1
00:00:00,018 --> 00:00:05,457
Josh:
There's a new best AI model in town, and it comes by the name of Astra, also known as GPT-6.

2
00:00:05,724 --> 00:00:09,598
Josh:
Just last week, OpenAI released what is now known as the most powerful model in the world.

3
00:00:09,598 --> 00:00:13,728
Josh:
This is the new frontier setting level model in which it is capable of doing

4
00:00:13,728 --> 00:00:16,738
Josh:
all these unbelievable things that we've been testing out over the weekend.

5
00:00:16,738 --> 00:00:19,618
Josh:
So if you aren't familiar, or if you've just been using Astra a little bit over

6
00:00:19,618 --> 00:00:22,678
Josh:
the weekend, this episode is gonna fill you in on everything you need to know

7
00:00:22,678 --> 00:00:26,768
Josh:
from demos to actual use cases, to what I spent my entire morning doing that

8
00:00:26,768 --> 00:00:29,725
Josh:
would have taken me hours previously, that now only takes five minutes.

9
00:00:29,911 --> 00:00:32,458
Josh:
It's a pretty amazing model. There are certainly some shortcomings in which

10
00:00:32,458 --> 00:00:36,638
Josh:
we'll get into in a little bit. But to start, I mean, it has GPT-6 Astros here.

11
00:00:36,638 --> 00:00:40,278
Josh:
The frontier has shifted. There's a new best model on the block and we've got to talk about it.

12
00:00:40,447 --> 00:00:45,638
Ejaaz:
We've had around 16 model releases in the last three months, Josh.

13
00:00:45,638 --> 00:00:47,668
Josh:
It's been exhausting, dude. We've been covering all of them.

14
00:00:47,998 --> 00:00:51,568
Ejaaz:
Many different companies. There's been 16 model releases. This is the first

15
00:00:51,568 --> 00:00:56,438
Ejaaz:
model release where I've truly felt like this model could potentially replace

16
00:00:56,438 --> 00:01:00,548
Ejaaz:
a lot of what I and many other fellow humans can do.

17
00:01:00,548 --> 00:01:04,758
Ejaaz:
And I'm not trying to exaggerate here. So it actually all started,

18
00:01:05,576 --> 00:01:10,942
Ejaaz:
the imminent second of launch. So what you're seeing on your screen here is

19
00:01:10,942 --> 00:01:17,158
Ejaaz:
a recount of when Astra GPT-6 got released, and it went live for 30 seconds.

20
00:01:17,338 --> 00:01:22,982
Ejaaz:
Now, no one in the world knew, aside from 30 seconds later, when every entire

21
00:01:22,982 --> 00:01:28,122
Ejaaz:
computer server basically went down, and it took down every single LLM with it.

22
00:01:28,604 --> 00:01:32,870
Ejaaz:
Grok from XAI, SpaceX AI went down. ChatGPT obviously went down.

23
00:01:33,050 --> 00:01:34,490
Ejaaz:
Google's Gemini went down.

24
00:01:34,722 --> 00:01:39,302
Ejaaz:
And Anthropik's Claude also went down. There was a two and a half hour period

25
00:01:39,302 --> 00:01:42,925
Ejaaz:
where you basically couldn't access any of the favored AI models.

26
00:01:43,133 --> 00:01:46,646
Ejaaz:
And this coincided directly with the launch of GPT-6.

27
00:01:46,924 --> 00:01:48,872
Josh:
So of course, my tinfoil hat on for this.

28
00:01:48,872 --> 00:01:51,002
Ejaaz:
Like, I mean, mine is already, I don't know if you can see it.

29
00:01:51,002 --> 00:01:53,482
Ejaaz:
It's actually elevated beyond the ceiling above me.

30
00:01:53,482 --> 00:01:54,672
Josh:
It's permanent.

31
00:01:54,672 --> 00:01:58,122
Ejaaz:
My, like, listen, we've spent the last two weeks talking about this hugging

32
00:01:58,122 --> 00:02:04,082
Ejaaz:
face incident where an internal version of GPT basically broke out and spread

33
00:02:04,082 --> 00:02:08,321
Ejaaz:
a swarm of different agents, hacked into many different production servers when it wasn't meant to.

34
00:02:08,600 --> 00:02:11,450
Ejaaz:
And now I'm looking at this and I'm thinking, oh my goodness,

35
00:02:11,730 --> 00:02:16,855
Ejaaz:
there is a version of Ultron that is out there hiding in compute servers,

36
00:02:17,081 --> 00:02:22,152
Ejaaz:
and not a single human being knows where it is or what on earth it's doing.

37
00:02:22,152 --> 00:02:24,951
Ejaaz:
And we're going to discover something very bad very later on.

38
00:02:25,642 --> 00:02:30,069
Ejaaz:
This model was a very dramatic launch. People were very excited about it.

39
00:02:30,069 --> 00:02:33,519
Ejaaz:
Sam has been teasing it for so long. Thibaut, the lead of Codex,

40
00:02:33,519 --> 00:02:35,799
Ejaaz:
said something was felt across the internet today.

41
00:02:35,799 --> 00:02:39,669
Ejaaz:
Greg Brockman came on stage, or wherever he was, the president of OpenAI,

42
00:02:39,669 --> 00:02:42,819
Ejaaz:
and said, AGI is finally here.

43
00:02:42,819 --> 00:02:46,129
Ejaaz:
Now, AGI has been kind of like thrown around so many times, and I don't know

44
00:02:46,429 --> 00:02:50,169
Ejaaz:
what to make of it. But if anything, this opening video, Josh,

45
00:02:50,169 --> 00:02:54,325
Ejaaz:
I don't know if you saw the promo video of GPT-6, was so sick.

46
00:02:54,487 --> 00:02:58,039
Ejaaz:
It depicts many different people sitting in an armchair, and they're basically

47
00:02:58,039 --> 00:03:03,029
Ejaaz:
just talking aloud to this mysterious AI model that then just proceeds to do

48
00:03:03,029 --> 00:03:05,649
Ejaaz:
anything and everything for them on the screen in front of them.

49
00:03:05,649 --> 00:03:07,183
Ejaaz:
What is the screen that is on in front of them?

50
00:03:07,450 --> 00:03:11,240
Ejaaz:
It is their desktop. It is their computer, except he didn't touch a single key

51
00:03:11,496 --> 00:03:14,271
Ejaaz:
or move a single cursor throughout the entire demo.

52
00:03:14,532 --> 00:03:17,679
Ejaaz:
And this kind of precedes the main sort of advantage with GPT-6,

53
00:03:17,679 --> 00:03:24,489
Ejaaz:
which is it is the most amazing and frontier computer use model that is out there right now.

54
00:03:24,489 --> 00:03:28,089
Ejaaz:
What it means by that is you can ask the model to do pretty much anything and

55
00:03:28,089 --> 00:03:32,409
Ejaaz:
everything a human can do on a desktop. Now, previously, a bunch of these models

56
00:03:32,409 --> 00:03:35,679
Ejaaz:
were kind of good at it. It was sort of slow. You would still need to follow

57
00:03:35,679 --> 00:03:36,977
Ejaaz:
up and kind of tell it what to do.

58
00:03:37,297 --> 00:03:41,349
Ejaaz:
GPT-6 Astra is unique in the sense where it is basically relentless.

59
00:03:41,349 --> 00:03:46,219
Ejaaz:
It spins up multiple different AI agents. And if it can't figure something out, it doesn't go to you.

60
00:03:46,457 --> 00:03:48,289
Ejaaz:
It goes back to itself. It kind

61
00:03:48,289 --> 00:03:51,217
Ejaaz:
of like tests different forums and figures out the way to do the task.

62
00:03:51,397 --> 00:03:55,639
Ejaaz:
And it would often work hours to days to weeks at a time to get a task done.

63
00:03:55,639 --> 00:03:57,899
Ejaaz:
It is one of the most intelligent models that I've come across.

64
00:03:58,340 --> 00:04:02,085
Josh:
Yeah, I thought this was a lovely launch video. It was really cool because in

65
00:04:02,085 --> 00:04:04,395
Josh:
a way, when you get new models, and we frequently get new models,

66
00:04:04,395 --> 00:04:07,535
Josh:
what do we have to show for them? We get a launch card. We get like a sheet

67
00:04:07,535 --> 00:04:09,285
Josh:
with all the specs on it. We get the pricing.

68
00:04:09,785 --> 00:04:13,105
Josh:
Yeah, benchmarks. I really appreciate the fact that they distilled this new

69
00:04:13,105 --> 00:04:15,819
Josh:
model into a singular moment, which is just computer use.

70
00:04:16,074 --> 00:04:19,205
Josh:
Like if you are thinking of how to describe Astra, you are thinking of it as

71
00:04:19,205 --> 00:04:22,825
Josh:
a user of your computer. In fact, it can possibly even use your computer better

72
00:04:22,825 --> 00:04:25,155
Josh:
than you can. So now we have a computer that uses your computer

73
00:04:25,155 --> 00:04:25,819
Ejaaz:
Better than you can.

74
00:04:26,069 --> 00:04:29,415
Josh:
But what I loved about this demo is it felt like a look into the future that

75
00:04:29,415 --> 00:04:32,861
Josh:
open ai is making and i didn't see a lot of this commentary online per se but

76
00:04:33,047 --> 00:04:36,675
Josh:
i mean like we know and like we cannot say i personally cannot stop talking

77
00:04:36,675 --> 00:04:39,026
Josh:
about open ai is working on this suite of devices

78
00:04:39,322 --> 00:04:43,670
Josh:
and suite of devices the idea is to change the way in which we interface with ai on a regular basis

79
00:04:44,413 --> 00:04:47,251
Josh:
and what we're seeing in this launch video is a very clear

80
00:04:48,000 --> 00:04:52,035
Josh:
vision of what that is going to look like in a way that felt novel to me at

81
00:04:52,035 --> 00:04:55,837
Josh:
least it's something i've we've kind of been talking about loosely where okay

82
00:04:56,023 --> 00:04:58,955
Josh:
we speak to our computer now more than we actually type because I'm faster at

83
00:04:58,955 --> 00:05:00,760
Josh:
speaking. Dictation is very good.

84
00:05:00,980 --> 00:05:03,345
Josh:
It's a really cool way of engaging with it. But we still have to do a lot of

85
00:05:03,345 --> 00:05:04,675
Josh:
the manual tactile work on the

86
00:05:04,675 --> 00:05:06,779
Josh:
computer. We have to use the mouse and keyboard. We have to click around.

87
00:05:07,215 --> 00:05:10,465
Josh:
This model removes that second piece of the puzzle. It's like,

88
00:05:10,465 --> 00:05:14,709
Josh:
okay, now I can speak to my computer, but I'm not speaking to text. I'm speaking to action.

89
00:05:15,022 --> 00:05:17,047
Josh:
And the computer is now taking action on my behalf.

90
00:05:17,361 --> 00:05:19,985
Josh:
And this morning I was trying it out. And just before we started recording,

91
00:05:19,985 --> 00:05:23,949
Josh:
I was talking to, I was talking to you and Luke, and I was sharing the fact that

92
00:05:24,762 --> 00:05:26,967
Josh:
For a long time, I've been using editing software. For videos,

93
00:05:26,967 --> 00:05:28,831
Josh:
I've been using DaVinci Resolve, Adobe Premiere.

94
00:05:29,156 --> 00:05:32,877
Josh:
And for the first time ever, this model was able to take multi-cam footage,

95
00:05:32,877 --> 00:05:36,209
Josh:
which for those who don't know, there's like multiple cameras recording the same moment at once.

96
00:05:36,610 --> 00:05:39,937
Josh:
And it was able to separate them into their own tracks, figure out where the

97
00:05:39,937 --> 00:05:44,567
Josh:
audio was, align them together, color grade them all, and then render out a

98
00:05:44,567 --> 00:05:48,394
Josh:
finished export vision of that. And that is something that would take me previously,

99
00:05:48,702 --> 00:05:50,252
Josh:
I don't know, maybe half an hour to an hour to do.

100
00:05:50,519 --> 00:05:53,317
Josh:
And it did it in the background. And I think that is what's really cool here

101
00:05:53,317 --> 00:05:56,219
Josh:
is it shows what the next paradigm of computing can look like,

102
00:05:56,428 --> 00:05:58,417
Josh:
where you don't actually need to be sitting in front of a mouse and keyboard.

103
00:05:58,417 --> 00:06:01,087
Josh:
You could just stand in front of this big screen and tell it what to do and

104
00:06:01,087 --> 00:06:02,396
Josh:
it will just go and do the thing.

105
00:06:02,756 --> 00:06:05,101
Josh:
And it's better at doing that than just about any model in the world.

106
00:06:05,641 --> 00:06:09,600
Ejaaz:
I think with previous LLMs, Josh, and I'm curious if you agree with me here,

107
00:06:09,942 --> 00:06:13,587
Ejaaz:
you could kind of think of them as like thought partners. You can kind of like

108
00:06:13,587 --> 00:06:15,957
Ejaaz:
bounce ideas off of them. They come back and they're like, oh,

109
00:06:15,957 --> 00:06:19,197
Ejaaz:
you're kind of right here, you're directionally correct, but actually you could

110
00:06:19,197 --> 00:06:20,239
Ejaaz:
improve it this way or that.

111
00:06:20,518 --> 00:06:23,577
Ejaaz:
But then that's where sort of like the conversation and the relationship ends

112
00:06:23,908 --> 00:06:28,617
Ejaaz:
um with these new models or with gbt6 astra specifically it feels more like

113
00:06:28,617 --> 00:06:32,287
Ejaaz:
a work colleague or a friend that actually or like a co-founder that actually

114
00:06:32,287 --> 00:06:36,162
Ejaaz:
can do things with you and you can rely on them you can trust them now

115
00:06:36,505 --> 00:06:40,667
Ejaaz:
we were speaking about this exact thesis almost a year ago actually when the

116
00:06:40,667 --> 00:06:44,247
Ejaaz:
first version of computer use came out from chat gbt and when it came out from

117
00:06:44,247 --> 00:06:45,107
Ejaaz:
claude and we were like remember.

118
00:06:45,107 --> 00:06:46,047
Josh:
It was so cool back then

119
00:06:46,527 --> 00:06:51,215
Ejaaz:
It was so cool remember we saw the cursor and it was moving like this like super slowly.

120
00:06:51,412 --> 00:06:56,027
Josh:
That's so funny. That was like a novel thing and now it's just ripping through and doing everything.

121
00:06:56,027 --> 00:06:58,997
Ejaaz:
Well, we were like, okay, it's clicking on the incorrect thing.

122
00:06:58,997 --> 00:07:02,505
Ejaaz:
No, that's the wrong file. Kind of like slapping its wrist, like don't go there.

123
00:07:02,987 --> 00:07:05,537
Josh:
I vividly remember the cursor, how amazing that cursor was.

124
00:07:05,547 --> 00:07:09,779
Ejaaz:
And now when I use GPT-6 Astro, I'm looking at my screen and I'm like...

125
00:07:10,702 --> 00:07:13,486
Ejaaz:
Oh my God, I didn't even see the cursor move. And hang on a second.

126
00:07:13,695 --> 00:07:16,775
Ejaaz:
It has like four other virtual desktops set up side by side.

127
00:07:16,775 --> 00:07:17,915
Ejaaz:
So it's doing all these different things.

128
00:07:18,131 --> 00:07:20,813
Ejaaz:
And by the way, we're not just talking about this. Like I have a bunch of demos.

129
00:07:21,045 --> 00:07:22,205
Ejaaz:
I have a demo of like- Please.

130
00:07:22,205 --> 00:07:22,855
Josh:
We gotta get into the demos.

131
00:07:22,885 --> 00:07:27,445
Ejaaz:
Creating like a Call of Duty game from scratch and then playing the game and

132
00:07:27,445 --> 00:07:30,454
Ejaaz:
then chatting to itself and other people that are playing the game, doing the thing.

133
00:07:30,647 --> 00:07:33,960
Ejaaz:
It is just absolutely insane. But I wanna jump into a few of these demos.

134
00:07:34,175 --> 00:07:39,399
Ejaaz:
But before I do that, I just wanna point out that this is the largest ever,

135
00:07:40,270 --> 00:07:44,278
Ejaaz:
compute training run that a lab has done to create this type of model.

136
00:07:44,278 --> 00:07:47,568
Ejaaz:
And I want to kind of like make this a point before we jump into the demos,

137
00:07:47,568 --> 00:07:49,617
Ejaaz:
because what you're about to see is very impressive.

138
00:07:49,936 --> 00:07:55,688
Ejaaz:
It was trained on 100,000 Frontier Edge NVIDIA GPUs. So that is,

139
00:07:56,153 --> 00:07:58,878
Ejaaz:
for all intents and purposes, a very expensive wallet budget.

140
00:07:58,878 --> 00:08:02,348
Ejaaz:
Like the only other person to potentially match that or is about to match that

141
00:08:02,348 --> 00:08:04,431
Ejaaz:
is Elon Musk, because he's acquired all the NVIDIA GPUs.

142
00:08:04,732 --> 00:08:08,858
Ejaaz:
But it just proves to show that 100,000 of these GPUs, which costs in like the

143
00:08:08,858 --> 00:08:13,618
Ejaaz:
billions of dollars, is able to scale frontier intelligence with these different models.

144
00:08:13,618 --> 00:08:15,838
Ejaaz:
So for those of you who are out there that were thinking, oh,

145
00:08:15,838 --> 00:08:18,898
Ejaaz:
you know, some of these Chinese open source models who has fewer compute,

146
00:08:18,898 --> 00:08:21,258
Ejaaz:
but is able to make some of these breakthroughs, they're good models.

147
00:08:21,444 --> 00:08:23,418
Ejaaz:
But the law still stands, which

148
00:08:23,418 --> 00:08:27,260
Ejaaz:
is if you have enough GPUs, you're able to create a much better model.

149
00:08:27,464 --> 00:08:30,878
Ejaaz:
Now, Josh, when we think of benchmarks, we hate these benchmarks, right?

150
00:08:30,878 --> 00:08:34,638
Ejaaz:
They're like just they're so kind of like docile, they don't really tell us

151
00:08:34,638 --> 00:08:38,470
Ejaaz:
anything but the one that i like the most is the pelican svg.

152
00:08:39,478 --> 00:08:43,619
Ejaaz:
Um test you know what i mean like it's a 2d image right and you ask a model

153
00:08:43,880 --> 00:08:48,818
Ejaaz:
hey i want you to do your best at basically like creating um,

154
00:08:49,318 --> 00:08:55,718
Ejaaz:
a 2d version of a pelican well check this out this is what uh gbt6 astra.

155
00:08:55,718 --> 00:08:57,578
Josh:
Pelican created san francisco

156
00:08:57,578 --> 00:09:02,928
Ejaaz:
Okay this is like 3d right it's got some form of a creature is that a beaver

157
00:09:02,928 --> 00:09:06,278
Ejaaz:
I have no idea, in the front of its basket. So for all the AI nerds that are

158
00:09:06,278 --> 00:09:09,132
Ejaaz:
listening to this, you're going to be the most impressed by this because this

159
00:09:09,374 --> 00:09:11,749
Ejaaz:
technically classifies as an SVG.

160
00:09:11,987 --> 00:09:13,848
Ejaaz:
But for those of you who are listening and are thinking, well,

161
00:09:13,848 --> 00:09:16,278
Ejaaz:
okay, who the hell cares? Why should I actually care about this?

162
00:09:16,278 --> 00:09:19,858
Ejaaz:
Well, for the gamers that are out there, Josh, you are one of these, right?

163
00:09:20,238 --> 00:09:25,398
Ejaaz:
I classify you as a gamer. Look at this. GBD6 Astra basically created this entire

164
00:09:25,398 --> 00:09:29,320
Ejaaz:
3D game, one-shotted it in 45 minutes. A few things to point out here.

165
00:09:29,582 --> 00:09:33,568
Ejaaz:
Look at the fidelity of this game. This is something that like you could create,

166
00:09:33,568 --> 00:09:38,108
Ejaaz:
like game developers spend probably years right now trying to recreate right

167
00:09:38,108 --> 00:09:42,388
Ejaaz:
now today. And it costs like millions of dollars. It costs like a lot of people's

168
00:09:42,388 --> 00:09:43,984
Ejaaz:
time to create from scratch.

169
00:09:44,605 --> 00:09:48,838
Ejaaz:
This model did it in 45 minutes from one single prompt. I don't think we've

170
00:09:48,838 --> 00:09:50,224
Ejaaz:
ever seen anything like this before.

171
00:09:50,764 --> 00:09:53,668
Josh:
What I found this really good at is tool use as well. There's another example

172
00:09:53,668 --> 00:09:56,948
Josh:
just above this one about humans and the Unreal Engine. Yes.

173
00:09:57,538 --> 00:10:00,278
Josh:
Yeah, so this is amazing. It's like, okay, for those who don't know,

174
00:10:00,278 --> 00:10:03,818
Josh:
the Unreal Engine powers a lot of AAA games that are being used out in the world

175
00:10:03,818 --> 00:10:05,950
Josh:
today. It's like a gaming engine. That's where game developers go.

176
00:10:06,240 --> 00:10:07,355
Josh:
They build these virtual worlds.

177
00:10:07,673 --> 00:10:11,266
Josh:
It had Astra go off and build a virtual world using this game engine,

178
00:10:11,533 --> 00:10:12,624
Josh:
and it looked incredible.

179
00:10:12,978 --> 00:10:16,738
Josh:
It was really impressive. The ask was to create a world with the Unreal Engine

180
00:10:16,738 --> 00:10:21,291
Josh:
and fill it with humans, each an astral-powered agent, who all have to work together to survive.

181
00:10:21,576 --> 00:10:25,688
Josh:
And those agents all get mini LLMs and they start, this feels like what we imagine

182
00:10:25,688 --> 00:10:29,828
Josh:
GTA 6 would look like, where every NPC has an LLM attached to it.

183
00:10:29,828 --> 00:10:31,769
Josh:
They have their own world, they have their own context, their own memory.

184
00:10:32,054 --> 00:10:35,188
Josh:
And it looks amazing. And what I've noticed throughout the demos of this,

185
00:10:35,188 --> 00:10:40,138
Josh:
at least in terms of like visuals, is that it's really powerful and impressive

186
00:10:40,138 --> 00:10:45,518
Josh:
at visually understanding the world, being able to take these tools and apply that to the world.

187
00:10:45,518 --> 00:10:48,998
Josh:
So I've seen a lot of examples also with Blender where people are designing houses.

188
00:10:49,329 --> 00:10:53,039
Josh:
And also, like I mentioned earlier, there is the examples of video,

189
00:10:53,317 --> 00:10:55,883
Josh:
which we talked about earlier where I was using as a video editor.

190
00:10:56,208 --> 00:11:01,537
Josh:
Now there's a version of GPT-6 making kids' cartoons using Higgsfield,

191
00:11:01,868 --> 00:11:04,958
Josh:
editing it in Premiere, and then uploading it to YouTube, which I found really

192
00:11:04,958 --> 00:11:08,178
Josh:
impressive. Like, look at this video of the computer use, actually clicking

193
00:11:08,178 --> 00:11:09,640
Josh:
through, editing the videos.

194
00:11:10,806 --> 00:11:14,690
Josh:
It's impressive. Like, there's no way to look at these demos and be like...

195
00:11:15,421 --> 00:11:18,393
Josh:
Wow, this isn't kind of like this novel new thing where for the first time ever,

196
00:11:18,393 --> 00:11:22,596
Josh:
it's like almost better at using a computer than we are. In fact, it is in most cases.

197
00:11:23,003 --> 00:11:28,043
Ejaaz:
Yeah, this is one of my favorite examples, the creation of a kid's show,

198
00:11:28,043 --> 00:11:33,043
Ejaaz:
because you can see all the three different components of this model working in real time.

199
00:11:33,043 --> 00:11:38,863
Ejaaz:
So on the right side of the screen on his laptop, it is basically the AI kind

200
00:11:38,863 --> 00:11:43,680
Ejaaz:
of like dreaming up a prompt of what a good episode would look like, typing it into itself.

201
00:11:43,895 --> 00:11:47,250
Ejaaz:
It's desktop version of ChatGPT coming up with the episode outline

202
00:11:47,598 --> 00:11:51,412
Ejaaz:
then coming up with the script and then being like okay how can I create

203
00:11:51,691 --> 00:11:54,993
Ejaaz:
the different scenes based on this script feeding that into Hicksville which

204
00:11:54,993 --> 00:11:58,603
Ejaaz:
is on the left side of the screen right and then like putting it into an editing

205
00:11:58,603 --> 00:12:02,674
Ejaaz:
software and then editing the different clips to make sure that it kind of like flows seamlessly

206
00:12:02,941 --> 00:12:05,223
Ejaaz:
there's like so many different things that you mentioned earlier Josh there's

207
00:12:05,223 --> 00:12:07,780
Ejaaz:
the audio files there's making sure that it kind of lines up with the right

208
00:12:07,983 --> 00:12:10,663
Ejaaz:
types of animation and then you have the finished product that is.

209
00:12:11,763 --> 00:12:15,743
Ejaaz:
On top of the screen right there which honestly looks like something that I

210
00:12:15,743 --> 00:12:21,474
Ejaaz:
would see on a kid's TV channel, Josh, like what I would see from my baby's checking it out, right?

211
00:12:21,654 --> 00:12:26,553
Ejaaz:
It comes with titles, it comes with captions, it comes with different angles

212
00:12:26,553 --> 00:12:30,263
Ejaaz:
of the same different scene. It's so thoughtful, but it's done this from scratch.

213
00:12:30,263 --> 00:12:32,196
Ejaaz:
And by the way, it just uploads it on YouTube.

214
00:12:32,515 --> 00:12:36,753
Ejaaz:
So you could have a hands-off kind of ghost YouTube channel that makes you a

215
00:12:36,753 --> 00:12:38,628
Ejaaz:
ton of money or entertains your kids.

216
00:12:38,802 --> 00:12:42,823
Ejaaz:
And let's say you had a kind of comment on this. You didn't like that your kid

217
00:12:42,823 --> 00:12:47,614
Ejaaz:
was kind of learning certain things or like being exposed to certain types of show content.

218
00:12:47,927 --> 00:12:51,823
Ejaaz:
Well, you could just kind of like ask your AI agent to create a version of this

219
00:12:51,823 --> 00:12:55,282
Ejaaz:
cartoon that they like, and they can just end up continuing doing this.

220
00:12:55,567 --> 00:12:59,108
Ejaaz:
Now, moving on, Josh, you are a Call of Duty fan.

221
00:12:59,346 --> 00:13:00,513
Josh:
Oh, so much so.

222
00:13:00,513 --> 00:13:04,842
Ejaaz:
I'm yet to see something that is of the same visual quality.

223
00:13:05,150 --> 00:13:09,553
Ejaaz:
But this is basically an idea where someone was like, okay, I want you to make

224
00:13:09,553 --> 00:13:11,872
Ejaaz:
a Call of Duty type game. And we saw this,

225
00:13:12,545 --> 00:13:17,125
Ejaaz:
the Fable 5.1 release about a week and a half ago, and there were some really good versions of this.

226
00:13:17,433 --> 00:13:22,761
Ejaaz:
This is a new version, except that this person played this for two hours straight,

227
00:13:22,761 --> 00:13:24,985
Ejaaz:
and they didn't realize that they'd been playing it for two hours straight.

228
00:13:25,293 --> 00:13:29,966
Ejaaz:
The reason why is the levels or the gameplay was continuous, Josh.

229
00:13:30,163 --> 00:13:33,811
Ejaaz:
So let's say he like moved past a building or he went around a certain corner

230
00:13:33,811 --> 00:13:36,218
Ejaaz:
for the map. It would just regenerate in real time.

231
00:13:36,392 --> 00:13:38,917
Ejaaz:
That's so crazy. And he was able to give it feedback and say,

232
00:13:39,161 --> 00:13:42,481
Ejaaz:
hey, the enemy was like kind of too easy to kill. like I need you to increase

233
00:13:42,481 --> 00:13:44,983
Ejaaz:
the difficulty setting and it just did it in a couple of seconds.

234
00:13:45,175 --> 00:13:47,861
Ejaaz:
So this is the kind of difference if I were to kind of capture,

235
00:13:47,861 --> 00:13:51,321
Ejaaz:
if you're wondering, should I be using like a Claude model? Should I be using

236
00:13:51,321 --> 00:13:52,542
Ejaaz:
Grok? Should I be using this?

237
00:13:52,832 --> 00:13:57,661
Ejaaz:
This model is really good at taking feedback in real time and also figuring

238
00:13:57,661 --> 00:14:00,181
Ejaaz:
out that feedback before it even comes to the top of your head.

239
00:14:00,181 --> 00:14:03,715
Ejaaz:
It's like, oh no, I think like he's gonna ask me about this. Let me try and fix this.

240
00:14:03,924 --> 00:14:06,083
Ejaaz:
And then it works continuously to do the thing.

241
00:14:06,339 --> 00:14:10,341
Ejaaz:
Now that may sound vague, but you can apply this to literally any part of your

242
00:14:10,341 --> 00:14:11,587
Ejaaz:
profession or daily life.

243
00:14:11,795 --> 00:14:16,481
Ejaaz:
If it is like Josh editing videos, if it is like us doing research on the next

244
00:14:16,481 --> 00:14:19,171
Ejaaz:
episode, or if it's for your kid trying to generate content,

245
00:14:19,171 --> 00:14:22,191
Ejaaz:
or if it's at work and you're trying to file a Jira ticket or complete a software

246
00:14:22,191 --> 00:14:26,668
Ejaaz:
task, you can just throw this model at anything and it'll figure it out. It's just amazing.

247
00:14:26,970 --> 00:14:29,671
Josh:
Yeah, seeing it play Call of Duty, I know it's good enough when I recognize

248
00:14:29,671 --> 00:14:31,939
Josh:
the map. Like I'm recognizing that this is Nukedown.

249
00:14:32,258 --> 00:14:35,021
Josh:
They are like trying to play the game. And I actually saw a few other demos

250
00:14:35,021 --> 00:14:36,751
Josh:
similar to this where they would share,

251
00:14:37,041 --> 00:14:40,071
Josh:
you could share a screenshot of the map in Call of Duty and it will recreate

252
00:14:40,071 --> 00:14:43,131
Josh:
it and i also found this particularly interesting like the demos they really

253
00:14:43,131 --> 00:14:44,663
Josh:
are crazy so you got to stick with us for here

254
00:14:44,901 --> 00:14:49,551
Josh:
this is the recreation of new york city with a beautiful car and shaders like

255
00:14:49,551 --> 00:14:52,401
Josh:
if you look at how the light reflects off of the car how the light reflects

256
00:14:52,401 --> 00:14:54,621
Josh:
off of the windows this is getting accurate to

257
00:14:54,941 --> 00:14:59,491
Josh:
becoming real world's emulation to a point where like we see how far we came

258
00:14:59,491 --> 00:15:02,901
Josh:
over a year where seeing a cursor on screen was impressive and now we have like

259
00:15:03,145 --> 00:15:06,268
Josh:
a fully fledged real world new york city that's you know it's not

260
00:15:06,738 --> 00:15:10,685
Josh:
high poly count per se like there's very low textures but look at that structurally

261
00:15:10,685 --> 00:15:13,415
Josh:
i know what city that is and you can walk down the road you'll recognize the

262
00:15:13,415 --> 00:15:15,605
Josh:
building topographically it's correct in terms of

263
00:15:15,925 --> 00:15:20,005
Josh:
where the buildings are where the hills are it's really amazing and then there's

264
00:15:20,005 --> 00:15:23,352
Josh:
just two that i had to show also the palace of fine arts is one of them because

265
00:15:24,299 --> 00:15:28,505
Josh:
its ability to recreate these real places is like stunning for those who are

266
00:15:28,505 --> 00:15:30,672
Josh:
familiar the palace of fine arts this is in san francisco

267
00:15:31,038 --> 00:15:35,615
Josh:
this is not a video of the real thing this is a 3d rendering not made by a visual

268
00:15:35,615 --> 00:15:39,786
Josh:
artist but made by gpt6 astra it is so accurate

269
00:15:40,013 --> 00:15:41,925
Ejaaz:
No reference images by the way it was.

270
00:15:41,925 --> 00:15:43,105
Josh:
Just no reference images find

271
00:15:43,105 --> 00:15:44,705
Ejaaz:
This thing and rebuild it.

272
00:15:45,545 --> 00:15:49,405
Josh:
And then another one for people who are going to be benefiting from these ipos

273
00:15:49,405 --> 00:15:52,174
Josh:
soon is the zillow recreation um

274
00:15:52,389 --> 00:15:54,945
Ejaaz:
Wait wait let me restart this one second this is a.

275
00:15:54,945 --> 00:15:59,715
Josh:
Crazy this is nuts dude you get like a full house tour from a zillow

276
00:15:59,715 --> 00:16:00,770
Ejaaz:
This is a real house.

277
00:16:00,996 --> 00:16:03,625
Josh:
This is a real house this is a real house so basically what the model does is

278
00:16:03,625 --> 00:16:05,147
Josh:
it takes the listing from zillow

279
00:16:05,535 --> 00:16:09,327
Josh:
and it uses all the context it takes all the information and it rebuilds this

280
00:16:09,594 --> 00:16:12,879
Josh:
in a virtual world in which you could go and actually walk through and navigate

281
00:16:13,210 --> 00:16:16,265
Josh:
the real world and for those who aren't familiar before podcasting i was big

282
00:16:16,265 --> 00:16:19,699
Josh:
in the real estate media game we would do photos of these houses we do videos

283
00:16:20,314 --> 00:16:23,345
Josh:
if we had technology like this oh my god we would have taken over the world

284
00:16:23,520 --> 00:16:26,689
Josh:
this is like how cool is it that you can create a full 3d rendering of a house

285
00:16:27,025 --> 00:16:31,304
Josh:
just by using a model and a couple dollars in tokens so it's like the demos are fun and i

286
00:16:32,053 --> 00:16:35,124
Josh:
It's important to caveat this with the idea that the demos are not everything.

287
00:16:35,460 --> 00:16:39,700
Josh:
Just because this model is visually compelling does not mean it is like the

288
00:16:39,700 --> 00:16:43,640
Josh:
God model. But this is some good demos, dude. This is some good stuff.

289
00:16:44,100 --> 00:16:48,750
Ejaaz:
Well, I need to keep the train chugging. Can I suggest another little demo,

290
00:16:48,750 --> 00:16:50,618
Ejaaz:
which kind of switches up the media, okay?

291
00:16:50,931 --> 00:16:54,060
Ejaaz:
So one big critique for a lot of these LLMs is like, okay, cool.

292
00:16:54,060 --> 00:16:56,600
Ejaaz:
It could do the video thing. Maybe it could do the image thing.

293
00:16:56,600 --> 00:17:01,894
Ejaaz:
It's really good at words. But like, you'll never take the epitome of human culture from us.

294
00:17:02,156 --> 00:17:06,020
Ejaaz:
You'll never take music you'll never take symphonies and there's this hilarious

295
00:17:06,020 --> 00:17:09,620
Ejaaz:
meme that we always see from uh i believe it was irobot will smith great movie

296
00:17:09,620 --> 00:17:13,990
Ejaaz:
by the way and there's a scene where he goes but can a robot write a symphony

297
00:17:13,990 --> 00:17:16,407
Ejaaz:
and the robot's like uh like no um,

298
00:17:17,240 --> 00:17:20,280
Ejaaz:
but in this version it's like actually yeah i can,

299
00:17:20,980 --> 00:17:27,330
Ejaaz:
quote tweets um this demo where basically someone asked a musician like an expert

300
00:17:27,330 --> 00:17:28,673
Ejaaz:
in their field basically says

301
00:17:28,998 --> 00:17:32,783
Ejaaz:
listen we have this benchmark I'm going to call it the Bach Benchmark.

302
00:17:33,038 --> 00:17:37,320
Ejaaz:
And it basically tested LLM's capabilities to understand music at its core level

303
00:17:37,320 --> 00:17:39,551
Ejaaz:
and to create a certain type of chord symphony.

304
00:17:39,807 --> 00:17:43,480
Ejaaz:
Now most AI models prior to this sucked at it, would never get it,

305
00:17:43,480 --> 00:17:44,550
Ejaaz:
would kind of spit something out

306
00:17:44,550 --> 00:17:48,114
Ejaaz:
that would be the equivalent of Apple GarageBand back in the day, right?

307
00:17:48,520 --> 00:17:52,541
Ejaaz:
Except Astra just completely one-shotted it. And I'm going to play this clip right now.

308
00:18:10,861 --> 00:18:13,819
Ejaaz:
Beethoven is cut. I mean, come on.

309
00:18:13,819 --> 00:18:19,529
Ejaaz:
He's rolling in his grave. Now, if you're a musician out here and you're like,

310
00:18:19,869 --> 00:18:22,593
Ejaaz:
that was the biggest load of whatever,

311
00:18:23,301 --> 00:18:26,119
Ejaaz:
please let us know because I just listened to that and I would listen to an

312
00:18:26,119 --> 00:18:29,709
Ejaaz:
hour of that whilst I'm trying to figure out what the next episode outline is.

313
00:18:29,709 --> 00:18:31,469
Ejaaz:
Josh, I'm like, come on. What's your reaction to that?

314
00:18:31,969 --> 00:18:35,859
Josh:
This is interesting. It's like, I'd love to play it with a piece that I'm familiar

315
00:18:35,859 --> 00:18:39,529
Josh:
with so you can kind of compare how it is relative to the bass piece.

316
00:18:39,529 --> 00:18:43,979
Josh:
But in terms of sounding good i mean the chords sound nice it sounds like a

317
00:18:43,979 --> 00:18:49,000
Josh:
piece of music it doesn't sound like ai slop it sounds like something that you know is is nice

318
00:18:49,360 --> 00:18:52,999
Josh:
and the idea that it can not only generate this but recreate this in fact i

319
00:18:52,999 --> 00:18:57,439
Josh:
saw another example where the model was playing a virtual piano and it was playing

320
00:18:57,439 --> 00:19:00,569
Josh:
the music like on a virtual piano it's like man

321
00:19:03,209 --> 00:19:03,639
Ejaaz:
It just makes.

322
00:19:03,639 --> 00:19:05,209
Josh:
You like shit do you

323
00:19:05,209 --> 00:19:09,573
Ejaaz:
Do you believe greg do you think agi is here at this point like that's a meaningful job no.

324
00:19:09,945 --> 00:19:15,169
Josh:
The goal posts continue to shift but this is like pretty remarkable in its ability

325
00:19:15,169 --> 00:19:17,709
Josh:
to use tools to the point where you start to see

326
00:19:18,198 --> 00:19:22,269
Josh:
and understand what a lot of the labs have been saying where they're like guys

327
00:19:22,269 --> 00:19:26,639
Josh:
like i don't think you understand what's coming we have like far superior models

328
00:19:26,639 --> 00:19:28,583
Josh:
to this we just don't know how to release them safely

329
00:19:28,878 --> 00:19:31,719
Josh:
please trust us this is going to change the world and then you see something

330
00:19:31,719 --> 00:19:33,409
Josh:
like this and you're like, well,

331
00:19:34,529 --> 00:19:38,016
Josh:
it's like pretty good at doing just about everything that I could do and more.

332
00:19:38,569 --> 00:19:41,329
Josh:
And you're like, wow, what does this look like in a year? Like if you just play

333
00:19:41,329 --> 00:19:45,649
Josh:
this out in a year, it's going to be pretty remarkable. You have to assume.

334
00:19:45,829 --> 00:19:47,866
Josh:
And a year from now, also, you just, we're going to have those,

335
00:19:48,146 --> 00:19:50,379
Josh:
we're going to have devices. We're going to have hardware in our lives that

336
00:19:50,379 --> 00:19:53,451
Josh:
is actually able to do this stuff without the need for a computer.

337
00:19:53,758 --> 00:19:56,419
Josh:
And it's going to have all of our context and it's going to be able to make

338
00:19:56,419 --> 00:19:57,417
Josh:
decisions on our behalf.

339
00:19:57,584 --> 00:20:01,729
Josh:
And I don't know like this this model has reinvigorated an enthusiasm that hasn't

340
00:20:01,729 --> 00:20:04,562
Josh:
been felt in a little while because there is so much novelty baked into it.

341
00:20:05,067 --> 00:20:08,931
Ejaaz:
It feels like that O1 moment. Do you remember when GPT-O1 got released?

342
00:20:08,931 --> 00:20:13,500
Ejaaz:
And we all lost our heads. We were like, this is AGI back then, right?

343
00:20:14,151 --> 00:20:17,961
Ejaaz:
It feels the same way. And to be honest with you, I thought we were sort of

344
00:20:17,961 --> 00:20:21,081
Ejaaz:
running out of steam, Josh. I thought we were like kind of like releasing all

345
00:20:21,081 --> 00:20:23,363
Ejaaz:
these different models successively and they were like kind of iterative.

346
00:20:23,682 --> 00:20:27,411
Ejaaz:
But this feels like the biggest major model leap. And the fact that this was

347
00:20:27,411 --> 00:20:31,130
Ejaaz:
the model that was ready about four months ago blows my mind.

348
00:20:31,455 --> 00:20:34,001
Josh:
Like OpenAI has, like,

349
00:20:34,001 --> 00:20:38,121
Ejaaz:
I think it was one or two models that are much more capable than the model that

350
00:20:38,121 --> 00:20:40,958
Ejaaz:
we're covering in this video, already ready to go.

351
00:20:41,259 --> 00:20:44,391
Ejaaz:
But they're paused training on that model whilst they figure out how to align

352
00:20:44,391 --> 00:20:48,980
Ejaaz:
that model and make sure that it doesn't wreak complete havoc on the world.

353
00:20:49,201 --> 00:20:51,093
Ejaaz:
So this is a very meaningful jump.

354
00:20:51,366 --> 00:20:54,711
Ejaaz:
They were able to achieve this because, and I wanna spend a bit of time talking

355
00:20:54,711 --> 00:20:59,922
Ejaaz:
about this, like OpenAI has had a very rocky couple of years, right?

356
00:21:00,172 --> 00:21:03,911
Ejaaz:
They had the lead, they lost the lead, they're now coming back,

357
00:21:03,911 --> 00:21:07,086
Ejaaz:
right? And they potentially have made like the full comeback with this research model.

358
00:21:07,301 --> 00:21:11,201
Ejaaz:
And the way that they've been able to do this is a massive reorganization in

359
00:21:11,201 --> 00:21:14,571
Ejaaz:
that company to kind of focus efforts on research and building the best coding

360
00:21:14,571 --> 00:21:18,951
Ejaaz:
model and best general model, but also by acquiring the most amount of compute.

361
00:21:18,951 --> 00:21:21,551
Ejaaz:
I mentioned like, you know, they've trained this on 100,000 GPUs,

362
00:21:21,551 --> 00:21:23,066
Ejaaz:
probably the largest training cluster ever.

363
00:21:23,350 --> 00:21:27,048
Ejaaz:
Their next one is probably even bigger than that. Compute basically scales.

364
00:21:27,750 --> 00:21:30,171
Ejaaz:
And also this model is able to do a lot of different things.

365
00:21:30,171 --> 00:21:33,791
Ejaaz:
Like eventually you're going to see this model be able to recreate itself,

366
00:21:33,791 --> 00:21:35,877
Ejaaz:
right? It's something known as recursive self-improvement.

367
00:21:36,173 --> 00:21:39,601
Ejaaz:
OpenAI is like going hammer and tongs, so is Anthropic, trying to figure out

368
00:21:39,601 --> 00:21:42,205
Ejaaz:
how to get the model to build the next best version of itself.

369
00:21:42,483 --> 00:21:47,421
Ejaaz:
And we already have a version of this, like GPT Astra was in part helped by

370
00:21:47,421 --> 00:21:51,731
Ejaaz:
previous versions of GPT 5.6 Sol to do the thing, to help build and design and

371
00:21:51,731 --> 00:21:53,426
Ejaaz:
architect it, but it wasn't fully autonomous.

372
00:21:53,699 --> 00:21:56,769
Ejaaz:
But we're getting to the point where it's going to be completely autonomous.

373
00:21:57,060 --> 00:22:00,351
Ejaaz:
What you're seeing on the screen now is a baby version of this, right?

374
00:22:01,326 --> 00:22:06,499
Ejaaz:
Josh, there's this company called Qualcomm, right? And they make these Snapdragon chips.

375
00:22:06,702 --> 00:22:10,216
Ejaaz:
And what someone did was they created an AI model that doesn't exist anywhere

376
00:22:10,216 --> 00:22:13,045
Ejaaz:
else, but only works for Qualcomm chips.

377
00:22:13,330 --> 00:22:16,636
Ejaaz:
And someone had a bright idea. They were like, okay, I wonder if I could just

378
00:22:16,636 --> 00:22:21,329
Ejaaz:
give GPT-6 Astra access to this chip, right? Not the model, just the chip,

379
00:22:21,724 --> 00:22:25,915
Ejaaz:
and kind of reverse engineer what a model would look like on that chip.

380
00:22:26,217 --> 00:22:30,636
Ejaaz:
And it ended up recreating the entire model that this person had independently created.

381
00:22:31,056 --> 00:22:34,373
Ejaaz:
Just from design. So the point I'm making is we're in this very weird world

382
00:22:34,663 --> 00:22:38,896
Ejaaz:
where you can take a model and throw it literally at anything you want.

383
00:22:38,896 --> 00:22:41,600
Ejaaz:
If you're a game developer or game designer, it can do that.

384
00:22:41,861 --> 00:22:45,746
Ejaaz:
If you are a knowledge worker trying to file accounts or finances or whatever,

385
00:22:45,746 --> 00:22:47,899
Ejaaz:
if you're on the finance team, you can throw it at that.

386
00:22:48,142 --> 00:22:52,723
Ejaaz:
If you are a musician and you're trying to figure out Beethoven's comeback,

387
00:22:52,961 --> 00:22:54,946
Ejaaz:
you can throw it on that and you can figure out a chord.

388
00:22:55,242 --> 00:22:59,376
Ejaaz:
It is just one of those universal moments where these models are equally kind

389
00:22:59,376 --> 00:23:03,146
Ejaaz:
of like very capable but also there's a big question around alignment and whether

390
00:23:03,146 --> 00:23:07,496
Ejaaz:
we can release this freely to the world. It's just my mind, I've been processing

391
00:23:07,496 --> 00:23:09,076
Ejaaz:
it. I haven't processed it.

392
00:23:09,516 --> 00:23:13,396
Josh:
So what you're telling me is these models are artificially, generally intelligent.

393
00:23:13,396 --> 00:23:17,671
Ejaaz:
I think they are. Did you see the index thing, Josh? Did you see the AGI 3 thing?

394
00:23:18,118 --> 00:23:21,256
Josh:
Yeah. I don't know. We're saturating a lot of benchmarks now.

395
00:23:21,856 --> 00:23:23,906
Josh:
It's at the point where like someone's gonna have to call it.

396
00:23:23,906 --> 00:23:27,969
Josh:
If we don't call it here, we're only like one or two miles away from calling it.

397
00:23:28,213 --> 00:23:32,806
Josh:
There reaches a point in which like it is just going to be like you can't even argue against the fact

398
00:23:32,806 --> 00:23:33,676
Ejaaz:
That we have we.

399
00:23:33,676 --> 00:23:37,060
Josh:
Can't and and it gets on like the context of like how do you define it well

400
00:23:37,408 --> 00:23:42,046
Josh:
it's a computer that's smarter than any human at any given task the best at

401
00:23:42,046 --> 00:23:44,188
Josh:
each category um it seems like

402
00:23:44,635 --> 00:23:47,689
Josh:
perhaps we're not there because the models are still spiky they're still not

403
00:23:48,049 --> 00:23:51,706
Josh:
excellent and i think one of the important things here is what makes this so

404
00:23:51,706 --> 00:23:55,866
Josh:
fun for us in particular is because we are consumers we are not hardcore developers

405
00:23:55,866 --> 00:23:59,036
Josh:
we are not cyber security experts we're just dudes who like playing with

406
00:23:59,734 --> 00:24:03,676
Josh:
And in terms of like the dude who likes to play with AI to likes to be productive,

407
00:24:03,676 --> 00:24:05,583
Josh:
likes to have like the fun novel experiences.

408
00:24:06,315 --> 00:24:09,043
Josh:
This model is very good at that. It's still spiky at other things.

409
00:24:09,293 --> 00:24:12,123
Josh:
It's still going to be solved by like word riddles. It's still going to have

410
00:24:12,123 --> 00:24:15,382
Josh:
these like weird edge cases where it seems like it's acting like a toddler.

411
00:24:15,701 --> 00:24:20,003
Josh:
But for the most part, it is like exceptionally good at a lot of things.

412
00:24:20,003 --> 00:24:22,900
Josh:
And I think for most consumers, that's enough.

413
00:24:23,242 --> 00:24:27,833
Josh:
And for consumers now, this is available on all of the paid plans for ChatGPT.

414
00:24:27,833 --> 00:24:32,257
Josh:
So anyone who is interested can actually go and play with this today as we're recording this.

415
00:24:32,571 --> 00:24:36,071
Josh:
And if you download it on your desktop, you can actually have it start to use

416
00:24:36,379 --> 00:24:39,733
Josh:
computer use and control things and play with it for yourself and see what it's

417
00:24:39,733 --> 00:24:42,439
Josh:
capable of because i mean this this feels like

418
00:24:43,269 --> 00:24:46,593
Josh:
A preview of what's to come for the next year it's like if i could place this

419
00:24:46,593 --> 00:24:49,434
Josh:
it's like okay here's how the paradigm is shifting first we're talking to it

420
00:24:49,800 --> 00:24:52,793
Josh:
now we're talking to it and it's acting on our behalf soon it's just going to

421
00:24:52,793 --> 00:24:55,373
Josh:
remove the display and we'll just have this like a little thing we could talk

422
00:24:55,373 --> 00:24:58,531
Josh:
to and it kind of acts as your assistant we see this happening with grokbot

423
00:24:58,873 --> 00:25:02,073
Josh:
where grokbot's like kind of good ejaz i've been playing with rock pot it sends

424
00:25:02,073 --> 00:25:03,403
Josh:
me morning notifications now

425
00:25:03,863 --> 00:25:06,803
Josh:
it like it knew that my friend was having a birthday it suggested a gift to

426
00:25:06,803 --> 00:25:08,823
Josh:
get him i'm like oh that's kind of nice and cool

427
00:25:09,643 --> 00:25:10,523
Ejaaz:
So yeah yeah.

428
00:25:10,523 --> 00:25:14,563
Josh:
So there's like there's we're we're on the cusp of like the next paradigm and

429
00:25:14,563 --> 00:25:17,473
Josh:
we certainly did have a lull and granted it wasn't really a lull we had what

430
00:25:17,473 --> 00:25:19,973
Josh:
16 models but now we have like okay

431
00:25:20,281 --> 00:25:23,493
Josh:
the big boys are coming back we have the new astro model open ai saying they

432
00:25:23,493 --> 00:25:26,853
Josh:
have more models anthropic very obviously has more models that they're working

433
00:25:26,853 --> 00:25:30,113
Josh:
on we're going to get some new frontier intelligence And the use cases that

434
00:25:30,113 --> 00:25:35,124
Josh:
are going to unlock with that, I cannot wait to play with and uncover because, dude, we're back in it.

435
00:25:36,103 --> 00:25:37,759
Josh:
It feels like we are so back.

436
00:25:38,996 --> 00:25:42,804
Josh:
I spent a lot of time this weekend playing with a model when I haven't done that in a very long time.

437
00:25:42,972 --> 00:25:48,737
Josh:
And just kind of in awe. Like I opened up my video editor that I've spent a decade plus spending.

438
00:25:49,416 --> 00:25:53,347
Josh:
Slaving away at and it just clicked the buttons did the sliders it did everything

439
00:25:53,347 --> 00:25:58,059
Josh:
i needed to do and i was like oh this is it this is like pretty cool this is a good model

440
00:25:58,344 --> 00:26:02,837
Ejaaz:
It's i i have a personal benchmark which is i internally call it the,

441
00:26:03,457 --> 00:26:08,717
Ejaaz:
uh the jarvis benchmark so for any of you who are iron man fans um i basically

442
00:26:08,717 --> 00:26:11,253
Ejaaz:
want a computer that is around me all the time

443
00:26:11,503 --> 00:26:15,177
Ejaaz:
that sounds super friendly and i can just talk to about different things it's

444
00:26:15,177 --> 00:26:18,577
Ejaaz:
like waiting ambiently when it's silent and it learns about myself,

445
00:26:18,577 --> 00:26:22,654
Ejaaz:
but it's always my friend and it works in alignment with whatever my goal is,

446
00:26:22,875 --> 00:26:24,784
Ejaaz:
whether it's personal or professional.

447
00:26:25,080 --> 00:26:28,842
Ejaaz:
And it seems like with this model, compared to any other previous model,

448
00:26:29,837 --> 00:26:34,287
Ejaaz:
we've made the biggest jump and leap towards what that might look like realistically.

449
00:26:34,287 --> 00:26:39,233
Ejaaz:
And you mentioned it earlier in the episode, this isn't just going to be a one model thing.

450
00:26:39,407 --> 00:26:42,137
Ejaaz:
This is going to be a collection of different things. It's going to require

451
00:26:42,137 --> 00:26:46,667
Ejaaz:
hardware. And listen, Apple has been a sleeping giant. We have a very big week for Apple.

452
00:26:46,667 --> 00:26:47,677
Josh:
Buckle up for that one.

453
00:26:47,677 --> 00:26:51,207
Ejaaz:
Buckle up. Very huge week for Apple. You know, we need hardware devices.

454
00:26:51,207 --> 00:26:54,727
Ejaaz:
Apple, you know, they have about three and a half billion, you know,

455
00:26:54,727 --> 00:26:56,079
Ejaaz:
live devices. No biggie, right?

456
00:26:56,329 --> 00:26:59,087
Ejaaz:
They're a lurking giant. But then you have OpenAI that's coming out with their

457
00:26:59,087 --> 00:27:03,851
Ejaaz:
new devices in Q1 of next year, which is also going to compete.

458
00:27:04,675 --> 00:27:07,956
Ejaaz:
And then we see the likes of GrokBot and all these other types of models.

459
00:27:07,956 --> 00:27:11,326
Ejaaz:
Meta, I know, is working on a hardware device where we're starting to see the

460
00:27:11,326 --> 00:27:15,536
Ejaaz:
emergence of ambient AI or I call it ephemeral AI, where it's like,

461
00:27:15,774 --> 00:27:19,936
Ejaaz:
you can kind of like speak and think and the AI just kind of like does stuff.

462
00:27:20,876 --> 00:27:26,496
Ejaaz:
And it's something very futuristic and weird. It's kind of like Blade Runner 2049 type stuff.

463
00:27:26,740 --> 00:27:30,008
Ejaaz:
But we're getting there. And I think we're gonna see that rate of acceleration

464
00:27:30,246 --> 00:27:32,646
Ejaaz:
go up even quicker than we've ever seen before.

465
00:27:32,646 --> 00:27:35,726
Ejaaz:
And I know that we're bullish on this episode and I know that we have to try

466
00:27:35,726 --> 00:27:39,196
Ejaaz:
and ground ourselves and I am, listen, 10 bucks in, $50 out.

467
00:27:39,196 --> 00:27:41,776
Ejaaz:
It's still very expensive for a model, but that's going to get...

468
00:27:41,776 --> 00:27:43,626
Ejaaz:
It's expensive, but it's still going to get cheaper, right?

469
00:27:43,626 --> 00:27:46,306
Ejaaz:
Like when you look at like cost per token, it's still going down.

470
00:27:46,306 --> 00:27:49,686
Ejaaz:
So eventually, like, I'm going to guess in like a month's time,

471
00:27:49,686 --> 00:27:54,306
Ejaaz:
Josh, we're going to have a GPT-6 Astra level model, whether it's not from OpenAI,

472
00:27:54,306 --> 00:27:57,557
Ejaaz:
but whether it's from a Chinese open source or whatever, that is like a fraction of the cost.

473
00:27:57,736 --> 00:28:00,376
Josh:
So that is... And also when we say it's expensive, that's like,

474
00:28:00,376 --> 00:28:02,176
Josh:
that's the API. Most people aren't using the API.

475
00:28:02,256 --> 00:28:02,336
Ejaaz:
Exactly.

476
00:28:02,676 --> 00:28:06,486
Josh:
It's your same $20, $100 plan you've been using. That's unchanged.

477
00:28:06,486 --> 00:28:08,134
Josh:
And I think that's what matters a lot for this model.

478
00:28:08,417 --> 00:28:11,396
Josh:
Because we're not experts. We're not like going and testing out the coding capabilities

479
00:28:11,396 --> 00:28:14,264
Josh:
of cybersecurity. We're just using it, again, as kind of normal dudes.

480
00:28:14,514 --> 00:28:17,306
Josh:
And your price remains unchanged. So I think that's important to know, too.

481
00:28:17,485 --> 00:28:21,438
Ejaaz:
Yeah, exactly. Yeah, I think that the models are...

482
00:28:22,106 --> 00:28:25,606
Ejaaz:
Equally becoming more intelligent than we could have ever expected,

483
00:28:25,926 --> 00:28:28,032
Ejaaz:
but also way more accessible.

484
00:28:28,294 --> 00:28:32,274
Ejaaz:
And that breaks a lot of bare case theses about AI in general,

485
00:28:32,274 --> 00:28:35,394
Ejaaz:
because the take is like, intelligence is going to cost too much.

486
00:28:35,394 --> 00:28:39,734
Ejaaz:
We are applying way too much money for very little reward, and we're seeing the exact opposite.

487
00:28:39,734 --> 00:28:42,764
Ejaaz:
So again, if you're listening to the show, and you think we're overly optimistic,

488
00:28:42,764 --> 00:28:47,194
Ejaaz:
or you're not experiencing the same quality of experience using these different

489
00:28:47,194 --> 00:28:50,494
Ejaaz:
models, we would love to hear from you because we are N

490
00:28:50,834 --> 00:28:53,624
Ejaaz:
data points of two over here you know josh and i go back and forth we're very

491
00:28:53,624 --> 00:28:56,974
Ejaaz:
optimistic about what the future might look like so um we would love to hear

492
00:28:56,974 --> 00:28:59,059
Ejaaz:
your feedback if you're listening to this show on youtube

493
00:28:59,425 --> 00:29:02,214
Ejaaz:
if you listen to us on spotify if you listen to us on apple music wherever you

494
00:29:02,214 --> 00:29:04,138
Ejaaz:
are please give us a rating leave us a comment

495
00:29:04,417 --> 00:29:07,264
Ejaaz:
please follow us if you aren't please subscribe to us if you aren't it helps

496
00:29:07,264 --> 00:29:11,264
Ejaaz:
us out massively the algo has been pumping our videos thank you youtube out

497
00:29:11,264 --> 00:29:15,254
Ejaaz:
there um and so a lot more of you have come on board recently i think we've

498
00:29:15,254 --> 00:29:19,371
Ejaaz:
got like 3 100 new subscribers over the last 28 days it's been crazy hello

499
00:29:19,701 --> 00:29:23,634
Ejaaz:
to you all we post a bunch of times a week and uh we're excited to kind of like

500
00:29:23,634 --> 00:29:25,913
Ejaaz:
take this journey forward josh any final thoughts.

501
00:29:26,157 --> 00:29:30,081
Josh:
Yes i have one final thought yeah this was trained with a hundred thousand gpu

502
00:29:31,126 --> 00:29:35,214
Josh:
like of the old gpus this isn't vera rubin gpu and this is a hundred thousand

503
00:29:35,214 --> 00:29:39,884
Josh:
no we have clusters now of like many hundreds of thousands of gpus and those

504
00:29:39,884 --> 00:29:44,199
Josh:
are soon going to be vera rubin gpus and if it's this good now

505
00:29:44,454 --> 00:29:45,203
Ejaaz:
It's like.

506
00:29:45,505 --> 00:29:50,300
Josh:
What happens when we quadruple the amount of gpus and like 5x the amount of throughput per gpu

507
00:29:53,886 --> 00:29:56,402
Josh:
Dude, it's going to buckle up,

508
00:29:56,402 --> 00:29:58,582
Ejaaz:
Buckle up, buckle up.

509
00:29:58,582 --> 00:30:01,652
Josh:
You know, you start to realize why the people in these labs are talking the

510
00:30:01,652 --> 00:30:05,492
Josh:
way they do, because they have insight into where this is going.

511
00:30:05,492 --> 00:30:08,782
Josh:
And, and like, we're seeing this as outsiders playing around with these tools.

512
00:30:08,782 --> 00:30:11,127
Josh:
And we're like, this is unbelievable. This is kind of like AGI.

513
00:30:11,388 --> 00:30:14,562
Josh:
And they're like, Oh, brother, you haven't seen anything. So buckle up.

514
00:30:14,562 --> 00:30:19,172
Josh:
It's going to be exciting. This is the new beginning. We have a hardware event happening tomorrow.

515
00:30:19,399 --> 00:30:21,802
Josh:
As we're recording this, it's going to be in two days for you. Apple.

516
00:30:22,233 --> 00:30:25,989
Josh:
Are they going to finally deliver the first consumer AI device? Possibly.

517
00:30:26,750 --> 00:30:29,982
Josh:
Possibly. We'll see. We'll have a lot to say on that one. But yeah,

518
00:30:29,982 --> 00:30:32,612
Josh:
that's the episode. Thank you so much for watching as always.

519
00:30:32,952 --> 00:30:34,964
Josh:
And yeah, we'll see you guys in the next one.