart people that
36:49 were maybe at Microsoft and let's let's
36:51 say it's less than that. Let's say it's
36:52 a hundred of them. Out of 15,000,
36:53 there's 100 people that are super
36:55 passionate about AI. One is Microsoft
36:59 should have kept them, right? I don't
37:00 know how they identified the 15,000. The
37:03 other one is those hundred people now
37:07 don't just have the skill that they were
37:09 hired for at Microsoft. They've got all
37:11 the other skills that they can tap into
37:13 with generative AI. They can do
37:15 research. They can do writing. They can
37:17 do art. They can do music. They can do
37:20 film. They can do all these different
37:22 things. They can do marketing plans.
37:26 And so what what I think we're going to
37:28 see is masses of people being forced
37:33 into entrepreneurship.
37:36 And those people that are like, "Well, I
37:38 don't I don't know how to start a
37:40 business and and they're not playing
37:42 with AI." They're going to get circles
37:44 run around them by people who do know AI
37:48 that are like, "Okay, I want to start an
37:50 Etsy shop. Okay, I need to come up with
37:52 designs. I need to come up with
37:53 manufacturing. I need to come up with
37:55 sourcing. I need to come up with a
37:57 marketing plan. I need to come up with
37:58 advertising. I need to figure out a way
38:00 to do content marketing because I don't
38:02 have a lot of money, but I do have a lot
38:03 of time because I just got laid off from
38:05 my [ __ ] job. Right? A single person.
38:08 If you if you want inspiration for the
38:11 kind of person
38:13 who's going to be incredibly successful
38:15 in the future,
38:18 take a look at Jim Ross, who's a part of
38:20 this organ, this community.
38:23 He's in the self- storage business. He
38:26 spends an hour every morning like like
38:29 drinking his cup of coffee playing with
38:31 AI doing things with his business for AI
38:34 or if he doesn't have anything he's just
38:36 out experimenting.
38:38 He comes through these lives on a
38:40 regular basis and when I teach
38:42 occasionally teach something and he
38:44 learns something new he's like huh let
38:45 me go try that for my business.
38:49 He's told me on multiple occasions that
38:51 he is functioning, he himself is
38:55 functioning like a 10-person company
38:58 because he's gone all in on AI.
39:02 That's what people aren't talking about.
39:04 They're just talking about the loss side
39:06 of it. The opportunity side of it is
39:08 [ __ ] huge because
39:10 because so many people are resisting AI,
39:13 pissed off at AI, avoiding AI,
39:16 dipping their toes in the water with AI,
39:19 waiting till someone else figures it out
39:20 with AI,
39:23 the people that embrace AI, the people
39:25 that are irregular enough to show up at
39:28 a channel like this five nights a week
39:31 to just be in the conversation to to
39:34 glean some knowledge through osmosis or
39:37 talking to others or watching me demo
39:40 something.
39:43 The work that they produce as a single
39:47 entity, as a single person,
39:51 is going to make them occur to other
39:54 people like [ __ ] wizards.
39:58 Wait, Jim, you what? What do you mean
40:00 you were on a call with a guy? Oh, this
40:03 was a thing that Jim Ross did. Here's
40:06 one th this is one where you appear like
40:08 a wizard.
40:10 So there was some big industry event for
40:15 self storage, right?
40:17 And there was a keynote speaker, the
40:20 main keynote speaker dropped out of this
40:24 big event
40:27 20 minutes
40:29 two minutes
40:31 before he was supposed to speak. sorry,
40:34 can't do it. So the the guy running the
40:37 event reached out to Jim Ross and said,
40:39 "Jim, can you do a keynote for this
40:42 event?" And Jim said, "I don't really
40:45 have a keynote."
40:47 You know what Jim did? This is [ __ ]
40:49 brilliant.
40:52 He went into I don't know Manis might
40:55 have been Manis or or GenSpark one of
40:58 them
40:59 and
41:01 the keynote started and he said hey you
41:04 know I'm here at the last minute for
41:06 this keynote. So, what I'm going to
41:08 actually do is I'm going to use AI to
41:11 write a keynote presentation about AI in
41:15 the self-s storage business.
41:18 And with people watching, he pushed the
41:21 button and this thing started writing
41:24 and generating slides for his keynote.
41:27 And so, five or 10 minutes later, it
41:30 generated this keynote. He gave the
41:32 keynote.
41:34 And then when when the thing was over,
41:37 what do you think people wanted to talk
41:39 about?
41:42 Self storage? No. They're like, "How did
41:44 you do that? What What was that? Is that
41:48 even possible? Did you Was that like a
41:51 movie? Like that's Did Did that just
41:54 happen?"
41:58 So like what's going to happen to the
41:59 programmers is the programmers that
42:03 understand
42:04 that the power dynamics
42:07 for all of employment are shifting and
42:10 they start adapting and start and start
42:13 doing everything they can like like one
42:15 thing that I think programmers could do
42:19 that is going to give them job longevity
42:21 for some time is I think there is a
42:25 whole industry
42:26 about to be born
42:29 of good engineers that can unfuck up
42:33 vibecoded applications,
42:37 right? Troubleshooting shitty code that
42:41 clueless people like me wrote, right? I
42:44 started a SAS company. I've got $100,000
42:47 in MM MR and I find out that my my SAS
42:51 service that I vibe coded is completely
42:53 insecure and people are getting their
42:55 credit cards stolen. Oh [ __ ] what do I
42:57 do? I don't know. Maybe call someone
42:59 that knows what the [ __ ] they're doing
43:00 with technology, right? That's going to
43:02 be a whole job.
43:05 Who's going to get that job? Not some
43:08 elitist programmer that's like vibe
43:10 coding's not real coding.
43:13 No, no. It's gonna be a vibe coder like
43:15 Andre Karpathy that's like, I'm gonna
43:17 I'm gonna learn how to do this vibe
43:19 coding stuff as well as I know how to do
43:21 that coding stuff
43:23 because I want to understand what it
43:24 does well, what it doesn't do well, and
43:26 I want to understand what the
43:27 vulnerabilities are. And I want to be
43:29 I'm you know what? I'm gonna start a
43:30 whole business around fixing things that
43:34 people messed up because I because what
43:36 I've learned in messing around with all
43:38 this is they always [ __ ] up the same
43:40 four things, right? So, I'll go in and
43:44 they'll be in a panic. They need me to
43:47 fix their stupid app that they made. And
43:50 because I know the four things that they
43:51 probably [ __ ] up, I can go in and look
43:53 at it and I'll charge them $10,000
43:55 because they're in a panic and I'll fix
43:57 it in an hour, right? Or a day or a
44:00 week, doesn't matter. Like there's a
44:03 whole business there. I I feel that's a
44:05 whole industry
44:07 like a whole sector.
44:13 It's developer approved.
44:17 Have you tried emergent yet? No, I have
44:20 not. Let me see. You know, it's hard to
44:23 type with a guitar in your lap.
44:26 That's scientific fact. There's a
44:29 research study. Oh, it's 8:44.
44:32 Uh, emergent.sh. If.sh means shell, if
44:37 if emerging is something I do in the
44:38 command line, Gareth, I I will you know
44:42 you know my opinion on emergent.sh.
44:50 Uh uh welcome to emergent build
44:53 ambitious app with AI. Uh
44:57 English not the first language of the
44:59 developers here. You know you could use
45:01 chat GPT and translate that a little
45:03 better, right?
45:07 I'm such a dick.
45:12 It's developer approved. All right. What
45:14 would you like to build today? Um, let's
45:17 do a Um,
45:20 okay. Um, I'd like to build.
45:25 Is this going to be one of these? Oh,
45:26 no. I've got 10 credits. Okay. I'd like
45:29 to build
45:32 a
45:38 story assistant
45:40 where
45:42 user picks
45:45 from
45:47 the top seven
45:50 storytelling
45:52 frameworks
45:54 and
45:57 then gives a simple
46:00 log line of their story
46:05 and picks a framework
46:10 and the site um
46:14 writes their outline, their complete
46:18 outline, their complete
46:20 outline.
46:25 We'll just start it there. Okay.
46:28 Bang. Emergent.sh.
46:31 This is looking decent. Setting up a
46:34 safe environment. Ah, safe environment.
46:37 See, not just setting up an environment,
46:39 setting up a safe environment. Very
46:41 cool. If if indeed it's safe, but but if
46:45 Gareth is recommending it, it probably
46:47 is configuration environment or
46:49 configuring environment and then
46:51 starting the agent.
46:55 Booting.
46:59 [Music]
47:11 Love you.
47:13 [Music]
47:36 Oh, good. Yeah, Brandon, keep keep No
47:39 stream yet. Okay. Yeah, keep an eye on
47:41 X. I I I have this sneaking suspicion
47:43 that they're not gonna they're not going
47:45 to go live at 9ine, but we'll see. All
47:48 right.
47:50 agent is thinking. So, I guess I guess
47:52 it set up a little environment for us.
48:00 Don't trust anything I recommend. This
48:03 is your music on hold. Oh man. Um, Side
48:08 Hustle Mimi is going to be back, Pate.
48:10 Yeah, it says safe right there. It's got
48:13 to be true or you couldn't put it on the
48:14 internet. That's that that's my
48:17 philosophy.
48:18 is is,
48:20 you know, if if someone took the time to
48:22 write it on the internet, then it's
48:25 probably fine. So, yeah, this said it
48:28 was safe. So, I think it's good. I think
48:29 we're good.
48:32 I mean, here's what safe at least says
48:35 to me is that they're at least aware
48:39 that not having a safe development
48:41 environment is a big problem in vibe
48:43 vibe coding. So, at some point,
48:45 someone's going to figure this [ __ ] out.
48:47 It'll probably be Google.
48:49 [Music]
49:00 You think Elon won't be accurate on his
49:03 release time? Gasp.
49:06 I just I I would have thought I would
49:09 have heard more of him bragging today
49:11 about what's coming with Grock 4. And
49:14 I'm hearing more about the Epstein
49:16 files. So, I don't know.
49:20 I just have this sneaking suspicion
49:22 they're not ready.
49:25 But they did they did flip um they did
49:29 flip Gro to this thing called Super
49:32 Grock.
49:38 The files are in Grock 4.
49:42 Oh. Oh, the FC files are in Gro 4.
49:54 That Roadster promised in 2016 will be
49:56 ready any day now. So will full
49:58 self-driving.
50:02 Um, what new features come with Grock
50:05 Fort? We really don't know. They and it
50:07 was this was supposed to be Grock 3.5
50:09 and then just all of a sudden they said
50:11 it was going to be Grock 4. Um, and we
50:14 haven't heard anything. So it looks like
50:16 my
50:18 my application is being written.
50:24 Introducing forks. Oh, that's cool. I
50:27 like forks.
50:29 A kind of weirdass version control.
50:31 Story assistant. Choose a framework.
50:33 Share your idea.
50:36 Start creating your story. 3X structure.
50:39 Hero's journey. Save the cat.
50:42 Can I select multiple of these? No. But
50:45 that's okay. We'll do save the cat.
50:47 Continue the story idea.
50:50 Or try one of these examples. A young
50:53 hacker discovers a conspiracy
50:58 that
51:01 the
51:04 list
51:07 of famous people
51:10 who
51:13 um did bad things
51:17 suddenly
51:19 went missing.
51:26 [Laughter]
51:28 Your story outline based on a young
51:30 hacker discovers cons. Wait, based on
51:34 the opening development would go here.
51:36 This is mock. Oh, so this is not using.
51:40 All right, this is all mock data. All
51:42 right, whatever. All right, cool. This
51:44 looks this looks very much in line with
51:47 uh with like replet agent things like
51:50 that. Thank you very much, Winston.
51:52 Thank you so much. That's very generous
51:53 of you. Um what's cooking? Um we're
51:56 we're minutes away, if you're to believe
51:59 what Elon Musk announced on Monday.
52:02 We're minutes away from
52:06 the Gro 4 announcement.
52:08 Do you use some kind of intro to the
52:10 prompt? What is this? Oh,
52:18 the texture of the image changes. I I'll
52:20 answer that. That's that's pretty cool.
52:22 That's a good question.
52:26 Yeah, Winston, thank you for the gift.
52:27 That that was very generous of you. I
52:29 appreciate that.
52:31 Uh
52:33 let's go to I wonder if it'll be Elon
52:36 Musk. Let's go to I guess we'll go to
52:37 X.AI AI
52:41 croc 4 search
52:51 gro 4
52:54 60 minutes left until Grock 4. I just
52:57 don't I don't think it's coming. Wait,
52:58 what's that? Grock 4 presentation
53:01 tomorrow is going to be hilarious.
53:05 Allegedly it will be atxai.
53:09 So is it x.ai x.ai.
53:16 Here you go.
53:17 Yeah.
53:17 So Elon put a tweet uh or a zit or
53:21 whatever we call them now a couple days
53:22 ago
53:24 saying that the live stream was going to
53:26 be on the XAI uh X handle or Twitter
53:30 handle.
53:31 Oh, the XI Twitter handle. Okay.
53:34 Wednesday at 8:00 PM Pacific. That being
53:37 said, I'm not seeing anything to
53:40 indicate that that would be going live
53:41 yet.
53:42 X AI
53:45 X AI. Okay.
53:49 And by the way, prediction is Oh, wait.
53:52 Breaking Grock to launch tonight at 11
53:55 p.m. But who the who's that? Who's son
53:59 of Walke?
54:04 Oh, 11 PM EST. Yeah. So, this is
54:07 [ __ ] Okay.
54:10 Um, where's Let's go to XAI.
54:14 I'm following them.
54:17 Live on X. Look, AI learning labs live
54:19 on X. Woo.
54:24 Gro for party.
54:28 Grock 4 launch party. Yeah, there's all
54:30 these launch parties. Where's the
54:32 launch, Elon? You got six minutes, Elon.
54:36 We're all waiting.
54:38 We're all waiting at this point. I've
54:40 worked with people coming
54:43 from going to every major AI player.
54:45 Wait. Oh, you've worked from for You've
54:48 worked with people coming from every
54:49 major AI player except X. Yeah.
54:52 Although, didn't X just poach a couple
54:54 of people from Open AI and then Open AI
54:57 poached a couple of people from X?
55:01 He's announcing his new party. Yeah. Um,
55:06 live search API.
55:09 Yeah, this is all old [ __ ]
55:25 [Music]
55:34 Yeah, we could do that. Um, if you're
55:37 while we're waiting for Elon to show up
55:39 with his [ __ ] um, if you're not a
55:42 member
55:44 of, uh, of the AI salon, you should be.
55:47 So, if you go to the salon.ai AI and
55:50 click on join our community.
55:53 That's going to drop you into this
55:55 awesome place full of awesome people. Um
55:59 you'll learn about things like the AI
56:01 readiness cycle, play first, mindfully
56:03 create, generously lead, um our values,
56:06 five stages of AI adoption. But if you
56:08 scroll down the lefth hand side here,
56:10 there's there's um first sort of seven
56:13 things you do are things like introduce
56:15 yourself, check out our values, check
56:16 out pest meetings. So you really get to
56:18 understand what the community is about.
56:20 Um then there's a community corner, a
56:22 news area, and then there's um a
56:25 mastermind area where there are clubs
56:28 and hubs and all sorts of things,
56:30 leadership lounges and things like that.
56:32 I want my crown. Um you should get a
56:34 tag. You should get a badge, Gareth, but
56:37 I don't think you get a crown. Just the
56:39 spaces have the crowns. Um the crowns
56:42 are part of the mastermind community.
56:43 Yeah. any anywhere where you see the
56:46 crown beside it, those are subscription
56:49 areas that are part of the mastermind.
56:51 Um, and the mastermind has very very low
56:53 special pricing through the end of 2025
56:57 and you get to become a founding member
56:59 of the AI salon mastermind. Um, but you
57:01 should go check it out. Another cool
57:03 thing on the AI salon is there's a a
57:06 free area called the Irregulars channel,
57:09 which is all the people that hang out
57:11 here. And here's Vicky's favorite
57:13 picture.
57:15 Did she make a Did you make a video of
57:17 them?
57:23 [Laughter]
57:30 That's pretty good.
57:32 You can see comedy like that right here.
57:34 Right here in the irregulars channel.
57:36 Look, there I am.
57:40 Artificial intelligence is no match for
57:42 natural stupidity. No [ __ ]
57:46 All right, back to uh back to X to see
57:48 if anything's happening anywhere.
57:52 [Music]
57:54 Ah, internet will break today after
57:56 seeing Grock 4. Uh, I'll believe it when
57:59 I see it. Let's go see if Grock 4 is
58:00 showing up anywhere.
58:02 Gro super. See where it says super
58:04 Grock? That's new, but it's still Grock
58:07 3.
58:09 Oh, go to grock.com.
58:15 Grock 3 enabled.
58:19 Custom instructions. Enable search.
58:35 [Music]
58:38 Good lord. So, what here's what we'll
58:41 do. We'll we'll make some cool [ __ ]
58:43 tonight. What?
58:46 What is that desk?
58:50 [Music]
58:59 [Music]
59:02 down.
59:03 [Music]
59:13 [Music]
59:24 [Music]
59:36 [Music]
59:47 All right, Elon, are we going? Are we
59:50 going? Are we going? Are we going?
59:55 I do not think we are going. It is nine
59:58 o'clock and I see nothing.
1:00:01 Uh let's look up Elon.
1:00:04 Elon Musk.
1:00:06 Hello, my name's Elon Musk. Grock to
1:00:09 understand
1:00:12 Grock for release
1:00:14 Wednesday at 800 PM. Ah, he didn't say
1:00:17 which Wednesday.
1:00:19 He didn't say which Wednesday, did he?
1:00:22 I think you need a paid subscription to
1:00:25 see Gro 4.
1:00:27 I've got a paid subscription.
1:00:31 This there's supposed to be an
1:00:32 announcement.
1:00:34 Gro 4 released live stream on Wednesday
1:00:36 at 8:00 PM
1:00:41 Pacific
1:00:45 Starlink.
1:00:52 Uh, I got nothing people.
1:00:56 I guess producer Brandon, why don't you
1:00:58 Hang on. I got to do physical dexterity
1:01:00 on TikTok. Um, why don't you just keep
1:01:04 checking on on uh Twitter and we'll just
1:01:06 go make some cool images and [ __ ] like
1:01:08 that. Um, let me I want to go actually
1:01:10 uh comment to this guy that asked a
1:01:12 question on on Twitter.
1:01:15 Um, did you use some kind of intro
1:01:17 prompt? Um, let's see. I made
1:01:21 this video
1:01:24 quickly on one of my lives as an
1:01:30 experiment.
1:01:34 Um,
1:01:36 so it has a
1:01:40 color grading
1:01:44 and um,
1:01:47 aspect ratio
1:01:49 glitch.
1:01:51 Here's why.
1:01:54 Um, one started
1:01:58 with a midjourney image.
1:02:03 Two, animated
1:02:06 that in midjourney
1:02:10 to make
1:02:14 the statue
1:02:17 look at the camera. Oh, here. Let me
1:02:20 show you what I'm talking about. Um,
1:02:23 oh wait, let me let me finish this and
1:02:25 I'll show you what I'm talking about.
1:02:26 camera.
1:02:28 Um,
1:02:30 took the last frame
1:02:36 of the animation
1:02:40 and used
1:02:43 it as start frame
1:02:49 for VO3
1:02:53 talking section.
1:02:57 Um, VO3
1:03:01 has a different
1:03:08 aspect
1:03:10 ratio
1:03:12 than MJ16 by9.
1:03:16 Um, and it
1:03:19 increased the contrast.
1:03:26 from the
1:03:28 original
1:03:34 AI fun. Yes.
1:03:36 First of all, that's not how you spell
1:03:37 aspect.
1:03:38 Uh or increased.
1:03:44 Second of all, um XAI just posted the
1:03:47 Gro 4 live stream will begin soon. Stay
1:03:50 tuned.
1:03:53 All right. So, let me go show you the uh
1:03:55 the the video that I made. Um, let me
1:04:01 This was the one. So, so someone just
1:04:02 commented on this. They asked the
1:04:04 question, "Why does the second half of
1:04:06 this video not look the same as the
1:04:08 first?"
1:04:10 And it's because I use two different
1:04:12 tools. Oops, that's not what I want to
1:04:13 be doing. Cancel.
1:04:20 So this was from last night.
1:04:23 We had
1:04:28 this image
1:04:31 in midjourney
1:04:34 and then we animated it for 5 seconds in
1:04:38 midjourney. then took the last frame of
1:04:40 that animation, took it into VO3 as a
1:04:43 start frame
1:04:45 so that we could get the statue to talk.
1:04:47 And so here's here's what we created
1:04:49 last night. And this was pretty quick.
1:04:51 This was like I don't know 10 15
1:04:53 minutes.
1:05:01 Oh, this curse, that vexing witch. When
1:05:05 will I be free?
1:05:09 Nice, right? But see how it's like flat
1:05:13 and then it goes high contrast.
1:05:14 This curse, that vexing witch. When will
1:05:19 I be free?
1:05:20 But it's still pretty [ __ ] cool. Like
1:05:22 if you if you did some color grading on
1:05:24 that to match them and and you know,
1:05:27 make sure it's not stuff's not jumping
1:05:29 around. Be pretty good. You're showing
1:05:31 your expiration date. Waha.
1:05:35 This is like Heraldo Rivera leading us
1:05:37 to Alapone's vault. If you don't
1:05:40 remember that, that was that was one of
1:05:41 the most hilarious
1:05:43 events in television history. All right,
1:05:46 let's go to Xi. Yes,
1:05:49 before you uh navigate away from this, I
1:05:52 sent you a DM on the hugging face
1:05:53 robots, and since you're already sharing
1:05:55 with sound, it would be a good idea to
1:05:56 bop over there.
1:05:58 Where? But is it just is it big and
1:06:01 prominent on hugging face or is it one
1:06:02 of the 8,000 links? No, no, no. It's
1:06:05 it's a it's a DM in your Twitter. It's a
1:06:07 tweet.
1:06:08 Oh,
1:06:08 with a video.
1:06:10 Oh, got it. Oh, okay. Because because
1:06:12 I'm sharing.
1:06:14 Great. Got it. Okay, cool.
1:06:18 So, for AI builders, so this is um
1:06:25 I guess Hugging Face is putting this
1:06:27 out, Brandon.
1:06:28 Yes.
1:06:28 Is that what you said? All right. So,
1:06:30 Hugging Face, if you don't know Hugging
1:06:32 Face, Hugging Face is a repository for
1:06:34 like open source models and weights and
1:06:37 all sorts of things. Like geeks use it
1:06:40 and you can go play with things there if
1:06:41 you want to get around dick around with
1:06:43 open source. It's a great it's a great
1:06:45 resource. Um, and apparently they're
1:06:48 putting out an open-source robot for 300
1:06:50 bucks. So, here we go.
1:06:53 [Music]
1:06:58 Let's
1:07:07 [Applause]
1:07:08 get everyone. You get everything you
1:07:11 want. If you want it, you can have
1:07:16 anything. If you want it,
1:07:21 you can have anything you want. In the
1:07:25 brand new
1:07:28 [Music]
1:07:41 cool reachi mini
1:07:44 cute
1:07:47 fascinating they're going after the
1:07:48 education market but you know
1:07:51 isn't that fascinating that like Where
1:07:54 education starts to go is go make a
1:07:57 robot. Go program a robot. Um, hugging
1:08:01 face powers everything in the AI world.
1:08:03 Um, if it's not the models themselves,
1:08:05 it's the data sets. Yeah, hugging. Yeah,
1:08:07 hugging face is really quite remarkable.
1:08:09 There's a lot a lot of stuff there,
1:08:11 which is why I avoid it. It makes my
1:08:13 head hurt. Um,
1:08:16 uh,
1:08:17 does anyone
1:08:20 high-five in real life? I still high
1:08:23 five. It's a Gen X thing.
1:08:26 Sometimes I high-five people and I I
1:08:29 tried to high-five a Gen Zer once and
1:08:31 she goes she goes she goes um I don't do
1:08:36 that. I find it aggressive.
1:08:43 Um I don't do that. I find it
1:08:46 aggressive.
1:08:47 [Laughter]
1:08:53 She's like, "Oh, no. No.
1:09:02 No live stream yet. We're waiting. We're
1:09:05 waiting. We're waiting."
1:09:08 Okay.
1:09:17 XAI. The Gro 4 live stream will begin
1:09:20 soon. Stay tuned. Link,
1:09:30 Google,
1:09:34 Gary Marcus is still dunking on all the
1:09:37 companies for not having their [ __ ]
1:09:39 together
1:09:41 like he'll do. Jav Lopez has got a new
1:09:44 upscaler coming.
1:09:46 Um
1:10:09 [Music]
1:10:16 Da da da. I found out that find vocal
1:10:21 fry is a normal expected part of
1:10:24 speaking Finnish. Hey
1:10:27 really
1:10:29 really
1:10:32 hey.
1:10:48 Teddy Rubkin was not safe for work.
1:10:53 You missed your chance for vocal fry.
1:10:58 Um,
1:11:00 by the way, if you're part of the AI
1:11:02 salon or if you just checked it out
1:11:03 because I told you to go check it out,
1:11:05 go to the events section next Tuesday.
1:11:08 We have an AI salon meet and greet where
1:11:10 you can come introduce yourself and meet
1:11:12 people in the community. It is very wise
1:11:16 in my opinion
1:11:18 to get your ass in an AI community.
1:11:22 Start getting curious and not an AI an
1:11:25 AI hater community
1:11:28 that that
1:11:30 like they're just going to get more and
1:11:31 more and more bitter. It's gonna it's
1:11:34 just I would avoid those. I would get in
1:11:36 one where people are like, "Well, it's
1:11:38 here. Might as well try to figure this
1:11:40 [ __ ] out, right?" Um, yeah, bring a a
1:11:44 friend or a friend of me. Yeah. Um, come
1:11:47 and introduce yourself. All right. Let's
1:11:50 see. Let's see. Let's see. Let's see.
1:11:52 Let's see. Let's see.
1:11:56 Hang on a sec. Oh, that's dead. Dang it.
1:12:10 Oh, wait. What's that? No. No.
1:12:16 11 minutes ago.
1:12:18 Stay tuned.
1:12:22 Breaking. Kanye no longer working with
1:12:24 Gro 4.
1:12:26 Mr. Strawberry. I'm super [ __ ]
1:12:27 excited. ASI baby. Shut up.
1:12:32 This is not going to be ASI.
1:12:35 This is probably going to be about as
1:12:37 good as GPT40
1:12:40 is my guess.
1:12:42 Maybe it'll have video as part of it. I
1:12:45 don't know. Maybe it'll be AGI. May
1:12:48 maybe Grock 4 is the first one to to
1:12:50 really hit it.
1:12:52 Kanye shouldn't be working with anyone.
1:12:59 Let's go. What are you guys predicting?
1:13:00 I predicted that this thing wouldn't
1:13:02 happen tonight. That's what that's my
1:13:04 prediction. My prediction is
1:13:11 Oh, here is also my prediction. If it
1:13:13 goes tonight, if there's a live stream
1:13:15 tonight,
1:13:17 it will be more awkward than any of the
1:13:20 Open AI tiny table presentations like 12
1:13:24 days of shipmiss,
1:13:26 right? Where lack of eye contact was the
1:13:29 feature. It was the feature, not the
1:13:31 bug.
1:13:41 Uh, it will have video creation
1:13:45 like VO3.
1:13:47 I heard that too. And and Elon sort of
1:13:50 hinted that, you know, video production,
1:13:52 you know, video generation is getting
1:13:54 really good. So, he kind of did a hint.
1:13:56 Um, and quite frankly, um,
1:14:00 when they dropped VO3, nobody was
1:14:02 expecting it to have image generation in
1:14:04 it that was any good. And it's, you
1:14:06 know, the image gen within V, uh, Gro 3
1:14:09 is pretty good.
1:14:11 Um, he's probably getting the video
1:14:13 upload error like I get. He's trying.
1:14:15 Yeah, that's what it is. They'd be live
1:14:17 streaming except if Elon can't get the
1:14:19 QuickTime video to upload to X.
1:14:26 Oh man. Give us five minutes. Elon had
1:14:28 to pee.
1:14:30 What's your hobby?
1:14:34 [Laughter]
1:14:39 Oh my god.
1:14:44 I've never understood.
1:14:51 I think I'm just too neurotic.
1:14:55 Because like I don't understand how you
1:14:57 could say, "Hey, we've got a major new
1:15:01 model coming out. Here's the here's the
1:15:04 launch date or the announcement date."
1:15:07 How do you miss the [ __ ]
1:15:09 announcement,
1:15:11 right? It's like like
1:15:14 I I just I feel like people's
1:15:16 relationships with deadlines anymore are
1:15:18 really [ __ ] annoying. Like, yeah,
1:15:21 we'll get this. Maybe it's 8. Maybe it's
1:15:23 8:15. Maybe, you know, maybe never. We
1:15:26 didn't Oh, you thought we meant this
1:15:28 Wednesday. That's That's [ __ ]
1:15:31 hilarious. Really? What? What made you
1:15:34 think that? Oh, because we announced it
1:15:36 on a Monday and we said it'd be live
1:15:39 Wednesday.
1:15:43 We didn't say which live. Which
1:15:44 Wednesday, did we? Huh? See how this
1:15:48 works?
1:15:51 been in politics. Now
1:15:54 the GPT store is also coming. Oh my
1:15:58 goodness. People don't care about time
1:15:59 and it drives me crazy. I know. It makes
1:16:02 me mad.
1:16:05 Madness. It's madness. I tell you,
1:16:08 madness.
1:16:11 Wednesday. Next Wednesday or next
1:16:13 Wednesday.
1:16:14 Oh, it was Wednesday Pacific time.
1:16:26 Vicki, I start my meetings on time. No
1:16:28 waiting for those who trickle in from
1:16:30 other meetings. Yeah, I know. I It's We
1:16:34 deal with agencies all the time and it's
1:16:36 just like
1:16:39 it's almost like they have Tourette's.
1:16:40 Oh, sorry I was late. I was in another
1:16:42 meeting.
1:16:44 Oh, oh, sorry I was late. I just had to
1:16:46 grab a bite to eat. Hey, sorry I was
1:16:47 late. I just don't respect your time.
1:16:53 If you're five minutes early, you're
1:16:55 late.
1:16:56 Yeah, it is true. If you have a meeting
1:16:58 with Vicky Baptiste, if you log in early
1:17:00 thinking you're going to be there first,
1:17:03 you're going to lose.
1:17:07 I have an itchy scalp. I wonder if it's
1:17:09 bugs.
1:17:13 You ever seeing that commercial? This
1:17:15 the GenX. Remember the Head and Shoulder
1:17:17 commercials where like the chick and the
1:17:20 dude are out on a date and she's like
1:17:23 she's thinking like she's like, "Oh,
1:17:25 he's really nice." And then he he kind
1:17:27 of does this and she goes, "Oh,
1:17:31 do you think he has dandruff?"
1:17:38 Funny comment on YouTube. Windows update
1:17:41 is almost done.
1:17:45 We should charge X for making us wait.
1:17:47 Like in a doctor's office. Yeah. Yeah.
1:17:50 Exactly. The The only thing less
1:17:51 reliable than tech announcements are
1:17:54 doctor's office announcements. You've
1:17:57 got to be there at 2:30 or, you know,
1:17:58 we'll cancel your appointment or we'll
1:18:00 kick you out of the practice. You get
1:18:02 there at 2:30. 4:15. They're like, Mr.
1:18:05 Shannon. Yeah. Well, there's no need to
1:18:08 be nasty. I've been here for four hours.
1:18:15 Oh, there is a reason to be nasty. There
1:18:17 is always a reason to be nasty.
1:18:21 Um,
1:18:29 why is the AI learning lab more
1:18:32 consistently on time than a what what's
1:18:36 XAI worth now? 1516 billion dollars.
1:18:40 Let's go ask Chat GBT what X AI is
1:18:43 worth.
1:18:44 Oh, you're not gonna be able to see
1:18:45 this, but I'll I'll just trust me. I'm
1:18:48 asking it. No, I'm good. Um, how much
1:18:54 is X AI valuation
1:19:02 [Music]
1:19:04 $80 billion?
1:19:07 Why is it that us with our poultry
1:19:10 little
1:19:12 20 million production studio made to
1:19:16 look like a shitty home office? Why are
1:19:18 we more on time than an 80 billion
1:19:21 company? Why? Huh? You tell me why. I
1:19:26 don't know why I just became Seinfeld.
1:19:29 You tell me why.
1:19:34 I don't get it. Elaine.
1:19:38 Um, they want you to be 15 minutes
1:19:41 early. They just don't respect our time,
1:19:45 you know.
1:19:47 Dear dear Mr. Musk,
1:19:51 I was very excited about the
1:19:52 announcement of of Gro 4. I have a
1:19:56 community of AI enthusiasts that show up
1:20:00 week nights, sometimes occasionally
1:20:03 around 800 PM Mountain time.
1:20:06 We're on that channel talking about AI
1:20:09 and I I reserved a very special episode
1:20:12 of the AI Learning Lab Live for for
1:20:15 Wednesday, July 9th, 2025 at 8:00 PM
1:20:19 Pacific time, 9:00 p.m. Mountain time.
1:20:22 That's an hour after we normally start.
1:20:24 We got there an hour early so we'd be
1:20:26 ready for your lives. Dear Mr. Musk,
1:20:29 to say that I am frustrated would be an
1:20:32 understatement. Sorry. Dear Mr. Musk to
1:20:35 to say that I am frustrated with the
1:20:40 lack of a live stream. Dear Mr. Musk,
1:20:44 congratulations on your big long hard
1:20:47 tall rocket flying into and coming back
1:20:52 down on the hard earth standing up
1:20:56 straight.
1:20:58 Um,
1:21:00 I waited for your live stream about Gro
1:21:03 4 because I heard it was going to change
1:21:05 everything and it never came. Dear Dear
1:21:09 Mr. Musk,
1:21:18 Dear Mr. Musk,
1:21:21 Dude, the start time now says 8:30.
1:21:24 [Laughter]
1:21:29 Dear dear Mr. Musk, um I I'm impressed
1:21:33 with how large and tall your rockets
1:21:36 are.
1:21:39 Wait,
1:21:41 dear Mr. Musk. Dear Mr. Musk, um
1:21:46 your electric cars are fast and sexy.
1:21:52 Your your boring company makes large
1:21:56 holes
1:21:59 and your rockets
1:22:02 are really big.
1:22:08 I forget what I was going to write here.
1:22:11 Um, congratulations,
1:22:14 Mr. Musk.
1:22:18 I hear you're making a lot of children
1:22:22 with the with the women with other women
1:22:24 with the women that you make you're
1:22:28 populating the because the birth rates
1:22:31 are right in Mars and making make
1:22:35 congratulations Mr. Musk I it's um
1:22:38 you're it's a it's a very impressive
1:22:40 rocket you have very very impressive
1:22:44 with the and with the children too.
1:22:46 Good. It's a it's a noble cause. It's a
1:22:50 noble cause.
1:22:52 Oh, sorry that was so loud. I think I
1:22:54 woke the cats.
1:23:02 I don't know what to say. Dear Mr. Musk,
1:23:05 please add the pink bow.
1:23:10 His rocket comes with report.
1:23:14 Well,
1:23:16 that went so high. It went so high in
1:23:18 the air. It's so exciting.
1:23:28 I guess I could have some some fancy
1:23:31 tequila.
1:23:38 [Music]
1:23:47 Okay, we're gonna go to I'm gonna change
1:23:51 my just producer Brandon, just let me
1:23:54 know when dear Mr. Musk gets his [ __ ]
1:23:57 live stream up and running. I'm gonna go
1:24:00 make some midjourney images. We're going
1:24:02 to start out with Dear
1:24:05 Mr. Musk. We're in midjourney, by the
1:24:09 way. Right now, the prompt is dear Mr.
1:24:12 Musk.
1:24:15 Um, we'll go dash dash sref style
1:24:20 reference space random.
1:24:24 We're going to turn our
1:24:27 stylization up to 9850.
1:24:32 We're going to go 16 by9 because I likes
1:24:34 me some 16 by9 aspect ratio. We're going
1:24:38 to do the black bar for producer
1:24:39 Brandon. He's being very patient
1:24:42 tonight.
1:24:44 And we're going to go dash dash
1:24:50 um
1:24:53 four.
1:24:55 And so R means repeat. So we're going to
1:24:57 create four sets of four. And then we're
1:24:59 going to go dash chaos.
1:25:04 We're going to do chaos of 65.
1:25:09 And we're going to turn on
1:25:11 personalization
1:25:12 because I want these things to look like
1:25:14 the [ __ ] I want them to look like. All
1:25:16 right, here we go.
1:25:18 And we're off.
1:25:32 planned to be. He is literally my love.
1:25:34 Oh, planned because he is literally my
1:25:36 love. Oh, come on, Elon.
1:25:41 Dear Elon, dear dear dear Mr. Musk. Oh,
1:25:46 look at that. She My god, he's a
1:25:50 handsome man. That is handsome. Look at
1:25:52 that. That South African chisel
1:25:58 looks like JFK there.
1:26:04 Oh, look. It's his little car. His
1:26:06 little cars are so cute.
1:26:09 They're so cute. They don't even have a
1:26:10 grill. Did you know that about the cars?
1:26:13 His car, they don't have a grill. You
1:26:15 know why?
1:26:17 Exactly. You don't have to cool down
1:26:19 that engine. It's just batteries. Just
1:26:22 batteries right along the floor. It's
1:26:24 just It's like a little miracle. It's
1:26:26 just a little miracle. He's just
1:26:27 amazing. The big tall rockets. Oh, I get
1:26:31 so excited.
1:26:33 Uh uh uh uh uh uh uh uh.
1:26:38 Dear Mr. Warhol, I would like to
1:26:41 commission you to make a self-portrait.
1:26:44 A portrait of myself in a little
1:26:46 spaceman outfit.
1:26:49 What? Oh, Mr. Warhole died years ago.
1:26:52 Huh? Well,
1:27:02 dear Mr. Mask, where's your [ __ ] live
1:27:05 stream? Wherefore art thon?
1:27:11 got a got a live stream tonight
1:27:14 precisely at 8:00 PM we launched rockets
1:27:18 so those are in pre precise timing
1:27:20 windows uh so when we launch something
1:27:23 with XAI similarly precise except not
1:27:26 tonight different we dealing with the
1:27:30 Epstein files we get them get my name
1:27:34 redacted it doesn't matter
1:27:39 that's actually a very nice image
1:27:41 DM'd you a link to the not let yet live
1:27:44 stream. Okay, good. Beautiful.
1:27:47 That's a cool style, too.
1:27:50 Which, by the way, so if we're going and
1:27:52 looking at these
1:27:54 um
1:28:01 see, right? So, you get your four images
1:28:03 here that all kind of look the same. And
1:28:05 right up here is the style reference.
1:28:08 You know, we said style reference random
1:28:10 and then we said do this four times. So,
1:28:13 this going to give us four different
1:28:14 style references.
1:28:18 Are they live?
1:28:19 No, sorry. I was just bringing it up in
1:28:21 the background so we can keep looking at
1:28:23 it.
1:28:23 Okay. Wait, am I not sharing? Oh, yeah.
1:28:26 Okay. Beautiful. Beautiful. Beautiful.
1:28:29 Beautiful. Beautiful.
1:28:31 Beautiful. Look at that. I like that. I
1:28:33 actually like that. Uh let's let's use
1:28:35 that SRE again. We use Chaos 65 again
1:28:41 and let's do um we'll do
1:28:45 um
1:28:50 Horatio
1:28:54 as the prompt
1:29:00 [Music]
1:29:04 low rent Robert Downey Jr.
1:29:13 Oh my god.
1:29:19 So much for efficiency. He's busy
1:29:21 starting his next failed political
1:29:23 party. He hasn't failed yet.
1:29:31 That's cool
1:29:33 for her ratio. That's pretty cool.
1:29:36 Oh, that's good. Looks like Philip
1:29:38 Seymour Hoffman.
1:29:43 How good is that?
1:29:50 Let's Let's animate some of these. All
1:29:53 right. So, we're going to animate that.
1:29:57 We'll animate that in high motion. We'll
1:30:00 do low motion there.
1:30:03 We'll do low motion there. We'll do high
1:30:05 motion there. All right.
1:30:08 We're making movies.
1:30:10 I got an ice cream and you can't
1:30:36 dear rock. How can one accelerate
1:30:38 emergence from a khole? asking for a
1:30:40 friend. Yes, Palindrome. Palenrome has
1:30:43 the right question. Mr. Musk. Mr. Musk.
1:30:45 Mr. Musk. Curious.
1:30:49 How do you accelerate emergence from a
1:30:51 khole?
1:30:52 You khole.
1:31:01 Seriously, when's the live stream?
1:31:04 My community is getting stir crazy. They
1:31:06 don't know what to do.
1:31:08 They're challenged in many ways. Neurode
1:31:10 divergent exception twice exceptional
1:31:13 three times except they so exceptional
1:31:16 we stopped counting the ways.
1:31:22 I understand you may have some neurode
1:31:24 divergency yourself, Mr. Musk. Mr. Musk.
1:31:26 Anyway, yes. No, I I've seen your
1:31:29 rockets. They're they're they're very
1:31:31 large. It's shiny. It's they glisten. Um
1:31:35 Yeah. No. Yes. I I'm sorry. I But no,
1:31:38 I'm here to hold you accountable for for
1:31:41 the live stream. You you promised us
1:31:43 8:00 p.m. precisely, Pacific Standard
1:31:46 Time.
1:31:49 It's not live.
1:31:52 All right.
1:31:54 [Music]
1:31:58 Let's see. How do I scrub?
1:32:04 Oh, here we go. All right. All right.
1:32:11 I think these two are better.
1:32:14 That's pretty cool, right?
1:32:17 How good are these?
1:32:19 I'm telling you, man. This midjourney
1:32:22 this midjourney [ __ ] animation is so
1:32:25 good.
1:32:29 Why Elon will not be president? He'll be
1:32:32 late on the debates.
1:32:38 Oh, yes. Well, there's no animation in
1:32:41 the arms there, so that's a loser.
1:32:44 Same here.
1:32:46 Huh. This is the first
1:32:51 This is the first
1:32:53 midjourney animation I've seen that I
1:32:55 feel like it it actually missed
1:32:57 something.
1:33:00 Like those arms should be doing all
1:33:01 sorts of crazy ass [ __ ]
1:33:04 Let's go try something. Let's go. Let's
1:33:06 go uh go back to that
1:33:10 and we'll do high motion manual and
1:33:13 we'll go um
1:33:17 the cat
1:33:19 tentacles
1:33:29 rise. as
1:33:32 the feline
1:33:36 looks menacingly
1:33:39 at the camera.
1:33:42 All right, let's try that. Let's see if
1:33:44 we can get it to work, people. I don't
1:33:46 know if we're going to get it to work.
1:33:48 I'm not a miracle worker here.
1:33:52 There there have been petitions signed.
1:33:54 I I can no longer claim that this is um
1:33:57 a university. Um I cannot say that if
1:34:01 you come to the AI learning lab you're
1:34:03 it's going to be an accredited
1:34:05 situation.
1:34:06 Um I am not legally allowed to give out
1:34:10 even a certificate at this point.
1:34:15 So I
1:34:16 but I think it is clear that you could
1:34:20 learn something here
1:34:23 potentially if the lawsuits stop.
1:34:28 L a captopus.
1:34:32 That's good.
1:34:35 Oh, wait. Although his head his head's
1:34:37 not moving. He's had Botox.
1:34:40 [Laughter]
1:34:47 Yeah. Yeah. See? Yeah.
1:34:54 See?
1:34:55 Yeah. Paluca.
1:34:57 Well, that was kind of cool. Except the
1:35:00 rain's not It's clearly raining. It's
1:35:03 got rain in front of his face and it's
1:35:04 not moving.
1:35:06 Like the famous turducken.
1:35:09 Elon needs to pull out Grock, blow on
1:35:11 it, and reinsert the cartridge. That's
1:35:14 Gen X video game reference.
1:35:17 Hey, have you seen my live stream? I I
1:35:21 had a streaming box. I think I left it
1:35:23 in the trunk of the car. Have you seen
1:35:24 it? Hey, have you do you have any
1:35:27 ketamine? Do you have any ketamine?
1:35:30 No. My teeth are
1:35:33 This is normal. This is what my teeth
1:35:35 look like.
1:35:38 [Music]
1:35:43 What?
1:35:45 Why are we not live? Why are we not
1:35:47 live? I've been I've been sleeping in
1:35:51 the in the artificial intelligence
1:35:53 factory. Where are the cars? Oh, sorry,
1:35:57 wrong company, Mr. Musk. Where are the
1:36:00 cars?
1:36:02 Uh uh uh. Oh, here we go. Yeah, there
1:36:04 you go. See? Oh, cool. That's what we're
1:36:07 looking for.
1:36:13 That's good.
1:36:15 It's so good.
1:36:18 So good.
1:36:26 Bad lips syncing AI midjourney edition.
1:36:34 All right, we're going to make this cat
1:36:36 talk
1:36:39 while we wait for Mr. Musk to get his
1:36:41 [ __ ] together.
1:36:43 I think it's going to be this one.
1:36:50 Or maybe it should be
1:37:04 No, not that one.
1:37:12 Could be that one. All right.
1:37:14 So, what we're going to do is we're
1:37:16 going to download this
1:37:31 We go to the last frame
1:37:38 and we're going to go
1:37:41 copy image
1:37:43 and then we're going to go to
1:37:46 flow.google.
1:37:50 Hello,
1:37:52 I'm with Flo.g
1:37:54 Google.
1:37:56 It's all Google Gemini. I guess I don't
1:37:58 have it open.
1:38:02 Um, flow.google.
1:38:09 All right. And then we're going to go
1:38:11 create with flow.
1:38:13 Then we're going to switch to the right
1:38:15 account.
1:38:22 [Music]
1:38:29 Elon does one more thing to make his
1:38:30 stock go down.
1:38:36 Ah, y'all are funny. I like comedy.
1:38:41 Comedy is not pretty. All right, new
1:38:44 project. So, we're going to do what we
1:38:46 did last night. We're going to do frames
1:38:47 to video.
1:38:50 We're going to paste a video. Oh, I
1:38:51 can't do that. I got to come over. Wait,
1:38:53 I got to go to preview.
1:38:56 Open that.
1:38:58 Do a new image.
1:39:00 Save that as cat.
1:39:06 Save.
1:39:09 And then we're going to go
1:39:14 here. We're going to go frames to video.
1:39:17 We're going to upload.
1:39:20 Where's our cat video?
1:39:24 No.
1:39:32 Where did I save it? What did I call it?
1:39:35 Cat.
1:39:38 Oh, there it is. Okay, there's cat.
1:39:42 Crop and save.
1:39:44 Crop and save.
1:39:47 Carrying on rather cherry. Pep pipep.
1:39:53 Um,
1:39:56 the
1:39:58 octo the
1:40:01 octo the cat
1:40:05 with tentacles
1:40:10 stops
1:40:12 in the middle
1:40:15 of the street
1:40:17 and says,
1:40:25 Something
1:40:27 funny
1:40:29 about
1:40:32 how ironic
1:40:35 it is
1:40:37 that
1:40:39 he has
1:40:41 cat taste
1:40:46 and
1:40:48 struggles
1:40:50 to not eat his
1:40:55 delicious
1:40:56 [Laughter]
1:41:01 octopus arms.
1:41:06 We We could have him say something
1:41:08 interesting, but no. Let's Let's do
1:41:11 something stupid.
1:41:14 Never mind. There's a weight list unless
1:41:15 you have Max. Yeah, the Perplexity
1:41:17 thing. Yeah, I know. Octopus found a
1:41:21 catapult one day. Uh uh uh. Is that Cat
1:41:26 Cthulhu?
1:41:37 OpenAI is also launching a browser. Yep.
1:41:39 Champy wants some cheese. Hang on. I got
1:41:41 to let him out. Champ, what's up? You
1:41:43 want out?
1:42:09 Watching
1:42:22 Damish Musk
1:42:26 hold Please,
1:42:31 [Music]
1:42:33 let me change my sharing so you can
1:42:35 actually hear this when it does it
1:42:36 because it's going to be cool, I think.
1:42:40 It might not understand the movement,
1:42:42 but that's all right. Um,
1:42:47 all right, we've got it.
1:43:00 All right, here we go. It's a cruel
1:43:02 irony, really. I have the refined pallet
1:43:04 of a feline, yet I'm constantly fighting
1:43:06 the urge to devour my own delicious
1:43:08 octopus arms. It's a cruel irony,
1:43:10 really. I have the refined
1:43:17 That's funny. Okay, let's go. We'll go
1:43:21 attach that to the end
1:43:24 of this.
1:43:26 Go add to end.
1:43:34 What happened?
1:43:44 Sit.
1:43:46 It's a cruel irony. Really?
1:43:50 Choose media. All right. Done.
1:43:54 Tabs. Oh, yeah.
1:43:59 [Music]
1:44:08 All right. So, I am in and blackboard. I
1:44:11 am in um QuickTime
1:44:15 player.
1:44:18 And so we made the first part of this
1:44:20 animation in midjourney
1:44:22 and then the hilarious
1:44:25 I have to resist eating my own delicious
1:44:27 octopus arms in V3.
1:44:32 All right, what happened?
1:44:35 All right, here we go.
1:44:43 Wait, that's not the right video. Oh,
1:44:48 okay. Here we go. We'll go full screen.
1:44:53 All right. You can all see this, right?
1:44:55 Beautiful.
1:44:56 You excited?
1:44:59 More excited than we are for Elon not
1:45:01 showing up.
1:45:05 Some creepy music behind it.
1:45:07 It's a cruel irony, really. I have the
1:45:10 refined pallet of a feline. Yet, I'm
1:45:12 constantly fighting the urge to devour
1:45:13 my own delicious octopus arms.
1:45:21 That's writing that's so bad that it's
1:45:23 good.
1:45:27 All right.
1:45:29 My own delicious octopus arms.
1:45:32 I'm constantly fighting the urge to eat
1:45:34 my own delicious octopus arms. All
1:45:37 right. Come on, Elon. Where the [ __ ] are
1:45:40 you?
1:45:42 Oh, you put the live stream here.
1:45:45 Live demo.
1:45:49 Try it on Grock.
1:45:53 Grock 3. No. Grock 4.
1:46:05 Grock for iOS.
1:46:17 Tune in for the live demo.
1:46:20 The world's most powerful AI assistant.
1:46:25 When is the live stream? The live stream
1:46:27 is now 46 minutes
1:46:31 into another dimension.
1:46:40 I guess we'll go till 10, then I'll call
1:46:41 it.
1:46:45 I was excited about this. I've done this
1:46:48 before. These things can be fun. But,
1:46:51 you know, I'm not going to be abused by
1:46:54 one Elon Musk.
1:47:00 That was rude. Okay. Did El Don fall out
1:47:03 of a window? I think he's just I don't
1:47:06 know snacking on some ketamine gummies
1:47:10 or something. I don't who knows.
1:47:30 Oh, let's go. Let's go to X. Let's go to
1:47:33 X and we'll go
1:47:36 we'll do a search for SREF. SR EF. Bang.
1:47:43 Or we do No, you know what we'll do?
1:47:44 We'll do midjourney. Mid
1:47:47 journey mood board.
1:47:56 An embedding space for M midjourney
1:47:58 SRFs.
1:48:04 Oh, that's a cool one.
1:48:08 Is this midjourney?
1:48:12 No sref
1:48:14 just a prompt.
1:48:27 Is that mine? Yeah, it's mine.
1:48:29 I remember when we made this. Remember
1:48:31 we we made this in here. I think this
1:48:34 has audio
1:48:36 [Music]
1:48:37 here. You want to see a movie we made?
1:48:39 When did we make this? We made this
1:48:42 think this is like January or something.
1:48:44 December. December of 2024. Gareth.
1:48:51 It's going to happen at 1001. I know
1:48:54 because I drop out at 10. Well, you know
1:48:56 me. I can't say goodbye anyway. But look
1:48:58 what we did six months ago
1:49:02 right here in this live. We did this.
1:49:13 So what I did was I made a mood board of
1:49:15 my sky photography and then I made a
1:49:18 bunch of midjourney images and then I
1:49:21 think we put them into I think it was
1:49:22 lumalabs to animate them. But now we
1:49:25 could do this right inside Midjourney,
1:49:27 which is super cool. Oh, this was Runway
1:49:29 ML. Runway ML did the animation. Sunno
1:49:33 did the music. Midjourney did the
1:49:34 images. My photography served as the
1:49:37 mood board.
1:49:39 And you know, being the photographer,
1:49:42 these these skies are in the
1:49:45 neighborhood of the kind of stuff I
1:49:46 would photograph.
1:49:53 [Music]
1:50:00 Heat.
1:50:03 [Applause]
1:50:05 [Music]
1:50:16 Heat.
1:50:19 [Music]
1:51:11 Nice. Nice. Nice. I like it.
1:51:16 It was very Philip Glass and Jeffrey
1:51:18 Reio. It was like Coana Scotsy without
1:51:21 the talent.
1:51:25 [Laughter]
1:51:29 [Music]
1:51:37 [Music]
1:51:52 Oh, here's another prediction. If they
1:51:55 go tonight, which they probably won't,
1:51:58 um, they won't even mention that they're
1:51:59 late.
1:52:06 Swanka dink dink dang swanka gank gad.
1:52:16 Oh, champ's back.
1:52:20 He got bored out there. He got bored.
1:52:23 There's never enough cheese.
1:52:27 Uhuh.
1:52:30 Actually, some of these styles
1:52:34 Here's
1:52:36 a cool Okay. Cinematic shot of
1:52:39 Let's grab this.
1:52:44 Let's go to midjourney.
1:52:46 We're going to the midjourney.
1:52:54 Okay. So, let's see.
1:52:58 All right.
1:53:00 Don't forget, one of the things that we
1:53:02 don't want to do in 2025 is use our
1:53:04 brain. So, we're gonna Oh, you can't see
1:53:07 any of this because I'm not sharing the
1:53:09 right tab. Dang it. You know,
1:53:14 okay, so I'm over in Midney. So, what I
1:53:16 did was I was over in the Twitter. I
1:53:18 found a look and feel that I liked. So,
1:53:22 an SREF and there there was a a little
1:53:24 bit of a prompt there. And so, here's
1:53:27 what I'm going to do. I'm going to copy
1:53:29 this. We're going to go to chat jpe.
1:53:32 We're going to assume that Brandon's
1:53:33 going to tell us if the live stream
1:53:34 starts. I'm going to say, "Hey, comma, I
1:53:39 found this prompt
1:53:42 for midjourney.
1:53:46 I want
1:53:49 you to create
1:53:52 a
1:53:53 20
1:53:55 image
1:53:57 storyboard
1:53:59 concept
1:54:01 where
1:54:04 no dialogue is necessary
1:54:09 and each shot is a
1:54:16 single subject
1:54:20 and then I'm going to give there's the
1:54:22 prompt.
1:54:25 Um, give me
1:54:29 an a
1:54:31 short overview
1:54:35 of the of
1:54:38 the story.
1:54:46 and then
1:54:48 20 different prompts
1:54:53 in the
1:54:55 format here
1:54:58 with that specific
1:55:01 sref
1:55:03 sref.
1:55:05 All right,
1:55:08 that's why I'm not watching tabs.
1:55:12 No, I'm good now. Right, we're watching
1:55:14 it now. The last lightkeeper. A desolate
1:55:17 world where oceans have swallowed nearly
1:55:20 everything. A lone lightkeeper maintains
1:55:22 the functional lighthouse on Earth.
1:55:25 Okay. Cinematic shot of a cloaked figure
1:55:29 standing alone. Ancient
1:55:33 lighthouse. Gloved hands.
1:55:39 Okay.
1:55:41 So now
1:55:46 I don't think we need that d-p, do we?
1:55:48 Oh, that's just personalized.
1:55:52 Personalization turned on. Okay. Copy.
1:55:55 Now we're going to go to midjourney.
1:55:59 Paste.
1:56:01 Go.
1:56:04 Chat GPT. Next shot, please. Copy. mid
1:56:08 journey paste. Oh, wait. What's my Do we
1:56:13 want stylization that high? Let's see
1:56:16 what the first ones look like. We might
1:56:17 go lower than that.
1:56:29 [Music]
1:56:31 A cloaked figure standing alone on a
1:56:33 cliffside as the waves crash below.
1:56:37 Yes, brother. Cheerio. Cheerio. Pep pep.
1:56:43 Carrying on. I think we need to go
1:56:46 lower. So, let's do change the prompt
1:56:49 back. We'll do this at
1:56:52 150 maybe.
1:56:55 And this is in draft mode. Let's leave
1:56:57 it in draft mode. Okay, there's that.
1:57:00 Here's the new one. Paste.
1:57:05 Go.
1:57:06 Go back here.
1:57:09 Did it give us the same? Yes, it kept
1:57:13 the SRF the same, which that was one of
1:57:15 my fears is it wouldn't. So now there's
1:57:18 our third shot.
1:57:20 Generate.
1:57:22 Let me go back here. We grab the fourth
1:57:24 shot. Uhoh. Looks like the live stream
1:57:27 might be there. Copy. Nope. Nothing.
1:57:29 False alarm. Carrying on. Cheerio. Pip
1:57:32 pip. Sir, may I inquire? Yes, sir. Go
1:57:34 on. Carry on. Oh yes, rather. Uh
1:57:38 would you be so kind uh to reveal if you
1:57:43 would not to be too forward?
1:57:50 What by chance are your qualifications,
1:57:52 sir?
1:57:54 My qualifications? Well, rather uh it's
1:58:00 a uh it's a rather direct question.
1:58:04 Uh,
1:58:07 you know, one could take offense to the
1:58:09 implication.
1:58:12 I'm not that man. I'm a far bigger man
1:58:15 than that. My qualifications. Yes. Well,
1:58:18 uh,
1:58:20 there
1:58:22 many years. Many years.
1:58:26 Uh, uh, uhuh.
1:58:29 Uh, do these look like the same
1:58:32 style? Yeah, they do. They do. All
1:58:36 right.
1:58:38 So, wait. So, we've got Okay. Shot one,
1:58:43 shot two, shot three, shot four. Did I
1:58:47 just copy shot five? No, I did not. Shot
1:58:51 five. Copying shot five.
1:58:54 What are you doing? Well, we are making
1:58:57 a piece of cinema. We're making cinema
1:59:01 using midjourney. Midjourney and cinema.
1:59:05 How we are doing that is we've had Chad
1:59:07 Gibb. I don't know if you've heard of
1:59:09 Chad GB. It's rather clever. You you
1:59:11 know the British uh thank you
1:59:14 Shakespeare are are quite quite uh the
1:59:17 linguists
1:59:19 and this chat GPT well it gives me a run
1:59:23 for my money. I do say the thesaurus has
1:59:27 worn thin.
1:59:32 Oh,
1:59:34 I'm clever. I am clever and humble.
1:59:38 Humble. Clever.
1:59:41 Dashing. I've been told
1:59:48 I I could use some some work on the
1:59:50 teeth. Yes. Yes. and the body and the
1:59:55 face, frankly, and the hair um are are
2:00:00 not quite
2:00:02 where apparently the uh adoring public
2:00:05 would like them to be. But
2:00:08 we are making cinema.
2:00:11 And as a cinema crafter,
2:00:14 one working on a chain of craft, as you
2:00:17 might say,
2:00:25 [Music]
2:00:30 it does look like they're going to be
2:00:31 going live tonight. All right, fine.
2:00:34 Well, let's keep making our movie.
2:00:38 Yes. Carrying on. rather cherryop pip.
2:00:41 So what I'm doing is I've got prompts
2:00:43 for
2:00:45 20 different shots.
2:00:49 I didn't look at them because I'm
2:00:52 committed to AI slop.
2:00:55 I don't want there to be anything on the
2:00:57 internet that isn't a waste of
2:00:59 everyone's time.
2:01:01 So therefore, I have chat GPT make a
2:01:05 random shot list of a random story
2:01:10 in a random style
2:01:12 and I'm going to tell that story and it
2:01:15 might be good.
2:01:17 It very well might be good. Uh but if it
2:01:20 good is good, uh it doesn't matter
2:01:23 because I've stolen it anyway. I've used
2:01:25 AI and if you use AI, you've stolen from
2:01:28 everyone. So, so Thievery Incorporated,
2:01:32 right?
2:01:33 Hey, Kyle, we're uh stream is up.
2:01:36 Stream is up. Okay, fine, fine, fine.
2:01:40 All right, let me go turn this off. Stop
2:01:44 acting silly. Share that. Uh, share
2:01:48 that.
2:01:48 It does still say it's going to be in
2:01:50 shortly, but now it's animated.
2:01:52 Oh, all right. Well, then I'll keep I'll
2:01:55 keep I'll go to it.
2:02:01 Okay,
2:02:03 there we go. We'll begin shortly.
2:02:06 I'll turn that off. Okay. Or maybe I'll
2:02:08 just make it low. But I think if I make
2:02:10 it low, it's going to even be
2:02:20 Okay. All right. Um, I'm going to You're
2:02:23 just going to see that. I'm I'm going to
2:02:25 keep going and making shots
2:02:32 for our cinema.
2:02:34 Our cinema.
2:02:36 I didn't realize you were in the cinema.
2:02:40 Pray tell. When did you become a
2:02:41 filmmaker? Oh, rather I went to chatbt
2:02:45 and I said, "How do you make a film?"
2:02:47 And it said, "Well, here you go. Just
2:02:49 paste this in and do nothing. You are a
2:02:52 talentless oath.
2:03:00 I embrace my talentlessness.
2:03:06 I wonder if I missed a shot. So, this is
2:03:09 17.
2:03:12 All right. So, let's see.
2:03:14 We've got one, two, three, four, five,
2:03:20 six, seven,
2:03:22 eight, nine, 10, 11, 12, 13, 14,
2:03:30 15, 16, 17. Great. Perfect.
2:04:03 [Music]
2:04:13 All right.
2:04:19 I'm making editorial choices without you
2:04:21 all. Hope you're good with that.
2:04:39 [Music]
2:05:22 Oh,
2:05:29 in a world where knowledge shapes
2:05:32 destiny, one creation dares to redefine
2:05:35 the future. From the minds at XAI,
2:05:38 prepare for Gro 4. This summer, the next
2:05:41 generation arrives faster, smarter,
2:05:44 bolder. It sees beyond the horizon.
2:05:47 The unasked and challenges the
2:05:49 impossible. Gro 4 unleash the truth.
2:05:53 Coming this summer.
2:05:55 That was cheesy.
2:05:58 All right, welcome to the Gro 4 release
2:06:00 here. Um, this is uh the smartest AI in
2:06:04 the world and we're going to show you
2:06:05 exactly how and why. Um, and uh, it
2:06:10 really is
2:06:12 remarkable to see the advancement of
2:06:14 artificial intelligence, how quickly it
2:06:16 is uh, evolving. Um,
2:06:22 I I sometimes think compare it to the
2:06:25 growth of
2:06:27 a human and how fast a human learns and
2:06:30 gains conscious awareness and
2:06:32 understanding and AI is advancing
2:06:37 just vastly faster than any human. Um,
2:06:42 I mean, we're going to take you through
2:06:44 a bunch of benchmarks that um that Gro 4
2:06:48 is able to uh achieve incredible numbers
2:06:52 on. Um, but it's it's it's actually
2:06:54 worth noting that that Grock 4 if if
2:06:58 given like the SAT would get perfect
2:07:01 SATs every time, even if it's never seen
2:07:03 the uh the questions before. Um and if
2:07:07 even going beyond that to say like uh
2:07:10 graduate student exams like the GRE uh
2:07:13 it will get near-perfect results in in
2:07:17 every discipline of uh of education. So
2:07:21 from the humanities to like languages,
2:07:25 math, physics, engineering, pick
2:07:27 anything and we're talking about
2:07:30 questions that it's never seen before.
2:07:31 These are not not on the internet and
2:07:34 it's
2:07:35 Grock for is smarter than
2:07:38 almost all graduate students uh in all
2:07:41 disciplines simultaneously
2:07:44 like it's actually just important to
2:07:46 appreciate the
2:07:47 I'm pretty sure that's AGI
2:07:49 uh really something
2:07:51 um
2:07:52 and uh the the reasoning capabilities of
2:07:56 Grock are incredible. Um, so there's
2:07:59 some people out there who who think AI
2:08:01 can't reason and look it they can reason
2:08:03 at superhuman levels. Um, so yeah, and
2:08:08 frankly it it only gets better from
2:08:10 here. So we'll we'll take you through uh
2:08:14 the Gro 4 release and uh
2:08:18 yeah um show you like the pace of pace
2:08:22 of progress here. Um like I guess the
2:08:26 first part is like in terms of the
2:08:28 training um
2:08:29 a little less awkward
2:08:30 going from grock two to Grock 3 uh to
2:08:33 Grock 4 we've essentially increased the
2:08:35 training by an order of magnitude in
2:08:36 each case. So it's a 100 times more
2:08:40 training than than Grock 2. Um and uh
2:08:44 and that that's only going to increase.
2:08:47 Um so
2:08:49 it's uh yeah frankly I mean I don't know
2:08:52 in some ways a little terrifying but uh
2:08:56 the growth of intelligence here is is
2:08:57 remarkable.
2:08:58 Yes it's important to realize there are
2:09:01 two types of training compute. One is
2:09:02 the pre-training compute that's from GR
2:09:05 two to GR three. Um but for from GR
2:09:08 three to GU four we're actually putting
2:09:10 a lot of compute in reasoning in RL
2:09:13 possible could be the lighting
2:09:15 just like you said this is literally the
2:09:17 fastest moving field and GR 2 is like
2:09:19 the high school student by today's
2:09:21 standard if you look back in the last 12
2:09:24 months GR 2 was only a concept we didn't
2:09:27 even have GR 2 12 months ago um and then
2:09:30 by training GR 2 that was the first time
2:09:32 we scale up like the pre-training we
2:09:34 realized that if you actually do the
2:09:35 data oblation really care
2:09:37 teacher loving using AI in education and
2:09:39 also the algorithm super super power
2:09:42 quite a lot by amount of 10x
2:09:44 other work yeah it's awesome
2:09:46 best pre model and that's why
2:09:49 critical thinking education model H100
2:09:54 and then with the best pre model and we
2:09:57 realized if you can collect these
2:09:58 verifiable outcome reward you can
2:10:01 actually train this model to start
2:10:02 thinking from the first principle start
2:10:03 to reason correct his own mistakes and
2:10:05 that's where the reasoning comes from.
2:10:07 And today we ask the question, what
2:10:09 happens if you take the expansion of
2:10:11 classes with all 200,000 GPUs, put all
2:10:14 these into RL 10x more compute than any
2:10:19 of the models out there on reinforcement
2:10:21 learning, unprecedented scale, what's
2:10:23 going to happen. So this is a story of
2:10:26 uh Gro 4 and uh you know Tony uh share
2:10:30 some uh insight with the audience.
2:10:32 Yeah. Um so yeah let's just talk about
2:10:35 how smart graph 4 is. So I guess um we
2:10:38 can start discussing this benchmark
2:10:40 called humanities last exam and this
2:10:43 this benchmark is a very very
2:10:45 challenging benchmark. Every single
2:10:47 problem is curated by subject matter
2:10:51 experts. Um it's in total 2500 problems
2:10:55 and it consists of many different
2:10:57 subjects mathematics natural sciences uh
2:11:00 engineering and also auto humanity
2:11:02 subjects. So um essentially when when it
2:11:06 was first released actually like earlier
2:11:08 this year uh most of the models out
2:11:11 there can only get singledigit accuracy
2:11:14 on this benchmark.
2:11:16 Yeah. So we can we can look at some of
2:11:18 th those examples. Um you know uh so
2:11:21 there there is this mathematical problem
2:11:24 which is about natural transformations
2:11:26 in category theory and there's this
2:11:29 organic chemistry problem that talks
2:11:31 about uh electro uh cyclic reactions and
2:11:35 also there's this linguistic problem
2:11:37 that tries to ask you about
2:11:39 distinguishing between closed and open
2:11:41 syllabus uh from a hub Hebrew source
2:11:44 text. So you can see also it's a very
2:11:48 wide range of problems and every single
2:11:51 problem is PhD or even advanced research
2:11:54 level problems.
2:11:55 Yeah. I mean these there are no humans
2:11:58 that can actually answer these can get a
2:12:00 good score. I mean if you actually say
2:12:02 like any given human um what like what's
2:12:06 the best that any human could score? I
2:12:09 mean I'd say maybe 5% optimistically.
2:12:13 Yeah.
2:12:13 Um, so this this is much harder than
2:12:17 than what any any human can do. It's
2:12:18 it's incredibly difficult and you can
2:12:20 see from the the types of questions like
2:12:22 you might be incredible in linguistics
2:12:24 or mathematics or chemistry or physics
2:12:27 or any one of a number of subjects, but
2:12:28 you're not going to be um at a postgrad
2:12:31 level in everything. And Grock 4 is a
2:12:34 postgrad level in everything.
2:12:36 Yeah.
2:12:37 Like it's it's just some of these things
2:12:39 are just worth repeating. like grock
2:12:41 boore is postgraduate like PhD level in
2:12:46 everything better than PhD but like most
2:12:48 PhDs would fail so it's better said I
2:12:51 mean at least with respect to academic
2:12:53 questions it I want to just emphasize
2:12:56 this point with respect to academic
2:12:58 questions gro is better than PhD level
2:13:01 in every subject no exceptions
2:13:04 wow
2:13:05 um now this doesn't mean that it's it
2:13:08 you know times it may lack common sense
2:13:11 and it has not yet invented new
2:13:13 technologies or discovered new physics
2:13:17 but that is just a matter of time.
2:13:19 Um if it I I I think it may discover new
2:13:22 technologies
2:13:23 uh as soon as later this year. Um and I
2:13:28 I would be shocked if it has not done so
2:13:29 next year. So I would expect Grock to
2:13:33 yeah literally discover new new
2:13:35 technologies that are actually useful no
2:13:36 later than next year and maybe end of
2:13:38 this year. Um, and it might discover new
2:13:41 physics next year. And within two years,
2:13:44 I'd say almost certainly
2:13:46 like, so just let that sink in.
2:13:48 Within two years, new physics.
2:13:53 Yeah.
2:13:57 So, yeah. Um, how Okay. So, I guess we
2:14:00 can talk about the the what what's
2:14:02 behind the scene of 4. As Jimmy
2:14:04 mentioned, uh, we actually throwing a
2:14:07 lot of compute into this training. You
2:14:10 know, when it started, it's only also a
2:14:13 single digit. Sorry. Um, the previous
2:14:15 slide, sorry. Yeah, it's only a single
2:14:18 digit uh number.
2:14:21 But as you start putting in more and
2:14:23 more training compute, it started to
2:14:25 gradually become smarter and smarter and
2:14:28 eventually solved a quarter of the HL
2:14:31 problems. And this is without any tools.
2:14:35 The next thing we did was to adding a
2:14:38 tools capabilities to the model. And
2:14:41 unlike GR 3, I think you know G3
2:14:44 actually is able to use C as well. But
2:14:46 here we actually make it more native in
2:14:48 a sense that we put the tools into
2:14:50 training. Uh G3 was only relying on
2:14:54 generalization. Here we actually put the
2:14:56 tools into
2:14:57 Okay. When they say they put the tools
2:14:58 into training, what they mean is if it
2:15:01 runs into a problem where it needs to
2:15:03 like write Python code to solve a
2:15:05 problem that's now it's been trained on
2:15:08 that. So it's probably going to be very
2:15:10 very good at
2:15:11 the G3 reasoning model uh but without
2:15:14 any specific training but we only ask it
2:15:17 to use those tools. So compared to this
2:15:20 it was much weaker in terms of its tool
2:15:22 capabilities
2:15:23 and unreliable
2:15:25 and unreliable. Yes. Yes. And and to be
2:15:27 clear like these are still I'd say
2:15:29 fairly this is still fairly primitive
2:15:31 tool use. If you compare it to say the
2:15:33 tools that are used at Tesla or SpaceX
2:15:35 uh where you're using um you know finite
2:15:39 element analysis and computational flow
2:15:42 dynamics and and you're you're able to
2:15:45 run uh or say like Tesla does like crash
2:15:48 simulations where the simulations are so
2:15:50 close to reality that if the test
2:15:52 doesn't match the simulation you assume
2:15:54 that the test article is wrong. That's
2:15:56 how good the simulations are. So, Grock
2:15:58 is not currently using any of the tools
2:16:01 that the really powerful tools that a
2:16:03 company would use, but but that is
2:16:05 something that we will provide it with
2:16:07 later this year. So, we'll have the
2:16:09 tools that that a company has. Um,
2:16:12 sophisticated tools later this year,
2:16:14 physics simulator. Um, ultimately the
2:16:18 the thing that will make the biggest
2:16:19 difference is being able to interact
2:16:21 with the real world via humanoid robots.
2:16:23 So you combine sort of Grock with with
2:16:25 Optimus and it can actually interact
2:16:27 with the real world.
2:16:28 So is Grock 4 on the spectrum? I think
2:16:31 definitely
2:16:32 formulate hypothesis and then confirm if
2:16:35 that hypothesis is is true or not.
2:16:38 Um
2:16:39 so we're really, you know, think about
2:16:42 like where we are today. We're we're at
2:16:44 the beginning of an immense intelligence
2:16:47 explosion. We're in we're in the
2:16:49 intelligence big bang right now.
2:16:52 Yep. Um,
2:16:54 and the mo we're at the most interesting
2:16:56 time to be alive of any time in history.
2:16:59 Yeah, agree.
2:17:01 Yeah. Now, that said, we need to make
2:17:04 sure that the AI is um a good AI.
2:17:09 Uh, good Brock. Um, and the the thing
2:17:12 that I think is most important for AI
2:17:14 safety, at least my biological neural
2:17:16 net tells me the most important thing
2:17:18 for AI is to be maximally truth
2:17:21 seeeking. So this is this is a very
2:17:24 fundamental um like you can think of AI
2:17:28 as this super genius child that
2:17:31 ultimately will outsmart you but you can
2:17:33 still in you can instill the right
2:17:35 values um and encourage it to be sort of
2:17:40 you know
2:17:42 uh truthful
2:17:45 uh
2:17:46 I don't know honorable you know good
2:17:48 good things like
2:17:52 the values you want to instill in in a
2:17:54 child that that that grow would grow
2:17:58 ultimately grow up to be incredibly
2:17:59 powerful.
2:18:00 Um
2:18:04 yeah.
2:18:06 So yeah, so these this is really I say
2:18:09 we say we say tools these are say still
2:18:11 primitive tools not the kind of tools
2:18:14 that um that serious commercial
2:18:17 companies use but we will provide it
2:18:20 with those tools and uh I think it will
2:18:22 be able to solve with those tools real
2:18:24 world technology problems. In fact I'm
2:18:26 I'm certain of it. It's just a question
2:18:27 of how long it takes.
2:18:29 Yes. Yes, exactly. Um,
2:18:31 so is it just compute all you need,
2:18:33 Tony?
2:18:35 Sorry.
2:18:35 Is it just compute all you need at this
2:18:37 point?
2:18:39 Well, you you need compute plus plus uh
2:18:42 the right tools.
2:18:43 Mhm.
2:18:43 And um and then ultimately to be able to
2:18:46 interact with physical world.
2:18:48 Yes.
2:18:48 Um
2:18:49 all right, let's talk about what it is
2:18:50 now and when we can play with it. I mean
2:18:52 we'll effectively have an economy that
2:18:54 is
2:18:56 well ultimately an economy that is
2:18:59 thousands of times bigger than our
2:19:00 current economy or maybe millions of
2:19:03 times. I mean if you if you think of
2:19:05 civilization as percentage completion of
2:19:08 the kadesv scale where kadeshv one is
2:19:13 using all the energy output of a planet
2:19:15 and kv two is using all the energy
2:19:18 output of a sun and three is all the
2:19:20 energy output of a galaxy. We're we're
2:19:23 only in my opinion probably like close
2:19:26 closer to 1% of kartev one um than we
2:19:30 are to 10%. So
2:19:33 like maybe a point or one one or two
2:19:35 percent of kadeshv one. So
2:19:39 um we we will get to
2:19:42 um most of the way like 80 90% kesh of
2:19:46 one and then hopefully movies
2:19:50 self annihilate the waffles
2:19:53 like it's the the actual notion of of a
2:19:56 human economy assuming civilization
2:19:58 continues to progress will seem very
2:20:00 quaint. Um in in retrospect,
2:20:03 um it will it will seem like uh
2:20:06 sort of cavemen throwing sticks into a
2:20:09 fire uh level of economy
2:20:12 um compared to what the future will
2:20:13 hold.
2:20:15 Um I mean it's very exciting. I mean, I
2:20:18 I've been at times kind of worried about
2:20:20 like, well,
2:20:23 you know, is this this this seems like
2:20:26 it's it's somewhat unnerving to have
2:20:28 intelligence created that is far greater
2:20:31 than our own. Um, and will this be bad
2:20:35 or good for humanity? Um,
2:20:39 it's like I I I think it'll be good.
2:20:41 Most likely it'll be good.
2:20:43 Jesus Christ.
2:20:45 My god. Yeah,
2:20:48 dude.
2:20:49 Yeah, but I somewhat reconciled myself.
2:20:51 Sober up fact that
2:20:53 even if even if it wasn't going to be
2:20:56 good, I'd at least like to be alive to
2:20:57 see it happen.
2:21:00 So yeah.
2:21:04 So actually one one Yeah. Yeah. I think
2:21:08 one one technical problem that we still
2:21:11 need to solve besides just compute is
2:21:13 how do we unblock the data uh data um
2:21:16 bottleneck u because um when we try to
2:21:19 scale up the RL uh in this case we did
2:21:23 invent a lot of new techniques
2:21:25 innovations to allow us to figure out
2:21:28 how to find a lot of a lot of
2:21:30 challenging RL problems to work on. It's
2:21:33 not just a problem itself needs to be
2:21:34 challenging but also it needs to be um
2:21:37 you also need to have like a reliable uh
2:21:40 signal to tell the model you did it
2:21:42 wrong you did it right. This is the sort
2:21:44 of the principle of reinforcement
2:21:45 learning and as the models get smarter
2:21:48 and smarter the number of cool problem
2:21:51 or challenging problems will be lesser
2:21:53 and less.
2:21:54 Yeah.
2:21:54 So it's going to be a new type of
2:21:57 challenge that we need to surpass
2:21:58 besides just compute. Yeah.
2:22:00 Yeah. And we actually are running out of
2:22:03 of of actual test questions to ask. Uh
2:22:05 so there's like even ridiculously
2:22:08 questions that are ridiculously hard if
2:22:10 not essentially impossible for humans
2:22:12 that are written down questions um are
2:22:15 uh becoming swiftly becoming trivial for
2:22:18 for AI.
2:22:20 Um so then there's
2:22:23 um but you know the the one thing that
2:22:25 is an excellent judge of things is
2:22:28 reality. So because physics is the law
2:22:31 ultimately everything else is a
2:22:32 recommendation. You can't break physics.
2:22:35 Um so the ultimate test I think for
2:22:37 whether an AI is um the the ultimate
2:22:41 reasoning test is reality.
2:22:43 Yes.
2:22:43 So you invent a new technology like say
2:22:45 improve the design of a car or a rocket
2:22:48 or um create a new medication.
2:22:52 Uh that's and and and does it work?
2:22:55 Yeah.
2:22:55 Um does does the rocket get to orbit?
2:22:57 Does the does the car drive? Does the
2:23:00 medicine work? Whatever the case may be.
2:23:03 Um, reality is the ultimate judge here.
2:23:05 Um, so it's it's going to be a
2:23:07 reinforcement learning closing loop
2:23:09 around reality.
2:23:14 We asked the question how do we even go
2:23:16 further? So um actually we are thinking
2:23:19 about now with single agent we are able
2:23:22 to solve 40% of a problem.
2:23:25 What if we have multiple agents running
2:23:27 at the same time? So this is what's
2:23:29 called test and compute. And as we scale
2:23:32 up the test and compute actually we are
2:23:35 able to solve almost more than 50% of
2:23:38 the uh taxon subset of the h problems.
2:23:41 So it's a remarkable achievement I
2:23:44 think. Yeah.
2:23:45 Yeah. This is this is this is insanely
2:23:46 difficult. These are it's it's so what
2:23:49 we're saying is like a majority of the
2:23:50 of the of the text based um of
2:23:54 humanities you know scarily named
2:23:56 humanities last exam um GR 4 can solve
2:23:59 and you and you can try it out for
2:24:01 yourself um and the with the GR 4 heavy
2:24:04 what what it does is it spawns multiple
2:24:06 agents in parallel and all of those
2:24:08 agents do do work independently and then
2:24:12 they compare their work and they they
2:24:15 decide which one like it's a study
2:24:17 group.
2:24:18 Um, and it's not as simple as a majority
2:24:21 vote because often only one of the
2:24:23 agents actually figures out the trick or
2:24:26 figures out the solution. Um, and and
2:24:29 and but once they share the the trick or
2:24:31 or or figure out what what the real
2:24:34 nature of the problem is, they share
2:24:35 that uh solution with the other agents
2:24:38 and then they compare they essentially
2:24:40 compare notes and then and and then
2:24:42 yield yield an answer. So that's that's
2:24:45 the the heavy part of Grful is is where
2:24:48 we you scale up the test time compute by
2:24:51 roughly an order of magnitude uh have
2:24:53 multiple agents
2:24:55 uh tackle the task and then they compare
2:24:57 their work and they they put forward
2:25:00 what they think is the best result.
2:25:03 Okay.
2:25:04 Yeah. So we will introduce graph 4 and
2:25:07 graph for heavy. Sorry can you click the
2:25:08 next four heavy?
2:25:10 Yes. So yeah, so basically Gu 4 is a
2:25:14 single version, a single agent version
2:25:16 and graph for heavy is the multi- aent
2:25:19 version.
2:25:21 So let's take a look how they actually
2:25:23 do on those exam problems and also some
2:25:26 real life problems.
2:25:27 Yeah.
2:25:28 So we're going to start out here and
2:25:29 we're actually going to look at one of
2:25:31 those HLE problems. This is uh actually
2:25:34 one of the easier math ones. Uh I don't
2:25:36 really understand it very well. I'm not
2:25:38 that smart. But I can launch this job
2:25:40 here and we can actually see how it's
2:25:42 going to go through and start to think
2:25:43 about this problem. Uh while we're doing
2:25:45 that, I also want to show a little bit
2:25:47 more about like what this model can do
2:25:49 and launch a uh Gro 4 heavy as well. So
2:25:52 everyone knows Poly Market. Um it's
2:25:54 extremely interesting. It's the you know
2:25:56 seek curve of truth. It aligns with what
2:25:58 reality is most of the time. And with
2:26:01 Grock, what we're actually looking at is
2:26:03 being able to see how we can try to take
2:26:05 these markets and see if we can predict
2:26:08 the the future as well. So, as we're
2:26:10 letting this run, we'll see how uh Gro
2:26:13 for Heavy goes about uh predicting the,
2:26:16 you know, the World Series odds for like
2:26:18 the current teams in the MLB. And while
2:26:20 we're waiting for these to process,
2:26:22 we're going to pass it over to Eric and
2:26:23 he's going to show you an example of
2:26:25 his.
2:26:29 Yeah. So uh I guess one of the coolest
2:26:32 things about uh Grapore is its ability
2:26:35 to understand the world and to solve
2:26:38 hard problems by leveraging tools like
2:26:41 Tony discussed. And I think one kind of
2:26:43 cool example of this we asked it to
2:26:46 generate a visualization of two black
2:26:49 holes colliding. Um and of course you
2:26:53 know it took some there are some
2:26:55 liberties. It's in my case actually
2:26:56 pretty clear in its thinking trace about
2:26:58 what these liberties are. Uh for
2:27:00 example, in order for it to actually be
2:27:02 visible, you need to really exaggerate
2:27:04 the the scale of the you know the uh the
2:27:09 uh the waves. And yeah, so here's like
2:27:13 you know this kind of inaction. Um it
2:27:17 exaggerates the scale in like multiple
2:27:19 ways. uh it drops off a bit less in
2:27:21 terms of amplitude um over distance and
2:27:27 uh but yeah we can kind of see uh the
2:27:30 basic effects that you know are actually
2:27:32 like you know correct. It starts with
2:27:34 the inspiral it merges uh and then you
2:27:38 have um the ring down and like this is
2:27:41 basically um largely correct um yeah uh
2:27:47 modulo some of the simplications I need
2:27:49 to do um you know it it's actually quite
2:27:52 explicit about this you know it uses
2:27:54 like post postneonian approximations
2:27:57 instead of actually like computing the
2:27:59 general relativistic effects at like uh
2:28:02 near the center of the black hole which
2:28:04 is you know incorrect um and you know
2:28:06 will lead to you know some incorrect
2:28:08 results but the overall you know
2:28:10 visualization is uh yeah is basically
2:28:14 there um and you can actually look at
2:28:16 the kinds of resources that it
2:28:18 references. So here it um it actually
2:28:22 you know it obviously uses search it
2:28:24 gathers results from a bunch of links
2:28:26 but also reads through a undergraduate
2:28:28 text in analytical analytic
2:28:32 gravitational wave models. Uh it um yeah
2:28:37 it um reasons quite a bit about u uh the
2:28:42 actual constants that it should use for
2:28:44 a realistic simulation. uh it references
2:28:47 uh I guess existing real world data. Um
2:28:51 and yeah I it yeah it's a it's a pretty
2:28:55 good model.
2:28:57 Yeah
2:28:58 but like actually going forward we can
2:29:00 we can we can play we can give it the
2:29:02 same model that physicists use. So it
2:29:04 can run the the same level of compute
2:29:07 that uh so leading physics researchers
2:29:10 are using and and give you a physics
2:29:12 accurate black hole simulation.
2:29:15 Exactly.
2:29:16 just right now is running in your
2:29:17 browser. So
2:29:18 yeah, this is just running in your
2:29:19 browser. Exactly.
2:29:20 Pretty simple.
2:29:21 So swapping back real quick here, we can
2:29:23 actually take a look. The math problem
2:29:25 is finished. Uh the model was able to uh
2:29:28 let's look at its thinking trace here.
2:29:30 So you can see how it went through the
2:29:32 problem. I'll be honest with you guys, I
2:29:34 really don't quite fully understand the
2:29:36 math, but what I do know is that ahead
2:29:38 of time
2:29:40 um and it did come to the the correct
2:29:42 answer here in the final part here.
2:29:44 Good lord. We can also come in and
2:29:46 actually take a look here at our uh our
2:29:48 World Series prediction. Um, and it's
2:29:51 still thinking through on this one, but
2:29:53 we can actually try some other stuff as
2:29:55 well. So, we can actually like try some
2:29:56 of the X integrations that we did. So,
2:29:58 we worked very heavily on working with
2:30:01 uh all of our X tools and building out a
2:30:03 really great X experience. So, we can
2:30:05 actually ask, you know, the model, you
2:30:07 know, find me the XAI employee that has
2:30:09 the weirdest profile photo. Um, so
2:30:12 that's going to go off and start with
2:30:13 that. And then we can actually try out
2:30:15 you know let's uh create a timeline
2:30:17 based on expost uh detailing the you
2:30:20 know changes in the scores over time and
2:30:22 we can see you know all the conversation
2:30:24 that was taking place as well. So we can
2:30:26 see who were the you know announcing
2:30:28 scores and like what was the reactions
2:30:30 at those times as well.
2:30:31 Um so we'll let that go through here and
2:30:34 process. And if we go back to uh uh this
2:30:38 was the uh Greg Yang photo here. So, uh,
2:30:41 if we scroll through here, whoops. So,
2:30:44 Greg Yang, of course, who has his
2:30:46 favorite, uh, photograph that he has on
2:30:49 his account. Um, that's actually not how
2:30:51 he looks like in real life, by the way.
2:30:52 Um, just to get aware, but is quite
2:30:55 funny,
2:30:55 but it had to understand that question.
2:30:57 Yeah.
2:30:57 Which is that that's the wild part. It's
2:30:59 like it understands what is a weird
2:31:01 photo.
2:31:02 What is a weird photo?
2:31:03 Yeah.
2:31:04 What is a less or more weird photo?
2:31:06 It goes through it has to find all the
2:31:08 team members. It has to figure out who
2:31:09 we all are, right? you know, searches
2:31:11 without access to the internal XAI
2:31:14 personnel logs. It's literally looking
2:31:15 at the just at the internet.
2:31:17 Exactly.
2:31:17 So, you could say like the weirdest of
2:31:19 any company.
2:31:20 Yeah.
2:31:20 To be clear.
2:31:21 Exactly. And uh we can also take a look
2:31:24 here at the uh question here for the
2:31:26 humanities last exam. So, it is still
2:31:28 researching all of the historical
2:31:30 scores. Um but it will have that final
2:31:33 answer here soon. But we can while it's
2:31:34 finishing up, we can take a look at one
2:31:36 of the ones that we set up here a second
2:31:38 ago. And we can see like you know it
2:31:40 defines the date that like Dan Hendricks
2:31:41 had initially announced it. We can go
2:31:43 through we can see you know uh OpenAI
2:31:45 announcing their score back in uh
2:31:47 February and we can see you know as
2:31:50 progress happens with like Gemini. We
2:31:52 can see like Kimmy uh and we can also
2:31:54 even see you know the leaked benchmarks
2:31:56 of uh what people are saying is you know
2:31:58 if it's right it's going to be pretty
2:31:59 impressive. So pretty cool. Um so yeah
2:32:03 I'm looking forward to seeing how
2:32:04 everybody uses these tools and gets the
2:32:06 most value out of them. Um but uh yeah,
2:32:09 it's been great.
2:32:10 Yeah. And we're gonna close the loop
2:32:12 around usefulness as well. So it's like
2:32:14 it's not just books smart, but actually
2:32:16 practically smart.
2:32:18 Exactly.
2:32:26 And we can go back to the uh the slides
2:32:29 here. Yeah. So
2:32:30 cool. Um so the so we actually evaluate
2:32:34 also on the multimodel subset. So on the
2:32:37 full set uh this is the number on the h
2:32:40 exam. Uh it you can see there's a little
2:32:44 dip on the numbers. Uh this is actually
2:32:46 something we're improving on which is
2:32:48 the multimodel uh understanding
2:32:50 capabilities. But I do believe uh in a
2:32:53 very short time we're able to really
2:32:55 improve and got much higher numbers on
2:32:58 this um even higher numbers on this
2:33:00 benchmark.
2:33:02 Yeah, this is the we we saw like you see
2:33:05 like what what is the biggest weakness
2:33:06 of Grock currently is that it's it's
2:33:08 sort of partially blind. It can't it's
2:33:11 it's understanding
2:33:13 um obviously in its image. Joe, I always
2:33:16 think I really want to watch these and
2:33:18 then I realize they're super [ __ ]
2:33:20 boring. Yeah, the the like I it is this
2:33:23 is unwatchable, but you know, um it
2:33:26 looks quite impressive. Um I mean the
2:33:29 problem is you've got people that are so
2:33:32 smart but so myopic in what they're
2:33:34 focusing on, they can't imagine how
2:33:37 other people are going to use it. So we
2:33:39 get these demos that are just, you know,
2:33:41 clever or relevant to them. uh here with
2:33:44 a heavy and we can see uh here we can
2:33:46 see all the tools and the process it
2:33:48 used to actually uh go through and find
2:33:51 the right answer. So it browsed a lot of
2:33:53 odd sites. It calculated its own odds
2:33:56 comparing to the market the market to
2:33:57 find its own alpha and edge. It walks
2:33:59 you through the entire process here and
2:34:02 it calculates the odds of the winner
2:34:04 being like the the Dodgers uh and it
2:34:07 gives them a 21.6%
2:34:09 chance of uh winning uh this year.
2:34:13 So it took approx approximately four and
2:34:16 a half minutes to compute.
2:34:17 Sure.
2:34:19 Yeah.
2:34:23 That it
2:34:25 when can we play with it?
2:34:26 You can also look at all the uh the
2:34:28 other benchmarks besides HE. Um, as it
2:34:31 turned out, GR 4 excelled on all the
2:34:34 reasoning benchmarks that people usually
2:34:36 test on. Um, including GBQA, which is a
2:34:40 PhD level, uh, problem sets. Uh, that's
2:34:44 easier compared to HRE. Um, on Amy 25,
2:34:48 America invitation mathematics exam, we
2:34:51 with graph for heavy, we actually got a
2:34:53 perfect score.
2:34:54 Wow.
2:34:54 Um, also on some of the coding benchmark
2:34:57 called live coding bunch. um and also on
2:35:00 uh HMMT, Harvard math uh MIT uh exam and
2:35:05 also USMO. Uh you can see actually um
2:35:08 for on all of those benchmarks we often
2:35:11 have a very large leap against the
2:35:14 second best uh model out there.
2:35:17 Yeah, it's I mean really we're going to
2:35:19 get to the point where uh it's going to
2:35:21 get every answer right in every exam. um
2:35:24 and where it doesn't get an answer
2:35:26 right, it's going to tell you what's
2:35:27 wrong with the question. Or if the
2:35:28 question is ambiguous, disambiguate the
2:35:31 question into answers A, B, and C and
2:35:33 tell you what what answers A, B, and C
2:35:36 would be with a disambiguated question.
2:35:38 Um so the only real test then will be
2:35:41 reality. Uh can I make useful
2:35:43 technologies, discover um new science?
2:35:47 That'll actually be the the only thing
2:35:49 left because human tests will simply not
2:35:52 be um meaningful. That's crazy.
2:35:54 You can make all
2:35:55 That's crazy. We're within a year.
2:35:57 Give it a
2:35:58 We're within a year of and and OpenAI is
2:36:01 about to launch something probably
2:36:03 similar. This is this is why the
2:36:05 frontier model companies have stopped
2:36:07 talking about AGI and moved on to
2:36:09 talking about ASI that there these
2:36:13 things are starting to accelerate and
2:36:15 Grock heavy tiers we're introducing
2:36:18 where you're able to access to both Gro
2:36:19 4 and Grock 4 heavy uh where you're
2:36:22 actually going to be the taskmaster of
2:36:23 bunch of little Grock research agent to
2:36:25 help you you know become smarter through
2:36:26 all the little research
2:36:29 of you know going through mundane tasks
2:36:32 uh and it's available right now. Uh so
2:36:35 yeah, so um we we did limit usage during
2:36:40 the the demo. So we didn't it didn't
2:36:42 break the demo because all these this
2:36:43 all this stuff is happening live. So
2:36:45 it's not there's no nothing canned about
2:36:46 the any of the tests that we're doing.
2:36:49 Um so uh after the after the uh demo is
2:36:52 done, we'll we'll allow we'll enable
2:36:55 more subscribers for Super Grock. So if
2:36:57 you can't subscribe right now, just try
2:36:59 in half an hour. It should work. Um, so,
2:37:02 uh, and now let's, let's get into voice.
2:37:07 Great,
2:37:09 so many of you have been enjoying our
2:37:11 voice mode and we've been working hard
2:37:12 to improve the experience over the past
2:37:14 couple months. Um, we have cut latency
2:37:18 in half to make it much snappier.
2:37:20 And today, we're excited to announce a
2:37:22 set of new voices that have exceptional
2:37:25 naturalenness and procity. You might
2:37:27 have noticed the movie trailer voice
2:37:30 that opened up the live stream. That is
2:37:31 one of our new voices, S, who's got that
2:37:34 epically deep tone. And we're also
2:37:37 excited to introduce Eve, a beautiful
2:37:41 British voice who's capable of rich
2:37:43 emotions. Come on, would you like to
2:37:46 introduce Eve?
2:37:47 Absolutely. Yeah, let's get into the
2:37:49 demo.
2:37:51 Hey, Eve. Millions of people are
2:37:53 watching us right now. Are you excited?
2:37:56 Oh, I'm thrilled. Really, it's like
2:37:59 being on stage at the Old Vic with a sea
2:38:01 of faces out there. Just imagine all
2:38:03 those eyes and ears tuned in. It's a
2:38:06 proper buzz, isn't it? What do you
2:38:08 reckon we should chat about to keep this
2:38:09 lot entertained?
2:38:10 You know, we have a lot to talk about,
2:38:12 but I'll be honest. I'm feeling a little
2:38:13 nervous. Can you whisper something
2:38:15 soothing to calm me down?
2:38:18 Take a deep breath, love.
2:38:20 We've got this. It's just you and me
2:38:23 having a quiet chant. Like we're tucked
2:38:26 away in a cozy corner of a Yorkshire
2:38:28 pub. The world's just a murmur out
2:38:31 there. Feel that calm wash over you.
2:38:35 Let's keep talking. Nice and easy.
2:38:38 What's on your mind?
2:38:39 Starf fighter. Oh, good lord. Fake
2:38:42 girlfriends are going to go bananas.
2:38:44 Can you give us an opera on Diet Coke?
2:38:49 Oh, an opera on Diet Coke. That's a
2:38:52 riot. Picture this. A shimmering silver
2:38:56 cannon glistening under stage lights.
2:39:00 ASMR diet coke. Thou elixer divine with
2:39:05 bubbles that dance in a sparkling line.
2:39:09 Thy crisp cool kiss on lips so fine. Oh
2:39:17 forever.
2:39:20 How's that for a mad little area?
2:39:23 Want me to belt out another verse or
2:39:24 switch up the tune?
2:39:26 Okay, let's try switching it up and just
2:39:28 start singing.
2:39:31 Right, then let's have a go. Oh, silver
2:39:34 can you gleam in the light diet coke my
2:39:38 heart's delight.
2:39:39 Okay, I won't touch it. It's very very
2:39:41 poetic. Um yeah, thank you Eve. Feeling
2:39:43 quite nice now.
2:39:45 Cool. Um so that's Eve. And um so
2:39:49 something else we do here is is a little
2:39:51 test each day to check on the natural
2:39:54 flow of our conversations against some
2:39:58 of our competitors. So we figured why
2:40:00 not try that live on the stream today?
2:40:05 Should I start?
2:40:09 Hey, I'm going to say a number and can
2:40:11 you say the same number back to me? One.
2:40:14 One. Got it. This is like a fun little
2:40:16 game of echo. What's next?
2:40:18 Two.
2:40:22 Oh, you said two.
2:40:26 Three.
2:40:30 Four. Nice and steady. What's next?
2:40:33 Um. Uh. Five.
2:40:38 Five. Nailed it. This is pretty fun,
2:40:40 actually. Feel free to keep going or
2:40:42 switch it up if you'd like.
2:40:47 Hey, Grock.
2:40:51 Hey, Grock.
2:40:53 Hello there. How can I help you today?
2:40:55 We're going to do a little test. Can you
2:40:56 repeat after me? One.
2:40:59 One. All right. What's next?
2:41:01 Two.
2:41:03 Two. What's on your mind?
2:41:05 Three.
2:41:07 Three. Need anything else?
2:41:09 Four.
2:41:10 Four. How can I assist you?
2:41:12 Five.
2:41:14 Five. What's next?
2:41:17 So, as you can see, um, Groc was, uh,
2:41:20 snappier. It didn't interrupt me. And
2:41:22 the procity is we made different design
2:41:24 choices. I think we shooting for
2:41:26 something more calm, smooth, more
2:41:27 natural versus something that's more
2:41:29 poppy or artificial. So, we'll keep
2:41:31 improving these ones.
2:41:33 Thanks, guys.
2:41:34 Yeah.
2:41:39 Yep. So since the launch of the voice
2:41:42 model uh we actually see the 2x faster
2:41:45 end to end latency uh in the last eight
2:41:48 weeks five different voices and also 10x
2:41:52 the active user. So Grock voice is
2:41:54 taking off now. All right I'm gonna mute
2:41:57 that. I'm going to go to bed. I can't
2:41:59 take it anymore. This is killing me. Um
2:42:01 we we'll figure this out tomorrow. Looks
2:42:03 like it's really [ __ ] smart. I assume
2:42:07 uh this is AGI or close to it. There's
2:42:10 the ARC AGI test. Gro 4 is above
2:42:13 everything else by a lot. Um I would be
2:42:17 dramatically surprised
2:42:20 if OpenAI doesn't come out with
2:42:22 something that tops this. Um I don't
2:42:24 think they're going to sit sit for for
2:42:26 long at this at this stage. But anyway,
2:42:30 all right. I'm done. Yeah, we need Steve
2:42:32 Jobs back, man. They got to figure out
2:42:34 how to do some presentations here. This
2:42:35 shit's this shit's awful. Um anyway, um
2:42:40 it does look impressive. I'm exhausted.
2:42:43 Peace out. Uh tomorrow is what's today?
2:42:46 Wednesday. Thursday. I'll see you at
2:42:48 eight o'clock. We'll be back. We'll
2:42:49 figure out what's worth talking about
2:42:51 with all this stuff. All right. I will
2:42:53 see you tomorrow. Bye.