1
00:00:00,100 --> 00:00:04,490
Josh:
Just last week, in the span of about 48 hours, three of the most powerful AI

2
00:00:04,490 --> 00:00:07,040
Josh:
labs on the planet all shipped brand new models.

3
00:00:07,240 --> 00:00:13,560
Josh:
Elon shipped Grok 4.5, OpenAI took 5.6 Global, and Meta, for the very first

4
00:00:13,560 --> 00:00:16,450
Josh:
time in history, put a price tag on its own frontier model.

5
00:00:16,450 --> 00:00:19,100
Josh:
And here's where it gets interesting. For the last five years or so,

6
00:00:19,210 --> 00:00:23,860
Josh:
the deal in AI was that every time a model shipped, it got smarter and cheaper at the same time.

7
00:00:23,860 --> 00:00:27,350
Josh:
But that kind of died this year with Anthropics Fable 5 and these new frontier

8
00:00:27,350 --> 00:00:31,900
Josh:
launches like GPT 5.6. So today, we're asking the question that probably everyone should.

9
00:00:32,090 --> 00:00:36,070
Josh:
When frontier intelligence costs less than a cup of coffee, when they cost just

10
00:00:36,070 --> 00:00:39,600
Josh:
a few pennies per millions of tokens, what does that look like in terms of your

11
00:00:39,600 --> 00:00:41,430
Josh:
costs and how much you use these models?

12
00:00:41,430 --> 00:00:44,400
Josh:
I mean, this is a totally different paradigm now. We have a very clear separation

13
00:00:44,620 --> 00:00:51,280
Josh:
between Fable and 5.6 Sol and Grok 4.5 and Meta's new model, MuseSpark 1.1.

14
00:00:51,490 --> 00:00:54,440
Josh:
And it seems like these models are kind of diverging in a way that's really

15
00:00:54,440 --> 00:00:56,320
Josh:
interesting, particularly centered around price.

16
00:00:56,510 --> 00:01:02,220
Ejaaz:
For the last couple of years, the ultimate validation of whether your AI model is good or not.

17
00:01:02,630 --> 00:01:07,830
Ejaaz:
Is how intelligent it is, how smart it is. And no one really cared about cost.

18
00:01:08,190 --> 00:01:14,560
Ejaaz:
And then Fable 5 kind of came on the scene. And I think it was like $20 or $30

19
00:01:14,560 --> 00:01:16,530
Ejaaz:
input and like $80 output.

20
00:01:16,900 --> 00:01:19,700
Ejaaz:
And companies that were spending tens to hundreds of millions of dollars started

21
00:01:19,700 --> 00:01:25,650
Ejaaz:
to think, is this right? Does this make sense? Do I need the smartest model to do every single task?

22
00:01:25,830 --> 00:01:30,800
Ejaaz:
And so we hear the likes of XAI, Elon's company. We look at the likes of Meta.

23
00:01:31,020 --> 00:01:33,320
Ejaaz:
They released these models and we compare it to Fable 5, we're like,

24
00:01:33,320 --> 00:01:34,710
Ejaaz:
they're not actually that smart.

25
00:01:35,100 --> 00:01:38,150
Ejaaz:
But that's the whole purpose of the models that they're releasing.

26
00:01:38,360 --> 00:01:41,970
Ejaaz:
They're not trying to be as smart as Fable. In fact, they're betting on the opposite.

27
00:01:41,970 --> 00:01:46,910
Ejaaz:
They're betting that the cheapest model per unit intelligence is the model that

28
00:01:46,910 --> 00:01:48,640
Ejaaz:
will ultimately fill the middle ground.

29
00:01:48,640 --> 00:01:52,040
Ejaaz:
That'll be the ultimate model that is embedded in every single enterprise and

30
00:01:52,040 --> 00:01:53,140
Ejaaz:
used by every single user.

31
00:01:53,140 --> 00:01:57,080
Ejaaz:
Because the truth is you don't need the most intelligent model to do your task.

32
00:01:57,080 --> 00:02:01,850
Ejaaz:
Maybe if it is 90 to 95% of the intelligence of the smartest model ever,

33
00:02:02,070 --> 00:02:04,540
Ejaaz:
but it costs one-tenth of the price.

34
00:02:04,810 --> 00:02:08,500
Ejaaz:
It's a no-brainer that you're using these models at scale. And like you said,

35
00:02:08,500 --> 00:02:11,780
Ejaaz:
Josh, over the last week, there have been two particular companies,

36
00:02:11,780 --> 00:02:14,030
Ejaaz:
Meta, who we've known and spoken

37
00:02:14,030 --> 00:02:17,840
Ejaaz:
about a lot on this show, who have spent upwards of, I think, $35 billion.

38
00:02:18,500 --> 00:02:23,690
Ejaaz:
To try and build the world's best model, came out with their new Meta Mew Spark 1.1.

39
00:02:23,690 --> 00:02:27,570
Ejaaz:
And then you had SpaceX AI, who recently IPO'd, and there's a lot of pressure

40
00:02:27,570 --> 00:02:32,300
Ejaaz:
writing on them, release their new Grok 4.5 models. Now, are they as good as Fable? No.

41
00:02:32,550 --> 00:02:37,130
Ejaaz:
But are they cheap enough to use at scale when you're using or spinning up agents

42
00:02:37,130 --> 00:02:40,180
Ejaaz:
or when you're trying to figure out that long, complex task that's going to

43
00:02:40,180 --> 00:02:43,650
Ejaaz:
take tens of hours and you don't want to burn very expensive Fable tokens?

44
00:02:44,060 --> 00:02:48,270
Ejaaz:
These models might be the ones to choose. And I think it's probably worth covering a bunch of them.

45
00:02:48,700 --> 00:02:52,070
Josh:
Yeah, there's a new meta almost in town where it's like a new model doesn't

46
00:02:52,070 --> 00:02:53,540
Josh:
necessarily mean higher intelligence.

47
00:02:53,540 --> 00:02:56,080
Josh:
It could just mean higher efficiency. And I think that's where we're going to

48
00:02:56,080 --> 00:03:00,350
Josh:
start with with the Grok 4.5 release, because this is kind of like an opus class

49
00:03:00,350 --> 00:03:03,960
Josh:
claim, but at a third of the price, which is a pretty big deal.

50
00:03:03,960 --> 00:03:09,210
Josh:
So this new Grok 4.5 model, it's built on XAIs or I guess SpaceX AIs,

51
00:03:09,500 --> 00:03:13,170
Josh:
their version nine foundation model, which is about 1.5 trillion parameters.

52
00:03:13,170 --> 00:03:16,380
Josh:
And for reference, the version we've been using all year, if you've used Grok

53
00:03:16,380 --> 00:03:20,720
Josh:
at all this year, that is the version eight small, which is about 500 billion parameters.

54
00:03:20,720 --> 00:03:23,080
Josh:
So we're looking at about a three times multiple in parameter count,

55
00:03:23,080 --> 00:03:27,720
Josh:
which generally speaking is three times better, but probably a little bit more.

56
00:03:27,720 --> 00:03:30,390
Josh:
It seems like this is a serious increase relative to what we've been using.

57
00:03:30,720 --> 00:03:35,850
Josh:
And Elon has called this Opus class model. But instead of just being this like

58
00:03:35,850 --> 00:03:40,630
Josh:
highly expensive, very slow model, it's much more quick and it's much more efficient

59
00:03:40,630 --> 00:03:43,800
Josh:
when it comes to how many tokens you're able to generate for that same dollar.

60
00:03:44,080 --> 00:03:49,600
Josh:
And this is very clearly the route that you could see SpaceX AI has been trying to go for a long time.

61
00:03:49,890 --> 00:03:52,610
Josh:
They're very hardcore engineers. They love the engineering challenge.

62
00:03:52,810 --> 00:03:56,830
Josh:
And what they're trying to do now is figure out how you can kind of sculpt these

63
00:03:57,100 --> 00:04:00,880
Josh:
GPUs that they're training on to get as efficient as possible.

64
00:04:01,060 --> 00:04:04,720
Josh:
We spoke a few weeks ago, Ijaz, about the etched guys, like the startup who

65
00:04:04,720 --> 00:04:09,530
Josh:
is building their own training architecture chip stack. And they're basically

66
00:04:09,530 --> 00:04:10,410
Josh:
building their own servers.

67
00:04:10,640 --> 00:04:13,420
Josh:
And within that, they're pretty big to make one specific thing work,

68
00:04:13,420 --> 00:04:14,330
Josh:
which is the transformer.

69
00:04:14,550 --> 00:04:17,590
Josh:
And we talked about how gpus are not very efficient they don't actually use

70
00:04:17,590 --> 00:04:23,100
Josh:
like sometimes up to 60 percent of the gpu isn't used what grok and the spacex ai team are doing

71
00:04:23,400 --> 00:04:26,680
Josh:
is they are taking that code base they're really getting down to the bare metal

72
00:04:26,680 --> 00:04:29,870
Josh:
to figure out how to squeeze the most juice out of it and that's what this model

73
00:04:29,870 --> 00:04:31,760
Josh:
is that's what 4.5 is it has

74
00:04:32,090 --> 00:04:35,940
Josh:
two dollars in six dollars out per million tokens generated

75
00:04:36,360 --> 00:04:40,940
Josh:
And it seems like it's incredibly efficient when comparing it to other models like Opus.

76
00:04:41,280 --> 00:04:47,460
Josh:
It appears as if it's up to four point times fewer tokens needed in order to reach the same task.

77
00:04:47,660 --> 00:04:51,280
Josh:
So that's like an adjusted multiple of what it says, 17 times less than Opus.

78
00:04:51,280 --> 00:04:55,060
Josh:
That's like a really big deal for a model as it relates to cost, at least.

79
00:04:55,060 --> 00:05:00,120
Ejaaz:
The major unlock that you have with Grok 4.5 is it's a model meant for building.

80
00:05:00,120 --> 00:05:02,830
Ejaaz:
So if you're a hobbyist out there, if you're a software engineer and you want

81
00:05:02,830 --> 00:05:05,480
Ejaaz:
to tinker with some of these models and try and build something,

82
00:05:05,480 --> 00:05:09,060
Ejaaz:
but you know it's costly and expensive. You don't want to use the Fable 5 API.

83
00:05:09,340 --> 00:05:13,120
Ejaaz:
This is a really, really good model to use because it's so efficient.

84
00:05:13,120 --> 00:05:15,510
Ejaaz:
So some of the stats you were referencing right there, Josh,

85
00:05:16,140 --> 00:05:21,250
Ejaaz:
basically it uses four times fewer tokens than Claude Opus 4.8.

86
00:05:21,520 --> 00:05:28,020
Ejaaz:
What that results in is a 17x less price or cost to do the same task.

87
00:05:28,280 --> 00:05:32,680
Ejaaz:
And the way that they've been able to achieve this is arguably the funnest or coolest part.

88
00:05:33,010 --> 00:05:37,210
Ejaaz:
They spin up a bunch of different agents to solve different problems of the

89
00:05:37,210 --> 00:05:40,760
Ejaaz:
tasks that you've asked it to at the same time in parallel.

90
00:05:40,760 --> 00:05:43,760
Ejaaz:
And this is a growing trend you'll see with the other models that we're going

91
00:05:43,760 --> 00:05:47,630
Ejaaz:
to talk about on this show today, where agentic coding or agentic reasoning

92
00:05:47,630 --> 00:05:52,570
Ejaaz:
or using agents to solve your problem has been a major unlock for these cheaper

93
00:05:52,570 --> 00:05:54,990
Ejaaz:
models to be cheap in the first place.

94
00:05:55,360 --> 00:05:57,900
Ejaaz:
The other thing I want to mention about Grok 4.5 in particular,

95
00:05:57,900 --> 00:06:03,310
Ejaaz:
because it's very unique to them and SpaceX AI is they just acquired a company called Cursor for...

96
00:06:03,860 --> 00:06:08,340
Ejaaz:
You know, a low price of $60 billion. And the major unlock that cursor gave

97
00:06:08,340 --> 00:06:13,740
Ejaaz:
them is they have all this data around how users use coding models.

98
00:06:13,740 --> 00:06:18,260
Ejaaz:
Now, it's very specific. It's not what they're coding. It is how they use the models.

99
00:06:18,540 --> 00:06:25,870
Ejaaz:
And this data was very important in training Grok 4.5 to become smarter at routing people's requests.

100
00:06:26,070 --> 00:06:32,650
Ejaaz:
So let's say you ask it to build an app that can change your wardrobe into something

101
00:06:33,070 --> 00:06:35,670
Ejaaz:
way better. And you send it like pictures of your camera or whatever that might

102
00:06:35,670 --> 00:06:38,950
Ejaaz:
be. We talked about this as an example on yesterday's episode.

103
00:06:39,260 --> 00:06:45,100
Ejaaz:
It'll be able to know which model to use at what time for how long and calculate

104
00:06:45,100 --> 00:06:48,290
Ejaaz:
the costs preemptively to make sure that it's not burning your wallet.

105
00:06:48,290 --> 00:06:51,930
Ejaaz:
It's a really, really smart model. And Elon himself has said,

106
00:06:52,250 --> 00:06:55,420
Ejaaz:
listen, this isn't gonna be the smartest model. We're working on better models,

107
00:06:55,610 --> 00:06:59,100
Ejaaz:
but it is a really good daily driver. It's a really good workhorse.

108
00:06:59,300 --> 00:07:02,840
Ejaaz:
And if you're anyone at a company that doesn't want to burn, you know,

109
00:07:03,110 --> 00:07:07,650
Ejaaz:
$30 in and $80 out on Fable 5 or whatever the cost is, something that's like

110
00:07:07,650 --> 00:07:12,090
Ejaaz:
10x more, you should use this model for 80 to 90% of the work that you're trying

111
00:07:12,090 --> 00:07:15,530
Ejaaz:
to do, and then use Fable to kind of orchestrate the plan or the design or whatever that might be.

112
00:07:15,920 --> 00:07:19,450
Ejaaz:
The final point I'm going to make in SpaceX AI's favor, because you might be

113
00:07:19,450 --> 00:07:22,390
Ejaaz:
thinking, okay, fine, whatever, but like, when's the next model going to come?

114
00:07:22,390 --> 00:07:25,450
Ejaaz:
They took like, whatever, nine months to release this upgrade.

115
00:07:25,970 --> 00:07:31,700
Ejaaz:
Elon is cooking up three more models that are in order of magnitude larger than

116
00:07:31,700 --> 00:07:33,140
Ejaaz:
the model that we're talking about today.

117
00:07:33,340 --> 00:07:37,610
Ejaaz:
So right now, in under a month, we're expecting to see Grok 5,

118
00:07:37,610 --> 00:07:40,170
Ejaaz:
and Grok 5.5 is also being cooked at the same time as well.

119
00:07:40,170 --> 00:07:44,360
Ejaaz:
And these are like 5 to 10 trillion parameter models. So it's feasible to say

120
00:07:44,360 --> 00:07:48,490
Ejaaz:
that we'll have a Mythos class model from SpaceX AI in a couple of months' time.

121
00:07:48,830 --> 00:07:51,230
Josh:
Which is awesome. Yeah, this is the part that's most exciting to me.

122
00:07:51,230 --> 00:07:55,700
Josh:
Is like it's very obvious that spacex ai is the best at the engineering part

123
00:07:55,700 --> 00:07:56,780
Josh:
of generating tokens they're

124
00:07:57,070 --> 00:07:59,760
Josh:
have the whole vertical integration now they have the ability to build out these

125
00:07:59,760 --> 00:08:03,150
Josh:
data centers they have the data centers running and now they have the harness

126
00:08:03,370 --> 00:08:07,510
Josh:
through cursor and they also have the data set that they've used through cursor in order to kind of

127
00:08:07,830 --> 00:08:11,740
Josh:
achieve this fully integrated stack and what's funny now is if they are continuing

128
00:08:11,740 --> 00:08:13,190
Josh:
to release models that are bigger and bigger

129
00:08:13,700 --> 00:08:17,030
Josh:
if they ever do run up against a compute wall they remember they have that deal

130
00:08:17,030 --> 00:08:20,760
Josh:
with anthropic and with google i believe and

131
00:08:21,250 --> 00:08:24,890
Josh:
they're gonna have to figure out who's who's gonna get the short end of that

132
00:08:24,890 --> 00:08:27,910
Josh:
stick because it doesn't seem like they're gonna have gpus for everyone but

133
00:08:27,910 --> 00:08:30,560
Josh:
like you mentioned the the exciting thing here is that like this is version

134
00:08:30,560 --> 00:08:33,690
Josh:
one of it feels like spacex ai 2.0 this is their

135
00:08:34,090 --> 00:08:37,600
Josh:
second try at getting to the frontier the first time they may have gotten there

136
00:08:37,600 --> 00:08:40,680
Josh:
for a couple days but it wasn't very long lived now we're getting new models

137
00:08:40,680 --> 00:08:42,610
Josh:
every single month and the goal

138
00:08:43,020 --> 00:08:48,120
Josh:
around august is a two trillion parameter model and then the goal for grok 5

139
00:08:48,120 --> 00:08:52,790
Josh:
which is coming hopefully not too long after is 6 to 10 trillion parameter models

140
00:08:52,790 --> 00:08:55,440
Josh:
like this is this is going to put them right up at the frontier and if they're

141
00:08:55,440 --> 00:08:56,750
Josh:
able to serve these tokens

142
00:08:57,050 --> 00:09:01,890
Josh:
at a fraction of the cost that say gpt6 is going to be or mythos 6 or whatever

143
00:09:01,890 --> 00:09:04,150
Josh:
comes next that's going to be a pretty serious

144
00:09:04,490 --> 00:09:07,050
Josh:
competitor in the ai space because they're going to be right at that frontier

145
00:09:07,340 --> 00:09:09,640
Josh:
with a very low cost model that i think a lot of people

146
00:09:10,280 --> 00:09:15,380
Josh:
Are going to find a lot of use for now that is the grok update there is a second

147
00:09:15,380 --> 00:09:19,710
Josh:
update that is just as noteworthy and probably even more noteworthy actually

148
00:09:19,710 --> 00:09:23,190
Josh:
because this is the first time that meta is charging for tokens

149
00:09:23,620 --> 00:09:28,190
Josh:
meta famously they have been the open source kings they have always wanted to

150
00:09:28,190 --> 00:09:32,290
Josh:
publish the models open source to move the needle forward to kind of make this

151
00:09:32,290 --> 00:09:34,300
Josh:
open source developer community thrive

152
00:09:34,660 --> 00:09:39,100
Josh:
unfortunately if you are a participant of that your time has ended because now

153
00:09:39,100 --> 00:09:41,200
Josh:
meta is closed source and they are releasing

154
00:09:41,510 --> 00:09:44,210
Josh:
these closed source models they are charging via the api it's not a lot of money

155
00:09:44,520 --> 00:09:49,320
Josh:
but it was big enough news for mark zuckerberg to come back on x after a what

156
00:09:49,320 --> 00:09:51,390
Josh:
was a three or four year hiatus and actually announce

157
00:09:51,670 --> 00:09:56,090
Josh:
muse spark 1.1 which he describes as a strong agentic and coding model at a

158
00:09:56,090 --> 00:10:00,480
Josh:
very low price it's available through our new meta model api and in the meta

159
00:10:00,480 --> 00:10:02,430
Josh:
ai so ejes the question i have for you

160
00:10:02,750 --> 00:10:05,870
Josh:
because you know i'm i haven't been the most excited about meta recently

161
00:10:06,200 --> 00:10:09,010
Josh:
their offerings have left a lot to be desired as it relates to AI.

162
00:10:09,300 --> 00:10:14,470
Josh:
Is this a serious model? Like, is this worth actually paying for relative to

163
00:10:14,470 --> 00:10:15,770
Josh:
all the other models that exist today?

164
00:10:16,230 --> 00:10:19,990
Ejaaz:
Short answer is yes. And this comes from a professional meta hater.

165
00:10:19,990 --> 00:10:22,620
Josh:
Which is shocking. Like this is this is a novel breakthrough.

166
00:10:22,620 --> 00:10:24,130
Josh:
This is an exciting announcement for meta.

167
00:10:24,320 --> 00:10:28,630
Ejaaz:
Well, I'm just happy to see something competitive enough for the,

168
00:10:29,170 --> 00:10:33,460
Ejaaz:
a lot of people to use in the AI market. And I'll tell you why they would use this model.

169
00:10:33,860 --> 00:10:40,440
Ejaaz:
Number one, the reason right at the top is it is the cheapest model per unit

170
00:10:40,440 --> 00:10:42,560
Ejaaz:
cost of intelligence. So what do I mean by that?

171
00:10:42,880 --> 00:10:46,910
Ejaaz:
You can have cheap models, but they're pretty crappy and you won't end up actually

172
00:10:46,910 --> 00:10:48,080
Ejaaz:
using it for serious work.

173
00:10:48,420 --> 00:10:53,900
Ejaaz:
This is a model that you'll end up using for serious work and it costs 25% of

174
00:10:53,900 --> 00:10:57,730
Ejaaz:
the price of the Frontier model. Now, can it do everything a Frontier model

175
00:10:57,950 --> 00:11:01,910
Ejaaz:
can do, like what Fable 5 does? No. And Mark Zuckerberg openly admits that.

176
00:11:02,170 --> 00:11:07,620
Ejaaz:
But it is an absolute workhorse. You can throw it at a task and it can work

177
00:11:07,620 --> 00:11:11,720
Ejaaz:
for hours, or you could throw it at multiple problems at once and it can figure it out.

178
00:11:12,010 --> 00:11:15,350
Ejaaz:
But there's a few other advantages that this model in particular has.

179
00:11:15,350 --> 00:11:22,160
Ejaaz:
So aside from the price, which is, by the way, $1.25 in per million tokens and $4.25 out.

180
00:11:22,390 --> 00:11:25,750
Ejaaz:
Now, if you want a comparison as to how crazy cheap that is,

181
00:11:26,050 --> 00:11:31,210
Ejaaz:
that is cheaper than GLM's model, GLM 5.2, which is the leading open source

182
00:11:31,210 --> 00:11:32,530
Ejaaz:
Chinese model right now.

183
00:11:32,760 --> 00:11:36,180
Ejaaz:
And they're known for being the cheapest model. So the fact that Mark Zuckerberg,

184
00:11:36,180 --> 00:11:38,840
Ejaaz:
there's some blissful irony there, actually, with the open source thing and

185
00:11:38,840 --> 00:11:40,790
Ejaaz:
the fact that, you know, he was competing with China and they beat him.

186
00:11:41,080 --> 00:11:45,770
Ejaaz:
To come back and offer the cheapest model is pretty amazing. But the second thing is.

187
00:11:46,120 --> 00:11:51,960
Ejaaz:
And like Grok 4.5, it uses agents very intelligently. So it has this thing called

188
00:11:51,960 --> 00:11:53,620
Ejaaz:
a master agent in this model.

189
00:11:53,940 --> 00:11:57,860
Ejaaz:
And the master agent reads your prompt and it thinks very diligently about a

190
00:11:57,860 --> 00:12:00,290
Ejaaz:
plan to answer and execute your prompt.

191
00:12:00,290 --> 00:12:02,900
Ejaaz:
Now that sounds very vague, right? You're like, oh, aren't the other models

192
00:12:02,900 --> 00:12:06,480
Ejaaz:
doing it? No, when you look at like Fable 5, when you look at GPT 5.6,

193
00:12:06,810 --> 00:12:11,460
Ejaaz:
it processes its entire model weights, which is gargantuan by the way.

194
00:12:11,670 --> 00:12:16,090
Ejaaz:
And that ends up being very costly. Meta found a sneaky little way to circumvent

195
00:12:16,090 --> 00:12:19,530
Ejaaz:
this, which is have a master agent, have it plan, and then have it delegate

196
00:12:19,530 --> 00:12:20,700
Ejaaz:
to a bunch of different agents.

197
00:12:21,000 --> 00:12:25,480
Ejaaz:
The third thing that is very impressive about this model is that it's an omni model.

198
00:12:25,650 --> 00:12:30,230
Ejaaz:
So it can take in video, it can take in images, it can take in text all at once

199
00:12:30,490 --> 00:12:33,440
Ejaaz:
and understand how to use that intelligently.

200
00:12:33,440 --> 00:12:37,600
Ejaaz:
So if you give it a video, you can extract that video, it knows that you want

201
00:12:37,600 --> 00:12:41,350
Ejaaz:
to list it or use it for a post that you have on Facebook Marketplace,

202
00:12:41,590 --> 00:12:43,830
Ejaaz:
and it'll be able to do that in one shot.

203
00:12:44,000 --> 00:12:47,470
Ejaaz:
And then the fourth and final thing that it's very good at is computer use.

204
00:12:47,470 --> 00:12:50,610
Ejaaz:
So this thing can take over your computer, take over your account or whatever

205
00:12:50,610 --> 00:12:54,500
Ejaaz:
that might be, and just know what you want it to do. Now, version one is very

206
00:12:54,500 --> 00:12:58,140
Ejaaz:
fine tuned to Meta's products, Instagram, WhatsApp, Facebook itself.

207
00:12:58,410 --> 00:13:02,510
Ejaaz:
But the idea is you can pretty much use this for any computer use work going

208
00:13:02,510 --> 00:13:05,550
Ejaaz:
forward. And I'm looking forward to version 1.2 and 1.3.

209
00:13:05,830 --> 00:13:10,770
Ejaaz:
They pulled this off, by the way, within a month of releasing Mew Spark 1.

210
00:13:11,560 --> 00:13:15,530
Ejaaz:
And they also released Mew's image and video in the same week,

211
00:13:15,530 --> 00:13:21,410
Ejaaz:
which tells me that the cycle of release similar to SpaceX AI is getting much quicker.

212
00:13:21,700 --> 00:13:26,560
Ejaaz:
And if you ask me why that is, it's because both Elon and Zuck have one thing

213
00:13:26,730 --> 00:13:29,280
Ejaaz:
that the two frontier labs, OpenA and Anthropic, don't have.

214
00:13:30,240 --> 00:13:36,210
Ejaaz:
A crap ton of compute. They have so much compute. They are the most aggressive

215
00:13:36,530 --> 00:13:40,720
Ejaaz:
hyperscalers and arguably the most successful hyperscalers. They have amassed the most GPUs.

216
00:13:41,020 --> 00:13:45,270
Ejaaz:
And if you still believe that compute scaling laws matter, Zuck and Elon are

217
00:13:45,270 --> 00:13:47,610
Ejaaz:
not out of the race. In fact, this proves that they're back into it.

218
00:13:47,900 --> 00:13:51,520
Josh:
One of the things that I found noteworthy of this is that this API presumably

219
00:13:51,520 --> 00:13:53,420
Josh:
runs on Meta's own silicon.

220
00:13:53,580 --> 00:13:56,990
Josh:
It's the chips that they have been making and producing and likely running in

221
00:13:56,990 --> 00:14:00,410
Josh:
fact there's a report that they are spending 250

222
00:14:01,010 --> 00:14:05,230
Josh:
billion dollars including chips over the next year in order to build out something

223
00:14:05,230 --> 00:14:07,610
Josh:
like 14 gigawatts of data center capability

224
00:14:07,920 --> 00:14:13,350
Josh:
and that is mostly going to be running their new mtia 400 chips which are basically

225
00:14:13,350 --> 00:14:19,310
Josh:
the in-house meta silicon which is 400 faster than the previous generation and

226
00:14:19,310 --> 00:14:22,730
Josh:
uses 51 percent more hbm so for the hbm folks

227
00:14:23,430 --> 00:14:27,040
Josh:
When we talk about all our investing videos um these chips are going to be using

228
00:14:27,040 --> 00:14:32,140
Josh:
a lot more of it and meta is now charging a quarter of the rival's prices which is a really

229
00:14:32,440 --> 00:14:36,160
Josh:
kind of interesting and noteworthy thing and that combined with the switch from

230
00:14:36,160 --> 00:14:39,760
Josh:
open source to closed source it kind of implies that like mark zuckerberg very

231
00:14:39,760 --> 00:14:43,160
Josh:
clearly thinks the model is now the product instead of the moat around

232
00:14:43,440 --> 00:14:47,470
Josh:
the model so i think traditional meta would have been no no we're going to release

233
00:14:47,470 --> 00:14:50,990
Josh:
the model we're going to build a moat around it and that is going to be product we monetize

234
00:14:51,240 --> 00:14:54,640
Josh:
this new shift implies like no actually the model is the product and we're going

235
00:14:54,640 --> 00:14:57,690
Josh:
to integrate it into all of our services but in order to do that we're going

236
00:14:57,690 --> 00:15:00,470
Josh:
to do this huge tremendous data center build out and they're going to

237
00:15:00,790 --> 00:15:04,000
Josh:
build their own proprietary chips and it seems like both those things are actually

238
00:15:04,300 --> 00:15:06,890
Josh:
going well and i have to ask it's like okay

239
00:15:07,330 --> 00:15:11,340
Josh:
what happens if they actually do it like what happens if there is 14 gigawatts

240
00:15:11,340 --> 00:15:15,200
Josh:
of compute running next year like that's that seems pretty considerable it's

241
00:15:15,200 --> 00:15:16,680
Josh:
particularly running on their own silicon

242
00:15:17,270 --> 00:15:21,420
Josh:
which we know has tremendous competitive advantages when it comes to training your own models.

243
00:15:21,990 --> 00:15:26,210
Ejaaz:
Well, here's the bet that like we're making, we've made on this show multiple

244
00:15:26,210 --> 00:15:32,200
Ejaaz:
times on previous episodes, is that the future of AI isn't you tapping a bunch

245
00:15:32,200 --> 00:15:34,850
Ejaaz:
of buttons and approvals every single second.

246
00:15:35,200 --> 00:15:39,980
Ejaaz:
It is you write a prompt, you send the prompt, and then the AI just kind of

247
00:15:39,980 --> 00:15:42,180
Ejaaz:
knows what to do. It autonomously works.

248
00:15:42,430 --> 00:15:46,500
Ejaaz:
If you believe in that world, then you believe in a world of AI agents.

249
00:15:46,710 --> 00:15:50,060
Ejaaz:
And if you believe in a world of AI agents, these agents are going to be making

250
00:15:50,200 --> 00:15:54,070
Ejaaz:
tens to hundreds of thousands of tool calls per month.

251
00:15:54,400 --> 00:15:56,620
Ejaaz:
You don't want to be there clicking approve the entire time.

252
00:15:56,800 --> 00:15:58,600
Ejaaz:
And those tool calls are pretty expensive.

253
00:15:58,850 --> 00:16:03,600
Ejaaz:
So you want to go the cheapest route when you're using an agent to get the work

254
00:16:03,600 --> 00:16:08,380
Ejaaz:
done. That's basically the entire thesis for why cheaper models are more effective.

255
00:16:08,700 --> 00:16:13,480
Ejaaz:
And therefore, the labs who have the most compute and can use it most effectively,

256
00:16:13,480 --> 00:16:15,350
Ejaaz:
you mentioned, you know, their custom chips.

257
00:16:15,920 --> 00:16:18,910
Ejaaz:
I believe Elon and SpaceX are also working on their own chips.

258
00:16:18,910 --> 00:16:21,940
Ejaaz:
Open Air is doing their own with Jalapeno. This is a growing trend.

259
00:16:22,290 --> 00:16:26,000
Ejaaz:
It makes sense that the people that have the most compute and the best chip

260
00:16:26,000 --> 00:16:31,780
Ejaaz:
architecture will end up winning. And that's basically the bet that Meta and Elon are going after.

261
00:16:32,240 --> 00:16:36,620
Ejaaz:
I have to say, if we ground ourselves for a second and look at the counter-thesis...

262
00:16:37,200 --> 00:16:40,000
Ejaaz:
I do think Zuck is heavily subsidizing this model.

263
00:16:40,200 --> 00:16:45,380
Ejaaz:
I don't think there's any chance in hell that it actually costs $125 and $425

264
00:16:45,380 --> 00:16:47,880
Ejaaz:
output. I think he's subsidizing this massively.

265
00:16:48,200 --> 00:16:51,620
Ejaaz:
And I think he needs to prove a point to his investors or shareholders that

266
00:16:51,890 --> 00:16:56,190
Ejaaz:
it is worth the AI CapEx spend that he's probably going to announce at the end of Q2.

267
00:16:56,190 --> 00:16:59,270
Ejaaz:
So that's my bet. I think it's strategic. I think it's the right move,

268
00:16:59,270 --> 00:17:04,040
Ejaaz:
but I don't think he's unlocked some kind of major architecture redesign just yet.

269
00:17:04,450 --> 00:17:09,130
Josh:
Regardless, these are two really solid models between Grok 4.5 and now MuseSpark 1.1.

270
00:17:09,460 --> 00:17:14,650
Josh:
And to kind of place them in a spot relative to others, we can go down the price

271
00:17:14,650 --> 00:17:18,450
Josh:
list of other models and kind of compare what they're like. So at the top of

272
00:17:18,450 --> 00:17:23,390
Josh:
this list is Claude Fable 5, and that's $10 in, $50 out per million tokens.

273
00:17:23,640 --> 00:17:29,290
Josh:
GPT 5.6 comes next at just a little bit, close to half at $5 in, $30 out.

274
00:17:29,640 --> 00:17:32,260
Josh:
Then Opus is $5 in, $25 out.

275
00:17:32,860 --> 00:17:35,770
Josh:
And then it kind of goes down the line until we get to Grok 4.5,

276
00:17:35,770 --> 00:17:42,760
Josh:
which is two dollars in six dollars out then we have muse spark 125 and 425

277
00:17:43,100 --> 00:17:46,530
Josh:
and at the very bottom believe it or not is open ai with a dollar

278
00:17:46,890 --> 00:17:53,550
Josh:
in and six dollars out for gpt 5.6 luna so the middle section and the upper

279
00:17:53,550 --> 00:17:57,790
Josh:
section is kind of where the war is it's like we have those two models at the top

280
00:17:58,220 --> 00:18:00,200
Josh:
we have claude and we have gpt 5.6 soul

281
00:18:00,870 --> 00:18:04,490
Josh:
Those are very much competing on the intelligence curve. But then below that,

282
00:18:04,490 --> 00:18:08,930
Josh:
where we have this like kind of cluster of, you can think Gemini and the smaller

283
00:18:08,930 --> 00:18:14,800
Josh:
GPT models and Grok and Meta 1.1, that's where we are seeing this like price competition.

284
00:18:14,800 --> 00:18:17,450
Josh:
And it's funny to see this kind of divergence in strategies.

285
00:18:17,450 --> 00:18:20,830
Josh:
And I think that's a good way of looking at the frontier when you evaluate these

286
00:18:20,830 --> 00:18:25,860
Josh:
new models is, okay, is this a frontier intelligence model or is this a frontier price model?

287
00:18:26,090 --> 00:18:28,870
Josh:
And those two things now are very different because they're going to be used

288
00:18:28,870 --> 00:18:31,510
Josh:
for a very different set of use cases. And I think one of the more interesting

289
00:18:31,510 --> 00:18:35,110
Josh:
applications that we're going to be following on the show is how people route

290
00:18:35,110 --> 00:18:36,720
Josh:
through these models to do different tasks.

291
00:18:37,030 --> 00:18:41,350
Josh:
Like you said, some agentic tasks don't require necessarily the highest frontier intelligence.

292
00:18:41,630 --> 00:18:45,210
Josh:
Maybe you can get away with using a MuseSpark 1.1 for a lot of that,

293
00:18:45,470 --> 00:18:49,600
Josh:
and perhaps using a Fable 5 for orchestration of those agents, things like that.

294
00:18:49,600 --> 00:18:53,760
Josh:
So we're going to see what I suspect is a new meta start to come into play of

295
00:18:53,990 --> 00:18:58,740
Josh:
this orchestration at a high level and using different models for more particular

296
00:18:58,740 --> 00:19:00,070
Josh:
things, more specific things.

297
00:19:00,920 --> 00:19:06,100
Ejaaz:
Josh, have you heard of something called the Silicon Token Expenditure Index?

298
00:19:07,160 --> 00:19:13,440
Ejaaz:
It's basically this index which tracks the spending of companies or enterprises.

299
00:19:13,830 --> 00:19:17,720
Ejaaz:
I think the Bloomberg ticker is SDLLMTK.

300
00:19:18,180 --> 00:19:23,750
Ejaaz:
But the point of this index is it captures how much money is being spent by

301
00:19:23,940 --> 00:19:27,580
Ejaaz:
the top Fortune 500 on AI specifically.

302
00:19:27,830 --> 00:19:30,750
Ejaaz:
So if you're listening to us talking about cheap models and And,

303
00:19:30,750 --> 00:19:34,700
Ejaaz:
you know, talking about this thesis of cheaper models will be used more and you don't believe it.

304
00:19:35,030 --> 00:19:38,700
Ejaaz:
Well, you have to look at the customers who are actually ingesting this AI and

305
00:19:38,700 --> 00:19:40,740
Ejaaz:
the movements and actions that they're taking.

306
00:19:41,110 --> 00:19:45,780
Ejaaz:
And if I show you the index, you'll notice a particular trend over the last

307
00:19:45,780 --> 00:19:52,090
Ejaaz:
couple of weeks and months, which is it's down 20%, which basically means people

308
00:19:52,090 --> 00:19:55,520
Ejaaz:
are spending much less on tokens or per token.

309
00:19:55,790 --> 00:19:59,680
Ejaaz:
And they're also spending much less overall. What that indicates is they are

310
00:19:59,680 --> 00:20:02,050
Ejaaz:
looking for cheaper alternatives.

311
00:20:02,050 --> 00:20:04,600
Ejaaz:
And we're seeing this from the headlines that we've seen from Uber,

312
00:20:04,870 --> 00:20:08,770
Ejaaz:
from Meta themselves, from Microsoft, who are now adopting Chinese open source

313
00:20:08,770 --> 00:20:11,110
Ejaaz:
models into their product, into their Copilot product.

314
00:20:11,440 --> 00:20:15,180
Ejaaz:
We are seeing this shift of enterprises realizing that it's not about using

315
00:20:15,180 --> 00:20:16,840
Ejaaz:
the smartest model for every single task.

316
00:20:17,110 --> 00:20:21,290
Ejaaz:
It's about finding that middle ground. It's about finding the right model or

317
00:20:21,290 --> 00:20:26,320
Ejaaz:
maybe the right types of models to use for the middle ground of tasks.

318
00:20:26,580 --> 00:20:28,450
Ejaaz:
And that brings me to another point which is.

319
00:20:29,160 --> 00:20:34,080
Ejaaz:
I don't think it's necessarily just going to be one cheap model that dominates everything.

320
00:20:34,280 --> 00:20:38,370
Ejaaz:
I think they're going to realize after they've shifted to a cheaper model that

321
00:20:38,670 --> 00:20:41,820
Ejaaz:
they could use certain models for specific tasks.

322
00:20:41,820 --> 00:20:46,190
Ejaaz:
And that takes us down the path of routing, which is what Cursor has infamously

323
00:20:46,190 --> 00:20:49,310
Ejaaz:
figured out and what SpaceX figured out and acquired them for $60 billion.

324
00:20:49,330 --> 00:20:52,440
Ejaaz:
So there's these really interesting conversations around this trend,

325
00:20:52,440 --> 00:20:57,390
Ejaaz:
but it's being validated by this index that companies are going to be looking for cheaper models.

326
00:20:57,600 --> 00:21:03,520
Ejaaz:
And the irony is, what is the paradox, Josh, that we spoke about early on in

327
00:21:03,520 --> 00:21:07,860
Ejaaz:
the AI thesis, where the cheaper something gets, the more you end up spending

328
00:21:07,860 --> 00:21:09,910
Ejaaz:
on it? What is the name of that paradox?

329
00:21:10,900 --> 00:21:14,600
Ejaaz:
Jevon's Paradox. Jevon's Paradox. That's it. So the cheaper something gets,

330
00:21:14,600 --> 00:21:18,860
Ejaaz:
I expect to see way more tokens being spent because the output will actually

331
00:21:18,860 --> 00:21:23,400
Ejaaz:
be worth it instead of spending a very expensive prompt and getting maybe a

332
00:21:23,400 --> 00:21:24,570
Ejaaz:
mediocre answer from it.

333
00:21:25,030 --> 00:21:28,640
Josh:
Yeah, and this is in line with expectations. Goldman, they published this prediction

334
00:21:28,870 --> 00:21:33,990
Josh:
that token consumption would multiply 24 times between 2026 and 2030.

335
00:21:33,990 --> 00:21:40,770
Josh:
That's like a tremendous amount. That's 120 quadrillion tokens per month, which is unbelievable.

336
00:21:40,910 --> 00:21:44,650
Josh:
So even if the price of these tokens does go down some extent,

337
00:21:44,900 --> 00:21:49,540
Josh:
a multiplier of that much is still increased spend, which I think is important to note.

338
00:21:49,540 --> 00:21:56,700
Josh:
It's like, I don't think we're going to see a Fable class model be 124th the price very shortly.

339
00:21:56,990 --> 00:22:00,350
Josh:
But they're expecting that much increased demand and every single kind of

340
00:22:01,100 --> 00:22:06,390
Josh:
Every single prevailing wind is pushing towards longer-term agentic coding sessions,

341
00:22:06,680 --> 00:22:10,060
Josh:
where I feel like rarely do people actually now, or people who are using it

342
00:22:10,060 --> 00:22:12,740
Josh:
for productive uses at least, are using it just as a chatbot.

343
00:22:12,740 --> 00:22:16,420
Josh:
They're using it as an agent to do longer and longer and longer-term tasks.

344
00:22:16,720 --> 00:22:20,030
Josh:
And what we notice with these cheaper models in particular is that they actually

345
00:22:20,030 --> 00:22:23,960
Josh:
do oftentimes consume more tokens to get the same output at a higher quality.

346
00:22:24,150 --> 00:22:28,270
Josh:
I know we were mentioning this with 5.6 the other day, where some of their lower

347
00:22:28,270 --> 00:22:32,020
Josh:
models, they actually, they cost less, but they do use a considerable amount

348
00:22:32,020 --> 00:22:33,840
Josh:
more tokens to get to the answer.

349
00:22:34,140 --> 00:22:36,780
Josh:
So we're in this weird crossroads where we have these, like,

350
00:22:36,780 --> 00:22:39,620
Josh:
forcing functions that are pushing people to want to generate more tokens.

351
00:22:39,930 --> 00:22:42,510
Josh:
Companies want these tokens to be highly intelligent because they don't want

352
00:22:42,510 --> 00:22:43,870
Josh:
subpar work if they're paying for it.

353
00:22:44,580 --> 00:22:48,030
Josh:
But then they have this like enterprise belt tightening that's kind of happening

354
00:22:48,030 --> 00:22:51,910
Josh:
where like amazon killed it's like i think they had their internal token leaderboard

355
00:22:51,910 --> 00:22:52,990
Josh:
and they totally removed it

356
00:22:53,320 --> 00:22:56,780
Josh:
um coinbase and walmart are setting up usage caps a lot of companies are kind

357
00:22:56,780 --> 00:23:03,020
Josh:
of like crunching down on usage of ai internally because they can't really figure out how to justify

358
00:23:03,340 --> 00:23:06,470
Josh:
the value that's been given to the company so there's a lot of these weird it

359
00:23:06,470 --> 00:23:09,490
Josh:
very much feels like we're at a crossroads now and there's this new paradigm

360
00:23:09,490 --> 00:23:12,820
Josh:
shift that is happening as it relates to cost in particular and it's going to

361
00:23:12,820 --> 00:23:14,060
Josh:
be really interesting to see how this plays out

362
00:23:14,930 --> 00:23:21,790
Ejaaz:
There is one metric that i will be tracking to see whether these cheap models

363
00:23:22,100 --> 00:23:25,000
Ejaaz:
are actually good enough and it's simple,

364
00:23:25,790 --> 00:23:29,670
Ejaaz:
are these companies making money? When do they turn profitable?

365
00:23:30,720 --> 00:23:35,200
Ejaaz:
With Meta, I'm almost certain that they're subsidizing their cost of their model.

366
00:23:35,690 --> 00:23:39,370
Ejaaz:
With SpaceX AI, I don't know, maybe they're doing something similar.

367
00:23:39,690 --> 00:23:44,770
Ejaaz:
But until these guys post a positive revenue earnings on their quarterly earnings

368
00:23:44,770 --> 00:23:49,210
Ejaaz:
or whatever that might be, I'm not going to be convinced that cheap models are the way forward.

369
00:23:49,210 --> 00:23:52,180
Ejaaz:
I think it will directionally play out, but that's going to be a big metric

370
00:23:52,180 --> 00:23:55,680
Ejaaz:
to track. Now, if you compare that to the likes of Anthropic,

371
00:23:55,680 --> 00:23:59,520
Ejaaz:
they're not a public company yet. They're rumored to IPO maybe later this year.

372
00:24:00,050 --> 00:24:05,640
Ejaaz:
But there are rumors that they have turned profitable in Q2.

373
00:24:06,420 --> 00:24:10,950
Ejaaz:
And if true or if confirmed, they'll be the first AI lab to do so.

374
00:24:11,230 --> 00:24:15,520
Ejaaz:
And the major way that they've been able to do that is not only creating amazing

375
00:24:15,520 --> 00:24:17,930
Ejaaz:
models that are used by every single enterprise in the world,

376
00:24:18,250 --> 00:24:22,230
Ejaaz:
but because it's expensive, because they pay for the cost that you're getting.

377
00:24:22,230 --> 00:24:25,140
Ejaaz:
And at the end of the day, if you're making a lot of money and people continue

378
00:24:25,140 --> 00:24:28,540
Ejaaz:
using it, maybe that is the right way to go through. So that that would be the

379
00:24:28,540 --> 00:24:30,310
Ejaaz:
only the counterpoint to this.

380
00:24:31,100 --> 00:24:34,310
Ejaaz:
The final thing I'll say on this topic, I think, is.

381
00:24:35,370 --> 00:24:40,560
Ejaaz:
What I was alluding to earlier, which is I'm now convinced that,

382
00:24:40,930 --> 00:24:44,920
Ejaaz:
number one, it's not just going to be OpenAI and Anthropic ruling the entire world.

383
00:24:44,920 --> 00:24:47,500
Ejaaz:
I think it's going to be multiple labs. And I think that's ultimately a very

384
00:24:47,500 --> 00:24:50,740
Ejaaz:
good future, right? Maybe five, six, maybe 10 labs, right?

385
00:24:51,270 --> 00:24:55,680
Ejaaz:
And I think they're going to be so many different models to use for so many

386
00:24:55,680 --> 00:24:58,050
Ejaaz:
different purposes. I mean, look at Metamuse Spark 1.1.

387
00:24:58,270 --> 00:25:00,780
Ejaaz:
It topped the health benchmark. Did you know that?

388
00:25:01,220 --> 00:25:05,090
Ejaaz:
Like a random social media company's model topped the health benchmark.

389
00:25:05,420 --> 00:25:09,260
Ejaaz:
Grok 4.5, really good at computer use and spinning up agents.

390
00:25:09,930 --> 00:25:13,830
Ejaaz:
Luna from OpenAIR, also really good at computer use. So all these different

391
00:25:13,830 --> 00:25:16,590
Ejaaz:
models would be good for various different tasks.

392
00:25:16,790 --> 00:25:20,350
Ejaaz:
And if you don't want to spend time thinking about which subscription to get

393
00:25:20,670 --> 00:25:24,140
Ejaaz:
or what model to use or when, I think routing companies.

394
00:25:24,280 --> 00:25:28,130
Ejaaz:
So what cursor basically built, which is you can type in a prompt and they decide

395
00:25:28,460 --> 00:25:33,690
Ejaaz:
which parts of the prompt gets used by which kind of models will be the killer platform or product.

396
00:25:33,980 --> 00:25:38,230
Ejaaz:
I don't know of any company that is building this that hasn't been acquired, aka Cursor.

397
00:25:38,450 --> 00:25:42,080
Ejaaz:
Maybe OpenRouter. We've had the founder, Alex Atala, on our show before.

398
00:25:42,350 --> 00:25:46,480
Ejaaz:
Maybe they pivot into some kind of a routing product. But I'm convinced now

399
00:25:46,480 --> 00:25:49,720
Ejaaz:
that it's going to be a layer that sits on top of these models, potentially.

400
00:25:50,290 --> 00:25:53,760
Josh:
Yeah. I think the way I'm looking at it is like, okay, now there's two markets.

401
00:25:53,760 --> 00:25:56,930
Josh:
This used to be a singular commodity, a singular race to the top.

402
00:25:57,220 --> 00:26:00,020
Josh:
Now there's these two things. And like cost per intelligence is the right lens

403
00:26:00,020 --> 00:26:05,000
Josh:
for looking at maybe 80% of the work that's become like kind of routine,

404
00:26:05,270 --> 00:26:09,340
Josh:
but it is the opposite and exactly the wrong lens for like the 20% where

405
00:26:09,710 --> 00:26:12,820
Josh:
capability and reliability and verification matter.

406
00:26:13,070 --> 00:26:15,900
Josh:
It's like if you are doing really complicated work where mistakes cost you a

407
00:26:15,900 --> 00:26:19,980
Josh:
tremendous amount of money and time, if you're doing frontier math or problem

408
00:26:19,980 --> 00:26:22,940
Josh:
solving, or you're building complicated things as a business,

409
00:26:23,200 --> 00:26:26,840
Josh:
there's almost no limit to the amount that you'll spend in order to accomplish your goals.

410
00:26:27,530 --> 00:26:30,720
Josh:
Because I'm sure many of these companies have a tremendous amount of ideas that

411
00:26:30,720 --> 00:26:31,870
Josh:
they want to implement to the market.

412
00:26:32,180 --> 00:26:35,650
Josh:
And the constraint is their workforce and knowledge base that could actually deploy that.

413
00:26:36,020 --> 00:26:39,320
Josh:
There is no limit to the amount that they will spend on these Fable models,

414
00:26:39,320 --> 00:26:42,250
Josh:
on these GPT 5.6 and GPT 6.0 models.

415
00:26:42,510 --> 00:26:45,810
Josh:
But for the rest of the world, maybe that's not the case. They are price sensitive.

416
00:26:45,810 --> 00:26:48,520
Josh:
They don't need the frontier intelligence for everything.

417
00:26:48,740 --> 00:26:51,580
Josh:
And that's where a lot of these models are going to really play a serious role.

418
00:26:51,580 --> 00:26:55,660
Josh:
And i think that is cool that like we no longer have intelligence as the singular

419
00:26:55,660 --> 00:26:59,650
Josh:
commodity it is now split between intelligence and price those two things both

420
00:26:59,650 --> 00:27:02,400
Josh:
matter and i think that's kind of where we are in the market right now it's

421
00:27:02,400 --> 00:27:06,220
Josh:
like we are at this crossroads the commodity has diverged into two

422
00:27:06,510 --> 00:27:10,510
Josh:
and we have some pretty serious players in the pricing game who have a very

423
00:27:10,510 --> 00:27:14,970
Josh:
serious trajectory of actually continuing to be serious threats spacex ai is

424
00:27:14,970 --> 00:27:17,630
Josh:
building a huge amount of data centers they have a very clear trajectory to a

425
00:27:17,890 --> 00:27:23,150
Josh:
20 trillion parameter model meta is going to put 15 gigawatts on the ground power it on

426
00:27:23,420 --> 00:27:26,250
Josh:
and they're going to do that vertically integrated with their own silicon and

427
00:27:26,250 --> 00:27:29,430
Josh:
like that's a pretty serious threat to the pricing war because you have to imagine

428
00:27:29,430 --> 00:27:30,090
Josh:
they're going to get some

429
00:27:30,410 --> 00:27:33,480
Josh:
amazing efficiency from that so yeah i think that's the update for

430
00:27:34,000 --> 00:27:38,200
Josh:
The state of low-cost models. It's been a very interesting week for them.

431
00:27:38,630 --> 00:27:42,300
Ejaaz:
As we're recording this episode, I'm realizing that we need to make another

432
00:27:42,300 --> 00:27:47,880
Ejaaz:
episode on these trends that we've spoken about during this talk.

433
00:27:48,960 --> 00:27:53,720
Ejaaz:
Primary ones being, okay, if the most expensive model lab or the most intelligent

434
00:27:53,720 --> 00:27:57,040
Ejaaz:
model lab isn't necessarily going to win over the next couple of years and it's

435
00:27:57,040 --> 00:28:00,170
Ejaaz:
going to be this middle ground, this cheap middle ground, which companies are

436
00:28:00,170 --> 00:28:01,800
Ejaaz:
there? Where are the bottlenecks?

437
00:28:02,330 --> 00:28:03,600
Ejaaz:
How are people using these things?

438
00:28:03,600 --> 00:28:07,340
Ejaaz:
What will agents look like when they're working on long horizon tasks?

439
00:28:07,570 --> 00:28:11,170
Ejaaz:
And most importantly, which companies, which customers are going to be using

440
00:28:11,170 --> 00:28:14,610
Ejaaz:
this thing? I think, I mean, you let me know, as you as the audience that are

441
00:28:14,610 --> 00:28:18,450
Ejaaz:
listening to this right now, would that be an interesting episode to sort of unpack and dig into?

442
00:28:18,850 --> 00:28:23,420
Ejaaz:
All of that said, I think that cheaper models are here to stay.

443
00:28:23,750 --> 00:28:26,790
Ejaaz:
And I think they're here to stay in a very big way. And I think they're going

444
00:28:26,790 --> 00:28:29,770
Ejaaz:
to come out, thankfully, from some frontier Western labs.

445
00:28:29,770 --> 00:28:34,270
Ejaaz:
We're shifted away from building the most expensive and intelligent model that

446
00:28:34,270 --> 00:28:38,640
Ejaaz:
is being kind of blocked to anyone and everyone to a $1.25 in,

447
00:28:38,640 --> 00:28:43,680
Ejaaz:
$4.25 out model from Meta, which surprisingly works and is intuitive.

448
00:28:43,940 --> 00:28:47,450
Ejaaz:
I'm also excited to see what people do with these cheap models.

449
00:28:47,450 --> 00:28:51,330
Ejaaz:
I don't think they're necessarily going to build the same things that people

450
00:28:51,330 --> 00:28:55,830
Ejaaz:
using Fable 5 are going to do simply because they're not prohibited by cost anymore.

451
00:28:55,830 --> 00:29:00,120
Ejaaz:
And I think that kind of unlocks a new line of creativity when you're building

452
00:29:00,120 --> 00:29:02,710
Ejaaz:
these apps. You're like, oh, you know what? Maybe I will try this crazy thing

453
00:29:02,710 --> 00:29:03,990
Ejaaz:
because it's only going to cost me 10 bucks.

454
00:29:04,280 --> 00:29:06,590
Ejaaz:
Heck, I'll just do it. So if you're listening to this.

455
00:29:07,170 --> 00:29:09,220
Ejaaz:
And you're inspired by some of the things that were spoken about,

456
00:29:09,510 --> 00:29:12,380
Ejaaz:
go and try some of these models. They're super cheap to use.

457
00:29:12,640 --> 00:29:15,920
Ejaaz:
And tell us what you end up building with it. Tell us if there's anything creative

458
00:29:15,920 --> 00:29:17,950
Ejaaz:
that you haven't seen being built by Fable 5.

459
00:29:18,200 --> 00:29:22,160
Ejaaz:
We would love to see it, demo it, heck, maybe even show it on an episode in the future.

460
00:29:22,450 --> 00:29:26,570
Ejaaz:
But if you enjoyed this, please, please, please share this with your friends.

461
00:29:26,570 --> 00:29:28,280
Ejaaz:
If you're not subscribed, please subscribe.

462
00:29:28,440 --> 00:29:31,230
Ejaaz:
If you haven't left us a comment, we always want to hear from you.

463
00:29:31,230 --> 00:29:32,510
Ejaaz:
We love hearing from you guys.

464
00:29:32,770 --> 00:29:35,630
Ejaaz:
And if you're listening to us on Spotify or Apple Music, please,

465
00:29:35,630 --> 00:29:40,340
Ejaaz:
please give us a rating. it helps us out pretty massively josh any any final thoughts.

466
00:29:40,690 --> 00:29:44,490
Josh:
No that's it thank you guys so much for watching um on sponsor front we're chatting

467
00:29:44,490 --> 00:29:46,950
Josh:
with some people this has been going well so for the people who have reached

468
00:29:46,950 --> 00:29:49,310
Josh:
out thank you we are working our way through those we are

469
00:29:49,520 --> 00:29:52,940
Josh:
slowly working towards becoming a self-sufficient entity which has been amazing

470
00:29:52,940 --> 00:29:55,940
Josh:
so thank you so much for the support on that again if you know anyone please

471
00:29:55,940 --> 00:29:56,990
Josh:
refer them send them our way

472
00:29:57,250 --> 00:29:59,360
Josh:
either on x email and description whatever it may be

473
00:29:59,640 --> 00:30:03,270
Josh:
but that is another episode next coming up is the roundup which is i think our

474
00:30:03,270 --> 00:30:05,850
Josh:
favorite episode of the week we just throw everything that we haven't been able

475
00:30:05,850 --> 00:30:10,270
Josh:
to talk about into one and do a full recap on everything that has happened this week which is a lot

476
00:30:10,270 --> 00:30:15,950
Ejaaz:
We got to talk about the data center bands dude like new york is my city why are we banning the day so.

477
00:30:15,950 --> 00:30:20,420
Josh:
We can have a lot of conversation about that one so stay tuned stay tuned for

478
00:30:20,420 --> 00:30:24,590
Josh:
that grieving session um but anyways that is the episode today thank you all

479
00:30:24,590 --> 00:30:27,060
Josh:
so much for watching as always and we will see you in the next one