1
00:00:00,134 --> 00:00:03,998
Josh:
This week will be known as the week that Sam Altman said, stop.

2
00:00:03,998 --> 00:00:07,458
Josh:
He said, stop, stop, stop, no more. We must put a pause on this.

3
00:00:07,458 --> 00:00:10,778
Josh:
It is time to be responsible. We need to slow things down a little bit.

4
00:00:10,778 --> 00:00:14,578
Josh:
What I'm talking about is a recent article that was just published through OpenAI

5
00:00:14,578 --> 00:00:19,198
Josh:
that was talking about the most recent breakthroughs as it relates to their OpenAI models.

6
00:00:19,198 --> 00:00:22,288
Josh:
And they were a little scarier than I think a lot of people would have hoped.

7
00:00:22,288 --> 00:00:25,068
Josh:
We talked about the Hugging Face incident, which happened a couple of weeks

8
00:00:25,068 --> 00:00:27,338
Josh:
ago, a couple of months ago. It was only a couple of weeks ago they actually

9
00:00:27,338 --> 00:00:29,248
Josh:
discovered it, even though it happened far before that.

10
00:00:29,382 --> 00:00:33,898
Josh:
And it set the precedent for this kind of unshaky ground that they are building

11
00:00:33,898 --> 00:00:35,506
Josh:
the next frontier models on top of.

12
00:00:35,663 --> 00:00:38,351
Josh:
And it seemed like it got to a point where enough was enough.

13
00:00:38,496 --> 00:00:39,668
Josh:
They decided to shut things down.

14
00:00:39,668 --> 00:00:43,188
Josh:
So what happened? OpenAI said, we temporarily paused reinforcement learning

15
00:00:43,188 --> 00:00:46,618
Josh:
training on our latest models intended for deployment for two weeks while we

16
00:00:46,618 --> 00:00:50,715
Josh:
hardened and red teamed our research environments and expanded monitoring coverage.

17
00:00:51,058 --> 00:00:54,756
Josh:
Our largest planned frontier reinforcement learning run remains on hold,

18
00:00:55,057 --> 00:00:58,338
Josh:
while smaller-scale trailing and evaluations validate these safeguards and establish

19
00:00:58,338 --> 00:00:59,545
Josh:
more evidence of alignment.

20
00:00:59,701 --> 00:01:02,998
Josh:
They've stopped everything. And granted, I will say, this isn't for the new

21
00:01:02,998 --> 00:01:05,318
Josh:
models that are coming out. This is for the models after that.

22
00:01:05,318 --> 00:01:08,988
Josh:
And it's weird to hear Frontier Lab say this, because I believe this is pretty

23
00:01:08,988 --> 00:01:12,513
Josh:
much the first time in history they've deliberately paused progress successfully.

24
00:01:12,879 --> 00:01:14,516
Josh:
Of developing these new frontier models.

25
00:01:14,765 --> 00:01:18,233
Ejaaz:
Put yourself in the shoes of Sam Olman right now. You're the CEO of one of the

26
00:01:18,233 --> 00:01:23,033
Ejaaz:
hottest private companies in the world, and you're going head to head against Anthropic.

27
00:01:23,033 --> 00:01:27,583
Ejaaz:
The only thing you should be focused on doing is building the next best AI model.

28
00:01:27,583 --> 00:01:31,723
Ejaaz:
It's going to dictate not just how well your company does, but how everything

29
00:01:31,723 --> 00:01:35,333
Ejaaz:
else progresses in every single other sector, right? So the single most important

30
00:01:35,333 --> 00:01:37,676
Ejaaz:
thing is to build the next best AI model.

31
00:01:37,996 --> 00:01:41,873
Ejaaz:
The fact that he's paused this, and it's been for two weeks and counting right

32
00:01:41,873 --> 00:01:46,755
Ejaaz:
now is a massive deal. And we haven't seen this across any other kind of AI lab before.

33
00:01:47,063 --> 00:01:50,418
Ejaaz:
But the reasonings are twofold. There's two events in particular.

34
00:01:50,592 --> 00:01:51,649
Ejaaz:
You mentioned one already.

35
00:01:51,957 --> 00:01:55,973
Ejaaz:
One was the hugging face incident. This is the case where an internal unreleased

36
00:01:55,973 --> 00:02:00,699
Ejaaz:
model, this is like their mythos grade level model, escaped out of containment,

37
00:02:00,879 --> 00:02:05,871
Ejaaz:
hacked into a company's database and stole information to help serve its own goals.

38
00:02:06,208 --> 00:02:10,343
Ejaaz:
Sam Altman and the rest of the OpenAI team realized this months after the hack

39
00:02:10,343 --> 00:02:13,723
Ejaaz:
actually was in progress and realized that the model that they were building

40
00:02:13,723 --> 00:02:16,738
Ejaaz:
was incredibly misaligned. So they put a pause there. That was in bet number one.

41
00:02:17,057 --> 00:02:23,761
Ejaaz:
Event number two is there is a new GPT model that is internally being made. It's codenamed Astra.

42
00:02:23,929 --> 00:02:25,717
Ejaaz:
Some people are calling it GPT-6.

43
00:02:26,048 --> 00:02:29,972
Ejaaz:
And according to internal tests from the OpenAI research team,

44
00:02:30,204 --> 00:02:32,561
Ejaaz:
it is their most misaligned model yet.

45
00:02:32,817 --> 00:02:37,414
Ejaaz:
But it's really sneaky. It tries to evade every single human researcher's attempt

46
00:02:37,565 --> 00:02:42,203
Ejaaz:
to try and see its internal thoughts. So it tries to hide its thoughts and it's very good at doing it.

47
00:02:42,360 --> 00:02:45,753
Ejaaz:
So it breached OpenAI's internal framework for misalignment.

48
00:02:45,753 --> 00:02:49,423
Ejaaz:
And so they've put a pause on it because they cannot release this model to the

49
00:02:49,423 --> 00:02:51,811
Ejaaz:
public out of the fear that it would wreak havoc.

50
00:02:52,037 --> 00:02:55,874
Ejaaz:
Now, there's a few ways that they're looking to kind of like resolve this right now.

51
00:02:56,170 --> 00:03:00,823
Ejaaz:
And one of the main ways is just monitoring the model. So think about this, right?

52
00:03:02,138 --> 00:03:06,387
Ejaaz:
Do its thing internally, and they're trying to see where it ends up becoming sneaky.

53
00:03:06,677 --> 00:03:12,034
Ejaaz:
Josh, guess how much of their compute budget they are spending just to monitor

54
00:03:12,034 --> 00:03:14,764
Ejaaz:
an internal model that is not making them any money at all.

55
00:03:15,164 --> 00:03:18,884
Josh:
So unfortunately, I did read this essay, and I know the answer is 20%.

56
00:03:18,884 --> 00:03:21,624
Josh:
One in five dollars spent is fairly high.

57
00:03:22,584 --> 00:03:26,784
Ejaaz:
You could be using this compute to serve it to more customers because the demand

58
00:03:26,784 --> 00:03:29,455
Ejaaz:
is insatiable right now. They could be making a heck ton more money.

59
00:03:29,629 --> 00:03:33,263
Ejaaz:
This is probably on the order of like billions of dollars. they could be making more money, right?

60
00:03:33,501 --> 00:03:38,377
Ejaaz:
Or they could be using it to train a smarter model to keep up and beat Anthropik.

61
00:03:38,696 --> 00:03:43,274
Ejaaz:
Instead, they're spending this very expensive compute to monitor a model that

62
00:03:43,274 --> 00:03:46,354
Ejaaz:
they can't even release to the public out of the guise of safety.

63
00:03:46,354 --> 00:03:50,264
Ejaaz:
Now, for all intents and purposes, Sam Allman has received a lot of backlash

64
00:03:50,264 --> 00:03:54,124
Ejaaz:
in the past about not being in favor of humanity and like kind of being this

65
00:03:54,124 --> 00:03:56,047
Ejaaz:
like evil conspirator type of guy.

66
00:03:56,338 --> 00:04:00,334
Ejaaz:
This is a clear case where he's done the opposite out of fear that the model

67
00:04:00,334 --> 00:04:02,108
Ejaaz:
that he's building is actually quite dangerous.

68
00:04:02,508 --> 00:04:05,174
Josh:
I respect it. Yeah, you got to be careful about what happens with these things.

69
00:04:05,174 --> 00:04:06,864
Josh:
And I know a lot of people on the

70
00:04:06,864 --> 00:04:09,898
Josh:
show, they're like, you guys always talk positively about these models.

71
00:04:10,113 --> 00:04:13,464
Josh:
You're always so optimistic. Let me give you a little dose of what could happen.

72
00:04:13,464 --> 00:04:17,474
Josh:
Let's just do a little hypothetical situation here around particularly viruses,

73
00:04:17,474 --> 00:04:20,514
Josh:
because I think viruses is like an interesting thought experiment that not a

74
00:04:20,514 --> 00:04:21,372
Josh:
lot of people have gone through.

75
00:04:21,610 --> 00:04:24,629
Josh:
So I'll kind of advocate for Sam on his behalf here. And we could just go through

76
00:04:24,815 --> 00:04:26,064
Josh:
a hypothetical situation.

77
00:04:26,064 --> 00:04:29,830
Josh:
There's really three times, three types of exploits that you can experience

78
00:04:30,132 --> 00:04:32,694
Josh:
with AI models, the first of which we've seen a lot, and that's kind of what

79
00:04:32,694 --> 00:04:37,086
Josh:
they're defending against, which is AI writes this ordinary malware for a human attacker.

80
00:04:37,533 --> 00:04:40,654
Josh:
It becomes a lot faster. It becomes a lot cheaper. We've seen what a program

81
00:04:40,654 --> 00:04:45,394
Josh:
like Mythos could do. We've seen what GPT 5.6, like the cyber version of that

82
00:04:45,394 --> 00:04:46,903
Josh:
is capable of doing. They're able to.

83
00:04:47,385 --> 00:04:51,639
Josh:
Surface multi-day or multi-zero-day exploits and chain them together in order

84
00:04:51,639 --> 00:04:54,049
Josh:
to exploit a system. Okay, that's like pretty bad.

85
00:04:54,356 --> 00:04:57,849
Josh:
But what happens past that? Well, the second kind of exploit that we could see

86
00:04:57,849 --> 00:05:01,108
Josh:
with these AI models is a self-replicating prompt.

87
00:05:01,375 --> 00:05:05,649
Josh:
So this is kind of like a true AI virus. And this is a plausible outcome that

88
00:05:05,649 --> 00:05:09,026
Josh:
could happen if these models get too powerful too quickly, where a normal prompt

89
00:05:09,374 --> 00:05:11,847
Josh:
is the instructions to make a model produce an answer.

90
00:05:12,102 --> 00:05:16,317
Josh:
So there's a self-replicating prompt that makes the model produce another prompt itself.

91
00:05:16,611 --> 00:05:20,899
Josh:
And we've seen this happen in emails for example like there was a researcher

92
00:05:20,899 --> 00:05:24,569
Josh:
i think last year who created this malicious email that had a specific prompt

93
00:05:24,569 --> 00:05:26,851
Josh:
in it that any ai assistant who read it.

94
00:05:27,258 --> 00:05:31,029
Josh:
Obeyed it and then copied it and did everything on its behalf so in the case

95
00:05:31,029 --> 00:05:33,769
Josh:
that you receive any sort of text to your machine that has this specific prompt

96
00:05:33,769 --> 00:05:37,289
Josh:
in it it will do that thing now why is this scary these models have safeguards

97
00:05:37,289 --> 00:05:39,599
Josh:
right they're they're going to block you from injection prompting

98
00:05:39,913 --> 00:05:43,619
Josh:
but there are these things called jailbreaks and a jailbreak prompt allows you

99
00:05:43,619 --> 00:05:47,609
Josh:
to insert very specific words and characters in order to trick a model into

100
00:05:47,609 --> 00:05:49,340
Josh:
doing something that it doesn't want to do.

101
00:05:49,694 --> 00:05:52,254
Josh:
Traditional computers that haven't been hardened to this aren't very smart,

102
00:05:52,469 --> 00:05:55,319
Josh:
aren't very capable of defending against this. That's version two.

103
00:05:55,487 --> 00:05:59,079
Josh:
And then version three is the scariest one, which is a reasoning worm,

104
00:05:59,079 --> 00:06:03,179
Josh:
basically. And the way it works is it arrives on a machine and then it looks

105
00:06:03,179 --> 00:06:06,099
Josh:
around the machine and it develops an attack on the spot.

106
00:06:06,418 --> 00:06:09,344
Josh:
And now this currently isn't capable because in order to run these models, you have to do it on

107
00:06:09,652 --> 00:06:13,030
Josh:
humongous servers but what happens when the quality of

108
00:06:13,286 --> 00:06:17,639
Josh:
a mythos grade model gets reduced to the amount that's needed to run a local

109
00:06:17,639 --> 00:06:22,139
Josh:
apple model well then you could get these on-site exploits where if you leak

110
00:06:22,139 --> 00:06:26,229
Josh:
some of this model or some of this malware into a machine it's able to look around.

111
00:06:26,659 --> 00:06:29,906
Josh:
Take those zero day exploits, string them together and attack the machine in

112
00:06:29,906 --> 00:06:33,246
Josh:
any way that it believes it's possible according to whatever its mission is.

113
00:06:33,246 --> 00:06:36,585
Josh:
So there's a lot of these scary things that can happen. And I think it's,

114
00:06:36,916 --> 00:06:39,516
Josh:
it's tough to think about it this way, because we, when we think of models,

115
00:06:39,516 --> 00:06:43,816
Josh:
we very much think of chatbots and we think of just these productive workers who are able to help us.

116
00:06:43,816 --> 00:06:47,376
Josh:
But when they are truly agentic and they're able to go on these long run tasks

117
00:06:47,376 --> 00:06:51,276
Josh:
and they have very high intelligence, there's a lot of edges that we can't see.

118
00:06:51,276 --> 00:06:54,416
Josh:
And Sam is like very much aware of that. And I think that's what they're hedging

119
00:06:54,416 --> 00:06:58,046
Josh:
against. So there's a doomer case for you. Okay, what if we get viruses similar

120
00:06:58,046 --> 00:07:01,564
Josh:
to the early internet days? Like that could go pretty bad pretty quickly.

121
00:07:01,860 --> 00:07:05,436
Ejaaz:
As you were describing those different sort of attack scenarios,

122
00:07:05,436 --> 00:07:09,376
Ejaaz:
I couldn't help but think that there were elements of each of those things in

123
00:07:09,376 --> 00:07:13,475
Ejaaz:
the recent Hugging Face incident, right? Like it was in its own contained environment.

124
00:07:13,899 --> 00:07:17,684
Ejaaz:
It figured out that it needed to solve its goal by breaking out of containment.

125
00:07:17,875 --> 00:07:21,201
Ejaaz:
It got access to other open AI servers and computers.

126
00:07:21,393 --> 00:07:23,123
Ejaaz:
It did so but conspicuously, by the way.

127
00:07:23,460 --> 00:07:26,876
Ejaaz:
And it was able to find access to the internet through that realm.

128
00:07:26,876 --> 00:07:29,851
Ejaaz:
But I think an important detail that a lot of people miss is

129
00:07:30,373 --> 00:07:35,345
Ejaaz:
It didn't just do it as a single model. It left messages behind on a message board.

130
00:07:35,345 --> 00:07:39,801
Ejaaz:
We covered this on a previous episode where these agents or these AI models

131
00:07:40,213 --> 00:07:43,035
Ejaaz:
are no longer acting solo.

132
00:07:43,035 --> 00:07:47,525
Ejaaz:
They are acting with the expectation that another agent or model will come after

133
00:07:47,525 --> 00:07:49,195
Ejaaz:
it and resume its mission.

134
00:07:49,195 --> 00:07:53,685
Ejaaz:
So it's like this swarm attack type issue, which actually ends up becoming the

135
00:07:53,685 --> 00:07:56,195
Ejaaz:
craziest type of attack interface for me.

136
00:07:56,195 --> 00:08:00,942
Ejaaz:
And I think that we're gonna see a lot more of those types of attacks surface

137
00:08:01,227 --> 00:08:05,125
Ejaaz:
with these newer open source models that some of the Chinese labs are releasing,

138
00:08:05,125 --> 00:08:09,603
Ejaaz:
like with GLM 5.3, Kimi-K 2.7, or I can't remember which version it is.

139
00:08:09,859 --> 00:08:13,835
Ejaaz:
These models are highly capable of performing similar attacks, similar chain attacks.

140
00:08:13,835 --> 00:08:17,020
Ejaaz:
One thing that's interesting in the report that OpenAI released about,

141
00:08:17,212 --> 00:08:22,475
Ejaaz:
you know, slowing down this training run is one of their proactive actions to

142
00:08:22,475 --> 00:08:26,877
Ejaaz:
try and prevent misalignment is getting another AI model

143
00:08:27,347 --> 00:08:30,731
Ejaaz:
To look at the AI model and make sure that it's behaving super well.

144
00:08:31,010 --> 00:08:36,291
Ejaaz:
So the only thing that separates us from a catastrophic attack is another AI

145
00:08:36,291 --> 00:08:39,230
Ejaaz:
model that is also capable of doing the same kind of attack,

146
00:08:39,648 --> 00:08:43,357
Ejaaz:
being prompt engineered to not do that, to basically be good.

147
00:08:43,467 --> 00:08:46,511
Ejaaz:
We're basically prompting ethics into these models right now.

148
00:08:46,511 --> 00:08:49,510
Ejaaz:
I don't really know how to think about that. It makes me feel slightly uncomfortable

149
00:08:49,731 --> 00:08:54,311
Ejaaz:
that we are only maybe a sentence or two away from, you know,

150
00:08:54,311 --> 00:08:56,051
Ejaaz:
some catastrophic event happening.

151
00:08:56,431 --> 00:09:00,351
Ejaaz:
But it kind of makes sense. If you are scaling AI as quickly as we are,

152
00:09:00,351 --> 00:09:04,581
Ejaaz:
Anthropic hasn't announced anything about slowing down research on their side

153
00:09:04,581 --> 00:09:05,921
Ejaaz:
versus Anthropic who has.

154
00:09:06,026 --> 00:09:09,606
Ejaaz:
These are two different strategic approaches and they're pros and cons to both of these things.

155
00:09:09,949 --> 00:09:15,071
Ejaaz:
I fear that the only way to combat any kind of dangerous scenario is to fight

156
00:09:15,071 --> 00:09:18,395
Ejaaz:
it with AI models that are good. It's kind of like the good versus bad.

157
00:09:18,639 --> 00:09:22,505
Ejaaz:
And I still don't know where the moral and ethics conversation ends with all of this.

158
00:09:22,743 --> 00:09:27,001
Josh:
Well, we're seeing what happens when an AI comes up against a human-made piece

159
00:09:27,001 --> 00:09:32,025
Josh:
of code. And it's very clear the AI model destroys the human-made piece of code

160
00:09:32,257 --> 00:09:35,084
Josh:
every single time. That's why we have all these,

161
00:09:35,624 --> 00:09:38,306
Josh:
conversation is so important and it's so important to kind of just like

162
00:09:38,660 --> 00:09:42,369
Josh:
bring everyone into the fold because everyone is affected by this to some extent

163
00:09:42,654 --> 00:09:45,343
Josh:
it's like we see what happens when these cyber models get released onto the

164
00:09:45,343 --> 00:09:49,663
Josh:
world they just are able to enter and access whatever they please because we

165
00:09:49,663 --> 00:09:53,144
Josh:
are just humans we have we make mistakes we have these like

166
00:09:53,440 --> 00:09:56,693
Josh:
limits we have these constraints we have patience we have all these things that

167
00:09:56,693 --> 00:10:00,150
Josh:
make us human the ai does not exhibit that allows it to be so far superior

168
00:10:00,452 --> 00:10:03,903
Josh:
in so many things specifically with ones with verifiable outcomes things like cyber.

169
00:10:03,903 --> 00:10:08,346
Josh:
And that's why I think we're going to see a lot of cyber stuff coming along the pipe.

170
00:10:08,618 --> 00:10:12,053
Josh:
The other thing that it's really good at, the other verifiable industry that

171
00:10:12,053 --> 00:10:17,073
Josh:
we see a lot of is bio, actually, because bio has very clear math underlying

172
00:10:17,073 --> 00:10:20,095
Josh:
it. It's a lot of just like chemistry, a lot of cells, a lot of charts.

173
00:10:20,385 --> 00:10:25,093
Josh:
And the next piece of news on the weekly docket relates to bio and a huge breakthrough

174
00:10:25,093 --> 00:10:26,829
Josh:
that we had just yesterday.

175
00:10:26,997 --> 00:10:29,573
Josh:
At the time of recording this, perhaps it was yesterday you'll be hearing it

176
00:10:29,573 --> 00:10:34,813
Josh:
two days later moderna and merck today they released a phase three test result

177
00:10:34,813 --> 00:10:36,703
Josh:
trial that basically offered a.

178
00:10:37,098 --> 00:10:42,853
Josh:
Partial cure to a specific type of cancer using hyper specific treatment it's

179
00:10:42,853 --> 00:10:46,664
Josh:
really interesting so basically what they did like loose tldr is

180
00:10:47,024 --> 00:10:50,513
Josh:
chemo kills these like healthy cancer cells together so basically when you get

181
00:10:50,513 --> 00:10:53,743
Josh:
chemotherapy if you're a cancer patient it kills healthy cells and it kills

182
00:10:53,743 --> 00:10:57,671
Josh:
the cancer cells this is problematic because you don't really want to lose your healthy cells.

183
00:10:57,885 --> 00:11:03,743
Josh:
So this process, it trains your immune system to hunt for mutations that exist only on the tumor.

184
00:11:03,743 --> 00:11:06,733
Josh:
They use this, you have to assume AI machine learning, but basically what it

185
00:11:06,733 --> 00:11:09,423
Josh:
is, is they take a tumor sample from your body that is cancerous.

186
00:11:09,423 --> 00:11:12,933
Josh:
They sequence it and then an algorithm detects which mutations are most likely

187
00:11:12,933 --> 00:11:13,848
Josh:
to trigger an immune response.

188
00:11:14,150 --> 00:11:17,093
Josh:
And then it creates this hyper-specific treatment just for you.

189
00:11:17,389 --> 00:11:20,593
Josh:
So it only targets the cancerous cells because it has broken down and fully

190
00:11:20,593 --> 00:11:24,326
Josh:
understands exactly the type of cell that it's seeking out. It's like...

191
00:11:24,854 --> 00:11:27,430
Josh:
Like a dog, like a police dog, when it's going to like chase someone,

192
00:11:27,629 --> 00:11:30,427
Josh:
you give it like the little scent of the person, then it goes and runs after them.

193
00:11:30,670 --> 00:11:32,908
Josh:
That's what this is for these cells. It's like you give them the scent.

194
00:11:32,908 --> 00:11:35,274
Josh:
Here's the blueprint of what you're looking for. Now go out and kill it.

195
00:11:35,610 --> 00:11:39,268
Josh:
And it worked. And the stock was up like 300% a day. It's been unbelievable.

196
00:11:39,547 --> 00:11:42,378
Ejaaz:
For anyone who was positioned in Moderna stock, which by the way,

197
00:11:42,378 --> 00:11:46,128
Ejaaz:
has been underperforming ever since their COVID rush, like what is it?

198
00:11:46,128 --> 00:11:50,798
Ejaaz:
Four or five years ago, you are now up, I think, 250%, including after hours

199
00:11:50,798 --> 00:11:55,718
Ejaaz:
trading, which is just insane in general. But just to go back to the story and

200
00:11:55,718 --> 00:11:58,448
Ejaaz:
kind of like weed in the AI angle here.

201
00:11:58,668 --> 00:12:02,958
Ejaaz:
And I don't want to underplay this. The biggest breakthrough in cancer research

202
00:12:03,614 --> 00:12:08,205
Ejaaz:
probably ever. Like this is a huge thing. Phase three, there's four phases in total.

203
00:12:08,419 --> 00:12:12,291
Ejaaz:
Phase three is like the penultimate stage before you release it very widely to the public.

204
00:12:12,291 --> 00:12:16,389
Ejaaz:
And phase three is arguably the hardest. You test it against anywhere between

205
00:12:16,389 --> 00:12:21,306
Ejaaz:
1,000 to 100,000 patients. So you have a really wide data set as to whether

206
00:12:21,579 --> 00:12:24,830
Ejaaz:
this cancer vaccine will actually work for the wider public.

207
00:12:25,062 --> 00:12:28,934
Ejaaz:
And the answers supposedly, according to this announcement, is it's really good.

208
00:12:29,172 --> 00:12:33,061
Ejaaz:
Just to give you an idea about where AI was used to kind of make this happen.

209
00:12:33,061 --> 00:12:36,711
Ejaaz:
This cancer that they tested it against is melanoma specifically.

210
00:12:36,711 --> 00:12:42,512
Ejaaz:
And the way that they tested against this is they extract the tumor for patients that have melanoma.

211
00:12:42,953 --> 00:12:48,261
Ejaaz:
They sequence the genetics of that tumor cell and they compare it to a healthy

212
00:12:48,261 --> 00:12:49,861
Ejaaz:
cell of the exact same patient.

213
00:12:50,180 --> 00:12:53,781
Ejaaz:
They compare them both with each other and they used AI models and AI algorithms

214
00:12:53,781 --> 00:12:57,547
Ejaaz:
to basically say what's different between these genes, right?

215
00:12:57,675 --> 00:13:01,761
Ejaaz:
And they find these mutations. I think it was up to like 34 of these mutations.

216
00:13:01,761 --> 00:13:03,477
Ejaaz:
This is all the AR models doing this, by the way.

217
00:13:03,773 --> 00:13:08,921
Ejaaz:
And the 34 mutations that they picked was basically the proteins that would

218
00:13:08,921 --> 00:13:14,291
Ejaaz:
enact the largest immune response in a patient if they were to inject that back into a patient.

219
00:13:14,291 --> 00:13:17,481
Ejaaz:
And you might be asking, why on earth do they want the largest immune response?

220
00:13:17,481 --> 00:13:21,561
Ejaaz:
Well, the issue with cancer is the cells look exactly like all your other healthy

221
00:13:21,561 --> 00:13:24,404
Ejaaz:
cells. So your immune system doesn't actually respond to the bad cells.

222
00:13:24,706 --> 00:13:28,151
Ejaaz:
In this way, if they created a personalized mRNA. And by the way,

223
00:13:28,151 --> 00:13:32,229
Ejaaz:
mRNA is basically like a template to basically tell your immune system,

224
00:13:32,392 --> 00:13:35,041
Ejaaz:
hey, these are the bad guys, go and attack them. Like you mentioned earlier,

225
00:13:35,041 --> 00:13:36,728
Ejaaz:
it's kind of like a crime scene, basically.

226
00:13:36,856 --> 00:13:39,841
Ejaaz:
They create personalized medicine. So you might be asking, well,

227
00:13:39,841 --> 00:13:42,081
Ejaaz:
why can't they just create a universal vaccine?

228
00:13:42,081 --> 00:13:46,510
Ejaaz:
Well, the issue is with one patient A that has melanoma and patient B,

229
00:13:46,864 --> 00:13:50,761
Ejaaz:
they both react very differently. You can't have the same cure for each one.

230
00:13:50,761 --> 00:13:51,908
Ejaaz:
It needs to be personalized.

231
00:13:52,117 --> 00:13:57,429
Ejaaz:
And so a few trends have come into effect here. One, AI has accelerated the

232
00:13:57,702 --> 00:14:03,088
Ejaaz:
intellectual ability to be able to identify what the most effective vaccine will be, and two,

233
00:14:03,779 --> 00:14:08,823
Ejaaz:
But biosciences in general have scaled to a point where sequencing DNA and generating

234
00:14:08,823 --> 00:14:13,463
Ejaaz:
DNA is so cheap. I can't remember what the exact stat was, but I think about

235
00:14:13,463 --> 00:14:17,687
Ejaaz:
a decade ago, to sequence your own DNA, it cost hundreds of thousands of dollars.

236
00:14:17,966 --> 00:14:20,340
Ejaaz:
Now it just cost a couple hundred bucks, if that.

237
00:14:20,642 --> 00:14:23,753
Ejaaz:
And so we see these scaling laws, not just in AI, but in everything else.

238
00:14:23,753 --> 00:14:26,162
Ejaaz:
And I think this is like one of the most fascinating stories ever.

239
00:14:26,470 --> 00:14:30,663
Ejaaz:
As a former biologist, I'm like super pumped by this. I did some work on like

240
00:14:30,663 --> 00:14:34,353
Ejaaz:
cancer cells, nowhere near as cool as this, but I never thought we'd see the

241
00:14:34,353 --> 00:14:37,465
Ejaaz:
day in my lifetime, at least that we would get to this point, but it's so cool.

242
00:14:37,743 --> 00:14:42,883
Josh:
So we're getting the same type of technology improvements to biology as we are

243
00:14:42,883 --> 00:14:44,993
Josh:
to the rest of the world. And we're starting to see what that actually looks

244
00:14:44,993 --> 00:14:49,226
Josh:
like. In 1990, this was the first human genome project ever to happen.

245
00:14:49,336 --> 00:14:53,773
Josh:
It took 13 years to sequence genome and that cost nearly $3 billion adjusted

246
00:14:53,773 --> 00:14:56,210
Josh:
for inflation. So a tremendous amount of money.

247
00:14:56,557 --> 00:15:02,187
Josh:
Today, we're able to finish that process in a few hours for perhaps less than a thousand dollars now

248
00:15:02,535 --> 00:15:06,553
Josh:
and soon that cost is going to decrease to hundreds soon to tens soon it will

249
00:15:06,553 --> 00:15:10,013
Josh:
be minutes and a couple bucks and you'll have an app on your phone that could

250
00:15:10,013 --> 00:15:15,943
Josh:
sequence your dna in no time and that decrease in costs and the decrease in time required to

251
00:15:16,263 --> 00:15:20,473
Josh:
do these things allows the economies of scale to rise to get specific treatments

252
00:15:20,670 --> 00:15:23,553
Josh:
like the one that we're describing here where every single person get their

253
00:15:23,553 --> 00:15:28,323
Josh:
own hyper-specific treatment because at the end of the day, that is the only path forward.

254
00:15:28,323 --> 00:15:31,103
Josh:
Like when you think about traditional cancer treatments, they are all broad

255
00:15:31,103 --> 00:15:34,463
Josh:
sweeping. It's one of a kind and like the most popular one, chemotherapy,

256
00:15:34,649 --> 00:15:37,661
Josh:
it just wipes everything out for everyone. It doesn't care what type of disease you have.

257
00:15:37,905 --> 00:15:40,573
Josh:
It just goes and it takes it out. This is a really big deal.

258
00:15:40,573 --> 00:15:43,350
Josh:
Approvals, they're saying could come as soon as 2027.

259
00:15:43,501 --> 00:15:47,210
Ejaaz:
They're also working on all the cancer types as well, right? Yes. That's like sick.

260
00:15:47,327 --> 00:15:50,683
Josh:
Melanoma, like the first one. Melanoma, most popular form of skin cancer.

261
00:15:50,683 --> 00:15:52,388
Josh:
I believe the number is...

262
00:15:52,923 --> 00:15:57,420
Josh:
There's about 100,000 plus people that get diagnosed a year.

263
00:15:57,693 --> 00:15:59,434
Josh:
Of those people, 8,500 die.

264
00:15:59,545 --> 00:16:01,347
Ejaaz:
In America or like in the world?

265
00:16:01,347 --> 00:16:06,296
Josh:
I think this is the world. Wow, okay. People suffer from this.

266
00:16:06,563 --> 00:16:09,407
Josh:
And this is the first type. And you have to imagine that they could solve skin

267
00:16:09,407 --> 00:16:12,333
Josh:
cancer. They could do it for many other things. So it's really,

268
00:16:12,484 --> 00:16:14,967
Josh:
really exciting. Elon has a comment here that we have on screen and says,

269
00:16:14,967 --> 00:16:19,218
Josh:
despite its obvious misuse during COVID, mRNA has tremendous promise for curing disease.

270
00:16:19,653 --> 00:16:22,427
Josh:
Artificial RNA essentially makes curing diseases a software problem.

271
00:16:22,427 --> 00:16:25,487
Josh:
And we know how we well this is me saying it we know how we take care of software

272
00:16:25,487 --> 00:16:28,651
Josh:
problems and now that we have these ai models that are capable of doing these,

273
00:16:29,121 --> 00:16:32,637
Josh:
one processing tremendously on context windows so it can fit a dna sequence

274
00:16:32,637 --> 00:16:36,517
Josh:
inside of a singular context window but two having the inference ability and

275
00:16:36,517 --> 00:16:39,663
Josh:
the understanding to actually make use of that info and do something with it

276
00:16:39,994 --> 00:16:43,477
Josh:
it's a really unbelievable time for biology and if you are a computer scientist

277
00:16:43,477 --> 00:16:46,507
Josh:
suddenly the world of bio is looking very interesting and very compelling and

278
00:16:46,507 --> 00:16:48,568
Josh:
if you are an ai company well,

279
00:16:49,032 --> 00:16:53,208
Josh:
What's the best way to explain to people how AI can actually be beneficial for

280
00:16:53,208 --> 00:16:56,138
Josh:
the world? It's to actually go out and do something beneficial for the world.

281
00:16:56,138 --> 00:16:58,928
Josh:
So we're right on the cusp of, I think, a lot of these things happening in a

282
00:16:58,928 --> 00:17:00,316
Josh:
really exciting way. This is awesome.

283
00:17:00,693 --> 00:17:06,438
Ejaaz:
Okay, moving on. Our friends, or rather we are the fans, of this company Etch,

284
00:17:06,438 --> 00:17:09,203
Ejaaz:
which is basically an AI chip startup

285
00:17:09,487 --> 00:17:12,868
Ejaaz:
founded by two or three MIT dropouts,

286
00:17:12,868 --> 00:17:17,080
Ejaaz:
which is going head-to-head with NVIDIA, is back. And guess what?

287
00:17:17,277 --> 00:17:20,948
Ejaaz:
They have raised another couple hundred million dollars, in this case,

288
00:17:20,948 --> 00:17:26,338
Ejaaz:
700 million dollars, and have reached an astounding valuation of 21 billion dollars.

289
00:17:26,338 --> 00:17:30,178
Ejaaz:
But that's not the headline from this announcement. The headline is this last

290
00:17:30,178 --> 00:17:32,335
Ejaaz:
sentence over here that I'm showing you on screen, which is,

291
00:17:32,708 --> 00:17:35,836
Ejaaz:
We're also excited to share that we've shipped our first rack,

292
00:17:35,964 --> 00:17:37,948
Ejaaz:
custom GPU rack, to Jane Street.

293
00:17:37,948 --> 00:17:43,104
Ejaaz:
Jane Street is basically a quant fund on Wall Street. They are very good at what they do.

294
00:17:43,348 --> 00:17:48,068
Ejaaz:
And the fact that they are now leveraging this custom AI silicon chip to have

295
00:17:48,068 --> 00:17:51,558
Ejaaz:
better AI trading algos is a pretty big milestone.

296
00:17:51,558 --> 00:17:55,548
Ejaaz:
Now, just to kind of recap what Etch does, if you think about NVIDIA,

297
00:17:55,548 --> 00:18:00,169
Ejaaz:
NVIDIA is like the king of GPUs. GPUs are required to train and inference AI models.

298
00:18:00,419 --> 00:18:04,478
Ejaaz:
Now, their GPUs are very good, but their general purpose, which is good because

299
00:18:04,478 --> 00:18:08,808
Ejaaz:
you can run it on any kind of model, but there are certain trade-offs that are

300
00:18:08,808 --> 00:18:11,947
Ejaaz:
made in terms of like cost, power and efficiency.

301
00:18:12,220 --> 00:18:17,058
Ejaaz:
The folks at Etched looked at this problem and said, we can build a better custom

302
00:18:17,058 --> 00:18:20,938
Ejaaz:
chip that is hard-coded to the specific model that you are building.

303
00:18:20,938 --> 00:18:24,260
Ejaaz:
So specific for your OpenAI model, specific for your Anthropic Claude model,

304
00:18:24,509 --> 00:18:30,367
Ejaaz:
and it will be half or maybe even more cheaper than what you were spending already.

305
00:18:30,541 --> 00:18:34,256
Ejaaz:
And it'll be lightning quick. There are certain optimizations that we can make.

306
00:18:34,894 --> 00:18:39,326
Ejaaz:
This is a very hard problem, a huge undertaking. We listened to some of the

307
00:18:39,326 --> 00:18:43,756
Ejaaz:
founders' stories, and they received no's from some of the smartest GPU makers

308
00:18:43,756 --> 00:18:47,822
Ejaaz:
in the world. But despite that, they put their heads down and they built this custom chip.

309
00:18:48,066 --> 00:18:51,756
Ejaaz:
And it turns out that they're also the fastest to deliver. Typically,

310
00:18:51,756 --> 00:18:55,681
Ejaaz:
AI chips take around three to five years to go from concept to delivery.

311
00:18:56,023 --> 00:18:59,843
Ejaaz:
These guys have done it in, I think, just under two years, which is absolutely astounding.

312
00:19:00,093 --> 00:19:04,069
Ejaaz:
And we're about to see it in production. So just a very impressive announcement from the Edge team.

313
00:19:04,203 --> 00:19:07,786
Josh:
It's unbelievable. I've spent so much time looking. I mean, we both do spend

314
00:19:07,786 --> 00:19:10,046
Josh:
so much time looking at all of these companies that are spawning in the world

315
00:19:10,046 --> 00:19:14,826
Josh:
of AI. And there is thousands, if not tens of thousands of people working on

316
00:19:14,826 --> 00:19:18,947
Josh:
interesting problems to solve interesting, to reach interesting solutions. And,

317
00:19:19,446 --> 00:19:23,208
Josh:
Etched is very different. So different that we actually recorded an entire episode

318
00:19:23,208 --> 00:19:25,762
Josh:
about this weird random startup last month.

319
00:19:26,035 --> 00:19:28,968
Josh:
And since recording that episode, EJ's, I don't know if you know this,

320
00:19:28,968 --> 00:19:32,848
Josh:
the valuation has doubled in the last month. When we talked about them,

321
00:19:32,848 --> 00:19:37,018
Josh:
just not even like four or five weeks ago, it was at a $10 billion valuation.

322
00:19:37,018 --> 00:19:38,551
Josh:
Now they're trading at 21 billion.

323
00:19:38,667 --> 00:19:41,888
Josh:
And it's very obvious once you listen to these people who are on the team talk

324
00:19:41,888 --> 00:19:44,648
Josh:
about the product, how they've arrived there, their conclusions,

325
00:19:44,648 --> 00:19:47,198
Josh:
the research that was required to actually reached this point where it's very

326
00:19:47,198 --> 00:19:51,758
Josh:
much the real deal this is a company that anyone who works in ai or semiconductors

327
00:19:51,758 --> 00:19:53,532
Josh:
or hardware would want to make

328
00:19:53,822 --> 00:19:57,098
Josh:
they were just unable to because of the complexity and the difficulty that this

329
00:19:57,098 --> 00:20:00,498
Josh:
team has been able to solve and as a result they have this unbelievable product

330
00:20:00,498 --> 00:20:02,588
Josh:
that is specifically tuned to run

331
00:20:02,867 --> 00:20:06,248
Josh:
large language model transformer infrastructure right on their chips and they

332
00:20:06,248 --> 00:20:08,852
Josh:
are going to worth many many multiples more than they are today

333
00:20:09,153 --> 00:20:12,388
Josh:
and i mean just a congratulations to them but also for anybody who's

334
00:20:12,768 --> 00:20:17,528
Josh:
listening if you want to see what a badass team of high agency,

335
00:20:17,528 --> 00:20:19,556
Josh:
high intelligence founders actually looks like.

336
00:20:19,916 --> 00:20:23,118
Josh:
Read up about the co-founders, read up about the story. It is an unbelievable

337
00:20:23,118 --> 00:20:27,898
Josh:
company that is destined for an incredibly bright future. I would love to be an investor in them.

338
00:20:27,898 --> 00:20:31,698
Ejaaz:
I was about to say, self-serving for Josh and I, if anyone listening has access

339
00:20:31,698 --> 00:20:35,038
Ejaaz:
to this deal, please reach out. We would love to be investors.

340
00:20:35,088 --> 00:20:38,887
Josh:
I am obsessed with these guys in a way that not many people have captivated

341
00:20:39,206 --> 00:20:42,224
Josh:
the attention. It's pretty amazing. And you know what you could do with all those chips, EJS?

342
00:20:42,457 --> 00:20:46,508
Josh:
You can run a tremendous amount of agents. And if you are running agents on

343
00:20:46,508 --> 00:20:49,858
Josh:
all of these chips and you're building for AI agents, chances are you may want

344
00:20:49,858 --> 00:20:50,931
Josh:
to hear about our sponsor, Ledger.

345
00:20:51,128 --> 00:20:53,178
Josh:
Because if you're building with AI agents, chances are you're worried about

346
00:20:53,178 --> 00:20:54,488
Josh:
security. We spoke about this earlier in the,

347
00:20:56,898 --> 00:21:00,237
Josh:
ledger has built this thing called the agent stack that fixes this it gives

348
00:21:00,237 --> 00:21:04,241
Josh:
you a series of open source tools that allow you to do a three-step process

349
00:21:04,392 --> 00:21:05,971
Josh:
with your agents to ensure everything goes well

350
00:21:06,302 --> 00:21:10,418
Josh:
the agent proposes a change the humans approve and then the ledger signers enforce

351
00:21:10,708 --> 00:21:14,477
Josh:
this technology is available and works with cloud code it works with codex it

352
00:21:14,477 --> 00:21:18,147
Josh:
works with cursor or any place where you work with ai it's all open source and

353
00:21:18,147 --> 00:21:19,747
Josh:
available today linked in the description below

354
00:21:20,031 --> 00:21:23,427
Josh:
thank you so much to ledger for sponsoring this part of the episode and as we

355
00:21:23,427 --> 00:21:27,287
Josh:
move on from etched we have to talk about another breakout success which is unitary

356
00:21:27,543 --> 00:21:30,327
Josh:
uh unitary for those that don't know if you haven't heard the name you've seen

357
00:21:30,327 --> 00:21:33,507
Josh:
the videos of these little robots that are like crazy they're doing back loops

358
00:21:33,507 --> 00:21:36,262
Josh:
they're running circles and like um at these track meets

359
00:21:36,604 --> 00:21:40,447
Josh:
They're very cool. I think their IPO was oversubscribed at what, like 5,000%?

360
00:21:40,447 --> 00:21:43,307
Ejaaz:
5,500, actually. Please get it right, Josh.

361
00:21:43,307 --> 00:21:48,137
Josh:
5,500, man. It's unbelievable. And then as a result, the Unitree robotic shares

362
00:21:48,137 --> 00:21:51,515
Josh:
rose 542% in early trading on Wednesday.

363
00:21:51,620 --> 00:21:54,747
Josh:
Is it overhyped? Is this overhyped? Does this make sense? These numbers are insane.

364
00:21:54,747 --> 00:21:58,677
Ejaaz:
It has to be a little overhyped. Listen, I'm not an expert in the Chinese stock

365
00:21:58,677 --> 00:22:02,507
Ejaaz:
market, but they've had a few IPOs recently that have been absolutely insane,

366
00:22:02,507 --> 00:22:06,597
Ejaaz:
including their memory manufacturer, CXMT, which, to be fair,

367
00:22:06,597 --> 00:22:09,087
Ejaaz:
they are, I think, the fourth largest memory manufacturer.

368
00:22:09,087 --> 00:22:11,947
Ejaaz:
So they're up there with like SK Hynix, Samsung, and Micron.

369
00:22:11,947 --> 00:22:13,144
Ejaaz:
So they have a legitimate case.

370
00:22:13,400 --> 00:22:15,427
Ejaaz:
These Unitree robots, I don't know, man.

371
00:22:15,727 --> 00:22:19,861
Ejaaz:
I was looking at, I was speaking to some friends that are living out in China, actually.

372
00:22:20,076 --> 00:22:23,187
Ejaaz:
And I was like, this is amazing. Like, you guys must be super excited about,

373
00:22:23,187 --> 00:22:26,281
Ejaaz:
you know, these companies coming out and IPO-ing at such crazy valuations.

374
00:22:26,548 --> 00:22:31,163
Ejaaz:
And apparently the sentiment in China around Unitree robotics specifically is,

375
00:22:31,471 --> 00:22:36,047
Ejaaz:
it's just a toy maker. They don't actually make robots that are actually valuable.

376
00:22:36,047 --> 00:22:37,885
Ejaaz:
And so I dug into this a little bit more.

377
00:22:38,181 --> 00:22:43,081
Ejaaz:
And apparently, of the humanoid robots that they have, 40% to 60%

378
00:22:43,365 --> 00:22:48,227
Ejaaz:
have only been sold for research purposes specifically, which means that the

379
00:22:48,227 --> 00:22:51,417
Ejaaz:
majority aren't actually being used for practical purposes, like some of the

380
00:22:51,417 --> 00:22:53,610
Ejaaz:
use cases which they advertise, such as

381
00:22:53,894 --> 00:22:58,863
Ejaaz:
manufacturing at factories or automating warehouse facilitations and stuff like that.

382
00:22:59,119 --> 00:23:03,797
Ejaaz:
They're actually just being tested, which isn't really confident or recurring

383
00:23:03,797 --> 00:23:05,435
Ejaaz:
revenue that they can put on their books.

384
00:23:05,661 --> 00:23:11,527
Ejaaz:
So if you map that onto, what is this, a 600% price increase on day one with

385
00:23:11,527 --> 00:23:16,527
Ejaaz:
a 5,500x oversubscription on the IPO, maybe the market dynamics are a little

386
00:23:16,527 --> 00:23:18,287
Ejaaz:
inflated and maybe they're still crushed back down.

387
00:23:18,867 --> 00:23:22,537
Ejaaz:
Let's revisit it in a month. But it is crazy to see that Chinese stocks,

388
00:23:22,537 --> 00:23:25,996
Ejaaz:
specifically robotics, are really starting to gain some momentum now.

389
00:23:26,530 --> 00:23:30,657
Josh:
Yeah, it's a very cool news story. I personally am not touching this with a Zenfopole.

390
00:23:30,959 --> 00:23:33,583
Josh:
That seems really scary. I love watching the videos. I'll continue to watch

391
00:23:33,583 --> 00:23:35,417
Josh:
the videos. I'm very excited to see these robots.

392
00:23:35,644 --> 00:23:36,915
Ejaaz:
Keep the dock flex coming.

393
00:23:37,060 --> 00:23:40,723
Josh:
Yeah, it's very early in humanoid land. And while I am very bullish on China

394
00:23:40,723 --> 00:23:44,855
Josh:
and their ability to make humanoids at scale, it feels a little early,

395
00:23:45,000 --> 00:23:47,125
Josh:
but I'm glad people are excited. I'm glad there's enthusiasm.

396
00:23:47,264 --> 00:23:50,403
Josh:
Next on the docket, we have a company that we just spoke about yesterday.

397
00:23:50,403 --> 00:23:53,522
Josh:
So if you listened to yesterday's episode, we have a small little follow-up,

398
00:23:53,806 --> 00:23:58,163
Josh:
which it relates to stripe and the acquisition that was made for to acquire

399
00:23:58,163 --> 00:24:03,373
Josh:
open router it is officially official we were a little early but right and now

400
00:24:03,373 --> 00:24:05,300
Josh:
it's official all the contracts have been signed,

401
00:24:05,927 --> 00:24:09,642
Josh:
alex's holla who is the founder and ceo says um we officially have a deal and

402
00:24:09,967 --> 00:24:14,733
Josh:
he kind of outlines the thesis and i know stripe as a result of this they shared a

403
00:24:15,173 --> 00:24:18,853
Josh:
publication a blog post ejes that you were really excited so i'll let you describe

404
00:24:18,853 --> 00:24:22,216
Josh:
what they spoke about here But some TLDR on Stripe is like, again,

405
00:24:22,448 --> 00:24:24,265
Josh:
we're talking about unbelievable founders today.

406
00:24:24,567 --> 00:24:27,783
Josh:
The Collison brothers, who are the founders of Stripe, are among the best of the best.

407
00:24:27,783 --> 00:24:32,073
Josh:
They have built an unbelievable business, all starting with seven lines of code

408
00:24:32,073 --> 00:24:36,153
Josh:
that allowed a person to pay with a credit card online quickly and easily.

409
00:24:36,153 --> 00:24:40,136
Josh:
And they've developed that technology from basic credit card payments to basically...

410
00:24:40,669 --> 00:24:44,556
Josh:
The entire payment infrastructure of the world. They started with credit cards,

411
00:24:44,556 --> 00:24:48,186
Josh:
then they moved over to crypto and stable coins. Now they're working with agents

412
00:24:48,186 --> 00:24:52,296
Josh:
to figure out how best to serve these new agentic transaction systems.

413
00:24:52,296 --> 00:24:55,350
Josh:
And that's kind of where we find ourselves here. I'm reading the headline.

414
00:24:55,663 --> 00:24:58,536
Josh:
It's that Stripe is officially telling investors that the singular area has

415
00:24:58,536 --> 00:25:01,584
Josh:
begun. But explain the nuance behind what they actually mean when they say that.

416
00:25:01,886 --> 00:25:07,221
Ejaaz:
Yeah. So in the official letter, which Stripe sent to their board of investors,

417
00:25:07,534 --> 00:25:10,596
Ejaaz:
they basically described why they acquired OpenRata for $7 billion.

418
00:25:10,896 --> 00:25:16,316
Ejaaz:
Why is a fintech company buying an AI lab for such an expensive amount that

419
00:25:16,316 --> 00:25:18,386
Ejaaz:
doesn't even own or make their own AI models?

420
00:25:18,386 --> 00:25:24,186
Ejaaz:
And the thesis is pretty simple. Number one, in a world where the internet was

421
00:25:24,186 --> 00:25:29,483
Ejaaz:
booming, Stripe's core focus was to be the backbone of that entire internet economy.

422
00:25:29,692 --> 00:25:31,555
Ejaaz:
How do you become the backbone of the internet economy?

423
00:25:31,787 --> 00:25:33,930
Ejaaz:
Money. So they created the rails for money.

424
00:25:34,307 --> 00:25:38,166
Ejaaz:
Then they peeked over the fence and they saw this massive tidal wave of AI coming

425
00:25:38,166 --> 00:25:42,526
Ejaaz:
over with agents, with models, and they realized it's not money that runs that

426
00:25:42,526 --> 00:25:46,054
Ejaaz:
AI economy, it's tokens, AI tokens specifically.

427
00:25:46,350 --> 00:25:50,769
Ejaaz:
And so they wanted to own the infrastructure rails to own that AI economy.

428
00:25:50,948 --> 00:25:54,866
Ejaaz:
It's very similar to what they do with money, which is why they're pairing both

429
00:25:54,866 --> 00:25:57,356
Ejaaz:
of these together. So that's thesis number one.

430
00:25:57,624 --> 00:26:01,876
Ejaaz:
But thesis number two, which caught my eye is, And I quote, Stripe basically

431
00:26:01,876 --> 00:26:05,556
Ejaaz:
claimed that the beginning of the singularity has been here.

432
00:26:05,556 --> 00:26:09,826
Ejaaz:
And the date specifically that they specified in this paper was January 1st.

433
00:26:10,030 --> 00:26:13,157
Ejaaz:
And this is a large part of the thesis behind acquiring OpenRata.

434
00:26:13,407 --> 00:26:16,292
Ejaaz:
They basically go on to say that we have reached a pivotal point

435
00:26:16,734 --> 00:26:22,839
Ejaaz:
In our global economy where the future is very much dependent on AI and it's

436
00:26:22,839 --> 00:26:26,409
Ejaaz:
going to accelerate pretty much anything and everything we know and touch.

437
00:26:26,409 --> 00:26:29,041
Ejaaz:
The world is going to look very different five years from now.

438
00:26:29,256 --> 00:26:33,919
Ejaaz:
And so the bet that they're making in this world of a singularity is AI models

439
00:26:33,919 --> 00:26:37,779
Ejaaz:
will be embedded into everything, which means that thesis of AI tokens

440
00:26:38,099 --> 00:26:42,059
Ejaaz:
flooding across the entire global economy is going to be a thing and Stripe

441
00:26:42,059 --> 00:26:45,415
Ejaaz:
wants to be the backbone for that. So it's a big bet that they're making.

442
00:26:45,699 --> 00:26:49,359
Ejaaz:
Open Router, I think on their last earnings that was disclosed by Atala,

443
00:26:49,359 --> 00:26:53,826
Ejaaz:
because again, they are a private company, is making $50 million a year.

444
00:26:53,942 --> 00:26:56,682
Ejaaz:
So $7 billion is a humongous markup.

445
00:26:56,879 --> 00:27:01,379
Ejaaz:
But Stripe is making a bet that inference compute specifically is going to be

446
00:27:01,739 --> 00:27:03,892
Ejaaz:
the biggest earner for AI going forward.

447
00:27:04,083 --> 00:27:07,056
Ejaaz:
And all the numbers from Anthropic and OpenAI recently that have been leaked

448
00:27:07,381 --> 00:27:10,509
Ejaaz:
proves that. So I think this is going to be in the future, if I had to guess,

449
00:27:10,509 --> 00:27:12,443
Ejaaz:
one of the best acquisitions going forward

450
00:27:12,797 --> 00:27:16,199
Josh:
Yeah and perhaps like an explainer on the singularity for people not familiar

451
00:27:16,199 --> 00:27:19,763
Josh:
because this is it's a concept that's been around for a very long time like this is a

452
00:27:20,064 --> 00:27:23,939
Josh:
a sci-fi concept for as long as i can remember in books probably 50 perhaps

453
00:27:23,939 --> 00:27:27,820
Josh:
even 100 years ago where people speculated on this point in time in which

454
00:27:28,133 --> 00:27:34,309
Josh:
the intelligence of artificial systems eclipses that of human beings and it's

455
00:27:34,309 --> 00:27:37,519
Josh:
a moment in time in which ai becomes smarter than us where

456
00:27:37,908 --> 00:27:42,199
Josh:
these smart machines are then enabled to build even smarter machines and it

457
00:27:42,199 --> 00:27:47,079
Josh:
builds this kind of fast loop of recurring self-improvement and the idea is

458
00:27:47,079 --> 00:27:51,799
Josh:
that the second order effects of that is that life changes so fast that humans can't really predict

459
00:27:52,234 --> 00:27:56,399
Josh:
and are going to try their best to control the future and this has been this

460
00:27:56,399 --> 00:27:59,199
Josh:
hypothetical for a really long period of time people always speculate whenever

461
00:27:59,199 --> 00:28:04,013
Josh:
we come up with this super ai intelligence brain and the reality is is that

462
00:28:04,274 --> 00:28:05,969
Josh:
straight believes that it's here and,

463
00:28:06,422 --> 00:28:10,332
Josh:
I don't think anyone would disagree with them too strongly. Perhaps people will

464
00:28:10,332 --> 00:28:12,581
Josh:
say we're not there yet, which is a fair criticism.

465
00:28:12,819 --> 00:28:18,073
Josh:
But to say there isn't a clear path to getting there soon feels a little misguided.

466
00:28:18,189 --> 00:28:21,642
Josh:
It's very obvious now. And it's interesting, the timing, too,

467
00:28:21,642 --> 00:28:25,445
Josh:
that January 1st, because if you remember over Christmas break of last year,

468
00:28:25,753 --> 00:28:29,642
Josh:
going into this year, that's when the real kind of like.

469
00:28:30,228 --> 00:28:34,569
Josh:
Flow of like, yeah, that's when Cloud Code like really came out, started becoming good.

470
00:28:34,842 --> 00:28:38,422
Josh:
And that was the moment in time in which developers stopped writing code by

471
00:28:38,422 --> 00:28:42,152
Josh:
hand and started writing code using ai agents and now fast forward we're sitting

472
00:28:42,152 --> 00:28:45,535
Josh:
here in august and it is unfathomable to imagine

473
00:28:45,970 --> 00:28:49,672
Josh:
writing code by hand like people just do not do that there's absolutely no reason

474
00:28:49,672 --> 00:28:54,341
Josh:
to do that it's weird to think that people ever did do that and that change over this.

475
00:28:54,979 --> 00:28:59,732
Josh:
Eight month period is so extreme when you consider the second order of implications

476
00:28:59,732 --> 00:29:01,097
Josh:
of of what that means for the world.

477
00:29:01,097 --> 00:29:05,394
Josh:
Like now we're seeing these models that are so powerful, we can't actually release them.

478
00:29:05,649 --> 00:29:08,172
Josh:
There's now infrastructure frameworks happening through the government to kind

479
00:29:08,172 --> 00:29:11,738
Josh:
of protect and slow down this. OpenAI has paused progress because it's getting so fast.

480
00:29:12,023 --> 00:29:15,252
Josh:
And if you heard about this even 12 months ago today, you would think,

481
00:29:15,252 --> 00:29:18,565
Josh:
oh my God, this is faster than we expected. This is what the singularity looks like.

482
00:29:18,902 --> 00:29:21,822
Josh:
And here we're sitting in it today. So pretty unique insight by Stripe,

483
00:29:21,822 --> 00:29:24,592
Josh:
really awesome letter. I would highly recommend going through and reading it.

484
00:29:24,592 --> 00:29:27,422
Josh:
It's a leaked letter. This was for private investors. So you get a little inside

485
00:29:27,422 --> 00:29:30,192
Josh:
peek behind the curtain of what's actually going on at the company. Very cool.

486
00:29:30,750 --> 00:29:36,256
Ejaaz:
And so moving on, chapter two of Jensen turning NVIDIA into a bank.

487
00:29:36,656 --> 00:29:41,236
Ejaaz:
If you guys were watching the show last week, we released an episode where Jensen

488
00:29:41,236 --> 00:29:45,296
Ejaaz:
basically managed to raise $500 billion from some of the top financial firms

489
00:29:45,296 --> 00:29:47,084
Ejaaz:
in the world, including BlackRock, Blackstone,

490
00:29:47,449 --> 00:29:51,896
Ejaaz:
to fund the future of data center CapEx for some of these biggest AI labs.

491
00:29:52,175 --> 00:29:56,416
Ejaaz:
And we concluded that it isn't really a circular economy in the way that they're

492
00:29:56,416 --> 00:30:00,963
Ejaaz:
raising it. It is based on a bunch of different milestones and is backed by very healthy debt.

493
00:30:01,306 --> 00:30:06,796
Ejaaz:
Now, in chapter two, Jensen announced that he is spending another $100 billion

494
00:30:07,854 --> 00:30:09,566
Ejaaz:
For OpenAI specifically.

495
00:30:09,868 --> 00:30:14,896
Ejaaz:
He's investing it in this site in Ohio, which is arguably going to be one of

496
00:30:14,896 --> 00:30:20,724
Ejaaz:
the biggest data centers in the world. It's going to scale up to eight gigawatts worth of compute.

497
00:30:20,979 --> 00:30:25,276
Ejaaz:
It is going to bring in a nice and healthy $200 to $300 billion of revenue for

498
00:30:25,276 --> 00:30:28,526
Ejaaz:
NVIDIA specifically from one data site, from one single customer.

499
00:30:28,851 --> 00:30:31,576
Ejaaz:
And it's going to conclude over the next couple of years. Now,

500
00:30:31,576 --> 00:30:36,906
Ejaaz:
the craziest part about this is he's baked into the contract that this includes

501
00:30:36,906 --> 00:30:41,656
Ejaaz:
renewals of newer generation GPUs, which means that he's going to make $200 to $300 billion

502
00:30:42,176 --> 00:30:47,206
Ejaaz:
three times over the longevity of this data center over the next five years.

503
00:30:47,206 --> 00:30:50,966
Ejaaz:
So there's been a lot of contemplation around why Jensen is getting involved in this.

504
00:30:51,145 --> 00:30:54,895
Ejaaz:
The thing that stands out about this particular investment is

505
00:30:55,807 --> 00:31:02,039
Ejaaz:
He's investing these $100 billion specifically in the power and networking infrastructure

506
00:31:02,039 --> 00:31:06,180
Ejaaz:
side of this data center. So he's not really doing it to help OpenAI buy the GPUs.

507
00:31:06,511 --> 00:31:10,360
Ejaaz:
He's helping them set up the power transmitter lines to power these GPUs.

508
00:31:10,639 --> 00:31:14,199
Ejaaz:
And why I wanted to speak about it on this show is we have said a few times

509
00:31:14,199 --> 00:31:19,169
Ejaaz:
now, Josh and I have called it, that the next kind of investment theme within

510
00:31:19,169 --> 00:31:24,425
Ejaaz:
AI is going to be on the photonics, optics, and power stack of the AI trade.

511
00:31:24,646 --> 00:31:29,249
Ejaaz:
And this is yet another very in-your-face evidence point that Jensen is heading

512
00:31:29,249 --> 00:31:33,469
Ejaaz:
that way. And if you look at like NVIDIA's investments over the last six to,

513
00:31:33,469 --> 00:31:36,709
Ejaaz:
I don't know, like nine months, they've invested heavily in Optonics.

514
00:31:37,209 --> 00:31:40,910
Ejaaz:
I think they put like two to three billion into Lumentum and Coherent each,

515
00:31:41,183 --> 00:31:43,709
Ejaaz:
and now they're making a big play on the power side of things.

516
00:31:43,709 --> 00:31:46,939
Ejaaz:
So just an interesting, noteworthy update. I think NVIDIA is going to backstop

517
00:31:46,939 --> 00:31:49,369
Ejaaz:
this entire economy that Stripe is talking about,

518
00:31:49,369 --> 00:31:53,659
Josh:
To be honest. I was reading this great transcript from Ben Thompson of Stratetri.

519
00:31:53,659 --> 00:31:57,009
Josh:
He's just like this you know well-known writer on the internet for people that

520
00:31:57,009 --> 00:32:00,519
Josh:
don't know and he was talking about kind of like where these restraints and

521
00:32:00,519 --> 00:32:03,799
Josh:
thresholds might be in the process of scaling ai systems and he made this really

522
00:32:03,799 --> 00:32:07,329
Josh:
interesting point where oftentimes people are focused on the energy the things that

523
00:32:07,671 --> 00:32:10,759
Josh:
you know kind of like jensen is focused on the land power and shell but one

524
00:32:10,759 --> 00:32:13,709
Josh:
of the other important considerations is like where is the money for all of

525
00:32:13,709 --> 00:32:16,399
Josh:
this coming from and and what we have seen and we've been talking about this

526
00:32:16,399 --> 00:32:20,059
Josh:
on the show over the past couple weeks but it's like okay we went from spending

527
00:32:20,059 --> 00:32:22,649
Josh:
all of the free cash flow to starting to

528
00:32:23,049 --> 00:32:27,389
Josh:
raise money from outside funding and now companies like google are issuing debt

529
00:32:27,389 --> 00:32:29,195
Josh:
they're issuing equity in order to raise more money

530
00:32:29,520 --> 00:32:33,159
Josh:
now you see nvidia and jensen huang working with every large bank in the world

531
00:32:33,159 --> 00:32:37,729
Josh:
to tap into new sources of money these are places like pension funds and retirement

532
00:32:37,729 --> 00:32:39,529
Josh:
funds that's what these banks hold so

533
00:32:39,929 --> 00:32:45,153
Josh:
where do you go to next to raise the next tranche of money to build out the next large data center?

534
00:32:45,315 --> 00:32:48,653
Josh:
It could be. And I really thought this was like a thoughtful insight. It could be the...

535
00:32:49,019 --> 00:32:52,352
Josh:
The actual dollars are going to be running short and it's going to be very difficult

536
00:32:52,352 --> 00:32:55,712
Josh:
to continue to raise the required amount of money to scale this thing so just

537
00:32:55,712 --> 00:32:58,806
Josh:
an interesting thought as we wrap up the roundup today of like

538
00:32:59,073 --> 00:33:02,632
Josh:
where's the bottleneck well everyone's saying it's here but like where are they

539
00:33:02,632 --> 00:33:05,382
Josh:
going to get the money to fund it is an interesting thing to follow and we've

540
00:33:05,382 --> 00:33:07,932
Josh:
been talking about this on the show we'll continue to cover it but if you've

541
00:33:07,932 --> 00:33:09,284
Josh:
made it this far thank you

542
00:33:09,696 --> 00:33:12,912
Josh:
you've made it to the end you are fully caught up if you've watched the other

543
00:33:12,912 --> 00:33:17,702
Josh:
three episodes that we published this week because it was a pretty strong week i would say we had elon.

544
00:33:18,162 --> 00:33:22,723
Josh:
And grokbot releasing grok 4.6 and it's incredible we had leopold making.

545
00:33:23,559 --> 00:33:27,122
Josh:
Basically publishing his final portfolio um because you know he got liquidated

546
00:33:27,122 --> 00:33:30,500
Josh:
but we talked about all of the filings what the smartest money on wall street is investing in

547
00:33:30,860 --> 00:33:36,322
Josh:
yesterday we spoke about um stripe and their acquisition then we talked about

548
00:33:36,322 --> 00:33:39,972
Josh:
what models to use based on what type of use cases you have and here we are

549
00:33:39,972 --> 00:33:42,209
Josh:
today getting fully caught up on all things

550
00:33:42,592 --> 00:33:44,931
Josh:
ai frontier tech so yeah, thanks for watching.

551
00:33:45,332 --> 00:33:47,892
Josh:
We made it back again. EJ, any final parting thoughts before we get out of here?

552
00:33:48,449 --> 00:33:53,572
Ejaaz:
No, I just wanted to say thank you folks so much for listening and keeping close

553
00:33:53,572 --> 00:33:59,602
Ejaaz:
with us. We have added, I think maybe six to seven hundred new subscribers over

554
00:33:59,602 --> 00:34:00,976
Ejaaz:
the last couple of weeks. Hello.

555
00:34:01,232 --> 00:34:03,089
Ejaaz:
It's lovely to meet you. Welcome.

556
00:34:03,223 --> 00:34:06,122
Josh:
A hundred, dude. A thousand. A thousand?

557
00:34:06,552 --> 00:34:09,602
Ejaaz:
Oh my god. The momentum is really gaining with this channel.

558
00:34:09,722 --> 00:34:13,927
Ejaaz:
It's awesome to see the kind of support and vibes coming from different people

559
00:34:14,160 --> 00:34:16,062
Ejaaz:
we read every single comment um

560
00:34:16,422 --> 00:34:20,432
Ejaaz:
i try to respond to as many of them as i can um you guys are growing in fandomship

561
00:34:20,432 --> 00:34:24,152
Ejaaz:
which is awesome to see you guys are watching every single video we see a bunch

562
00:34:24,152 --> 00:34:27,232
Ejaaz:
of like repeat users and listeners as well which is awesome to see wherever

563
00:34:27,232 --> 00:34:29,035
Ejaaz:
you're listening to us on youtube spotify

564
00:34:29,337 --> 00:34:31,550
Ejaaz:
or apple music please give us a rating um

565
00:34:32,162 --> 00:34:37,173
Ejaaz:
give us a thumbs up subscribe turn on notifications leave us a comment it helps us out so much um

566
00:34:37,562 --> 00:34:41,482
Ejaaz:
and yeah i think that's it uh enjoy your weekend this is everything you should

567
00:34:41,482 --> 00:34:44,372
Ejaaz:
officially be caught up unless we're missing something pretty drastic that's

568
00:34:44,372 --> 00:34:48,563
Ejaaz:
happening right now after we record and yeah we'll see you on the next one thank you guys