1
00:00:00,120 --> 00:00:06,460
Ejaaz:
For years, the battle between open source and centralized AI models has been very one-sided.

2
00:00:06,460 --> 00:00:11,710
Ejaaz:
Anthropic open air has always dominated the field until this week where Moonshot

3
00:00:11,710 --> 00:00:15,990
Ejaaz:
Labs, an AI lab out of China, has finally released their next model,

4
00:00:16,240 --> 00:00:19,900
Ejaaz:
Kimi K3, and it is Fable 5 worthy.

5
00:00:19,900 --> 00:00:23,810
Ejaaz:
Now, the craziest part about this is Chinese and open source models in general

6
00:00:23,810 --> 00:00:27,570
Ejaaz:
have been behind American frontier labs. They haven't been able to close the

7
00:00:27,570 --> 00:00:31,340
Ejaaz:
gap. Dario recently said that that gap is roughly around eight months,

8
00:00:31,340 --> 00:00:32,860
Ejaaz:
having closed from about a year.

9
00:00:33,190 --> 00:00:36,890
Ejaaz:
Today, with Kimi K3, that model has closed it to today.

10
00:00:37,060 --> 00:00:42,290
Ejaaz:
It is as good at one-shotting visual prompts as Fable 5 is. It's amazing at

11
00:00:42,290 --> 00:00:45,830
Ejaaz:
reasoning in general. And we're going to get into a bunch of cool demos that

12
00:00:45,830 --> 00:00:47,380
Ejaaz:
we see on our screen today.

13
00:00:47,600 --> 00:00:50,320
Ejaaz:
Other big news in open source, Thinking Machine Labs, which,

14
00:00:50,320 --> 00:00:53,840
Ejaaz:
funnily enough, raised the largest seed round ever, $2 billion,

15
00:00:54,380 --> 00:01:00,030
Ejaaz:
headed by ex-OpenAI CTO Miro Murati, released their first model called Inkling.

16
00:01:00,220 --> 00:01:02,650
Ejaaz:
It's small, yet pretty mighty. We're going to get into that.

17
00:01:02,650 --> 00:01:07,060
Ejaaz:
And then OpenAI themselves have revealed some secretive details about their

18
00:01:07,060 --> 00:01:10,050
Ejaaz:
new hardware device. If you've watched the show at all, you know Joshua and

19
00:01:10,050 --> 00:01:11,120
Ejaaz:
I are obsessed with this.

20
00:01:11,360 --> 00:01:14,280
Ejaaz:
We're excited to get into the details and show you what is coming soon.

21
00:01:14,600 --> 00:01:17,800
Josh:
Yeah, so to start, we have a new king on the block, and that comes in the form

22
00:01:17,800 --> 00:01:20,710
Josh:
of Kimi K3. I mean, this very much feels like...

23
00:01:21,160 --> 00:01:24,820
Josh:
I wouldn't say the DeepSeek moment, but something somewhat familiar in the sense

24
00:01:24,820 --> 00:01:30,630
Josh:
that this is a really novel, leading edge, open source model coming out of the Kimmy team.

25
00:01:30,880 --> 00:01:33,430
Josh:
And it was paired with this really lovely launch video that actually.

26
00:01:34,170 --> 00:01:37,050
Josh:
I was shocked that it was coming from this lab because this looks like something

27
00:01:37,230 --> 00:01:39,850
Josh:
Apple would release as a product launch video.

28
00:01:40,120 --> 00:01:43,210
Josh:
It was beautiful. The sound design was amazing. The visuals were great.

29
00:01:43,600 --> 00:01:45,900
Josh:
And it's funny, they launched this before they even officially launched the

30
00:01:45,900 --> 00:01:49,150
Josh:
model. In fact, at the time of recording, they haven't even officially stated

31
00:01:49,420 --> 00:01:50,530
Josh:
that the model is live but

32
00:01:50,820 --> 00:01:53,910
Josh:
it's available it is available in the model picker for you to go and test it

33
00:01:53,910 --> 00:01:58,410
Josh:
now the model comes in the form of about a 2.8 trillion parameter model which

34
00:01:58,410 --> 00:02:01,460
Josh:
if you listen to the episode yesterday featuring grok that's about double what

35
00:02:01,460 --> 00:02:03,620
Josh:
grok 4.5 was in terms of parameter count

36
00:02:04,010 --> 00:02:06,880
Josh:
based on the early reviews that i've seen a lot of people have been using this

37
00:02:06,880 --> 00:02:11,550
Josh:
model kind of early it seems like this is pretty incredible um it seems like

38
00:02:11,550 --> 00:02:15,410
Josh:
people have been able to do really impressive long horizon tasks they've been able to do really

39
00:02:15,830 --> 00:02:21,300
Josh:
impressive coding generation gaming generation 3d generation and a lot of people are comparing this to

40
00:02:21,720 --> 00:02:26,850
Josh:
close to a fable worthy model and early testers they put it above gpt 5.6 actually

41
00:02:26,850 --> 00:02:30,320
Josh:
on coding and almost fable worthy at coding which puts it in a really solid

42
00:02:30,320 --> 00:02:34,930
Josh:
spot because the pricing of this thing is going to be very low and the quality

43
00:02:34,930 --> 00:02:36,290
Josh:
of it is incredibly high and to

44
00:02:36,610 --> 00:02:39,810
Josh:
our point yesterday talking about the two frontiers there is the

45
00:02:40,190 --> 00:02:44,050
Josh:
intelligence frontier and there is the cost Frontier, this is placing a new

46
00:02:44,190 --> 00:02:48,950
Josh:
point on that Pareto curve of cost and intelligence at a part of the curve that

47
00:02:48,950 --> 00:02:50,650
Josh:
I don't think anyone's ever made to before.

48
00:02:50,650 --> 00:02:53,990
Josh:
This is like very much a frontier model in terms of cost per intelligence.

49
00:02:53,990 --> 00:02:54,760
Josh:
It's really impressive.

50
00:02:55,070 --> 00:02:59,840
Ejaaz:
I think this tweet that I have over here summarizes it the best.

51
00:02:59,840 --> 00:03:04,280
Ejaaz:
He goes, the more I test K3, the more it feels like another DeepSeek R1 moment.

52
00:03:04,620 --> 00:03:09,880
Ejaaz:
It is often Fable level, maybe a little worse, but consistently better than 5.6.

53
00:03:09,880 --> 00:03:15,110
Ejaaz:
And that's a recurring theme across a lot of different takes from early testers,

54
00:03:15,110 --> 00:03:19,150
Ejaaz:
that it's not quite as good as Fable. It's as good as Fable when it comes to

55
00:03:19,150 --> 00:03:20,830
Ejaaz:
certain things, but not everything.

56
00:03:21,140 --> 00:03:25,690
Ejaaz:
But it is certainly better or on par with GPT 5.6, which is a bold claim,

57
00:03:25,690 --> 00:03:28,610
Ejaaz:
obviously, because OpenAI has invested billions and billions of dollars into

58
00:03:28,610 --> 00:03:33,800
Ejaaz:
their thing. Now, enough of us claiming theory. Let's talk about some actual examples.

59
00:03:34,040 --> 00:03:37,600
Ejaaz:
So on your screen right now, you're looking at the promotional video from the

60
00:03:37,600 --> 00:03:38,860
Ejaaz:
Moonshot Labs themselves.

61
00:03:39,080 --> 00:03:45,120
Ejaaz:
Now, what I'm going to shift to right now is Kimi K3 one-shotting the video entirely from scratch.

62
00:03:45,120 --> 00:03:48,440
Josh:
Whoa. And building it from scratch. Oh, that's really cool.

63
00:03:48,830 --> 00:03:53,280
Ejaaz:
Exactly. This is one single prompt. It was fed the promotional video that Moonshot

64
00:03:53,280 --> 00:03:56,970
Ejaaz:
Labs themselves created from scratch using humans and all that kind of stuff.

65
00:03:57,260 --> 00:04:00,420
Ejaaz:
And it only took 25 minutes to do.

66
00:04:00,660 --> 00:04:04,570
Ejaaz:
Now, rumor, the cost to create this thing. Josh, actually, take a guess.

67
00:04:04,790 --> 00:04:06,030
Ejaaz:
How much did it cost to create this video?

68
00:04:06,330 --> 00:04:10,740
Josh:
Oh, my God. One shot. probably nothing right because the cost per million tokens

69
00:04:10,740 --> 00:04:15,300
Josh:
is 95 cents i think so what would this take maybe five million tokens or say

70
00:04:15,300 --> 00:04:17,370
Josh:
five five to ten dollars no more than 10 bucks.

71
00:04:17,920 --> 00:04:20,810
Ejaaz:
It costs under 10 bucks, which is just insane.

72
00:04:20,810 --> 00:04:22,690
Josh:
Oh, that's so cool. It's insane.

73
00:04:22,690 --> 00:04:29,140
Ejaaz:
Now, like if you have used products from Anthropic or OpenAire that are relevant

74
00:04:29,140 --> 00:04:32,130
Ejaaz:
to artifacts, the visual diagrammatics that they help you create,

75
00:04:33,000 --> 00:04:36,890
Ejaaz:
it takes a while. It takes a couple of bits of prompting, sequential prompting.

76
00:04:37,170 --> 00:04:39,660
Ejaaz:
With Kimi, it seems that it could just one-shot this completely.

77
00:04:39,660 --> 00:04:41,610
Ejaaz:
And it's not just one example I have.

78
00:04:41,860 --> 00:04:46,400
Ejaaz:
I have quite a few. If you look at this, you're looking at a 3D simulation.

79
00:04:46,400 --> 00:04:50,010
Ejaaz:
Now, you would be excused if you thought that this was Fable 5 worthy,

80
00:04:50,010 --> 00:04:54,650
Ejaaz:
because if you rewind a few episodes ago where we covered all of Fable 5's cool

81
00:04:54,650 --> 00:04:57,800
Ejaaz:
demos and stuff, we showed you a Hogwarts demo, a bunch of other things like that.

82
00:04:58,100 --> 00:05:03,350
Ejaaz:
It was of the same quality and par. Kimi K3 can now compete on the same level.

83
00:05:03,350 --> 00:05:07,460
Ejaaz:
And it's not just for simulations. It's recreated Mario Kart from scratch.

84
00:05:07,460 --> 00:05:11,020
Ejaaz:
Now, granted, it looks like Minecraft, but, you know, what AI model doesn't do that?

85
00:05:11,020 --> 00:05:14,660
Ejaaz:
And you could imagine with a few more prompts and like visual graphics added,

86
00:05:14,890 --> 00:05:20,340
Ejaaz:
it could get to, you know, maybe not a AAA, but like a AA type game vibe.

87
00:05:20,340 --> 00:05:24,300
Ejaaz:
So the point is, it's a very impressive model and it's physically accurate.

88
00:05:24,300 --> 00:05:28,460
Ejaaz:
It gets the physics of how things work, the mechanics of wings that you're seeing

89
00:05:29,470 --> 00:05:32,130
Ejaaz:
On your screen right now, how it works. It nails it.

90
00:05:32,350 --> 00:05:36,670
Ejaaz:
Now, where it falls or where it's a little lackluster is general reasoning,

91
00:05:36,950 --> 00:05:41,510
Ejaaz:
writing and coding in general. Now, coding for visual prompts, it's very good.

92
00:05:41,790 --> 00:05:45,770
Ejaaz:
Coding for hard specifically developer tasks, Fable 5 is still good,

93
00:05:46,080 --> 00:05:50,250
Ejaaz:
but it matches the likes of GPT 5.6 Terra and Luna.

94
00:05:50,440 --> 00:05:54,290
Ejaaz:
Maybe not quite Sol, but Terra and Luna and certainly Sol at least on medium

95
00:05:54,290 --> 00:05:57,310
Ejaaz:
effort. So if you're looking for a comparison as to whether you should use this

96
00:05:57,310 --> 00:06:00,390
Ejaaz:
model, it's pretty up there and you should be surprised that they've been able

97
00:06:00,390 --> 00:06:01,750
Ejaaz:
to close the gap so quickly.

98
00:06:02,040 --> 00:06:06,370
Ejaaz:
The final thing I'll say is on cost, it is super cheap. It is under a dollar

99
00:06:06,370 --> 00:06:09,820
Ejaaz:
input tokens, and I think it is a couple dollars output tokens.

100
00:06:09,990 --> 00:06:13,070
Ejaaz:
Again, none of this is officially confirmed. Just yet we're getting like leaks

101
00:06:13,070 --> 00:06:14,410
Ejaaz:
as we're recording this episode.

102
00:06:15,200 --> 00:06:20,330
Ejaaz:
That would price it at the same value as sonnet five or sonnet the sonnet model

103
00:06:20,330 --> 00:06:23,250
Ejaaz:
from from claude themselves so we're not even talking about opus level here

104
00:06:23,250 --> 00:06:25,680
Ejaaz:
we're talking about like sonnet cost which is like cheapest chips

105
00:06:25,830 --> 00:06:29,980
Josh:
Yeah very impressive i mean it's funny as we cover this model i think about the open claw community

106
00:06:30,370 --> 00:06:34,070
Josh:
and people who use these long-running agentic tasks i don't know where they

107
00:06:34,070 --> 00:06:36,750
Josh:
are now i haven't seen many of the open clock community publicly in a while

108
00:06:37,090 --> 00:06:42,040
Josh:
but this is such a dream model for them where it is so cheap it is so capable

109
00:06:42,040 --> 00:06:45,200
Josh:
of these long-running agentic tasks it has all the visual understanding it seems

110
00:06:45,200 --> 00:06:46,290
Josh:
like a very good companion

111
00:06:46,660 --> 00:06:51,890
Josh:
and this seems like a legitimate kind of dent in the moat of these companies

112
00:06:51,890 --> 00:06:54,530
Josh:
that are kind of holding the frontier of that pareto curve in terms of cost

113
00:06:54,530 --> 00:06:57,220
Josh:
in terms of intelligence so it's a really impressive model coming out of china

114
00:06:57,220 --> 00:06:59,770
Josh:
i'd love to know how this came to be

115
00:07:00,180 --> 00:07:05,290
Josh:
what types of models were distilled if any kind of whose ip are you taking here

116
00:07:05,540 --> 00:07:07,980
Josh:
whose stolen gpus are you training this on listen.

117
00:07:07,980 --> 00:07:11,630
Ejaaz:
Josh that's actually important context right you you raise a good point because

118
00:07:11,630 --> 00:07:15,480
Ejaaz:
um moonshot labs the guys that created this model have been accused by not just

119
00:07:15,480 --> 00:07:20,000
Ejaaz:
Anthropite, but also OpenAI and Google for distilling their frontier models.

120
00:07:20,000 --> 00:07:23,030
Ejaaz:
So we have no proof whether they've done the same thing here,

121
00:07:23,030 --> 00:07:25,650
Ejaaz:
but maybe it can be assumed. I don't know.

122
00:07:25,650 --> 00:07:28,550
Josh:
I mean, yeah, I find it hard to believe that it's anything otherwise than they've

123
00:07:28,550 --> 00:07:32,720
Josh:
just distilled the frontier models and turned them into cheaper models that are more capable.

124
00:07:33,070 --> 00:07:36,860
Josh:
It seems like this is kind of well known, that they have smuggled GPUs into

125
00:07:36,860 --> 00:07:38,900
Josh:
the country, that they have like kind of,

126
00:07:39,950 --> 00:07:43,120
Josh:
nudge the numbers on how much the training costs were on a relative basis where

127
00:07:43,120 --> 00:07:44,790
Josh:
they say it's a lot cheaper than it actually is

128
00:07:45,140 --> 00:07:48,520
Josh:
there's a lot of malpractice going on behind these models which is why it's

129
00:07:48,520 --> 00:07:53,590
Josh:
a little difficult to support them but man on just a pure open weight basis

130
00:07:53,590 --> 00:07:56,720
Josh:
for them to publish a model like this it's pretty impressive and i think a lot

131
00:07:56,720 --> 00:07:58,790
Josh:
of people are going to get a lot of use out of this so,

132
00:07:59,250 --> 00:08:01,910
Josh:
i am going to go and try this i'm going to check my model picker after this

133
00:08:01,910 --> 00:08:04,920
Josh:
just to test maybe we'll have a follow-up episode showing some of the demos

134
00:08:05,200 --> 00:08:07,750
Josh:
all the interesting things you could do with it but if you are watching this

135
00:08:07,750 --> 00:08:11,430
Josh:
chances are it is live it's available. Go try it out and let us know what you think.

136
00:08:11,870 --> 00:08:15,230
Josh:
That brings us to our second topic of the day. And this one I am particularly

137
00:08:15,230 --> 00:08:18,210
Josh:
excited about because, Ejaz, if you remember, 18 months ago,

138
00:08:18,210 --> 00:08:24,140
Josh:
after leaving OpenAI, the co-founder of OpenAI, Mur Murati, she left and she

139
00:08:24,140 --> 00:08:26,740
Josh:
went to go start a company named Thinking Machines.

140
00:08:27,000 --> 00:08:30,600
Josh:
She is one of many co-founders who have left the company, unfortunately,

141
00:08:30,790 --> 00:08:33,000
Josh:
for better or worse, and went off and tried to do her own thing.

142
00:08:33,490 --> 00:08:38,070
Josh:
During this process, she raised the largest seed round in venture capital history.

143
00:08:38,260 --> 00:08:42,010
Josh:
Now, seed rounds, generally speaking, I remember back in the day, a seed round of like,

144
00:08:42,460 --> 00:08:44,850
Josh:
a couple million dollars if you raise like five to ten million dollars.

145
00:08:44,850 --> 00:08:48,570
Josh:
That's a remarkable seed round. This was two billion dollars.

146
00:08:48,910 --> 00:08:51,730
Josh:
And for those not familiar, the seed round is the first money into the company.

147
00:08:52,080 --> 00:08:54,720
Josh:
You have an idea and you say, I'm going to go build this idea and an investor

148
00:08:54,720 --> 00:08:55,580
Josh:
gives you some money to do it.

149
00:08:55,890 --> 00:08:59,360
Josh:
She raised two billion dollars at a twelve billion dollar valuation right off

150
00:08:59,360 --> 00:09:03,150
Josh:
the bat from companies like Andreessen Horowitz, NVIDIA, AMD and Jane Street.

151
00:09:03,150 --> 00:09:05,120
Josh:
These are the largest ones in the game.

152
00:09:05,370 --> 00:09:08,380
Josh:
So for the last 18 or so months, we have been asking ourselves the question,

153
00:09:08,380 --> 00:09:10,290
Josh:
well, what are they actually working on?

154
00:09:10,290 --> 00:09:13,870
Josh:
Two billion dollars is a tremendous amount of money to go off and build something

155
00:09:14,180 --> 00:09:17,340
Josh:
and mira and the team are clearly very talented because they were one of the

156
00:09:17,340 --> 00:09:19,690
Josh:
day ones at open ai who were building the gpt

157
00:09:20,070 --> 00:09:24,570
Josh:
early gpt days and as of today we finally have an answer and it comes in the

158
00:09:24,570 --> 00:09:30,150
Josh:
form of this model named ship shipling inkling what is this what did they actually release.

159
00:09:30,390 --> 00:09:34,370
Ejaaz:
It to be fair it doesn't roll off the tongue but it's called inkling and it is a

160
00:09:34,370 --> 00:09:35,390
Josh:
Model that falls.

161
00:09:35,670 --> 00:09:39,430
Ejaaz:
Yeah yeah it's a model that falls just under a trillion parameters in size,

162
00:09:39,430 --> 00:09:42,320
Ejaaz:
so relatively small given the landscape today.

163
00:09:42,640 --> 00:09:48,330
Ejaaz:
But would it surprise you if I said it's a pretty unremarkable model that is

164
00:09:48,530 --> 00:09:53,760
Ejaaz:
remarkable in a very specific way that not many people see? So let me explain what I mean by that.

165
00:09:54,030 --> 00:10:00,600
Ejaaz:
It's trained on about 45 trillion tokens of text, image, video, and audio.

166
00:10:00,600 --> 00:10:05,680
Ejaaz:
Now, the reason why that is pretty cool is that it's completely omni-model.

167
00:10:05,680 --> 00:10:11,250
Ejaaz:
It's like a multi-modal model that you can send text to, you can send audio

168
00:10:11,250 --> 00:10:13,570
Ejaaz:
to, you can speak to it, doesn't interrupt you.

169
00:10:14,790 --> 00:10:18,160
Ejaaz:
It's that type of model that is very intuitive. It feels very human.

170
00:10:18,160 --> 00:10:22,100
Ejaaz:
And Mira Murati released a really interesting blog post earlier this week,

171
00:10:22,370 --> 00:10:24,490
Ejaaz:
which described her vision for Thinking Machine Labs.

172
00:10:24,490 --> 00:10:29,600
Ejaaz:
It's not to be the frontier LLM, it's to be the frontier companion for humans.

173
00:10:29,600 --> 00:10:33,180
Ejaaz:
And so she's building these models in a very specific way. So when we look at

174
00:10:33,180 --> 00:10:35,750
Ejaaz:
the model itself, it was trained in under nine months.

175
00:10:36,000 --> 00:10:39,910
Ejaaz:
It is quite good at reasoning. It is okay at coding.

176
00:10:40,110 --> 00:10:45,660
Ejaaz:
It is nowhere near Fable 5 worthy. But where it gets super powerful is if you take their model,

177
00:10:46,190 --> 00:10:50,740
Ejaaz:
And then you fine tune it for your own specific task department or team.

178
00:10:50,960 --> 00:10:54,280
Ejaaz:
So what that looks like is, and there's a live example of this actually happening

179
00:10:54,280 --> 00:10:58,330
Ejaaz:
right now, where you can take this model and let's say you operate like one

180
00:10:58,330 --> 00:11:00,860
Ejaaz:
of the world's top trading hedge funds, right?

181
00:11:01,160 --> 00:11:04,050
Ejaaz:
You could feed it data. You could run it locally, by the way,

182
00:11:04,050 --> 00:11:05,650
Ejaaz:
feed it data, all encrypted,

183
00:11:05,880 --> 00:11:09,980
Ejaaz:
train the model to understand how your team company operates,

184
00:11:09,980 --> 00:11:13,390
Ejaaz:
what your investment theses are, feed it maybe even proprietary information

185
00:11:13,390 --> 00:11:16,330
Ejaaz:
because, by the way, this model is completely open-weight.

186
00:11:16,330 --> 00:11:20,050
Ejaaz:
So, you know, open sources come back to America. Well done, Mira Morati.

187
00:11:20,430 --> 00:11:24,770
Ejaaz:
And you can hyper-focus this model on your daily trading activities if you are

188
00:11:24,770 --> 00:11:25,790
Ejaaz:
in that hedge fund position.

189
00:11:25,790 --> 00:11:29,230
Ejaaz:
Now, if you don't believe me, there is a live example of this actually happening

190
00:11:29,230 --> 00:11:31,660
Ejaaz:
from none other than thinking machines themselves.

191
00:11:31,660 --> 00:11:35,530
Ejaaz:
They're working reportedly pretty closely with Bridgewater Capital,

192
00:11:35,530 --> 00:11:41,200
Ejaaz:
which is one of the biggest funds in the entire world using their Tinker API to fine-tune models.

193
00:11:41,200 --> 00:11:45,310
Ejaaz:
So the long story short is this model isn't meant to be remarkable on its own.

194
00:11:45,310 --> 00:11:48,980
Ejaaz:
It's meant to be taken by you, whether you're a startup founder or a hobbyist

195
00:11:48,980 --> 00:11:52,660
Ejaaz:
or whatever that might be, and fine-tune it to do the specific tasks that you can do.

196
00:11:52,940 --> 00:11:56,200
Ejaaz:
If you end up doing that, you end up with a more powerful model that is less

197
00:11:56,200 --> 00:12:00,130
Ejaaz:
generalized and more specific than a classic anthropic or open AI model.

198
00:12:00,410 --> 00:12:05,390
Ejaaz:
And it is way, way cheaper than if you use any of these other generalized models.

199
00:12:05,390 --> 00:12:08,560
Ejaaz:
And as we've spoken about on yesterday's episode and a lot of episodes recently,

200
00:12:08,560 --> 00:12:12,110
Ejaaz:
actually, cheaper models might actually end up being the models that are used

201
00:12:12,110 --> 00:12:14,570
Ejaaz:
more than the intelligent models themselves.

202
00:12:15,110 --> 00:12:17,990
Josh:
Yeah, I think this is a really interesting angle. And I'm excited to see that

203
00:12:17,990 --> 00:12:20,790
Josh:
this is the way that they're going. They're not trying to compete with the frontier

204
00:12:20,790 --> 00:12:24,280
Josh:
labs. In fact, to use their own words, they said, it's not the strongest overall

205
00:12:24,280 --> 00:12:28,260
Josh:
model available today, open or closed. So they're very much aware that they are not on the frontier.

206
00:12:28,520 --> 00:12:32,090
Josh:
The interesting thing is addressing this common concern that we see talked about a lot.

207
00:12:32,380 --> 00:12:37,180
Josh:
If you remember, there's this viral clip on CNBC a couple weeks ago with Palantir CEO Alex Karp.

208
00:12:37,640 --> 00:12:41,400
Josh:
And he was describing how close frontier tools are too expensive.

209
00:12:41,400 --> 00:12:44,670
Josh:
And they have very somewhat murky IP protection where you kind of feed these

210
00:12:44,670 --> 00:12:47,950
Josh:
models. your contacts, and then perhaps they go off and use them to train new models.

211
00:12:48,240 --> 00:12:51,430
Josh:
And what Thinking Machines is doing is basically building this infrastructure

212
00:12:51,430 --> 00:12:56,140
Josh:
for you to build and own the IP custom to you. And I think this is the interesting

213
00:12:56,140 --> 00:12:59,680
Josh:
thing where when you think about monetization for a company like this, they raised $2 billion.

214
00:12:59,980 --> 00:13:02,710
Josh:
Now they're just giving away their technology for free. That's the reality to

215
00:13:02,710 --> 00:13:06,380
Josh:
an extent because this model is open source. You can actually go and download

216
00:13:06,380 --> 00:13:08,830
Josh:
it and use it for your own, but they have this platform called Tinker.

217
00:13:09,040 --> 00:13:12,870
Josh:
That's meant to hyper optimize these models specific for your use case with

218
00:13:12,870 --> 00:13:15,860
Josh:
a much less amount of data than traditional models would.

219
00:13:16,050 --> 00:13:20,560
Josh:
So traditionally, these models need like terabytes of data in order to go through

220
00:13:20,560 --> 00:13:24,500
Josh:
this reinforcement learning process and to become hyper specifically good at a certain task.

221
00:13:24,790 --> 00:13:27,990
Josh:
This model and using this Tinker product, you're actually able to feed it a

222
00:13:27,990 --> 00:13:30,210
Josh:
lot less context to get a very specific outcome.

223
00:13:30,210 --> 00:13:34,540
Josh:
So if you want to train it to be a financial wizard, like Blackwater was perhaps

224
00:13:34,540 --> 00:13:37,400
Josh:
doing, or if you want to train to be a professional private chef,

225
00:13:37,400 --> 00:13:38,250
Josh:
if that's what you're doing.

226
00:13:38,780 --> 00:13:42,790
Josh:
It requires a lot less context and you could kind of mold it to the form that

227
00:13:42,790 --> 00:13:45,960
Josh:
you want specific to you and i think that's a really powerful thing this is

228
00:13:45,960 --> 00:13:51,780
Josh:
mostly focused on consumer and enterprise but very specific use cases for it and i think this is a,

229
00:13:52,260 --> 00:13:55,770
Josh:
novel angle in the sense that a lot of the frontier labs their general purpose

230
00:13:55,770 --> 00:13:58,400
Josh:
you go to them with any question you want and they will answer to the best of

231
00:13:58,400 --> 00:14:01,440
Josh:
their abilities generally better than everyone else but this lab will allow

232
00:14:01,440 --> 00:14:03,450
Josh:
you to do a very specific answer

233
00:14:03,870 --> 00:14:08,060
Josh:
with a very low cost using open weight models without any potential leaking

234
00:14:08,060 --> 00:14:09,480
Josh:
of any of your intellectual property.

235
00:14:09,480 --> 00:14:12,580
Josh:
And I think that's a pretty novel use case and really cool to see out of the

236
00:14:12,580 --> 00:14:13,330
Josh:
Thinking Machines team.

237
00:14:13,900 --> 00:14:17,680
Ejaaz:
I think an obvious question to come from this as well is how were they able

238
00:14:17,680 --> 00:14:22,030
Ejaaz:
to achieve this? Well, I'm pleased to say that I think the Chinese are getting

239
00:14:22,240 --> 00:14:23,930
Ejaaz:
a little taste of their own medicine.

240
00:14:24,640 --> 00:14:27,730
Ejaaz:
What I'm showing on the screen here is in order to train this model,

241
00:14:27,730 --> 00:14:31,880
Ejaaz:
thinking machines used DeepSeq's V3 architecture.

242
00:14:31,880 --> 00:14:35,630
Ejaaz:
Now, if you've heard of DeepSeq before, it comes from an AI lab in China,

243
00:14:35,950 --> 00:14:40,160
Ejaaz:
funnily enough, a hedge fund that built their Chinese model called DeepSeq.

244
00:14:40,160 --> 00:14:43,880
Ejaaz:
And it uses this architectural mixture of experts. I'm not going to bore you

245
00:14:43,880 --> 00:14:47,640
Ejaaz:
with the details, but basically, you can have a really large model and only

246
00:14:47,640 --> 00:14:51,030
Ejaaz:
utilize the necessary parts of the model when you're answering a prompt.

247
00:14:51,030 --> 00:14:51,890
Ejaaz:
It's way more efficient.

248
00:14:52,040 --> 00:14:54,690
Ejaaz:
It's way more cheaper. And it allows you to customize it way more.

249
00:14:54,690 --> 00:14:56,430
Ejaaz:
That's how they were able to pull this off.

250
00:14:56,720 --> 00:15:00,680
Ejaaz:
And it's kind of nice to see this reverse distillation. And I say that justingly

251
00:15:00,680 --> 00:15:02,710
Ejaaz:
because, obviously, all the Chinese models are open source.

252
00:15:02,980 --> 00:15:06,680
Ejaaz:
But moving on, Miro Morati is my favorite topic.

253
00:15:07,170 --> 00:15:12,910
Ejaaz:
Mira Marathi's former boss has been up to some good and some no good.

254
00:15:12,910 --> 00:15:16,910
Ejaaz:
If you watched our previous episode earlier this week, you'll know that Sam's

255
00:15:16,910 --> 00:15:18,360
Ejaaz:
in a bit of hot water right now.

256
00:15:19,340 --> 00:15:25,640
Ejaaz:
The big boys at Apple are suing Sam and OpenAI for stealing hardware trade secrets

257
00:15:25,830 --> 00:15:28,670
Ejaaz:
to build their new AI hardware device.

258
00:15:29,000 --> 00:15:33,020
Ejaaz:
And a question on Josh and I's minds was, what on earth is this device?

259
00:15:33,400 --> 00:15:35,960
Ejaaz:
Josh, Maybe we have some light now. Can you give us some details?

260
00:15:35,960 --> 00:15:38,340
Josh:
Oh my God. How many times have we spoken about this on the show?

261
00:15:38,340 --> 00:15:41,180
Josh:
We got to be approaching like double digits now. This is probably the 10th time

262
00:15:41,180 --> 00:15:42,410
Josh:
we've talked about this freaking device.

263
00:15:42,600 --> 00:15:44,900
Josh:
We still don't know what it looks like. I am so desperate to know.

264
00:15:44,900 --> 00:15:48,740
Josh:
I've never wanted to hold a product in my hand so bad in my entire life than

265
00:15:48,740 --> 00:15:52,860
Josh:
I do want to hold this Johnny Ive OpenAI collab device.

266
00:15:53,160 --> 00:15:57,170
Josh:
And it of course has not come without its troubles. They are currently experiencing

267
00:15:57,170 --> 00:16:00,460
Josh:
lawsuits. I'm wondering if those lawsuits have impacted the trajectory of this

268
00:16:00,460 --> 00:16:02,670
Josh:
suite and family of products that they're going to launch.

269
00:16:02,960 --> 00:16:07,680
Josh:
But we do seem to have somewhat of a close confirmation to what the very first

270
00:16:07,680 --> 00:16:11,030
Josh:
device in this family of devices will look like and when we'll actually be able

271
00:16:11,030 --> 00:16:12,940
Josh:
to get our hands on it now mark german who

272
00:16:13,360 --> 00:16:17,260
Josh:
some people might know as the apple leaker he is kind of the guy who you go

273
00:16:17,260 --> 00:16:20,810
Josh:
to when you want to know about leaks he has very reliable sources he is generally

274
00:16:21,040 --> 00:16:24,610
Josh:
right when he publishes these things for his publication i think he works with

275
00:16:24,610 --> 00:16:27,850
Josh:
bloomberg i'm pretty sure that's where he publishes all of these and

276
00:16:28,150 --> 00:16:33,590
Josh:
we have an idea of what this looks like and drumroll please it is coming in a form of a speaker.

277
00:16:33,930 --> 00:16:35,310
Ejaaz:
Which is interesting.

278
00:16:35,560 --> 00:16:41,180
Josh:
It is described as a screen-free smart speaker that will act as a new type of

279
00:16:41,180 --> 00:16:43,410
Josh:
home computer for the AI era.

280
00:16:43,670 --> 00:16:47,280
Josh:
So it's screenless, which is important to note, but it's certainly not senseless.

281
00:16:47,280 --> 00:16:51,580
Josh:
The idea is that it will have a camera plus these environmental sensors, probably like light.

282
00:16:51,870 --> 00:16:55,450
Josh:
It'll have microphones. It'll have a rechargeable battery, which I found interesting

283
00:16:55,450 --> 00:16:58,100
Josh:
because that means that this will be a portable device.

284
00:16:58,280 --> 00:17:02,390
Josh:
And then, Ijaz, I was asking you about this yesterday, is, does that mean that

285
00:17:02,570 --> 00:17:06,110
Josh:
it's going to be something that fits in my pocket? Like, is this the puck?

286
00:17:06,410 --> 00:17:10,150
Josh:
Or is this going to be, I think, of the Amazon Echo, like that little...

287
00:17:10,580 --> 00:17:13,370
Josh:
Little block that sits on your desk and you could kind of carry it from room

288
00:17:13,370 --> 00:17:16,600
Josh:
to room. I'm kind of now hypothesizing of what this is going to look like.

289
00:17:16,950 --> 00:17:22,090
Ejaaz:
Well, the secret is revealed in this tweet from Mark Gurman and none themselves.

290
00:17:22,090 --> 00:17:26,770
Ejaaz:
He says, as one of his first descriptors of this device, that it's a mobile device.

291
00:17:27,390 --> 00:17:29,920
Ejaaz:
And the idea, if you open up this article and read through it,

292
00:17:30,200 --> 00:17:32,680
Ejaaz:
is he talks about this device being somewhat portable.

293
00:17:32,840 --> 00:17:35,900
Ejaaz:
It's meant to live in your home. It's meant to ingest and listen to everything

294
00:17:36,160 --> 00:17:39,570
Ejaaz:
that is said, the conversations that you have, you can also speak to it.

295
00:17:39,850 --> 00:17:44,270
Ejaaz:
But the goal, the stated goal of what OpenAIR is trying to build here is,

296
00:17:44,750 --> 00:17:48,500
Ejaaz:
A companion, a companion that can sit by your side, that you can speak to,

297
00:17:48,500 --> 00:17:51,960
Ejaaz:
that sounds very human, that understands the work or tasks or goals,

298
00:17:51,960 --> 00:17:54,980
Ejaaz:
whatever you're trying to achieve for that day and over a set number of days,

299
00:17:55,300 --> 00:17:57,200
Ejaaz:
and it can work productively with you.

300
00:17:57,200 --> 00:18:03,170
Ejaaz:
Now, a big question for me is, how different is this from just opening up the

301
00:18:03,170 --> 00:18:05,110
Ejaaz:
ChatGPT app on my phone and speaking to it?

302
00:18:05,470 --> 00:18:08,760
Ejaaz:
And I think I've answered my own question, which is, it is quite different.

303
00:18:08,760 --> 00:18:14,270
Ejaaz:
There is a big delta between having a device that is omnipresent,

304
00:18:14,480 --> 00:18:17,620
Ejaaz:
that is ambient, that listens to everything you say, that can chime in when

305
00:18:17,620 --> 00:18:21,950
Ejaaz:
necessary, like a real human does, like a companion or an advisor or a mentor does,

306
00:18:22,240 --> 00:18:25,410
Ejaaz:
versus having something that is a little more archaic, manual,

307
00:18:25,410 --> 00:18:28,040
Ejaaz:
that you have to trigger when you think of it.

308
00:18:28,040 --> 00:18:34,010
Ejaaz:
And I think this kind of like jump going from kind of like manual devices where

309
00:18:34,010 --> 00:18:38,070
Ejaaz:
you have to click buttons, swipe and go to the specific app versus having this

310
00:18:38,070 --> 00:18:41,010
Ejaaz:
omnipresent artificial in terms.

311
00:18:41,010 --> 00:18:43,660
Ejaaz:
It's funny. It's like I'm describing like the Terminator plot right now.

312
00:18:43,660 --> 00:18:46,810
Ejaaz:
And I'm listening to myself and I'm like, oh, God, like maybe this isn't good for us.

313
00:18:47,040 --> 00:18:50,430
Ejaaz:
But I get that that's where we are moving to. Now, of course,

314
00:18:50,430 --> 00:18:52,180
Ejaaz:
all of this is really exciting to see.

315
00:18:52,340 --> 00:18:56,550
Ejaaz:
But if there is a massive Apple lawsuit looming over their heads,

316
00:18:56,810 --> 00:19:01,640
Ejaaz:
there are grounds for this potentially being delayed. Now, the target date is

317
00:19:01,780 --> 00:19:07,300
Ejaaz:
early 2027, specifically February. Sam wants to ship around 100 million units initially.

318
00:19:07,540 --> 00:19:10,040
Ejaaz:
Josh and I are probably going to buy up 50 million of those units.

319
00:19:10,040 --> 00:19:14,450
Ejaaz:
But the point is that this is going to be really in demand, but it might get

320
00:19:14,450 --> 00:19:18,160
Ejaaz:
delayed to later on in the year if Apple has grounds for them stealing secrets.

321
00:19:18,480 --> 00:19:22,990
Josh:
I really hope not. The reports look like it's going to come out at around $200 to $300.

322
00:19:23,220 --> 00:19:25,870
Josh:
And you have to imagine if they're building a hundred million of these,

323
00:19:25,870 --> 00:19:28,810
Josh:
a couple hundred million of these, it's going to be a fairly simple design.

324
00:19:28,810 --> 00:19:31,950
Josh:
There won't be a ton of moving parts. There won't be a lot of moving pieces.

325
00:19:31,950 --> 00:19:34,990
Josh:
This will kind of be a solid machine is my assumption.

326
00:19:35,300 --> 00:19:38,700
Josh:
If I know anything from my obsession of Johnny AI, his design.

327
00:19:39,130 --> 00:19:44,980
Josh:
Entire goal is to reverse the damage that has been done through the smartphone era in the sense that

328
00:19:45,340 --> 00:19:50,940
Josh:
smartphones have taken so much of our lives away from reality and they've placed

329
00:19:50,940 --> 00:19:55,030
Josh:
them into this digital world and i know he is on a mission hell-bent on reversing that

330
00:19:55,430 --> 00:19:58,570
Josh:
and part of that is removing the screen but the second part of that is trying

331
00:19:58,570 --> 00:20:01,740
Josh:
to make the experience more human and i think we got an early release version

332
00:20:01,740 --> 00:20:04,640
Josh:
and look at this with the new chat model from chat gpt

333
00:20:04,960 --> 00:20:08,730
Josh:
on how conversational it is how it feels very warm and friendly and i expect

334
00:20:08,730 --> 00:20:10,420
Josh:
this to very much be an extension of that

335
00:20:10,740 --> 00:20:15,420
Josh:
in a way it's kind of converging on the voice models and miramirati miramirati's

336
00:20:15,420 --> 00:20:18,980
Josh:
like tinkering model where it will kind of build up this context window about

337
00:20:18,980 --> 00:20:23,240
Josh:
you it will be able to be your personal assistant and allow you to remove yourself

338
00:20:23,240 --> 00:20:25,680
Josh:
away from the matrix that he has,

339
00:20:26,180 --> 00:20:30,830
Josh:
created such a large amount of through his hardware devices almost unintentionally

340
00:20:31,190 --> 00:20:34,610
Josh:
and to me that's super exciting and i cannot wait to get my hands on it in early

341
00:20:34,610 --> 00:20:39,260
Josh:
2027 so we will be keeping our eyes peeled any more rumors you know we're going

342
00:20:39,260 --> 00:20:41,370
Josh:
to talk about it because we can't help but talk about it this is like the 10th

343
00:20:41,370 --> 00:20:44,890
Josh:
time but that is the update on the open ai device coming soon.

344
00:20:45,350 --> 00:20:51,220
Ejaaz:
I i will say that um apple is planning to at least announce their ai hardware

345
00:20:51,220 --> 00:20:55,900
Ejaaz:
devices plural by the end of this year so they might just be doing this whole

346
00:20:55,900 --> 00:20:57,850
Ejaaz:
lawsuit thing just to to front run them.

347
00:20:57,850 --> 00:21:01,970
Ejaaz:
And I know that Apple is working on, I believe, some kind of a disc pendant,

348
00:21:02,580 --> 00:21:06,670
Ejaaz:
AirPods with cameras, as well as some smart glasses. So I'm excited to see these

349
00:21:06,670 --> 00:21:10,470
Ejaaz:
two honestly go at it because the end benefit is us. We get to buy a bunch of

350
00:21:10,470 --> 00:21:12,730
Ejaaz:
cool gadgets and hopefully wear them every day.

351
00:21:13,400 --> 00:21:15,780
Ejaaz:
Moving on to the next topic.

352
00:21:16,300 --> 00:21:20,090
Ejaaz:
Elon Musk, surprise, surprise, is in a bit of trouble.

353
00:21:20,410 --> 00:21:26,530
Ejaaz:
Now, he has this product called Grok Build, which is basically their Claude Code competitor.

354
00:21:26,710 --> 00:21:29,920
Ejaaz:
It is Grok used specifically for coding and building different apps.

355
00:21:30,180 --> 00:21:35,990
Ejaaz:
And some people leaked that XAI's Grok Build CLI, so their API basically,

356
00:21:36,210 --> 00:21:40,070
Ejaaz:
or their front-end interface, was taking people's code,

357
00:21:40,390 --> 00:21:46,960
Ejaaz:
uploading it to private servers for SpaceX AI and then using that data to train new models.

358
00:21:46,960 --> 00:21:50,940
Ejaaz:
Now, if you've ever been in this model game for a while, personal data,

359
00:21:50,940 --> 00:21:53,260
Ejaaz:
privacy, and encryption are incredibly important.

360
00:21:53,260 --> 00:21:55,760
Ejaaz:
Anthropic and Open AI take this very, very seriously because,

361
00:21:55,760 --> 00:21:59,000
Ejaaz:
of course, you want to make sure that they're not taking your personal information

362
00:21:59,000 --> 00:22:02,880
Ejaaz:
that you're imparting to these AI models and using it for nefarious ways or

363
00:22:02,880 --> 00:22:04,380
Ejaaz:
taking advantage of them.

364
00:22:05,310 --> 00:22:10,010
Ejaaz:
Grok, or rather SpaceX AI and Elon were caught doing this And they had to do

365
00:22:10,010 --> 00:22:14,270
Ejaaz:
a lot of damage control, which ended up being, a few days later,

366
00:22:14,550 --> 00:22:17,470
Ejaaz:
them open sourcing the entire Grok build.

367
00:22:17,470 --> 00:22:22,340
Ejaaz:
They apologized and basically said, zero data retention policies are super important

368
00:22:22,340 --> 00:22:25,580
Ejaaz:
where we don't retain any of the data. And if we do, we will be upfront.

369
00:22:25,580 --> 00:22:29,330
Ejaaz:
You can turn this off in settings, but it sounded like it was on by default,

370
00:22:29,330 --> 00:22:30,410
Ejaaz:
which is the warning part.

371
00:22:30,740 --> 00:22:34,130
Ejaaz:
But in order to retaliate, they've said, we've open sourced this entire product,

372
00:22:34,330 --> 00:22:38,500
Ejaaz:
take it, fine tune it, do whatever you want with it, and see the code yourself

373
00:22:38,500 --> 00:22:41,130
Ejaaz:
and see that we're not taking any of your stuff.

374
00:22:42,630 --> 00:22:45,550
Ejaaz:
It's a bit of a roller coaster for them. I'm glad they're being able to do damage

375
00:22:45,550 --> 00:22:49,930
Ejaaz:
control, especially as like Elon is like hurling insults to his enemy at Sam

376
00:22:49,930 --> 00:22:53,000
Ejaaz:
and OpenAI, which arguably haven't done this yet.

377
00:22:53,000 --> 00:22:55,550
Josh:
I respect it. It feels like the early days of Facebook, the move fast and break

378
00:22:55,550 --> 00:22:58,870
Josh:
things where they're just going balls to the wall, full speed.

379
00:22:59,110 --> 00:23:01,570
Josh:
If they make a mistake, they correct it. And in fact, I feel like they really

380
00:23:01,570 --> 00:23:02,940
Josh:
even over-indexed on correcting it.

381
00:23:02,940 --> 00:23:06,040
Josh:
They open source everything and they had zero data retention policies.

382
00:23:06,040 --> 00:23:09,580
Josh:
They really went like fully locked down, almost Apple status,

383
00:23:09,580 --> 00:23:12,520
Josh:
where now there is zero data that is going to cross the line.

384
00:23:12,520 --> 00:23:15,210
Josh:
I think this is going to prove to be a very unique advantage.

385
00:23:15,210 --> 00:23:19,590
Josh:
When you think about the Apple edge, a lot of it has to do with guaranteed privacy of data.

386
00:23:19,900 --> 00:23:23,940
Josh:
And if Grok is going to be the only person who is going to provide that guarantee,

387
00:23:24,320 --> 00:23:28,440
Josh:
that is a huge benefit because remember, they have the cost per token down about

388
00:23:28,440 --> 00:23:31,070
Josh:
as low as it gets. They're on the frontier in terms of cost per intelligence.

389
00:23:31,420 --> 00:23:34,250
Josh:
And now if they have privacy baked in, that's a very compelling product that

390
00:23:34,250 --> 00:23:35,620
Josh:
I think is noteworthy to watch.

391
00:23:35,940 --> 00:23:40,690
Josh:
There is a second bit of news about elon and it is about elon not about his

392
00:23:40,690 --> 00:23:42,310
Josh:
company spacex not about tesla

393
00:23:42,540 --> 00:23:47,370
Josh:
but about elon musk the individual who just acquired a company himself for an

394
00:23:47,370 --> 00:23:50,940
Josh:
estimated one billion dollars which i found interesting i was like why is

395
00:23:51,180 --> 00:23:53,470
Josh:
elon the person acquiring this company

396
00:23:53,760 --> 00:23:57,300
Josh:
and i mean surprise surprise it's an energy company named apr energy.

397
00:23:57,950 --> 00:24:02,980
Josh:
And they are a Jacksonville fleet of trailer-mounted mobile gas and diesel turbines.

398
00:24:02,980 --> 00:24:04,920
Josh:
Why on earth would he want to buy this?

399
00:24:05,080 --> 00:24:09,510
Josh:
Well, it's because they have a gigawatt plus of capacity. And to buy a gigawatt

400
00:24:09,510 --> 00:24:13,400
Josh:
of capacity for a billion dollars, it turns out that's a pretty good deal.

401
00:24:13,580 --> 00:24:18,030
Josh:
And as we know, Grok has these Memphis data centers. They have Colossus 1,

402
00:24:18,030 --> 00:24:20,030
Josh:
they have Colossus 2, they need to spin them up.

403
00:24:20,330 --> 00:24:25,060
Josh:
And one of those big constraints has been power. And we've seen how much in-demand

404
00:24:25,060 --> 00:24:29,040
Josh:
power has been in the ability to deploy quickly through companies like Bloom

405
00:24:29,040 --> 00:24:31,030
Josh:
Energy. We've talked about Bloom Energy a lot on the show.

406
00:24:31,320 --> 00:24:34,310
Josh:
Their stock has gone basically vertical recently because they are one of the

407
00:24:34,310 --> 00:24:38,200
Josh:
few people who can modularly deploy energy to these data centers quickly.

408
00:24:38,610 --> 00:24:42,970
Josh:
This company that Elon bought now is able to stand up these projects in about

409
00:24:42,970 --> 00:24:47,960
Josh:
15 to 30 days, which is really exciting because these grid interconnection queues

410
00:24:47,960 --> 00:24:51,010
Josh:
to tap into the mainline grid normally take a couple of years.

411
00:24:51,230 --> 00:24:55,880
Josh:
So this really shrinks that time to get energy to the data center by like a

412
00:24:55,880 --> 00:25:00,190
Josh:
order of magnitude and i think that's a really big deal so this is exciting

413
00:25:00,190 --> 00:25:01,960
Josh:
it was kind of under the radar it didn't get published

414
00:25:02,300 --> 00:25:05,390
Josh:
anywhere it was just kind of discovered through these filings that elon had

415
00:25:05,390 --> 00:25:08,510
Josh:
made the purchase and i think it's a testament to how fast they are moving they

416
00:25:08,510 --> 00:25:12,220
Josh:
need power elon's going out he said i'll get you a gigawatt myself he bought

417
00:25:12,220 --> 00:25:16,040
Josh:
it they're going to deploy it in 15 to 30 days they're going to bring way more gpus online

418
00:25:16,420 --> 00:25:20,550
Josh:
and that's a testament to the velocity in spacex ai so again really bullish

419
00:25:20,550 --> 00:25:24,940
Josh:
news like keep an eye out on them they're they're making their way up to the frontier very quickly.

420
00:25:25,420 --> 00:25:30,210
Ejaaz:
The problem Elon's solving with this is he has way too many GPUs to the tune

421
00:25:30,210 --> 00:25:35,260
Ejaaz:
of like 1 to 1.5 million of all the latest NVIDIA GPUs, but he has no means

422
00:25:35,260 --> 00:25:36,310
Ejaaz:
of bringing them online.

423
00:25:36,650 --> 00:25:43,800
Ejaaz:
The actual wait time to get your GPUs hooked up to the energy grid is around five years right now.

424
00:25:43,800 --> 00:25:48,080
Ejaaz:
And New York City, or rather New York State, this week became the first state

425
00:25:48,080 --> 00:25:53,580
Ejaaz:
to pause or ban data center development and it's following a trend of a few other states.

426
00:25:53,730 --> 00:25:58,580
Ejaaz:
If this continues to happen, workarounds like this, like acquiring companies

427
00:25:58,580 --> 00:26:04,090
Ejaaz:
like this to get portable gas turbines and powering your GPUs may be the only way to work around it.

428
00:26:04,090 --> 00:26:06,870
Ejaaz:
And the important part about this, why is this important at all,

429
00:26:06,870 --> 00:26:09,850
Ejaaz:
is you need compute at the end of the day to build the best models.

430
00:26:09,850 --> 00:26:14,670
Ejaaz:
We've seen Meta release MuseSpark 1.1, which isn't quite Frontier, but is pretty good.

431
00:26:14,840 --> 00:26:19,100
Ejaaz:
And you've seen Grok 4.5 be able to compete with the big boys unexpectedly.

432
00:26:19,100 --> 00:26:21,990
Ejaaz:
And the way they've been able to do this is because they've acquired so much

433
00:26:21,990 --> 00:26:25,390
Ejaaz:
compute, but none of your GPUs make sense. They're worthless pieces of metal

434
00:26:25,390 --> 00:26:27,660
Ejaaz:
if you can't power them up in the first place.

435
00:26:27,980 --> 00:26:31,980
Ejaaz:
One thing I will say is we've spoken about Bloom Energy a lot on this show.

436
00:26:32,750 --> 00:26:35,740
Ejaaz:
Leopold Aschenbrenner, who runs one of the biggest AI funds,

437
00:26:35,740 --> 00:26:38,850
Ejaaz:
24-year-old kid, has made a big investment in Bloom Energy.

438
00:26:39,050 --> 00:26:43,190
Ejaaz:
This is their direct competitor. And Elon just went out and looked at Bloom

439
00:26:43,190 --> 00:26:45,830
Ejaaz:
Energy and thought, I don't want to have to deal with this. I'll just acquire

440
00:26:45,830 --> 00:26:49,430
Ejaaz:
the company themselves and integrate them into what I'm doing here at SpaceX AI.

441
00:26:49,730 --> 00:26:53,280
Ejaaz:
And I think it's an incredibly smart move. Now, moving on.

442
00:26:53,280 --> 00:26:58,720
Ejaaz:
Yeah, there is a popular product that I believe both of us like to use quite

443
00:26:58,720 --> 00:27:00,540
Ejaaz:
a lot called Whisperflow.

444
00:27:00,540 --> 00:27:04,840
Ejaaz:
And the way that that works is instead of typing prompts, it takes a lot of

445
00:27:04,840 --> 00:27:09,260
Ejaaz:
time, you can tap a button, hold it down on your laptop or on your computer.

446
00:27:09,460 --> 00:27:14,890
Ejaaz:
And you can speak into a microphone or speak to your computer and all your words are recorded.

447
00:27:15,130 --> 00:27:19,430
Ejaaz:
And it has been a major unlock for me. Voice to AI is just the way forwards

448
00:27:19,430 --> 00:27:23,260
Ejaaz:
and I honestly can't see anywhere else where it doesn't result in that.

449
00:27:23,720 --> 00:27:28,830
Ejaaz:
Now, this guy, Farza, released a new product from his company that is called

450
00:27:28,830 --> 00:27:30,630
Ejaaz:
Screen Aware Dictation.

451
00:27:30,630 --> 00:27:34,740
Ejaaz:
Now, one problem that I have when I hold down the button in Whisperflow and

452
00:27:34,740 --> 00:27:38,030
Ejaaz:
I speak into it and it records exactly everything that I say is,

453
00:27:38,890 --> 00:27:40,400
Ejaaz:
I'm also thinking in real time.

454
00:27:40,510 --> 00:27:47,050
Ejaaz:
I'm like, oh, I want to say something in a tone that is kind but not too assertive.

455
00:27:47,430 --> 00:27:51,390
Ejaaz:
And I want to talk about this topic. I don't know the definition of this topic,

456
00:27:51,390 --> 00:27:55,060
Ejaaz:
but I need to do some research on that. And if I was using Whisperflow,

457
00:27:55,060 --> 00:27:59,050
Ejaaz:
it would record everything I say verbatim and put it in that email.

458
00:27:59,350 --> 00:28:03,460
Ejaaz:
With this new feature, it listens to you, it understands when you don't understand

459
00:28:03,460 --> 00:28:08,140
Ejaaz:
something, it goes away, does the research, comes back and formulates the reply

460
00:28:08,140 --> 00:28:09,470
Ejaaz:
or the response that you want.

461
00:28:09,730 --> 00:28:13,870
Ejaaz:
And it's this massive jump up where you now have this feature which kind of

462
00:28:13,870 --> 00:28:16,670
Ejaaz:
like simultaneously listens to what you want to say exactly,

463
00:28:17,720 --> 00:28:21,100
Ejaaz:
also understands when you don't want to say something or what you don't know,

464
00:28:21,260 --> 00:28:24,130
Ejaaz:
goes away, does the research, pulls all the context from your personal files

465
00:28:24,130 --> 00:28:27,120
Ejaaz:
that it's connected to, and then comes up with the perfect response.

466
00:28:27,120 --> 00:28:31,620
Ejaaz:
You can literally say, hey, I want to create a Google slide deck that is to

467
00:28:31,620 --> 00:28:34,420
Ejaaz:
do with this document that I have on another screen or another tab.

468
00:28:34,630 --> 00:28:39,260
Ejaaz:
It'll go away, research that, come back, and put that into a very well-presented

469
00:28:39,490 --> 00:28:44,140
Ejaaz:
slide deck. So it's a really amazing product or feature. I've been using it a bit this morning.

470
00:28:44,440 --> 00:28:49,160
Ejaaz:
I think he got overloaded since he announced this literally under 25 uh 24 hours

471
00:28:49,160 --> 00:28:52,250
Ejaaz:
ago and it's got like 3 million views but um i don't know i just want to highlight

472
00:28:52,250 --> 00:28:53,930
Ejaaz:
this feature because i think it's it's really cool

473
00:28:54,170 --> 00:28:56,690
Josh:
Cool yeah i'm excited to give it a shot this feels like something that like

474
00:28:56,690 --> 00:28:58,740
Josh:
open ai can release tomorrow and kind of

475
00:28:59,170 --> 00:29:01,960
Josh:
knock off the the grid and call it a day i think this is a

476
00:29:02,380 --> 00:29:06,260
Josh:
thematically interesting project though because again everyone is really trying

477
00:29:06,260 --> 00:29:09,760
Josh:
to collect the context if you think about what thinking machines released they

478
00:29:09,760 --> 00:29:13,150
Josh:
are just collecting context so that they can give you a custom tailored experience

479
00:29:13,510 --> 00:29:16,650
Josh:
the open ai speaker that is probably going to be coming out its job is just

480
00:29:16,650 --> 00:29:20,170
Josh:
to collect context about you and to then customize your experience same thing

481
00:29:20,170 --> 00:29:23,930
Josh:
with this so much of the ai story now is kind of pivoting to context because

482
00:29:24,240 --> 00:29:27,480
Josh:
the more an ai understands about you understands about your preferences the

483
00:29:27,480 --> 00:29:30,780
Josh:
better it can serve you and this is another interesting example of it i saw

484
00:29:30,780 --> 00:29:33,980
Josh:
there's a few other kind of rapid fire points that we can get through the next

485
00:29:33,980 --> 00:29:37,210
Josh:
one being deep seek is preparing an ipo which is kind of crazy

486
00:29:37,470 --> 00:29:39,960
Josh:
um that's going to be interesting to talk about,

487
00:29:40,840 --> 00:29:43,710
Josh:
are there any interesting talking points with this ipo or is this kind of a

488
00:29:43,720 --> 00:29:44,710
Josh:
rumor you guys do you know.

489
00:29:44,930 --> 00:29:49,990
Ejaaz:
Um it's it's kind of an unofficial confirmed thing at this point a lot of the

490
00:29:49,990 --> 00:29:53,540
Ejaaz:
main investors have kind of been like referring to this ipo and referring to

491
00:29:53,540 --> 00:29:56,640
Ejaaz:
this new raise the reason why this is like compelling news is

492
00:29:56,950 --> 00:30:01,410
Ejaaz:
they literally raised i think 15 billion dollars a week and a half ago

493
00:30:01,860 --> 00:30:06,580
Ejaaz:
so as soon as that round closed they've raised yet another round that values them at like almost

494
00:30:07,080 --> 00:30:09,490
Ejaaz:
one and a half X higher than they were before.

495
00:30:09,820 --> 00:30:13,090
Ejaaz:
And now they're touting the IPO rumor, which you know, they're going to join

496
00:30:13,090 --> 00:30:17,790
Ejaaz:
cheapu and a bunch of other Chinese AI companies in that public race.

497
00:30:17,940 --> 00:30:20,520
Ejaaz:
Another interesting fact is the founder

498
00:30:21,040 --> 00:30:25,840
Ejaaz:
Liang Wenfeng of DeepSeek owns around, I think it's 94% of the company,

499
00:30:26,110 --> 00:30:28,780
Ejaaz:
which means that his net worth is roughly around $26 billion,

500
00:30:28,780 --> 00:30:32,050
Ejaaz:
making him one of the most wealthiest AI CEOs out there.

501
00:30:32,280 --> 00:30:37,510
Ejaaz:
This random guy from China who was a hedge quant fund trader literally about a year and a half ago.

502
00:30:37,540 --> 00:30:41,260
Josh:
It's insane. Good for him, man. He's doing well. And then with that,

503
00:30:41,260 --> 00:30:44,050
Josh:
I think that is pretty much all the news for this week. That's it.

504
00:30:44,100 --> 00:30:44,990
Josh:
Yeah, we got it all covered.

505
00:30:45,220 --> 00:30:47,980
Josh:
It's been an eventful week. There's a lot going on.

506
00:30:48,460 --> 00:30:51,910
Josh:
It's amazing. I'm always surprised at how every week we can sit down here and

507
00:30:51,910 --> 00:30:54,630
Josh:
somehow manage to find so many topics that are interesting to talk about

508
00:30:54,920 --> 00:30:59,070
Josh:
the just frontier and the world of ai and technology is moving so rapidly so

509
00:30:59,220 --> 00:31:01,890
Josh:
if you've made it this far through the episode thank you for watching if you've

510
00:31:01,890 --> 00:31:05,380
Josh:
listened to the three prior episodes this week which i encourage you to do so

511
00:31:05,380 --> 00:31:06,540
Josh:
we had some pretty interesting ones

512
00:31:06,820 --> 00:31:11,750
Josh:
one was about the gpt 5.6 release the other one was about open ai getting sued

513
00:31:11,750 --> 00:31:16,240
Josh:
by apple and the third being about um why the price war actually matters and

514
00:31:16,240 --> 00:31:19,350
Josh:
we referenced that episode quite a bit about how the new frontier is kind of

515
00:31:19,350 --> 00:31:21,770
Josh:
being battled on price instead of intelligence

516
00:31:22,030 --> 00:31:23,830
Josh:
so those were all really great episodes if you enjoy this one

517
00:31:24,100 --> 00:31:27,260
Josh:
please do not forget to share with your friends and if you did make it through

518
00:31:27,260 --> 00:31:30,800
Josh:
all of that congratulations you're caught up you can go touch grass enjoy your weekend

519
00:31:31,140 --> 00:31:34,300
Josh:
and yeah each has any final parting thoughts before we head out for the weekend.

520
00:31:34,940 --> 00:31:39,900
Ejaaz:
I want to know what people are doing on the weekend. Are you using any of these models?

521
00:31:40,640 --> 00:31:43,770
Ejaaz:
What are we tinkering with? You know, pun very much intended.

522
00:31:43,980 --> 00:31:47,550
Ejaaz:
We, I think 90% of our content this week has been about new models.

523
00:31:47,550 --> 00:31:51,420
Ejaaz:
There's OpenAI releasing literally three different models of three different modes.

524
00:31:51,670 --> 00:31:55,340
Ejaaz:
We've got all these new open source models and they're super cheap to use.

525
00:31:55,340 --> 00:31:59,040
Ejaaz:
So I'm wondering what you folks listening to this are tinkering with.

526
00:31:59,040 --> 00:32:02,020
Ejaaz:
What are you doing? What are you building? What are you doing with these models? Are they even useful?

527
00:32:02,210 --> 00:32:06,040
Ejaaz:
Are you still just kind of like ignoring all of this and using Fable 5 or GBT

528
00:32:06,040 --> 00:32:08,420
Ejaaz:
5.6, let us know in the comments, DM us.

529
00:32:08,680 --> 00:32:12,150
Ejaaz:
And the final thing I'll say is we are on the lookout for sponsors.

530
00:32:12,840 --> 00:32:16,620
Ejaaz:
Limitless has spun out. We are officially on our own. We're keeping the lights on through our own.

531
00:32:16,620 --> 00:32:20,860
Ejaaz:
And we would love to partner with someone that has a product or service that

532
00:32:20,860 --> 00:32:24,260
Ejaaz:
they feel incredibly passionate about and want to share with our audience of

533
00:32:24,260 --> 00:32:27,390
Ejaaz:
users who love to use these different things. So if that sounds like you,

534
00:32:27,390 --> 00:32:30,840
Ejaaz:
or if there's a friend that sounds like that, please reach out to them, reach out to us.

535
00:32:30,980 --> 00:32:34,280
Ejaaz:
Our email is in the description below. You can DM us on X or wherever we are.

536
00:32:34,570 --> 00:32:37,350
Ejaaz:
And yeah, we'll see you on the next one. You're all caught up. Yeah.

537
00:32:37,650 --> 00:32:39,790
Josh:
Awesome. Well, have a great weekend, everyone. See you next time.