1
00:00:00,020 --> 00:00:04,070
Ejaaz:
I've been waiting so long to say this. Elon is officially back.

2
00:00:04,070 --> 00:00:08,350
Ejaaz:
The first major AI model released from SpaceX AI since they went public,

3
00:00:08,590 --> 00:00:12,230
Ejaaz:
since the $60 billion cursor acquisition is officially live.

4
00:00:12,230 --> 00:00:15,740
Ejaaz:
It's called Grok 4.5 and it's pretty damn good.

5
00:00:15,860 --> 00:00:21,320
Ejaaz:
It ranks around three after Fable 5 and GPT 5.5, but it's really good at coding

6
00:00:21,320 --> 00:00:26,060
Ejaaz:
and it's astonishingly cheap. It's around half the cost of Opus 4.8 and it's

7
00:00:26,060 --> 00:00:27,930
Ejaaz:
their quickest model yet.

8
00:00:28,630 --> 00:00:32,910
Ejaaz:
Elon only got to enjoy this for about a day or maybe even a couple of hours

9
00:00:33,180 --> 00:00:37,650
Ejaaz:
because GPT 5.6 was officially released from OpenAI. And if you're listening

10
00:00:37,650 --> 00:00:40,530
Ejaaz:
to this, you probably have access to it right now, a general release.

11
00:00:40,720 --> 00:00:43,910
Ejaaz:
It's their best model yet, extremely good energetic reasoning,

12
00:00:43,910 --> 00:00:49,050
Ejaaz:
but not quite fable worthy. But they are training GPT 6, which is going to be

13
00:00:49,050 --> 00:00:52,050
Ejaaz:
released supposedly a month from now. If you're releasing models,

14
00:00:52,050 --> 00:00:54,090
Ejaaz:
you're probably thinking that's amazing for the AI world.

15
00:00:54,200 --> 00:00:59,410
Ejaaz:
But this comes right at a time where our main adversary in China has been releasing

16
00:00:59,410 --> 00:01:01,690
Ejaaz:
so many models that are open source and everyone gets to use it.

17
00:01:01,690 --> 00:01:06,320
Ejaaz:
They're super cheap, but they might be considering cutting off the West from using their models.

18
00:01:06,580 --> 00:01:11,110
Ejaaz:
Meanwhile, Meta, who spent around $25 billion trying to build the best model

19
00:01:11,110 --> 00:01:14,030
Ejaaz:
and have seemingly failed, just released two new models there,

20
00:01:14,030 --> 00:01:17,560
Ejaaz:
Metamuse Image and Metamuse Video Models. We're going to get into all of this

21
00:01:17,790 --> 00:01:19,250
Ejaaz:
and more on today's episode.

22
00:01:19,480 --> 00:01:22,340
Josh:
There's a lot going on. It feels like nothing happens for a week or two,

23
00:01:22,340 --> 00:01:25,620
Josh:
and then five different releases happen all in one day. And that's what just

24
00:01:25,620 --> 00:01:30,100
Josh:
happened. We got new JATGPT. We have a new voice interface, which I love that we'll get into.

25
00:01:30,240 --> 00:01:33,690
Josh:
But I think we do have to start with Grok because I mean, like you said, Elon is back.

26
00:01:33,920 --> 00:01:37,810
Josh:
The SpaceX AI team, which is now officially formalized, they have a logo,

27
00:01:37,810 --> 00:01:40,030
Josh:
they have their official Twitter handle on X.

28
00:01:40,220 --> 00:01:44,190
Josh:
This is their first release in a long time and the results are pretty remarkable.

29
00:01:44,190 --> 00:01:49,430
Josh:
They're kind of classifying this as Opus class on a relative basis compared to Claude Opus 4.8.

30
00:01:49,780 --> 00:01:52,870
Josh:
And it seems to have actually accomplished that on just about all of the benchmarks.

31
00:01:52,870 --> 00:01:57,040
Josh:
In fact, it even exceeds it slightly in some benchmarks for a fraction of the

32
00:01:57,040 --> 00:02:00,150
Josh:
cost. And I think this is one of the more interesting things about this model

33
00:02:00,150 --> 00:02:03,580
Josh:
is just how efficient it is for how much intelligence you get out.

34
00:02:03,840 --> 00:02:07,470
Josh:
And I know that the XAI team or the SpaceX AI team, they've always been really

35
00:02:07,470 --> 00:02:09,280
Josh:
insistent on maximizing the.

36
00:02:09,760 --> 00:02:13,610
Josh:
Amount of intelligence per watt of compute. And it seems like this is the first

37
00:02:13,610 --> 00:02:14,950
Josh:
sign of that actually working.

38
00:02:15,110 --> 00:02:18,580
Josh:
For the same amount of power and electricity, they're able to generate tokens

39
00:02:18,770 --> 00:02:21,410
Josh:
that have more intelligence per token than just about anyone else.

40
00:02:21,540 --> 00:02:25,210
Josh:
Like you mentioned, Ejaz, the new Grok model is being trained.

41
00:02:25,210 --> 00:02:28,450
Josh:
It is currently cooking in the oven. They have all the GPUs humming.

42
00:02:28,700 --> 00:02:32,620
Josh:
And when that comes out, I really believe that they're going to have a shot at the top.

43
00:02:32,850 --> 00:02:38,150
Josh:
And we talked about this way in the past, maybe like two or three months ago, about how the spacex ai

44
00:02:38,490 --> 00:02:43,030
Josh:
opportunities really just in their ability to build out infrastructure quickly

45
00:02:43,340 --> 00:02:47,040
Josh:
and as a result they actually had too much infrastructure for their use cases

46
00:02:47,040 --> 00:02:50,250
Josh:
they were selling it off to amazon they were selling it off to anthropic because

47
00:02:50,250 --> 00:02:52,530
Josh:
they just had access and it created this huge business model for them

48
00:02:52,830 --> 00:02:55,330
Josh:
but now they actually have a use case for this they're starting to take these

49
00:02:55,330 --> 00:02:58,360
Josh:
gpus they're starting to train the models and the results are actually great

50
00:02:58,610 --> 00:03:01,270
Josh:
and it's like okay wow they were they're acquired for a couple months but now

51
00:03:01,270 --> 00:03:03,880
Josh:
here they are right back on the frontier and you have to imagine

52
00:03:04,210 --> 00:03:07,550
Josh:
that the next model that's being trained on however many trillions of parameters

53
00:03:07,550 --> 00:03:10,830
Josh:
is going to be right up there at least very close to the frontier for what i

54
00:03:10,830 --> 00:03:12,850
Josh:
would imagine is a very very competitive price.

55
00:03:13,030 --> 00:03:16,570
Ejaaz:
I think i think the most hilarious part about this is this is probably elon's

56
00:03:16,570 --> 00:03:20,850
Ejaaz:
most humble model launch ever he was uh super up front saying you know this

57
00:03:20,850 --> 00:03:25,740
Ejaaz:
is opus 4.8 class grade model it's not as good as fable but we are training

58
00:03:25,740 --> 00:03:27,550
Ejaaz:
a much larger model speaking of size

59
00:03:27,960 --> 00:03:32,330
Ejaaz:
The size of this model, GROC 4.5, is only around 1.4 trillion parameter models.

60
00:03:32,330 --> 00:03:37,390
Ejaaz:
Now, if you compare that to Fable 5, that's around 10 to 15 trillion parameters in size.

61
00:03:37,570 --> 00:03:41,430
Ejaaz:
So the fact that they were able to achieve number three on some of these coding

62
00:03:41,430 --> 00:03:43,640
Ejaaz:
benchmarks, like the one that we're showing on your screen right now,

63
00:03:43,640 --> 00:03:47,790
Ejaaz:
DeepSwee Bench, is pretty impressive. And you can only imagine what a model,

64
00:03:47,790 --> 00:03:51,510
Ejaaz:
maybe five times the size of this, could probably be quite competitive.

65
00:03:51,810 --> 00:03:56,960
Ejaaz:
And this brings me to the absolute Hail Mary move that Elon and SpaceX AI pulled

66
00:03:56,960 --> 00:04:02,230
Ejaaz:
off a few months ago, or actually less than a month ago, which was the $60 billion acquisition of Cursor.

67
00:04:02,230 --> 00:04:04,850
Ejaaz:
Now, if you're wondering... Did that pay off? Yeah, it paid off massively.

68
00:04:04,850 --> 00:04:10,780
Ejaaz:
Like, just to bring people up to speed here, at the time of their public IPO,

69
00:04:11,100 --> 00:04:15,520
Ejaaz:
It was pretty much a foregone conclusion that SpaceX had fallen behind in the race.

70
00:04:15,700 --> 00:04:18,970
Ejaaz:
They hadn't released a meaningful model. Grok 4.2 kind of sucked.

71
00:04:19,160 --> 00:04:22,430
Ejaaz:
And OpenAI and Anthropic seem to be just running away with it.

72
00:04:22,720 --> 00:04:26,710
Ejaaz:
With 4.5, they have officially caught up and it's because they've been using

73
00:04:27,020 --> 00:04:29,530
Ejaaz:
this moat, this data moat that Cursor has.

74
00:04:29,530 --> 00:04:31,920
Ejaaz:
For those of you who don't know, for those of you who haven't used Cursor,

75
00:04:32,190 --> 00:04:36,770
Ejaaz:
Cursor is this amazing platform that kind of like pulls prompts across many

76
00:04:36,770 --> 00:04:39,710
Ejaaz:
different models and routes it to the right model at the right time to do the

77
00:04:39,710 --> 00:04:42,880
Ejaaz:
right types of tasks. It's also extremely cost efficient.

78
00:04:43,110 --> 00:04:46,700
Ejaaz:
Now that kind of data is really valuable towards training new models.

79
00:04:46,700 --> 00:04:49,360
Ejaaz:
It's not just about how much compute you put into train a model.

80
00:04:49,570 --> 00:04:53,980
Ejaaz:
It is about how you figure out how that model thinks. And Cursor's data mode

81
00:04:53,980 --> 00:04:58,590
Ejaaz:
is really useful for that. So that $60 billion acquisition seems to have paid off pretty damn well.

82
00:04:59,360 --> 00:05:02,450
Ejaaz:
Now, if you're wondering whether I should switch to GROC 4.5,

83
00:05:02,680 --> 00:05:07,870
Ejaaz:
if you're using your LLMs to do Google searches and do basic research, I would say no.

84
00:05:08,100 --> 00:05:11,470
Ejaaz:
But if you are a software engineer, if you're trying to build a new app,

85
00:05:11,610 --> 00:05:16,660
Ejaaz:
I would say absolutely yes, give it a go, because this model was designed to build.

86
00:05:16,660 --> 00:05:20,080
Ejaaz:
Some of the demos, which we'll show shortly, has been pretty awesome.

87
00:05:20,080 --> 00:05:22,390
Ejaaz:
It builds apps from scratch, builds simulators from scratch,

88
00:05:22,590 --> 00:05:24,840
Ejaaz:
and it's super cheap and fast to use.

89
00:05:27,410 --> 00:05:31,950
Ejaaz:
I it's around 17 times cheaper than opus 4.8 puttas it's pretty awesome

90
00:05:32,390 --> 00:05:37,000
Josh:
Yeah i was looking at the the sw bench um benchmarks that they offer and it

91
00:05:37,000 --> 00:05:41,620
Josh:
showed that the average is solved in about 16 000 tokens versus opus which is

92
00:05:41,620 --> 00:05:46,500
Josh:
solved in 67 000 tokens so it's about a 4.2 times gap

93
00:05:46,810 --> 00:05:49,980
Josh:
on efficiency which is like really impressive if you're paying for these tokens

94
00:05:50,230 --> 00:05:53,420
Josh:
and you want something more efficient this is a model at least to try to put

95
00:05:53,420 --> 00:05:55,270
Josh:
it in your toolkit to test it out to see where it walls.

96
00:05:55,530 --> 00:05:59,500
Josh:
And hey, if you don't like this one, there's some better news for you because Elon says that,

97
00:06:00,150 --> 00:06:05,300
Josh:
spacex ai will ship a completely new from scratch foundation model every single

98
00:06:05,300 --> 00:06:07,650
Josh:
month through the end of 2026 which is

99
00:06:08,040 --> 00:06:10,710
Josh:
pretty unbelievable like that's one of those where i'll believe it when i see

100
00:06:10,710 --> 00:06:14,200
Josh:
it because like a foundation model from scratch to build every single month

101
00:06:14,200 --> 00:06:16,380
Josh:
seems like a lot they're gonna have.

102
00:06:16,380 --> 00:06:19,340
Ejaaz:
To be training a lot of these in parallel he has around 500 000 gpus which

103
00:06:19,340 --> 00:06:21,210
Josh:
Is so crazy it is the largest cluster.

104
00:06:21,610 --> 00:06:25,880
Ejaaz:
But he he has enough gpus to feasibly train these models i think it's it might be true yeah

105
00:06:26,110 --> 00:06:29,360
Josh:
Yeah so i mean if you think of it he's got this giant bakery and he's got like

106
00:06:29,360 --> 00:06:32,520
Josh:
a couple different ovens going at once all cooking various size

107
00:06:32,930 --> 00:06:36,360
Josh:
desserts those desserts coming in the form of grok so i'm really excited that

108
00:06:36,360 --> 00:06:39,820
Josh:
they're back i'm excited to see how they evolve over the next few months it

109
00:06:39,820 --> 00:06:41,460
Josh:
seems like now that they've launched the first one

110
00:06:41,700 --> 00:06:44,770
Josh:
the next ones are going to come in pretty rapid iteration and we will have our

111
00:06:44,770 --> 00:06:48,150
Josh:
finger on the pulse but that is not the only model launch in fact that is certainly

112
00:06:48,150 --> 00:06:50,420
Josh:
not the best model launch we had this week because,

113
00:06:50,880 --> 00:06:55,320
Josh:
gpt 5.6 is here finally it was kind of here it was here last week or two weeks

114
00:06:55,320 --> 00:06:58,650
Josh:
ago whenever they were released it but it unfortunately got a little blocker from the government,

115
00:06:59,140 --> 00:07:02,100
Josh:
they've cleared things up with the government they said okay your model is good

116
00:07:02,100 --> 00:07:05,440
Josh:
to go so now it is good to go and if you were listening to this podcast,

117
00:07:05,970 --> 00:07:09,100
Josh:
chances are you can go and use it right now we are a little early to the news

118
00:07:09,100 --> 00:07:11,930
Josh:
it hasn't quite released publicly for the world at the time of recording so

119
00:07:11,930 --> 00:07:13,160
Josh:
we don't have any interesting demos

120
00:07:13,550 --> 00:07:17,000
Josh:
we will probably record a lot of cool demos showcasing all the capabilities

121
00:07:17,000 --> 00:07:20,320
Josh:
model at some point in the near future but for now we have the news that it's

122
00:07:20,320 --> 00:07:25,760
Josh:
out and this is all three they have gpt 5.6 soul terra and luna all broadly

123
00:07:25,760 --> 00:07:28,300
Josh:
available, and they are very much at the frontier.

124
00:07:28,300 --> 00:07:33,730
Josh:
Their costs are, I think, $5 per million input, $30 per million output,

125
00:07:33,730 --> 00:07:38,300
Josh:
so cheaper than Fable, and seems like it's a pretty good model.

126
00:07:38,300 --> 00:07:41,440
Josh:
The results on X seem like people are really enjoying this so far.

127
00:07:42,410 --> 00:07:47,280
Ejaaz:
Yeah, so I was scouring across all the early tested reviews from Theo,

128
00:07:47,280 --> 00:07:48,700
Ejaaz:
from a bunch of other folks.

129
00:07:48,700 --> 00:07:53,040
Ejaaz:
And one thing stood out pretty massively to me, which is OpenAI seems to have

130
00:07:53,040 --> 00:07:56,560
Ejaaz:
solved the toughest problem that they've or the toughest critique that they've

131
00:07:56,560 --> 00:07:58,640
Ejaaz:
been facing, which is taste.

132
00:07:58,900 --> 00:08:01,570
Ejaaz:
So when you use ChatGPT, I don't know if you've used it recently,

133
00:08:01,570 --> 00:08:06,170
Ejaaz:
Josh, it still seems pretty sycophantic. It still sounds like I'm talking to

134
00:08:06,170 --> 00:08:10,430
Ejaaz:
an AI bot versus a human that can actually help me or like a companion that

135
00:08:10,430 --> 00:08:12,200
Ejaaz:
can like help me complete my task.

136
00:08:12,460 --> 00:08:19,960
Ejaaz:
With GPT 5.6, it seems that the feedback is overwhelmingly in favor of taste. It sounds better to use.

137
00:08:20,150 --> 00:08:22,670
Ejaaz:
It's more of a friend to kind of consult with.

138
00:08:22,930 --> 00:08:28,720
Ejaaz:
And that has swayed people to switch from Fable 5 to GPT 5.6.

139
00:08:28,720 --> 00:08:33,030
Ejaaz:
Now, it's important to mention that it isn't really a direct competitor to GPT

140
00:08:33,030 --> 00:08:34,840
Ejaaz:
5.6, according to these reviews.

141
00:08:34,840 --> 00:08:39,910
Ejaaz:
Apparently, it still lacks when it comes to certain complex tasks at coding,

142
00:08:40,130 --> 00:08:44,410
Ejaaz:
when it comes to certain complex tasks at reasoning, but it is much, much better.

143
00:08:44,410 --> 00:08:48,440
Ejaaz:
It's a big jump from GPT 5.5. Now, if you're wondering what the fable competitor

144
00:08:48,440 --> 00:08:51,500
Ejaaz:
from OpenAI is, it's called GPT 6.

145
00:08:51,660 --> 00:08:55,820
Ejaaz:
And according to Greg Brockman, the president of OpenAI, he said that it's aiming

146
00:08:55,820 --> 00:08:58,810
Ejaaz:
to come out in under a month, potentially.

147
00:08:58,810 --> 00:09:04,700
Ejaaz:
So again, we're looking at a new paradigm, a new era right now where the top

148
00:09:04,700 --> 00:09:09,890
Ejaaz:
model labs are basically releasing a new model every single month.

149
00:09:09,890 --> 00:09:11,920
Ejaaz:
Now, if you compare that to what that looked like,

150
00:09:12,470 --> 00:09:16,040
Ejaaz:
I don't know, a year ago, we had to wait months. We had to wait months.

151
00:09:16,700 --> 00:09:19,820
Ejaaz:
There were draughts between the news cycles, if you remember, Josh.

152
00:09:19,820 --> 00:09:22,660
Josh:
I don't know how we did it. It was so long between each model launch.

153
00:09:22,660 --> 00:09:23,800
Josh:
It was literally multiple months.

154
00:09:24,160 --> 00:09:27,840
Ejaaz:
It's so funny because I remember we had a lot of time to test these different

155
00:09:27,840 --> 00:09:31,720
Ejaaz:
models. Remember we attributed a personality to each of these different models?

156
00:09:31,850 --> 00:09:32,520
Josh:
Now they're kind.

157
00:09:32,520 --> 00:09:37,170
Ejaaz:
Of converging across this one uniform kind of face and voice,

158
00:09:37,170 --> 00:09:38,780
Ejaaz:
and they're kind of flitting between each other.

159
00:09:39,060 --> 00:09:43,670
Ejaaz:
I honestly am It is the best time alive if you are a builder,

160
00:09:43,940 --> 00:09:47,280
Ejaaz:
if you are someone that's interested in using these models and testing them

161
00:09:47,280 --> 00:09:50,970
Ejaaz:
out, you now get access to all of this stuff for a $20 a month subscription.

162
00:09:50,970 --> 00:09:53,950
Ejaaz:
So I'm super excited to actually get my hands on this and use it.

163
00:09:53,950 --> 00:09:56,410
Ejaaz:
If you're listening to this right now, I probably already am doing this,

164
00:09:56,610 --> 00:10:00,470
Ejaaz:
but you can expect an episode as soon as next week to show you what you can do with it.

165
00:10:01,010 --> 00:10:03,410
Josh:
And this is also great news for people who aren't developers,

166
00:10:03,410 --> 00:10:05,970
Josh:
who aren't builders, people who just want to use the model.

167
00:10:05,970 --> 00:10:10,390
Josh:
Not only is it more intelligent but also you have a new voice model today as

168
00:10:10,390 --> 00:10:14,090
Josh:
well which is really exciting so the voice model is something that we did have

169
00:10:14,090 --> 00:10:18,040
Josh:
time to play around with i was testing it last night i was chatting with it and it's um

170
00:10:18,470 --> 00:10:21,700
Josh:
pretty good i love the launch video that they have we're showing it can we play

171
00:10:21,700 --> 00:10:24,270
Josh:
a clip yeah yeah please play some audio it's very funny.

172
00:10:24,270 --> 00:10:25,370
Ejaaz:
Let's see let's see

173
00:10:26,730 --> 00:10:33,460
Ejaaz:
Today, we are announcing the all-new chat GPT Voice, powered by GPT Live 1,

174
00:10:33,460 --> 00:10:39,610
Ejaaz:
a full duplex conversational partner that is the most powerful voice model ever built.

175
00:10:39,610 --> 00:10:41,010
Josh:
Can you simplify that?

176
00:10:41,010 --> 00:10:45,280
Ejaaz:
Sure can. Can you explain what a full duplex voice model is?

177
00:10:45,280 --> 00:10:47,580
Ejaaz:
Yeah. Imagine a normal call with a friend.

178
00:10:47,940 --> 00:10:51,620
Ejaaz:
You can listen and talk at the same time. That's full duplex.

179
00:10:52,800 --> 00:10:56,090
Josh:
There you go. And that's the news. It's like you can now listen and talk at

180
00:10:56,090 --> 00:11:00,370
Josh:
the same time and you could talk over this voice in a way that you never were able to before.

181
00:11:00,610 --> 00:11:04,110
Josh:
And I will say this isn't novel technology. GPT launched this kind of as like

182
00:11:04,110 --> 00:11:08,180
Josh:
a research preview for more developers a month or two ago where it's kind of

183
00:11:08,180 --> 00:11:11,790
Josh:
live intelligence. But now it's rolled into the actual application and it's awesome.

184
00:11:12,020 --> 00:11:15,130
Josh:
You can yell at it. You can argue at it. You could have it think about things

185
00:11:15,130 --> 00:11:16,810
Josh:
while you are talking about something else.

186
00:11:17,170 --> 00:11:20,830
Josh:
I was doing a little demo where I was asking it to do some research about something.

187
00:11:21,110 --> 00:11:24,310
Josh:
And then we continued the conversation in parallel while it was researching in the background.

188
00:11:24,630 --> 00:11:27,200
Josh:
And then once it had the correct answer, it came back and it said,

189
00:11:27,200 --> 00:11:29,400
Josh:
hey, by the way, that thing you asked me, here's the answer.

190
00:11:29,400 --> 00:11:32,500
Josh:
And it felt very natural in a way that I've never felt before.

191
00:11:32,870 --> 00:11:35,640
Josh:
And watching this demo and getting the opportunity to try it,

192
00:11:36,580 --> 00:11:38,820
Josh:
I noticed that there were some new visual elements as well.

193
00:11:39,110 --> 00:11:41,100
Josh:
So when I asked like, hey, what's the weather in New York City?

194
00:11:41,100 --> 00:11:43,910
Josh:
I wanted to see how fast I could pull that up. It was close to instant.

195
00:11:44,160 --> 00:11:47,350
Josh:
And then it shows this like really pretty card on screen that gives you just

196
00:11:47,550 --> 00:11:49,910
Josh:
brief overview of all the weather that's going on.

197
00:11:49,910 --> 00:11:54,200
Josh:
And I'm thinking to myself, this is the first version of whatever their new

198
00:11:54,200 --> 00:11:57,070
Josh:
operating system is going to look like for this hardware device.

199
00:11:57,070 --> 00:11:59,590
Josh:
We have all the rumors in the world about what it's going to look like,

200
00:11:59,590 --> 00:12:02,410
Josh:
what it's going to feel like, and we know that voice is going to be at the core of it.

201
00:12:02,770 --> 00:12:06,030
Josh:
So getting an opportunity to interface with a new voice, I could very clearly

202
00:12:06,030 --> 00:12:09,980
Josh:
see this working well with an AirPods-like device. It's like when I'm walking down the street,

203
00:12:10,690 --> 00:12:12,790
Josh:
You could chat with it very naturally. It talks back to you.

204
00:12:12,790 --> 00:12:16,220
Josh:
It has the context of everything around you. And it's really exciting to try.

205
00:12:16,220 --> 00:12:18,840
Josh:
So I would encourage everyone who has a ChatGPT account, give it a shot.

206
00:12:18,840 --> 00:12:20,160
Josh:
It's like pretty cool technology.

207
00:12:21,120 --> 00:12:24,980
Ejaaz:
This might be a hot take, but I think this is OpenAI's biggest release so far

208
00:12:24,980 --> 00:12:29,820
Ejaaz:
of this year. And it's because I think voice is just the way people are going

209
00:12:29,820 --> 00:12:34,260
Ejaaz:
to communicate with AI going forwards. I can't type as quick as I can think.

210
00:12:34,560 --> 00:12:38,060
Ejaaz:
And in fact, like, I can't remember the last time I typed towards my LLM.

211
00:12:38,440 --> 00:12:43,210
Ejaaz:
I use whisper AI or I use the microphone and I just kind of like speak at Claude

212
00:12:43,210 --> 00:12:46,830
Ejaaz:
or I speak at GPT and it comes back with very useful insights.

213
00:12:46,830 --> 00:12:48,700
Ejaaz:
It's a much more natural way of interacting.

214
00:12:49,060 --> 00:12:50,930
Ejaaz:
Now, if you listen to this and you're thinking,

215
00:12:51,450 --> 00:12:55,590
Ejaaz:
Why is this significant? Like I've been using voice the entire time.

216
00:12:55,880 --> 00:13:01,360
Ejaaz:
Well, I hate to inform you, but you might have been using a lesser lower grade model.

217
00:13:01,670 --> 00:13:06,040
Ejaaz:
Voice typically works on older models, not the cutting edge models.

218
00:13:06,040 --> 00:13:09,240
Ejaaz:
It takes a while to kind of like catch up and re-implement with like the GPT

219
00:13:09,240 --> 00:13:14,020
Ejaaz:
5.5, GPT 5.6. For example, I doubt this is supported by 5.6.

220
00:13:14,260 --> 00:13:18,610
Ejaaz:
So you're already accessing a lower level of intelligence. With this new voice

221
00:13:18,610 --> 00:13:22,240
Ejaaz:
upgrade, you now get access to the higher set of things.

222
00:13:22,930 --> 00:13:28,880
Ejaaz:
It also finally implements a really useful feature that I enjoyed from Thinking

223
00:13:28,880 --> 00:13:33,000
Ejaaz:
Machine Labs. So Mira Murati, who was the former CTO of OpenAI,

224
00:13:33,000 --> 00:13:34,830
Ejaaz:
left to set up her own AI lab.

225
00:13:35,250 --> 00:13:39,850
Ejaaz:
She created this model called the interaction model under Thinking Machine Labs.

226
00:13:40,170 --> 00:13:43,950
Ejaaz:
And it does a lot of what is shown on this video right now, where you can kind of like interrupt it.

227
00:13:44,220 --> 00:13:47,370
Ejaaz:
You can kind of like talk to yourself and it knows it's not talking to you.

228
00:13:47,370 --> 00:13:49,810
Ejaaz:
Very natural human type interactions.

229
00:13:50,180 --> 00:13:53,490
Ejaaz:
We've finally got this feature in OpenAI. So now I'm thinking about her lab

230
00:13:53,490 --> 00:13:55,120
Ejaaz:
and what she's built and what her mode is.

231
00:13:55,120 --> 00:13:58,500
Ejaaz:
And I'm thinking, what's the point of using that when I can just do this on OpenAI?

232
00:13:58,500 --> 00:14:04,100
Ejaaz:
And again, it's this trend repeated constantly of these frontier AI labs that

233
00:14:04,100 --> 00:14:07,770
Ejaaz:
are basically looking over the fence, seeing what competitors are doing and

234
00:14:07,770 --> 00:14:09,000
Ejaaz:
saying, you know what, that's pretty cool.

235
00:14:09,000 --> 00:14:13,660
Ejaaz:
I'll get AI to build this in a couple of weeks and launch it ourselves. Well done from OpenAI.

236
00:14:14,000 --> 00:14:16,590
Josh:
Yeah, I like the point of it being the most important release it very much feels

237
00:14:16,590 --> 00:14:20,650
Josh:
like this is the critical release like this is the future way that we are going to interface with ai

238
00:14:21,040 --> 00:14:23,920
Josh:
and for them to really nail this and to make it a great product in the mobile

239
00:14:23,920 --> 00:14:27,040
Josh:
app is step one before it gets moved everywhere else and you could i wonder

240
00:14:27,040 --> 00:14:30,000
Josh:
i have a feeling that's probably why they used older people these are all grandmas they

241
00:14:30,380 --> 00:14:34,130
Josh:
three grandmas it's like okay as for everyone of all ages anyone could speak

242
00:14:34,130 --> 00:14:37,020
Josh:
anyone could have a conversation now you have an ai to converse with.

243
00:14:37,340 --> 00:14:42,200
Ejaaz:
My fiance's grandma uses this every day and I'm not joking because I looked

244
00:14:42,200 --> 00:14:47,970
Ejaaz:
at her tracking usage, three to four hours on average per day. She talks to it.

245
00:14:49,290 --> 00:14:53,780
Ejaaz:
As a friend, as a companion, as a consultant, as a gossip artist,

246
00:14:53,780 --> 00:14:57,010
Ejaaz:
when she's gossiping with her friends, it's insane.

247
00:14:57,290 --> 00:14:59,680
Josh:
That's unbelievable. People use this. That's high signal too,

248
00:14:59,680 --> 00:15:03,520
Josh:
coming from an older audience, because I don't think anyone's really tapped into that just yet.

249
00:15:03,520 --> 00:15:07,700
Josh:
So this is definitely a place to follow along closely. And my God,

250
00:15:07,700 --> 00:15:11,660
Josh:
we are halfway there to the end of the year when we are going to be getting

251
00:15:11,660 --> 00:15:13,480
Josh:
a preview of whatever this new hardware device is.

252
00:15:13,710 --> 00:15:17,410
Josh:
I am certain that it will be running some form of this voice mode.

253
00:15:17,750 --> 00:15:23,500
Josh:
Next up on the docket, Chinese AI labs are building their own chips while buying NVIDIAs.

254
00:15:23,800 --> 00:15:27,390
Josh:
Ejaz, as our resident Chinese expert, what the hell's going on?

255
00:15:29,180 --> 00:15:34,860
Ejaaz:
Hi, reporting from Beijing over here. Okay, so a narrative that we've consistently

256
00:15:34,860 --> 00:15:38,340
Ejaaz:
talked about on this show is China versus the US.

257
00:15:38,340 --> 00:15:42,320
Ejaaz:
If you are a new listener, China versus the US is probably the biggest air battle

258
00:15:42,320 --> 00:15:43,310
Ejaaz:
that's happening out there.

259
00:15:43,310 --> 00:15:48,470
Ejaaz:
And the biggest kind of question is, can China catch up to America's frontier

260
00:15:48,470 --> 00:15:51,540
Ejaaz:
of AI models? So far, they've been able to keep up pretty well,

261
00:15:51,790 --> 00:15:55,890
Ejaaz:
but they've lacked in one main thing, which is their chip architecture.

262
00:15:56,180 --> 00:16:00,370
Ejaaz:
They've had to buy and source and rely on NVIDIA chips.

263
00:16:00,810 --> 00:16:05,250
Ejaaz:
Until maybe the last couple of months where they announced that companies like

264
00:16:05,250 --> 00:16:09,840
Ejaaz:
Huawei are building their own AI chips and the government has mandated all Chinese

265
00:16:09,840 --> 00:16:12,080
Ejaaz:
AI labs to use Chinese chips.

266
00:16:12,400 --> 00:16:15,690
Ejaaz:
Okay, maybe there needs to be some kind of political resolve there.

267
00:16:15,980 --> 00:16:20,830
Ejaaz:
But now we got announcements for not one, but two AI labs, including the major

268
00:16:20,830 --> 00:16:25,010
Ejaaz:
AI lab from China, DeepSeek, that they're developing their own AI chip,

269
00:16:25,010 --> 00:16:26,760
Ejaaz:
according to a bunch of people that are familiar

270
00:16:27,050 --> 00:16:32,020
Ejaaz:
with the matter, in order to reduce its reliance on NVIDIA and Huawei chips alone.

271
00:16:32,020 --> 00:16:37,680
Ejaaz:
So even just relying less on Chinese manufacturers and focusing on their own AI chips.

272
00:16:37,960 --> 00:16:41,520
Ejaaz:
Now, if I had to guess why an AI lab would do this, it would be twofold.

273
00:16:41,520 --> 00:16:45,070
Ejaaz:
One, you don't want to rely on export controls and restrictions.

274
00:16:45,270 --> 00:16:48,450
Ejaaz:
You know, Trump or any other future US president might say, I don't want to

275
00:16:48,450 --> 00:16:50,790
Ejaaz:
sell you any more GPUs and your model is dead in the water.

276
00:16:51,130 --> 00:16:55,330
Ejaaz:
But number two, the most important one is, I think these AI labs are trying

277
00:16:55,330 --> 00:17:00,010
Ejaaz:
to create a vertically integrated stack. What I mean by that is it's not only

278
00:17:00,010 --> 00:17:02,680
Ejaaz:
good enough to create your own intelligent model right now.

279
00:17:02,900 --> 00:17:06,190
Ejaaz:
You kind of want to own the hardware and infrastructure that it runs on because

280
00:17:06,190 --> 00:17:09,050
Ejaaz:
then you can optimize it specifically for your model.

281
00:17:09,250 --> 00:17:14,090
Ejaaz:
So it could be cheaper, it could work much faster, and it could arguably be more intelligent.

282
00:17:14,370 --> 00:17:16,770
Ejaaz:
If you don't believe me, look at what OpenAI is doing. Well,

283
00:17:16,770 --> 00:17:18,960
Ejaaz:
look at what Meta is doing. They're working on their own chips.

284
00:17:18,960 --> 00:17:21,970
Ejaaz:
OpenAI is releasing their new jalapeno chip nine months from now.

285
00:17:22,160 --> 00:17:25,650
Ejaaz:
So if they end up pulling this off, I think we see a world, honestly,

286
00:17:25,650 --> 00:17:29,670
Ejaaz:
a year from now, where a lot of these labs create their own chips and they run

287
00:17:29,670 --> 00:17:31,950
Ejaaz:
it on their own infrastructure. Less reliance on NVIDIA.

288
00:17:32,550 --> 00:17:36,360
Josh:
I found it really interesting how they've been so quiet about which chips or

289
00:17:36,360 --> 00:17:37,670
Josh:
if they're even making chips at all.

290
00:17:37,950 --> 00:17:40,410
Josh:
And then within 24 hours or within even hours of each other,

291
00:17:40,770 --> 00:17:47,290
Josh:
we got the leaks from two places, both DeepSeek and Jipu, both building their own custom chips.

292
00:17:47,500 --> 00:17:50,310
Josh:
And I think it's this interesting coordinated effort that we're beginning to

293
00:17:50,310 --> 00:17:56,180
Josh:
see with this rotation, perhaps from open source to more closed source as they

294
00:17:56,180 --> 00:17:57,760
Josh:
feel like they're getting more powerful.

295
00:17:58,170 --> 00:18:03,800
Josh:
So as they get the capability to build these custom asic chips these custom inference chips and

296
00:18:04,160 --> 00:18:07,590
Josh:
i'll note that these aren't really pre-training chips these are custom built

297
00:18:07,880 --> 00:18:11,150
Josh:
accelerator chips similar to that episode that we filmed with etched about etched

298
00:18:11,150 --> 00:18:14,940
Josh:
uh the other day where these are very purpose-built chips built for optimizing

299
00:18:14,940 --> 00:18:18,990
Josh:
a specific stack and it's very clear that china is now in this race and they're trying to

300
00:18:19,310 --> 00:18:23,120
Josh:
get themselves as close to the frontier as possible and in a way they're basically

301
00:18:23,120 --> 00:18:27,110
Josh:
now just mirroring the u.s one-to-one it's like open ai has jalapeno.

302
00:18:27,720 --> 00:18:30,430
Josh:
Now deep seek is going to build their own custom asic chip and they're doing

303
00:18:30,430 --> 00:18:33,800
Josh:
it the same place they're probably using some of the same suppliers i have to

304
00:18:33,800 --> 00:18:37,860
Josh:
imagine at least the same memory supplier so they are now at war in terms of

305
00:18:37,860 --> 00:18:42,200
Josh:
supply chain with the united states so now traditionally the u.s versus china has been

306
00:18:42,580 --> 00:18:46,560
Josh:
fought on a software front this is the first instances in which there's probably some

307
00:18:46,960 --> 00:18:50,150
Josh:
going to be some serious competition on the hardware front at least as it relates

308
00:18:50,150 --> 00:18:53,790
Josh:
to supply chain and manufacturing these things because the capacity to do this

309
00:18:54,120 --> 00:18:56,640
Josh:
is so limited so really interesting developments coming out of china and we'll

310
00:18:56,640 --> 00:19:00,500
Josh:
see what they can build with these now we know the process is very long to go from

311
00:19:00,800 --> 00:19:05,340
Josh:
chip architecture to taping out a chip to building a chip and actually producing

312
00:19:05,340 --> 00:19:07,050
Josh:
it at scale then plugging it in

313
00:19:07,340 --> 00:19:11,010
Josh:
and then training a model and then publishing a model so that takes a very long

314
00:19:11,010 --> 00:19:14,200
Josh:
time it's going to be a while until we get the first models that have actually

315
00:19:14,200 --> 00:19:17,520
Josh:
been trained on these chips but the seeds are being planted now for pretty,

316
00:19:18,220 --> 00:19:20,850
Josh:
i would say impressive future from china if they continue on this path.

317
00:19:21,080 --> 00:19:24,950
Ejaaz:
I i personally think it's a it's a bigger problem than people are making out

318
00:19:24,950 --> 00:19:28,530
Ejaaz:
like i remember two years ago when people were like oh china would never catch

319
00:19:28,530 --> 00:19:31,450
Ejaaz:
up and back then it was true they were like three years behind

320
00:19:31,840 --> 00:19:35,440
Ejaaz:
now that gap is only really six months and if you can project into the future

321
00:19:35,440 --> 00:19:38,730
Ejaaz:
let's say that they create a chip that's maybe not nvidia level but it's like

322
00:19:38,730 --> 00:19:40,680
Ejaaz:
80 to 90 percent nvidia level

323
00:19:41,130 --> 00:19:45,860
Ejaaz:
you could probably train a fable 5 or mythos 5 grade model from that and the

324
00:19:45,860 --> 00:19:50,610
Ejaaz:
chinese His Chibu founder has literally said, we'll have this ready in a few

325
00:19:50,610 --> 00:19:52,600
Ejaaz:
months, definitely before the end of the year.

326
00:19:52,880 --> 00:19:58,260
Ejaaz:
Now, remember, China isn't the US. There may not be any strict restrictions

327
00:19:58,260 --> 00:19:59,980
Ejaaz:
on this. They may use it nefariously.

328
00:20:00,340 --> 00:20:04,970
Ejaaz:
They may use it maliciously. And these models are pretty capable of some dangerous stuff.

329
00:20:04,970 --> 00:20:09,650
Ejaaz:
So I think one of the biggest missteps that we've made, at least on the Western

330
00:20:09,650 --> 00:20:13,550
Ejaaz:
side of things, is imposing these export controls because it's meant that they

331
00:20:13,550 --> 00:20:16,060
Ejaaz:
have been forced to create their own chip architecture. and

332
00:20:16,720 --> 00:20:19,740
Ejaaz:
that's what China relied on a lot. I think if they, if we could have just sold

333
00:20:19,740 --> 00:20:24,150
Ejaaz:
them a few more NVIDIA GPUs, we would have made them more reliant on the architecture,

334
00:20:24,150 --> 00:20:26,310
Ejaaz:
but now it's going to become a heavily siloed thing.

335
00:20:26,310 --> 00:20:28,900
Ejaaz:
And I think, like you said, we're going to lose access to these open source

336
00:20:28,900 --> 00:20:32,630
Ejaaz:
models. I bet you they're going to close source a bunch of the future meter-square models.

337
00:20:32,950 --> 00:20:37,170
Josh:
Yeah. Well, I have another company for you that used to be open source and they've gone close source.

338
00:20:37,610 --> 00:20:41,040
Josh:
So maybe we have a peek into the future because that company comes in the form of meta.

339
00:20:41,320 --> 00:20:45,040
Josh:
Remember them when they were so hell bent on creating open source models that were going to

340
00:20:45,270 --> 00:20:48,430
Josh:
change the world be accessible to everyone well it turns out that didn't happen

341
00:20:48,640 --> 00:20:52,110
Josh:
uh the same way that the metaverse didn't happen but we still call them meta

342
00:20:52,110 --> 00:20:55,310
Josh:
sorry i i feel bad for roasting them but oh my god it's so easy to every single

343
00:20:55,310 --> 00:20:59,980
Josh:
time but anyways meta is back in the news because they have done something interesting and novel in fact

344
00:21:00,410 --> 00:21:03,870
Josh:
um well i will say i didn't actually realize this was the news ej has filmed

345
00:21:03,870 --> 00:21:06,850
Josh:
me right before we recorded i was like oh wow they did release something this

346
00:21:06,850 --> 00:21:12,120
Josh:
week and that comes in the form of muse image and muse spark you'll remember the Muse models.

347
00:21:12,120 --> 00:21:15,320
Josh:
Oh, I feel so bad roasting them. But like the last experience we had with Muse

348
00:21:15,320 --> 00:21:17,530
Josh:
models, that was when everyone's Instagram accounts got hacked.

349
00:21:17,530 --> 00:21:22,230
Josh:
So they're back at it again with a new model. This one now focused on visuals and videos.

350
00:21:22,560 --> 00:21:26,360
Josh:
What is different about this model than the previous one? Because I remember

351
00:21:26,360 --> 00:21:31,020
Josh:
we had that like AI slop TikTok feed thing that kind of came and went from meta.

352
00:21:31,300 --> 00:21:33,660
Josh:
Is this kind of the next iteration on that model?

353
00:21:34,770 --> 00:21:36,760
Ejaaz:
It's not the next iteration, but it's the...

354
00:21:37,360 --> 00:21:41,330
Ejaaz:
It's the power infrastructure. It's the thing that will power all those future

355
00:21:41,330 --> 00:21:42,680
Ejaaz:
MetaBibes apps and stuff.

356
00:21:42,910 --> 00:21:48,150
Ejaaz:
So the quick TLDR is, this is their new image and video model,

357
00:21:48,150 --> 00:21:53,150
Ejaaz:
and it is a lot, lot better than their previous image and video models.

358
00:21:53,150 --> 00:21:56,780
Ejaaz:
Now, the bar admittedly was set pretty low back then.

359
00:21:57,500 --> 00:22:00,280
Ejaaz:
I would say this is where Grok Imagine currently is.

360
00:22:00,280 --> 00:22:04,100
Ejaaz:
When I look at this, when I look at Grok Imagine, they're relatively the kind

361
00:22:04,100 --> 00:22:07,880
Ejaaz:
of same grade of quality, which is impressive for Meta because this is a big

362
00:22:07,880 --> 00:22:13,650
Ejaaz:
jump from their previous model. Now, why I like this model specifically is how it works.

363
00:22:13,890 --> 00:22:18,870
Ejaaz:
There's been this recent trend across any LLM in OpenAI, Anthropic,

364
00:22:19,480 --> 00:22:24,550
Ejaaz:
with the recent CROC 4.5 model, which is called agentic reasoning or agentic thinking.

365
00:22:24,550 --> 00:22:29,950
Ejaaz:
What that means is they allow the model to, one, access more tools to do its

366
00:22:29,950 --> 00:22:33,830
Ejaaz:
thinking, and two, to think more. They give it more tokens. Meta has one of

367
00:22:33,830 --> 00:22:35,450
Ejaaz:
the largest data centers out there.

368
00:22:35,680 --> 00:22:39,830
Ejaaz:
So that's what they did with this model. So it's not necessarily an intelligent model,

369
00:22:40,340 --> 00:22:44,840
Ejaaz:
thinks more. It uses web search to look at potential images that might make

370
00:22:44,840 --> 00:22:46,470
Ejaaz:
sense for the request that you have.

371
00:22:46,700 --> 00:22:50,520
Ejaaz:
It then gets inspired from those images and it generates its own version and

372
00:22:50,520 --> 00:22:53,790
Ejaaz:
then it looks at it and it thinks, hmm, will Josh like this?

373
00:22:54,080 --> 00:22:57,930
Ejaaz:
If it's not good enough to Josh's expectations, I'm going to either edit this

374
00:22:57,930 --> 00:23:01,710
Ejaaz:
using my editing tools or I'm going to completely regenerate this from scratch.

375
00:23:01,930 --> 00:23:06,220
Ejaaz:
So you may have to wait a little bit longer, but you end up getting a much better output. And guess what?

376
00:23:06,500 --> 00:23:10,760
Ejaaz:
It's cheaper because the tokens that they serve from their own data centers are cheaper to serve.

377
00:23:10,760 --> 00:23:15,280
Ejaaz:
So all around a good engineering effort from Meta. I don't think they're anywhere

378
00:23:15,280 --> 00:23:18,830
Ejaaz:
near Fable 5 LLM worthy, but I actually don't think it matters.

379
00:23:18,830 --> 00:23:21,940
Ejaaz:
I think their main focus should be on niche things like this to do what they're

380
00:23:21,940 --> 00:23:23,170
Ejaaz:
good at, which is social media.

381
00:23:23,530 --> 00:23:27,810
Josh:
Yeah. I mean, it seems like it looks pretty good. This is certainly far from the frontier.

382
00:23:28,050 --> 00:23:30,090
Josh:
And I'm starting to realize now through all of these releases,

383
00:23:30,090 --> 00:23:33,120
Josh:
how difficult it is to compete at the frontier.

384
00:23:33,120 --> 00:23:36,200
Josh:
It's just like, I mean, I think of Google, I think of Meta, I think of,

385
00:23:36,650 --> 00:23:40,780
Josh:
spacex ai these are all incredible companies and they are putting all of their

386
00:23:40,780 --> 00:23:42,430
Josh:
weight behind getting to this frontier

387
00:23:42,750 --> 00:23:45,510
Josh:
and yet they're still having a difficult time of actually reaching it at the

388
00:23:45,510 --> 00:23:48,530
Josh:
same rate that some other companies like anthropic and open ai are doing that

389
00:23:48,530 --> 00:23:51,890
Josh:
said i mean this is a good model for what it is i love the idea that it has

390
00:23:51,890 --> 00:23:56,680
Josh:
reasoning i think with open ai and their most recent image model that they they published the,

391
00:23:57,140 --> 00:24:00,640
Josh:
novel and really exciting thing was that it had reasoning abilities baked in

392
00:24:00,960 --> 00:24:03,730
Josh:
it didn't just take your text prompt at face value and generate an image.

393
00:24:04,000 --> 00:24:05,090
Josh:
It actually thought about it.

394
00:24:05,390 --> 00:24:08,410
Josh:
It works through the physics. It had this chain of thought that you kind of

395
00:24:08,410 --> 00:24:12,680
Josh:
trace to get to a final image. And therefore the outputs were significantly better.

396
00:24:12,940 --> 00:24:16,470
Josh:
This is another version of that where it's very smart. So I imagine we get some

397
00:24:16,470 --> 00:24:19,000
Josh:
fun kind of like mobile based...

398
00:24:19,580 --> 00:24:23,420
Josh:
Optimized features with this where it can recognize details and images really

399
00:24:23,420 --> 00:24:25,890
Josh:
well it can perhaps like if you have

400
00:24:26,190 --> 00:24:28,990
Josh:
for example on this image there is something sitting on the table you could

401
00:24:28,990 --> 00:24:32,010
Josh:
swap out that element in particular i'm sure there's a lot of customization

402
00:24:32,010 --> 00:24:35,400
Josh:
being built around this reasoning ability to understand what it's looking at

403
00:24:35,400 --> 00:24:37,070
Josh:
and kind of reason through the images that you want

404
00:24:37,440 --> 00:24:40,990
Josh:
and yeah overall pretty cool release um are we going to be using it is this

405
00:24:40,990 --> 00:24:44,200
Josh:
something in the daily driver toolkit maybe thumbnail.

406
00:24:44,200 --> 00:24:45,740
Ejaaz:
Creation no no why

407
00:24:45,740 --> 00:24:49,610
Josh:
Would you do that when you got Chachi Petit image gen baby that's true or nano banana,

408
00:24:50,230 --> 00:24:54,240
Josh:
like these are both just significantly better so there's no real use case for

409
00:24:54,240 --> 00:24:56,360
Josh:
anyone who wants to generate images

410
00:24:56,730 --> 00:25:00,580
Josh:
if you just have a $20 a month subscription anywhere else but it is cool to

411
00:25:00,580 --> 00:25:03,770
Josh:
see more progress coming out of meta I mean we have brain wave reading and now

412
00:25:03,770 --> 00:25:07,310
Josh:
we have image generating this week from them so they are pushing the needle forward.

413
00:25:07,430 --> 00:25:11,520
Ejaaz:
We have two lost items on the docket and then we're going to round this episode

414
00:25:11,520 --> 00:25:15,230
Ejaaz:
out since we're on the topic of image gen I hate to break it to you,

415
00:25:15,230 --> 00:25:19,260
Ejaaz:
but there might be a replacement for GPT Image 2 coming out soon.

416
00:25:19,260 --> 00:25:23,740
Ejaaz:
CDream from ByteDance, those are the owners and creators of TikTok,

417
00:25:23,740 --> 00:25:27,560
Ejaaz:
if you didn't know, has come out with their first image model that competes

418
00:25:27,560 --> 00:25:29,320
Ejaaz:
directly with GPT Image 2.

419
00:25:29,540 --> 00:25:32,910
Ejaaz:
And honestly, if this demo is anything to go by, it is incredibly impressive,

420
00:25:32,910 --> 00:25:37,290
Ejaaz:
but not because of the quality necessarily, but one for the cost.

421
00:25:37,290 --> 00:25:41,580
Ejaaz:
It's super cheap to create and also really easy and intuitive to edit.

422
00:25:41,580 --> 00:25:44,630
Ejaaz:
I mean, look at some of these things. Like I could create websites de novo from

423
00:25:44,630 --> 00:25:47,530
Ejaaz:
scratch using some of these things. I could use different languages,

424
00:25:47,530 --> 00:25:50,060
Ejaaz:
whatever I want. I could swap in animals for other types of animals.

425
00:25:50,060 --> 00:25:51,520
Ejaaz:
I can go for a certain palette color.

426
00:25:51,830 --> 00:25:54,230
Ejaaz:
It just seems like a creator or

427
00:25:54,230 --> 00:25:58,600
Ejaaz:
designer's dream. Definitely one worth experimenting with and one that I'm

428
00:25:59,200 --> 00:26:00,480
Ejaaz:
honestly excited to get my hands on

429
00:26:00,970 --> 00:26:05,050
Josh:
Yeah this this one is really good i think anything that by dance releases has

430
00:26:05,050 --> 00:26:08,200
Josh:
been like pretty impressive and this is no shortage the most interesting thing

431
00:26:08,200 --> 00:26:11,670
Josh:
like you said is the ability to edit things quickly and to kind of workshop things to your

432
00:26:12,030 --> 00:26:15,420
Josh:
own ability you can separate elements almost as if it's photoshop so if you're

433
00:26:15,420 --> 00:26:18,730
Josh:
a designer who's been incredibly frustrated because you can't edit specific

434
00:26:18,730 --> 00:26:23,000
Josh:
features this is the model for you you can now basically select any part of

435
00:26:23,000 --> 00:26:25,360
Josh:
the image you want update it change it to however you'd like

436
00:26:25,660 --> 00:26:27,800
Josh:
really powerful for customization so

437
00:26:28,140 --> 00:26:32,370
Josh:
i'd say this probably is less for the average person more for the pro or the

438
00:26:32,370 --> 00:26:35,720
Josh:
designer who wants to get a little bit more out of the image who wants a very specific output,

439
00:26:36,320 --> 00:26:37,230
Josh:
very impressive overall.

440
00:26:37,480 --> 00:26:41,030
Ejaaz:
And the final item, Cloudflare, which is not really a company that you hear

441
00:26:41,030 --> 00:26:45,550
Ejaaz:
often because it's in the back end of the internet, has opened up a new monetization

442
00:26:45,550 --> 00:26:47,160
Ejaaz:
gateway, a new payments API.

443
00:26:47,160 --> 00:26:51,530
Ejaaz:
Now, if you operate any kind of website, it's likely that Cloudflare is the

444
00:26:51,530 --> 00:26:53,960
Ejaaz:
guys behind it that is making sure there's no bots attacking you,

445
00:26:53,960 --> 00:26:58,510
Ejaaz:
that there's no spam accounts, et cetera, et cetera, so they can govern visits of a website.

446
00:26:58,510 --> 00:27:01,780
Ejaaz:
The reason why this is important is, let's say you create a cool bit of content.

447
00:27:01,780 --> 00:27:04,340
Ejaaz:
Let's say you create an awesome video on limitless that, you know,

448
00:27:04,340 --> 00:27:09,340
Ejaaz:
might be for paid subscribers in the future, you can now have an AI agent troll

449
00:27:09,340 --> 00:27:11,780
Ejaaz:
this site and pay a little micropayment.

450
00:27:12,030 --> 00:27:14,910
Ejaaz:
Typically, websites in the internet aren't built for this.

451
00:27:14,910 --> 00:27:19,190
Ejaaz:
But with protocols like x402 from our friends over on the Web3 side of the world,

452
00:27:19,770 --> 00:27:21,120
Ejaaz:
it enables micropayments.

453
00:27:21,120 --> 00:27:23,730
Ejaaz:
So you can pay a couple of cents to get access to an article instead of having

454
00:27:23,730 --> 00:27:27,030
Ejaaz:
to pay like a $100 per year subscription and stuff like that.

455
00:27:27,030 --> 00:27:30,040
Ejaaz:
So the reason why I wanted to point this out is I thought it was cool.

456
00:27:30,040 --> 00:27:34,590
Ejaaz:
I think it probably marks the open of a new type of economy which is agents

457
00:27:34,590 --> 00:27:37,690
Ejaaz:
paying for stuff without you even knowing about it and having the means to do

458
00:27:37,690 --> 00:27:39,470
Ejaaz:
so just thought it was worthy of an update

459
00:27:40,040 --> 00:27:43,440
Josh:
Yeah i found it interesting how cloudflare uh they say 52 percent of crawler

460
00:27:43,440 --> 00:27:46,710
Josh:
requests are now for ai training which is funny so half of the crawlers in the

461
00:27:46,710 --> 00:27:51,430
Josh:
world are used just for training better models which is up from 22 percent in spring of 2025,

462
00:27:52,330 --> 00:27:57,630
Josh:
um and now ai crawlers request content 100 to 10 000 times more than they refer

463
00:27:57,630 --> 00:27:58,740
Josh:
traffic back so this is kind of

464
00:27:58,740 --> 00:28:03,290
Josh:
built around that thesis that idea it's interesting it's cool i think um

465
00:28:03,620 --> 00:28:06,650
Josh:
it's a testament to where things are to come we just had the cloud fair publication

466
00:28:06,830 --> 00:28:10,130
Josh:
a few weeks ago that said the internet's now dead uh over half of the traffic

467
00:28:10,130 --> 00:28:13,170
Josh:
is bought so this is a continuation of that okay well if half the

468
00:28:13,550 --> 00:28:17,040
Josh:
internet traffic is bought let's build a payments gateway for them and that's

469
00:28:17,040 --> 00:28:21,700
Josh:
exactly what this release is so just a continuing of the trend on this topic and with that.

470
00:28:22,300 --> 00:28:27,150
Josh:
That is all for this week. You are fully caught up in the know with all things

471
00:28:27,150 --> 00:28:29,730
Josh:
AI, frontier technology. This is our fourth episode of the week.

472
00:28:29,730 --> 00:28:32,230
Josh:
If you missed any previous ones, go back and check them out.

473
00:28:32,230 --> 00:28:34,440
Josh:
We had one about this company that we love, Etched.

474
00:28:34,680 --> 00:28:36,950
Josh:
We had the Fable 5 Return. We had,

475
00:28:37,560 --> 00:28:41,920
Josh:
ai finding all the trades that we missed over the last 200 episodes of recording

476
00:28:41,920 --> 00:28:44,620
Josh:
which was a very fun episode so there's some good stuff for you to dig into

477
00:28:44,620 --> 00:28:46,900
Josh:
this week if you haven't caught up if you have caught up

478
00:28:47,170 --> 00:28:50,370
Josh:
go and enjoy the weekend touch grass it is an amazing weekend at least in new

479
00:28:50,370 --> 00:28:53,740
Josh:
york it's looking sunny and beautiful i hope wherever you are the same is true

480
00:28:54,080 --> 00:28:57,040
Josh:
and yeah if you enjoyed this episode don't forget to share it with a friend

481
00:28:57,160 --> 00:28:59,260
Josh:
don't forget to rate us on your favorite podcast platform,

482
00:28:59,850 --> 00:29:02,260
Josh:
he does any final parting thoughts before we head out.

483
00:29:02,700 --> 00:29:05,940
Ejaaz:
I think the final thing to say is if you love our episodes, if you enjoy our

484
00:29:05,940 --> 00:29:10,500
Ejaaz:
content, and if you would like to support us in any way, we are in the market for sponsors.

485
00:29:10,820 --> 00:29:13,810
Ejaaz:
We're looking for companies and features and products that we're excited to

486
00:29:13,810 --> 00:29:17,860
Ejaaz:
use and talk about to our audience of hundreds and hundreds and hundreds of

487
00:29:17,860 --> 00:29:19,280
Ejaaz:
thousands of people per month.

488
00:29:19,460 --> 00:29:23,410
Ejaaz:
So if that is you or if that is someone that you know that think might be them,

489
00:29:23,590 --> 00:29:26,010
Ejaaz:
please let them know. Please send them a message. Send us a message.

490
00:29:26,010 --> 00:29:29,850
Ejaaz:
We're available on X, on email, whatever. We'll leave it in the description below.

491
00:29:30,500 --> 00:29:31,450
Ejaaz:
And yeah, hopefully speak to you

492
00:29:31,450 --> 00:29:34,040
Ejaaz:
soon. Thank you so much for listening and we'll see you on the next one.

493
00:29:34,360 --> 00:29:34,750
Josh:
See ya.