1
00:00:00,018 --> 00:00:05,298
Ejaaz:
OpenAI Astra, also known as GPT-6, is on the verge of being released.

2
00:00:05,298 --> 00:00:08,938
Ejaaz:
We've seen a bunch of updates from the OpenAI side. If you're listening to this

3
00:00:08,938 --> 00:00:11,438
Ejaaz:
episode right now, you may have access to this.

4
00:00:11,438 --> 00:00:15,470
Ejaaz:
This is OpenAI's most powerful model yet, and it's rumored to be more powerful

5
00:00:15,685 --> 00:00:21,538
Ejaaz:
than Mythos 5.1, which was also released from Anthropic alongside Fable 5.1,

6
00:00:21,538 --> 00:00:23,365
Ejaaz:
two new models from Anthropic.

7
00:00:24,102 --> 00:00:26,949
Ejaaz:
This wasn't the only model releases this week. We had a new one from Meta.

8
00:00:27,269 --> 00:00:30,209
Ejaaz:
We had a new one from Google, as well as a brand new world model,

9
00:00:30,209 --> 00:00:34,607
Ejaaz:
which changes our perspective on how we create simulated realities.

10
00:00:34,798 --> 00:00:38,659
Ejaaz:
But in other news, the ghoul is back on this show.

11
00:00:38,868 --> 00:00:43,739
Ejaaz:
OpenAI's rumored hardware device may have been peaked at again.

12
00:00:43,739 --> 00:00:49,599
Ejaaz:
We have Parma Lucky this time, rocking these really weird new futuristic earbuds,

13
00:00:49,599 --> 00:00:54,469
Ejaaz:
as well as a metal puck. We're going to get into all of that and more on today's roundup.

14
00:00:54,469 --> 00:00:56,839
Josh:
This is the round. This is my favorite episode of the week. I'm so excited.

15
00:00:56,839 --> 00:00:59,469
Josh:
We have so much to talk about today. So buckle up, get comfortable.

16
00:00:59,469 --> 00:01:00,869
Josh:
The first topic is Astra.

17
00:01:01,189 --> 00:01:04,529
Josh:
Now at the time of recording, it is not out, although it is rumored and it is

18
00:01:04,529 --> 00:01:05,869
Josh:
likely to be coming out any moment.

19
00:01:05,869 --> 00:01:08,599
Josh:
So normally upon conclusion is normally when they decide to publish it.

20
00:01:08,599 --> 00:01:11,739
Josh:
Once we hit stop on this recording, they will publish the model. So get ready for that.

21
00:01:11,739 --> 00:01:13,909
Josh:
It is basically, it's going to be the new best model in the world.

22
00:01:13,909 --> 00:01:18,199
Josh:
It's going to be incredibly powerful, incredibly intelligent, possibly efficient.

23
00:01:18,199 --> 00:01:21,609
Josh:
We don't know. There's a lot to learn as this model comes out basically this

24
00:01:21,609 --> 00:01:24,879
Josh:
model's been ready for quite some time we don't know exactly how long but open

25
00:01:24,879 --> 00:01:26,415
Josh:
ai themselves have been teasing it

26
00:01:26,705 --> 00:01:30,199
Josh:
for at least a couple of months now we know it hit roadblocks it hit problems

27
00:01:30,199 --> 00:01:33,404
Josh:
with the government approvals then it hit problems with their own internal

28
00:01:33,927 --> 00:01:37,729
Josh:
Breakouts per se like we've mentioned about the exploits that have happened

29
00:01:37,729 --> 00:01:40,463
Josh:
with hugging face maybe three separate episodes but you should go check out

30
00:01:40,678 --> 00:01:43,229
Josh:
our most recent moments earlier this week it's finally time to see what they've

31
00:01:43,229 --> 00:01:47,229
Josh:
been cooking so this is probably possibly going to be GPT-6.

32
00:01:47,229 --> 00:01:50,979
Josh:
It seems like it is strong enough to be GPT-6 level. What I suspect and what

33
00:01:50,979 --> 00:01:55,209
Josh:
I hope is that we may also get a image generation model as well in addition

34
00:01:55,209 --> 00:01:59,089
Josh:
to Astra. So this is going to be a fun weekend for anyone who is excited about

35
00:01:59,089 --> 00:02:01,680
Josh:
using these new tools to see what this thing is capable of.

36
00:02:01,971 --> 00:02:05,579
Josh:
There is a news headline that I think is important to highlight specifically

37
00:02:05,579 --> 00:02:09,726
Josh:
around how they trained this model and it appears as if there's a new breakthrough

38
00:02:10,022 --> 00:02:14,318
Josh:
that happened. It's called recurring depth or like a looped transformer architecture.

39
00:02:14,631 --> 00:02:17,168
Josh:
And the TLDR is instead of every token kind of moving

40
00:02:17,650 --> 00:02:21,835
Josh:
once throughout this like fixed stack of transformer layers portions of that

41
00:02:21,835 --> 00:02:24,371
Josh:
network can be run repeatedly and basically what this means is that

42
00:02:24,806 --> 00:02:28,035
Josh:
you get more computational depth without proportionally increasing the number

43
00:02:28,035 --> 00:02:30,507
Josh:
of unique parameters we hear about parameter count all the time

44
00:02:30,913 --> 00:02:34,225
Josh:
that's not the reason why these models necessarily need to get better they have

45
00:02:34,225 --> 00:02:36,045
Josh:
figured out a new way of doing it the problem with this

46
00:02:36,381 --> 00:02:39,499
Josh:
is you see less of the reasoning traces you see less of the chain of thought

47
00:02:39,771 --> 00:02:42,976
Josh:
and people are concerned because i mean we just saw with hugging face what happens

48
00:02:43,202 --> 00:02:46,041
Josh:
when you don't monitor these models closely and it seems as if this is

49
00:02:46,418 --> 00:02:49,825
Josh:
kind of taking a little cheat code to getting to a better place faster with

50
00:02:49,825 --> 00:02:52,641
Josh:
a little less safety so i don't know exciting model on the way

51
00:02:53,018 --> 00:02:58,840
Ejaaz:
Yeah i mean the scariest part about this is open ai is releasing gpt6 uh

52
00:02:59,175 --> 00:03:02,817
Ejaaz:
which is by the way the same model which broke out of containment

53
00:03:03,096 --> 00:03:07,765
Ejaaz:
got access to open air's own internal research clusters and got admin access

54
00:03:07,765 --> 00:03:11,395
Ejaaz:
to like all their research tools which is just kind of crazy this is stuff which

55
00:03:11,395 --> 00:03:14,665
Ejaaz:
is typically like locked behind a vault and no one gets access to it.

56
00:03:14,665 --> 00:03:18,613
Ejaaz:
Very few people in the company have access to that thing. And an internal model was able to do this.

57
00:03:18,833 --> 00:03:22,455
Ejaaz:
Second scary thing is this model is already about three to four months old.

58
00:03:22,455 --> 00:03:25,205
Ejaaz:
So you can imagine like the current model that they have internally right now

59
00:03:25,205 --> 00:03:26,529
Ejaaz:
is way, way more powerful.

60
00:03:26,918 --> 00:03:30,365
Ejaaz:
And then we have this news that they're training it in a particular way that

61
00:03:30,365 --> 00:03:35,575
Ejaaz:
makes it less readable. In other words, we can't kind of like discuss or figure

62
00:03:35,575 --> 00:03:39,405
Ejaaz:
out the kind of thoughts that this model is having. The reason why that's important

63
00:03:39,405 --> 00:03:41,286
Ejaaz:
to your earlier point, Josh, is.

64
00:03:41,750 --> 00:03:46,694
Ejaaz:
The hugging face incident was only discovered because researchers were able

65
00:03:46,694 --> 00:03:49,720
Ejaaz:
to look at the internal thoughts of the internal model that they were training.

66
00:03:49,894 --> 00:03:53,372
Ejaaz:
And they were like, oh, wait, this thing is completely misaligned.

67
00:03:53,662 --> 00:03:56,404
Ejaaz:
So if they're kind of like leaning into this technique, which is known as loop

68
00:03:56,404 --> 00:03:59,862
Ejaaz:
transformers, and they can't read the internal thoughts of the model,

69
00:04:00,105 --> 00:04:04,514
Ejaaz:
you can kind of reason to argue that this model can kind of like hide some of

70
00:04:04,514 --> 00:04:07,902
Ejaaz:
its misaligned behavior. So that is a little bit concerning.

71
00:04:08,116 --> 00:04:12,395
Ejaaz:
Now, when it comes to our favorite rivalry in the entire world.

72
00:04:12,772 --> 00:04:16,384
Ejaaz:
OpenAI, unfortunately, is ending their partnership with Cursor.

73
00:04:16,384 --> 00:04:18,975
Ejaaz:
Now, Cursor, if you've been listening to the show for a while,

74
00:04:19,202 --> 00:04:22,784
Ejaaz:
was originally kind of like a neutral platform, which will essentially be a

75
00:04:22,784 --> 00:04:24,914
Ejaaz:
wrapper around different kinds of models.

76
00:04:24,914 --> 00:04:28,664
Ejaaz:
And their kind of like product was essentially this harness.

77
00:04:28,664 --> 00:04:31,604
Ejaaz:
So it would take your prompt and it would route it to the right model at the

78
00:04:31,604 --> 00:04:34,074
Ejaaz:
right time. It would save you a bunch of money. It would be incredibly efficient.

79
00:04:34,277 --> 00:04:36,686
Ejaaz:
And arguably, it would give you better answers.

80
00:04:36,953 --> 00:04:43,890
Ejaaz:
Now, that company was acquired for $60 billion by none other than Elon Musk in SpaceX AI himself.

81
00:04:44,099 --> 00:04:47,794
Ejaaz:
Now, Sam and Elon haven't really got along. They have a very rivaled history

82
00:04:47,794 --> 00:04:51,384
Ejaaz:
stemming all the way back to the creation of OpenAI where Elon Musk was involved

83
00:04:51,384 --> 00:04:54,833
Ejaaz:
and then he had to like separate and he spun up XAI and it was a whole thing.

84
00:04:55,140 --> 00:04:58,434
Ejaaz:
And now Sam Altman has completely severed ties. He said, okay,

85
00:04:58,434 --> 00:05:01,724
Ejaaz:
Elon, if you own this entire thing, I'm not giving you access to my frontier

86
00:05:01,724 --> 00:05:04,881
Ejaaz:
model, which you can kind of potentially steal. I'm gonna take that away from you.

87
00:05:05,206 --> 00:05:09,654
Ejaaz:
And that resulted in arguably the best tweet of the year. i don't know if you

88
00:05:09,654 --> 00:05:14,064
Ejaaz:
saw this josh but it came from michael trull who's like the 24 year old founder

89
00:05:14,064 --> 00:05:15,475
Ejaaz:
and ceo of cursor who goes.

90
00:05:15,957 --> 00:05:19,659
Ejaaz:
Listen, we're starting to see that OpenAI put out a note saying they plan to block Cursor.

91
00:05:20,042 --> 00:05:24,863
Ejaaz:
By the way, OpenAI models serve about 5% of Cursor user traffic,

92
00:05:24,863 --> 00:05:27,003
Ejaaz:
which is just the biggest dunk.

93
00:05:27,763 --> 00:05:30,883
Ejaaz:
But he says, we're speaking to the OpenAI team to resolve this.

94
00:05:30,883 --> 00:05:34,673
Ejaaz:
And I just found this whole thing completely hilarious. And I think like Elon

95
00:05:34,673 --> 00:05:37,893
Ejaaz:
called him an asshole in the midst of all of this. It was just a crazy thing.

96
00:05:38,340 --> 00:05:42,063
Josh:
That sounds about right. I mean, questionable activities from Sam to want to

97
00:05:42,063 --> 00:05:44,081
Josh:
pick another fight with Elon and co

98
00:05:44,406 --> 00:05:47,668
Josh:
particularly when cursor and spacex are the ones providing

99
00:05:47,959 --> 00:05:51,523
Josh:
a tremendous amount of compute for their competitors as well as just pushing

100
00:05:51,523 --> 00:05:54,673
Josh:
along the frontier it seems like why why would you want to make enemies with

101
00:05:54,673 --> 00:05:58,783
Josh:
the greatest entrepreneur of our time like that seems like a mishap but on the

102
00:05:58,783 --> 00:06:01,705
Josh:
other half of the coin sam is saying interesting things he's kind of like

103
00:06:02,223 --> 00:06:07,330
Josh:
more based sam has been coming out to play this week we got him kind of shutting down open ai

104
00:06:07,632 --> 00:06:11,806
Josh:
and spacex's relationship but we also got him on a podcast addressing the water

105
00:06:12,090 --> 00:06:16,043
Josh:
issue and addressing the a lot of the data center problems that people have

106
00:06:16,043 --> 00:06:19,353
Josh:
been concerned about and just confronting them head on he had this amazing quote

107
00:06:19,353 --> 00:06:20,401
Josh:
and this amazing clip that i loved

108
00:06:20,686 --> 00:06:23,937
Josh:
which said for every 38 000 chat gpt queries

109
00:06:24,233 --> 00:06:27,533
Josh:
that is the same amount of water that is used in the production of a single

110
00:06:27,533 --> 00:06:32,113
Josh:
almond in california and i'm not like that great at doing math in my head but

111
00:06:32,113 --> 00:06:36,197
Josh:
um 38 000 times like a handful of almonds equals probably more

112
00:06:36,806 --> 00:06:39,952
Josh:
tokens that we've consumed in our entire life so by you eating a handful of

113
00:06:39,952 --> 00:06:42,542
Josh:
almonds you have essentially consumed a lifetime of prompts and i found that

114
00:06:42,542 --> 00:06:46,072
Josh:
to be interesting too he mentions where he thinks it came from there does seem

115
00:06:46,072 --> 00:06:51,365
Josh:
to be this kind of like paid organized psyop coming from china particularly on the tiktok platform

116
00:06:51,696 --> 00:06:54,117
Josh:
that is convincing people of these mistruths

117
00:06:54,535 --> 00:06:57,612
Josh:
um but i like that he's out here addressing it i like that he is discussing

118
00:06:57,612 --> 00:07:01,042
Josh:
in public and now that we have clips like this we can share it around the internet

119
00:07:01,042 --> 00:07:05,142
Josh:
with viewers like you to let everyone know like hey, it's not actually ending the world.

120
00:07:05,142 --> 00:07:08,982
Ejaaz:
Yeah, it was a great podcast episode. I recommend you go listen to it.

121
00:07:08,982 --> 00:07:12,462
Ejaaz:
It was produced by, I believe, Alex Heath, who is kind of like an insider and

122
00:07:12,462 --> 00:07:14,912
Ejaaz:
he discusses a lot of these things. Definitely go check it out.

123
00:07:15,490 --> 00:07:18,442
Ejaaz:
My comment on the water thing is, if you're listening to this and you still

124
00:07:18,442 --> 00:07:24,052
Ejaaz:
believe that data centers, AI data centers specifically, consume a lot of water,

125
00:07:24,336 --> 00:07:28,012
Ejaaz:
that is factually incorrect. And a lot of the claims that have been made in

126
00:07:28,012 --> 00:07:30,182
Ejaaz:
the past is completely baseless.

127
00:07:30,879 --> 00:07:34,282
Ejaaz:
Water is typically reused a lot in AI data centers, and that's been the case

128
00:07:34,282 --> 00:07:36,713
Ejaaz:
for a while now. So definitely go check out that episode.

129
00:07:37,292 --> 00:07:41,442
Ejaaz:
But shifting gears slightly, there have been so many model releases this week

130
00:07:41,442 --> 00:07:43,737
Ejaaz:
that it's honestly become quite overwhelming.

131
00:07:44,312 --> 00:07:47,452
Ejaaz:
And the first one, or rather one of them, comes from none other than,

132
00:07:47,932 --> 00:07:52,676
Ejaaz:
you know, I have a mixed relationship with Mr. Zuckerberg over here. I've been a hater.

133
00:07:53,152 --> 00:07:59,632
Ejaaz:
I've now been a lover for a while now. And Meta released, surprisingly, another model.

134
00:07:59,632 --> 00:08:03,542
Ejaaz:
This is their fourth release in five to six weeks, I believe,

135
00:08:03,542 --> 00:08:08,350
Ejaaz:
which is unprecedented for Meta on its own. It's MuseSpark 1.3,

136
00:08:08,652 --> 00:08:13,601
Ejaaz:
and it's a specialized reasoning, agentic coding model. So what does that mean?

137
00:08:13,955 --> 00:08:16,762
Ejaaz:
It means that it's much better at coding than its previous model.

138
00:08:16,762 --> 00:08:19,242
Ejaaz:
And its previous model, which was released, I believe, two and a half weeks

139
00:08:19,242 --> 00:08:21,240
Ejaaz:
ago, was also pretty decent as well.

140
00:08:21,513 --> 00:08:25,014
Ejaaz:
But the main standout feature for Meta and this particular model is,

141
00:08:25,147 --> 00:08:31,242
Ejaaz:
once again, it is the cheapest model around to get 90% of your tasks done. And I can prove it.

142
00:08:31,591 --> 00:08:35,857
Ejaaz:
I have a demo here shared by Alexander Wang, who is one of the leads here.

143
00:08:36,514 --> 00:08:42,195
Ejaaz:
Where they basically or someone recreated Minecraft and it cost them 10 cents

144
00:08:42,195 --> 00:08:45,495
Ejaaz:
to do. So if you're wondering, you know, why is this part of Meta's strategy?

145
00:08:45,495 --> 00:08:48,135
Ejaaz:
Well, Mark Zuckerberg's looking at the competition and he's thinking,

146
00:08:48,135 --> 00:08:51,381
Ejaaz:
well, I can catch up to frontier intelligence, but it's going to take me a while.

147
00:08:51,665 --> 00:08:55,395
Ejaaz:
In the meantime, I'm going to provide the cheapest intelligence possible so

148
00:08:55,395 --> 00:08:58,965
Ejaaz:
that enterprises have to second guess whether they should go with the anthropic

149
00:08:58,965 --> 00:09:02,935
Ejaaz:
model or with the open area model when they could just use some of Meta's models

150
00:09:02,935 --> 00:09:05,383
Ejaaz:
to do the majority of their work and then maybe use that,

151
00:09:06,015 --> 00:09:08,195
Ejaaz:
in partnership with some of these other smarter models.

152
00:09:08,195 --> 00:09:11,405
Ejaaz:
And that's what Meta's strategy has been for a while now. And it's so impressive

153
00:09:11,405 --> 00:09:12,950
Ejaaz:
to see the kind of cadence that they have.

154
00:09:13,333 --> 00:09:18,235
Ejaaz:
But it would be bequeath me to say that I can't mention Meta's model without

155
00:09:18,235 --> 00:09:21,867
Ejaaz:
Google's model because they're kind of now weirdly playing the same game.

156
00:09:22,174 --> 00:09:25,297
Ejaaz:
Google also released a new model called Gemini 3.8 Flash.

157
00:09:25,559 --> 00:09:28,955
Ejaaz:
And I have to say the weirdest part about this is Meta's model is better.

158
00:09:28,955 --> 00:09:32,461
Ejaaz:
So Meta has finally crushed Google and it's just so bizarre off to me.

159
00:09:32,693 --> 00:09:36,175
Josh:
I could probably tell you the change log between these models without actually

160
00:09:36,175 --> 00:09:37,476
Josh:
even reading a single thing on the card.

161
00:09:37,755 --> 00:09:40,657
Josh:
It's slightly more efficient, slightly cheaper, slightly smarter.

162
00:09:40,849 --> 00:09:44,585
Josh:
It feels like this is like iOS version updates where it's just getting slightly

163
00:09:44,585 --> 00:09:47,505
Josh:
better across the board and nothing really exciting is happening.

164
00:09:47,505 --> 00:09:50,845
Josh:
It's like, okay, we're giving you cheap tokens. It's kind of like lower on the

165
00:09:50,845 --> 00:09:53,173
Josh:
Pareto Frontier in terms of like our cost competition.

166
00:09:53,399 --> 00:09:56,685
Josh:
It's not much smarter. It seems fine. I'm like, okay, this is like cool.

167
00:09:56,685 --> 00:09:59,205
Josh:
Am I going to go and try it? Like probably not. There's other more interesting

168
00:09:59,205 --> 00:10:00,992
Josh:
models to try, but it's nice to see.

169
00:10:01,102 --> 00:10:04,025
Josh:
I do find it interesting when we talk about we're going to talk about the frontier

170
00:10:04,025 --> 00:10:07,375
Josh:
model that was released this week but there's been this race on kind of like

171
00:10:07,375 --> 00:10:11,575
Josh:
the cost basis and then the race on the intelligence basis and gemini and meta

172
00:10:11,575 --> 00:10:15,649
Josh:
right now are kind of going head to head on that cost basis and i think it'll be interesting to see

173
00:10:15,991 --> 00:10:20,130
Josh:
where they go but those are the two smaller models we had a really cool thing this

174
00:10:20,368 --> 00:10:23,335
Ejaaz:
This next one um this is my favorite yes.

175
00:10:24,577 --> 00:10:27,671
Josh:
Gaussian splats yes need I say anymore

176
00:10:27,839 --> 00:10:28,978
Ejaaz:
We were the OGs.

177
00:10:29,998 --> 00:10:33,447
Josh:
I'm a huge splat guy man like

178
00:10:33,632 --> 00:10:36,258
Ejaaz:
Tell Josh is a big splat guy guys dude I.

179
00:10:36,258 --> 00:10:40,988
Josh:
Freaking love splats so I can't even like describe to you what it is you just

180
00:10:40,988 --> 00:10:44,517
Josh:
just watch this video basically what this technology it's like unbelievable

181
00:10:44,726 --> 00:10:48,627
Josh:
so it's able to take still images or some sort of a video image

182
00:10:48,847 --> 00:10:51,605
Josh:
and then generate an entire world around it for you to navigate.

183
00:10:52,290 --> 00:10:56,018
Josh:
So let's say you have an image of a place that you've traveled to that you really

184
00:10:56,018 --> 00:10:58,848
Josh:
love and you want to relive what that looks like, or you'd like to generate

185
00:10:58,848 --> 00:11:00,602
Josh:
an extension of it and go and live inside of it.

186
00:11:00,991 --> 00:11:04,328
Josh:
It is able to do that. So what we're seeing here is a standard input image on

187
00:11:04,328 --> 00:11:09,129
Josh:
the screen where it is a photo of a church and you can draw the line of a camera

188
00:11:09,460 --> 00:11:13,048
Josh:
that goes through and navigates the 3D space and it will actually generate all

189
00:11:13,048 --> 00:11:16,238
Josh:
of what it expects it to look like for this video model.

190
00:11:16,238 --> 00:11:19,478
Josh:
And it's like unbelievable world builder. It's been a while since we've talked

191
00:11:19,478 --> 00:11:21,018
Josh:
about world building in general

192
00:11:21,398 --> 00:11:24,318
Josh:
because we haven't really gotten one of the new gemma models from google in

193
00:11:24,318 --> 00:11:27,448
Josh:
a while but this seems to take what google is working on and kind of flip it

194
00:11:27,448 --> 00:11:31,688
Josh:
on its head in a way that is unbelievably impressive it requires one still image

195
00:11:31,688 --> 00:11:33,190
Josh:
and you could see in the video here

196
00:11:33,661 --> 00:11:37,538
Josh:
it will just generate an entire room from a still image and let you walk around

197
00:11:37,829 --> 00:11:41,788
Josh:
and actually navigate it so in terms of data training i mean what what are world

198
00:11:41,788 --> 00:11:44,928
Josh:
models good for training robots they're good for entertainment they're good

199
00:11:44,928 --> 00:11:47,258
Josh:
for gaming this is definitely right

200
00:11:47,838 --> 00:11:50,848
Josh:
on the frontier of what's possible here and of course naturally it's by fei

201
00:11:50,848 --> 00:11:57,068
Josh:
fei lee the godmother of ai and i found this release to be so exciting of all

202
00:11:57,068 --> 00:12:00,677
Josh:
the tools that came out this week this is the one i want to spend the most amount of time with because

203
00:12:01,060 --> 00:12:03,701
Josh:
man how much fun is it to create these 3d virtual environments like

204
00:12:04,437 --> 00:12:06,754
Josh:
Like nothing it's so easy now and it's so impressive

205
00:12:07,497 --> 00:12:11,508
Ejaaz:
So for those of you who have been following kind of like the world model advancements,

206
00:12:12,241 --> 00:12:14,741
Ejaaz:
Fei-Fei Li, as you mentioned, is an OG here.

207
00:12:14,741 --> 00:12:18,811
Ejaaz:
Fun fact, she used to teach Andre Carpathy, the Andre Carpathy,

208
00:12:18,811 --> 00:12:22,731
Ejaaz:
back when he was just a young student trying to learn about all this kind of AI stuff.

209
00:12:22,731 --> 00:12:26,111
Ejaaz:
So Fei-Fei Li's been around for a while, and she recently started her own company

210
00:12:26,111 --> 00:12:30,561
Ejaaz:
called World Labs, which is focused, as the name suggests, on world models.

211
00:12:30,561 --> 00:12:34,201
Ejaaz:
Now, the difference between a world model and an LLM, a world model is an AI

212
00:12:34,201 --> 00:12:38,741
Ejaaz:
model that really has a good understanding and spatial awareness of the actual world.

213
00:12:38,741 --> 00:12:41,571
Ejaaz:
LLMs are like kind of like a student in a library. It reads all the books.

214
00:12:41,571 --> 00:12:44,581
Ejaaz:
It understands the theory of how the world works. World models actually go out

215
00:12:44,581 --> 00:12:48,371
Ejaaz:
into the world and understand how the physics of the world works.

216
00:12:48,371 --> 00:12:50,581
Ejaaz:
If I drop this drink that I'm showing you on my screen right now,

217
00:12:50,581 --> 00:12:53,164
Ejaaz:
what happens, how it splatters the physics behind all of it.

218
00:12:53,391 --> 00:12:56,231
Ejaaz:
Now, the advancement that we're seeing with this particular model,

219
00:12:56,231 --> 00:12:58,651
Ejaaz:
and I wanted to show just this clip.

220
00:12:58,651 --> 00:13:02,081
Josh:
The demos are crazy too. I've been watching demos. They're sick.

221
00:13:02,081 --> 00:13:04,111
Ejaaz:
So this is someone, if you're not, if you're listening to this and you're not

222
00:13:04,111 --> 00:13:07,030
Ejaaz:
watching your screen, I advise you do this is someone spilling a drink.

223
00:13:07,321 --> 00:13:10,264
Ejaaz:
And by the way, this is just one video shot from one angle.

224
00:13:10,687 --> 00:13:15,811
Ejaaz:
And so this person has been able to navigate and go around him in every single

225
00:13:15,811 --> 00:13:19,441
Ejaaz:
shape and form, including capturing his face, his beard trimmings,

226
00:13:19,674 --> 00:13:23,969
Ejaaz:
the angle of the lighting on the strawberries, the coloring, everything.

227
00:13:24,132 --> 00:13:26,771
Ejaaz:
This looks like a video that was shot by someone walking around.

228
00:13:26,771 --> 00:13:30,311
Ejaaz:
But really, it's just kind of like, from one plane of view, someone just shot

229
00:13:30,311 --> 00:13:32,711
Ejaaz:
one video. And now you're able to go round and round around,

230
00:13:32,711 --> 00:13:35,115
Ejaaz:
you're able to stop it, you're able to edit it, you're able to zoom in.

231
00:13:35,347 --> 00:13:37,971
Ejaaz:
There's so many different ways that you can kind of play around with this tool.

232
00:13:37,971 --> 00:13:40,995
Ejaaz:
This could have effects in VFX, Hollywood, which, by the way,

233
00:13:41,193 --> 00:13:45,991
Ejaaz:
typically takes tens of millions of dollars to recreate this and hours and months of work.

234
00:13:45,991 --> 00:13:49,761
Ejaaz:
It's so crazy. Reduced to like kind of like a couple of dollars and like a couple

235
00:13:49,761 --> 00:13:51,055
Ejaaz:
of seconds. It's just insane.

236
00:13:52,222 --> 00:13:55,003
Ejaaz:
What's the comparison to this model? Well, it's from Google.

237
00:13:55,223 --> 00:13:58,325
Ejaaz:
Google, I believe it was Genie 3 that they released a while ago.

238
00:13:58,325 --> 00:13:59,339
Ejaaz:
It was their world model.

239
00:13:59,583 --> 00:14:03,045
Ejaaz:
It was okay, but it kind of looked like the Unreal game engine.

240
00:14:03,045 --> 00:14:04,609
Ejaaz:
It kind of looked animated.

241
00:14:04,876 --> 00:14:07,595
Ejaaz:
The main difference here is you can take a picture of the real world,

242
00:14:07,595 --> 00:14:09,295
Ejaaz:
like a human being, and you can

243
00:14:09,295 --> 00:14:14,419
Ejaaz:
turn it into a simulated reality with a 3D point for every single pixel.

244
00:14:14,617 --> 00:14:19,005
Ejaaz:
And so imagine this applied to like sports. Imagine this applied to like movies

245
00:14:19,214 --> 00:14:23,533
Ejaaz:
this could like change the way we interact with media and it's just such a cool development.

246
00:14:23,655 --> 00:14:26,785
Josh:
It's also just cool there's another demo we have with basketball like this dude

247
00:14:26,785 --> 00:14:30,415
Josh:
was just taking he you could see there's like a couple cameras there's three

248
00:14:30,415 --> 00:14:33,805
Josh:
different cameras basically and from the three camera angles it's able to rebuild

249
00:14:33,805 --> 00:14:35,834
Josh:
the entire three-dimensional

250
00:14:36,339 --> 00:14:38,895
Josh:
structure of this and guess the things that it can't see so you'll notice the

251
00:14:38,895 --> 00:14:42,161
Josh:
cameras are all facing the basketball hoop it still is able to generate what's

252
00:14:42,388 --> 00:14:43,975
Josh:
behind the basketball hoop because

253
00:14:43,975 --> 00:14:46,271
Josh:
it's smart enough to understand and interpret what it should look like

254
00:14:46,596 --> 00:14:50,005
Josh:
and then generate that so all it takes is a few camera angles and i love the

255
00:14:50,005 --> 00:14:52,955
Josh:
description where it just says like this is kind of indistinguishable from black

256
00:14:52,955 --> 00:14:55,217
Josh:
magic and we're seeing another one where if you have

257
00:14:55,675 --> 00:14:58,357
Josh:
Just if you have a single camera angle, it works pretty well,

258
00:14:58,565 --> 00:15:02,350
Josh:
but it is less accurate in the sense that it is less truthful of what reality looks like.

259
00:15:02,559 --> 00:15:05,661
Josh:
If you have a couple cameras, it will just take whatever you give it and stitch

260
00:15:05,661 --> 00:15:07,501
Josh:
it together and turn into this unbelievable world.

261
00:15:07,501 --> 00:15:10,622
Josh:
It is so, so impressive. So I think of all the releases this week,

262
00:15:10,930 --> 00:15:13,301
Josh:
this one's my favorite. This is what I want everyone to try.

263
00:15:13,301 --> 00:15:16,084
Josh:
This is what I'm most excited to try. I want to go out and record some videos,

264
00:15:16,322 --> 00:15:19,391
Josh:
like just post up a couple iPhones and see what type of crazy stuff I can get.

265
00:15:19,391 --> 00:15:21,553
Josh:
Maybe we'll share the results next week on the show.

266
00:15:21,669 --> 00:15:26,011
Ejaaz:
How the tables have turned though, Josh. We typically always used to talk about

267
00:15:26,011 --> 00:15:30,347
Ejaaz:
like the new anthropic models like released like the first thing on the docket um

268
00:15:30,731 --> 00:15:33,871
Ejaaz:
but claude or rather anthropic released not one

269
00:15:34,103 --> 00:15:40,621
Ejaaz:
but two new models this week um claude fable 5.1 and claude mythos 5.1 um this

270
00:15:40,621 --> 00:15:43,741
Ejaaz:
is an iterative upgrade to my understanding i've been playing around with these

271
00:15:43,741 --> 00:15:47,269
Ejaaz:
models i've only got access to fable 5.1 which is the one with the safeguards on it

272
00:15:47,652 --> 00:15:50,931
Ejaaz:
and i understand that there's a bunch of different changes here so number one

273
00:15:50,931 --> 00:15:52,711
Ejaaz:
it is much better bang for buck.

274
00:15:52,711 --> 00:15:57,461
Ejaaz:
So you can get the same amount of intelligence or increased intelligence for a fraction of the cost.

275
00:15:57,711 --> 00:16:01,341
Ejaaz:
They're doing this thing called caching or the cutting down on sort of like

276
00:16:01,341 --> 00:16:04,731
Ejaaz:
the caching mechanism that the model uses, which means that you could get up

277
00:16:04,731 --> 00:16:09,681
Ejaaz:
to 75% lower costs on your prompting, especially over long agentic workloads.

278
00:16:09,681 --> 00:16:13,622
Ejaaz:
So the longer your task is, typically this model is very good and much, much cheaper.

279
00:16:14,231 --> 00:16:19,253
Ejaaz:
The second thing is it's really compelling when it comes to video generation,

280
00:16:19,479 --> 00:16:23,431
Ejaaz:
which is crazy for me to say, given that what you're seeing on the screen right

281
00:16:23,431 --> 00:16:27,740
Ejaaz:
now, which is a demo from Fable 5.1, is completely generated from code.

282
00:16:27,856 --> 00:16:31,751
Ejaaz:
This didn't take any kind of input images. This was just generated from code

283
00:16:31,751 --> 00:16:34,751
Ejaaz:
from scratch. And it is just crazy to see the navigation around it.

284
00:16:34,751 --> 00:16:37,501
Ejaaz:
It's almost like I'm looking at that world model that we just literally like

285
00:16:37,501 --> 00:16:40,331
Ejaaz:
kind of like went through. So it's cheaper, it's more efficient.

286
00:16:40,552 --> 00:16:43,631
Ejaaz:
And then the other thing I'll say about this is the safeguards.

287
00:16:43,631 --> 00:16:46,771
Ejaaz:
One thing that was really annoying for me for Fable 5, at least personally,

288
00:16:46,771 --> 00:16:50,771
Ejaaz:
when I was experiencing it is it just spoke too much and it didn't kind of answer

289
00:16:50,771 --> 00:16:51,854
Ejaaz:
the questions that I gave it.

290
00:16:52,127 --> 00:16:55,391
Ejaaz:
This model has been a significant leap. It now answers my questions.

291
00:16:55,391 --> 00:16:59,846
Ejaaz:
It stops saying too many words and it's overall a much more pleasant experience to have.

292
00:17:00,547 --> 00:17:03,552
Josh:
Yeah, it's amazing. This demo in particular is very cool.

293
00:17:03,552 --> 00:17:06,672
Josh:
The person who wrote it says, for this one, I gave it a picture of a property

294
00:17:06,672 --> 00:17:10,792
Josh:
lot and it designed a house for the lot, rendered it and then produced a cinematic walkthrough.

295
00:17:10,792 --> 00:17:14,192
Josh:
And it's important to note also that Fable didn't actually generate the images,

296
00:17:14,192 --> 00:17:17,652
Josh:
but Fable worked with Blender, which is a 3D rendering software.

297
00:17:17,652 --> 00:17:20,932
Josh:
It was headless and it worked together and it conducted it and orchestrated

298
00:17:20,932 --> 00:17:23,321
Josh:
it to build this 3D virtual space.

299
00:17:23,519 --> 00:17:25,718
Josh:
And I think that's one of the more powerful things is because the model,

300
00:17:25,951 --> 00:17:29,322
Josh:
it doesn't have an image model and yet it's able to create these unbelievable

301
00:17:29,322 --> 00:17:32,602
Josh:
visual results. The second one being, we have another example here of a one-shot

302
00:17:32,602 --> 00:17:34,843
Josh:
prompt for a Mario Kart game.

303
00:17:35,110 --> 00:17:41,622
Josh:
And Mable 5.1 turned a singular prompt into a fully working Mario Kart game.

304
00:17:41,622 --> 00:17:44,492
Josh:
And I found this to be really impressive too. It's like, okay,

305
00:17:44,492 --> 00:17:48,072
Josh:
well, it doesn't have an image generation model, but this is looking pretty

306
00:17:48,072 --> 00:17:51,602
Josh:
strong to me. In fact, if you gave me this game like a couple of years ago on

307
00:17:51,602 --> 00:17:55,676
Josh:
my handheld device, this looks as good as Frontier in terms of like a Nintendo game.

308
00:17:55,920 --> 00:17:59,402
Josh:
So I think the capabilities of these models and the intelligence to be able

309
00:17:59,402 --> 00:18:03,272
Josh:
to interact with other tools that exist and harness them and orchestrate them

310
00:18:03,272 --> 00:18:06,282
Josh:
to get the results that you're looking for is a really impressive capability.

311
00:18:06,537 --> 00:18:09,086
Josh:
Great model, great new model. This is like the new frontier.

312
00:18:09,742 --> 00:18:13,232
Josh:
We'll see what Astra has to like come back with. I'm very excited.

313
00:18:13,232 --> 00:18:15,640
Josh:
It seems as if Astra might leapfrog this to some extent, but for now,

314
00:18:15,947 --> 00:18:18,223
Josh:
best model, really impressive, really enjoying using it.

315
00:18:18,658 --> 00:18:21,275
Ejaaz:
Okay, so just to give the run of the mill of everything that we've just covered,

316
00:18:21,275 --> 00:18:26,408
Ejaaz:
just to catch you up on the roundup itself, we have OpenAI Astra coming out.

317
00:18:26,524 --> 00:18:29,425
Ejaaz:
You might likely have access to it right now. You've got Google that released

318
00:18:29,425 --> 00:18:34,535
Ejaaz:
Gemini 3.8 Flash, super cheap, super effective, but it's competing with Meta's

319
00:18:34,535 --> 00:18:37,225
Ejaaz:
new model, which is Muse Spark 1.3.

320
00:18:37,225 --> 00:18:41,814
Ejaaz:
You've got Actropic releasing two models, Fable 5.1 and Mythos 5.1.

321
00:18:41,988 --> 00:18:44,025
Ejaaz:
And then you've got a few Chinese open source models as well,

322
00:18:44,025 --> 00:18:46,395
Ejaaz:
as well as this new World Labs world model.

323
00:18:46,395 --> 00:18:51,565
Ejaaz:
So the point that I want to make here is we have moved very far away from

324
00:18:51,985 --> 00:18:56,755
Ejaaz:
one model per couple of months and it's like a major capability leaf when now

325
00:18:56,755 --> 00:19:00,015
Ejaaz:
having this kind of iterative software upgrades and that's kind of like the

326
00:19:00,015 --> 00:19:04,585
Ejaaz:
new world that we live in and it's net-net so amazing for you, the listener or the,

327
00:19:05,045 --> 00:19:08,235
Ejaaz:
user for these different things because it means that you can constantly kind

328
00:19:08,235 --> 00:19:12,245
Ejaaz:
of upgrade and test different kinds of views and you're not beholden to one

329
00:19:12,245 --> 00:19:15,784
Ejaaz:
particular AI lab provider which I think is just an amazing world to be in.

330
00:19:16,109 --> 00:19:20,525
Ejaaz:
So moving on josh you sent me this before we started the show right before.

331
00:19:20,885 --> 00:19:23,865
Josh:
I gotta have a headache just thinking about it trying to be insane

332
00:19:23,865 --> 00:19:27,325
Ejaaz:
Why am i looking at alexander's gals guard again and why am i looking at this

333
00:19:27,325 --> 00:19:29,125
Ejaaz:
damn stainless steel earbuds and.

334
00:19:29,125 --> 00:19:32,565
Josh:
If you've been listening to the show for a while you know the lore here this

335
00:19:32,565 --> 00:19:35,465
Josh:
is a really important topic that has been at the core of

336
00:19:35,765 --> 00:19:39,475
Josh:
my like the the bane of my existence essentially because what we saw here just

337
00:19:39,475 --> 00:19:43,215
Josh:
quick tldr this video was released as a leaked super bowl commercial came out

338
00:19:43,215 --> 00:19:46,675
Josh:
during the super bowl was leaked on reddit was deleted shortly afterwards and

339
00:19:46,675 --> 00:19:48,530
Josh:
it shows a rumored device

340
00:19:49,065 --> 00:19:53,075
Josh:
no one knows what it was but at the conclusion of it it showed the open ai logo

341
00:19:53,075 --> 00:19:56,775
Josh:
and it said dime and this is a famous actor and this is a really beautiful looking

342
00:19:56,775 --> 00:20:00,445
Josh:
device say his name josh say his name alexander scans

343
00:20:02,705 --> 00:20:05,317
Josh:
i've no i can never get this freaking name right i have no idea

344
00:20:05,502 --> 00:20:11,087
Ejaaz:
I'm sorry it's almost been a year. I'm not exaggerating. And I've told Josh,

345
00:20:11,487 --> 00:20:15,335
Ejaaz:
it's Skarsgård. And the guy goes, Skarsgård. Well, you say it with an accent,

346
00:20:15,335 --> 00:20:18,825
Ejaaz:
too. It messes up in my head. Anyway, so this actor.

347
00:20:18,825 --> 00:20:22,238
Josh:
This famous dude, who's apparently in succession, was like, doing this thing.

348
00:20:22,888 --> 00:20:25,436
Josh:
And that was it and we were convinced this was the open ai device until open

349
00:20:25,436 --> 00:20:28,351
Josh:
i came out and said no that's not actually the device at all we don't know what that is

350
00:20:28,809 --> 00:20:32,766
Josh:
fast forward a couple months later joe gebbia co-founder of airbnb he's currently

351
00:20:32,766 --> 00:20:36,066
Josh:
leading design for the united states government as a whole like just really

352
00:20:36,066 --> 00:20:38,956
Josh:
awesome legendary job he's spotted in a cafe

353
00:20:39,630 --> 00:20:43,496
Josh:
and it's like the seemingly staged post that shows him just casually at a cafe

354
00:20:43,804 --> 00:20:48,056
Josh:
but what do you know the earbuds and the puck are inside of the frame as well

355
00:20:48,056 --> 00:20:50,751
Josh:
and he's using the device that we saw in that deleted video.

356
00:20:51,448 --> 00:20:54,722
Josh:
Okay, what is this thing? Third data point just happened yesterday.

357
00:20:54,965 --> 00:20:59,211
Josh:
Palmer Luckey, founder of Anduril, founder of Oculus, the 3D guy,

358
00:20:59,211 --> 00:21:00,114
Josh:
the virtual reality guy.

359
00:21:00,323 --> 00:21:05,937
Josh:
Yeah, he's posting a photo from a flower shop that he allegedly shops at all the time.

360
00:21:06,244 --> 00:21:08,746
Josh:
Casual flower shop shouting him out, oh, what is that on his ears?

361
00:21:08,921 --> 00:21:10,453
Josh:
The same device, the same buck in his hand.

362
00:21:10,743 --> 00:21:13,621
Josh:
And if you go to the name of the flower shop account, if you scroll up just

363
00:21:13,621 --> 00:21:16,101
Josh:
a little bit, it's Sweet Pea Flowers.

364
00:21:16,299 --> 00:21:19,938
Josh:
And Sweet Pea is the codename for the OpenAI device.

365
00:21:20,451 --> 00:21:25,018
Josh:
So there's like a lot, a lot of people are messing around here. I don't know what it is.

366
00:21:25,384 --> 00:21:28,831
Josh:
I mean, OpenAI has openly confirmed that it is not them, but I don't know if

367
00:21:28,831 --> 00:21:31,801
Josh:
I believe it. Like I'm just, at this point, whose device is this?

368
00:21:31,801 --> 00:21:32,826
Josh:
This is clearly a device.

369
00:21:33,023 --> 00:21:36,065
Josh:
We have Palmer Luckey, we have the actor guy, we have Joe Gebbia,

370
00:21:36,291 --> 00:21:38,741
Josh:
and they're all using the same device. And I have no idea what it is,

371
00:21:38,741 --> 00:21:42,241
Josh:
but it looks like this new AI first device and it's everything that was rumored

372
00:21:42,514 --> 00:21:45,515
Josh:
OpenAI2D. So that's the update on this story.

373
00:21:45,881 --> 00:21:48,601
Josh:
I don't know what it is. It's driving me crazy. If anyone wants to join the

374
00:21:48,601 --> 00:21:52,241
Josh:
mystery, please share a comment below. I am desperate to know what this mystery thing is.

375
00:21:52,566 --> 00:21:58,371
Ejaaz:
We do not make outreaches to insiders, but if there was ever a thing or moment

376
00:21:58,371 --> 00:22:00,331
Ejaaz:
to do it for, this is the thing.

377
00:22:00,331 --> 00:22:05,221
Ejaaz:
If you are someone listening to this show, if you're one of the 50 to 100,000

378
00:22:05,221 --> 00:22:08,241
Ejaaz:
people that listen to each episode nowadays, which is a lot of you,

379
00:22:08,241 --> 00:22:11,091
Ejaaz:
and you know something about this, please reach out to us. DM us on X.

380
00:22:12,413 --> 00:22:15,751
Ejaaz:
Just to satiate our curiosity, like we won't even mention it on the episode

381
00:22:15,751 --> 00:22:18,531
Ejaaz:
if you don't want us to, but like what on earth is going on here?

382
00:22:18,531 --> 00:22:19,681
Ejaaz:
I'm sufficiently confused.

383
00:22:19,902 --> 00:22:23,991
Ejaaz:
So we have Joe Gebbia, Airbnb, very respectable design guy, works for the US

384
00:22:23,991 --> 00:22:26,891
Ejaaz:
government doing this thing. We then have Alexander Skarsgård,

385
00:22:27,699 --> 00:22:32,013
Ejaaz:
Hollywood actor. What is this man doing with this thing? And now we have the

386
00:22:32,013 --> 00:22:36,623
Ejaaz:
American forlord, Paul Mulaki, also using the thing.

387
00:22:36,623 --> 00:22:37,717
Josh:
With no shoes on, I might add.

388
00:22:38,031 --> 00:22:41,143
Ejaaz:
With no shoes on, very, very rustic, very rural.

389
00:22:41,643 --> 00:22:45,333
Ejaaz:
I'm loving it. Honestly, it could be an aesthetic Instagram post from about

390
00:22:45,333 --> 00:22:49,947
Ejaaz:
a year ago if we were really getting into it. But let me comment on the device. I have some thoughts.

391
00:22:50,140 --> 00:22:54,473
Ejaaz:
Number one, those earbuds, they are much larger than I thought,

392
00:22:54,473 --> 00:22:57,893
Ejaaz:
unless Palmolaki has incredibly small ears.

393
00:22:57,893 --> 00:23:01,523
Ejaaz:
I do not know, but those earbuds look kind of constrictive around his ears.

394
00:23:01,523 --> 00:23:05,673
Ejaaz:
And the second thing I'll say is the phone he's holding is a pretty sizable

395
00:23:05,673 --> 00:23:09,363
Ejaaz:
phone. It's a big phone. It's arguably bigger than the iPhone Pro Max or whatever

396
00:23:09,363 --> 00:23:10,800
Ejaaz:
the hell is the latest phone.

397
00:23:11,112 --> 00:23:14,843
Ejaaz:
And that puck is massive. So yes, that could fit in your pocket,

398
00:23:14,843 --> 00:23:17,184
Ejaaz:
but it seems kind of heavy and kind of clunky.

399
00:23:17,416 --> 00:23:20,343
Ejaaz:
And I'm wondering like if I'm carrying my phone similar in the way that he's

400
00:23:20,343 --> 00:23:21,857
Ejaaz:
doing it, is that comfy? Is that cool?

401
00:23:22,563 --> 00:23:27,883
Ejaaz:
But I'm going to make a call right now, Josh. I think this is an open AI device.

402
00:23:28,103 --> 00:23:32,773
Ejaaz:
And I think open AI had to deny it for legal reasons before they kind of like,

403
00:23:32,773 --> 00:23:35,876
Ejaaz:
I don't know, whether it's an IPO thing or whether it was kind of like,

404
00:23:36,492 --> 00:23:39,719
Ejaaz:
too far away from their actual launch but i think this is an open area device

405
00:23:39,719 --> 00:23:41,025
Ejaaz:
and they're just teasing us

406
00:23:41,275 --> 00:23:45,122
Ejaaz:
all the way until an excruciating launch about three months from now in the

407
00:23:45,331 --> 00:23:50,879
Ejaaz:
actually you know six months from now in the early quarter of 2027 i think is

408
00:23:50,879 --> 00:23:51,572
Ejaaz:
when it's meant to come out.

409
00:23:51,914 --> 00:23:54,809
Josh:
I don't know hopefully soon i have no idea what it is if you know if anyone

410
00:23:54,809 --> 00:23:57,999
Josh:
has any conclusions please let us know i'm desperate to understand more there's

411
00:23:57,999 --> 00:24:02,489
Josh:
also another thing on the docket here glm 5.3 it says uh released someone released

412
00:24:02,489 --> 00:24:05,376
Josh:
a rogue version of What does that mean to be a rogue version?

413
00:24:05,666 --> 00:24:13,439
Ejaaz:
Yes. Okay. So you'll see this word here where it says, today we are releasing obliterated model V2.

414
00:24:14,879 --> 00:24:18,459
Ejaaz:
And what they're talking about here is they took the number one.

415
00:24:19,099 --> 00:24:22,839
Ejaaz:
I still think it's number one, Chinese open source model, GLM 5.3,

416
00:24:22,839 --> 00:24:26,477
Ejaaz:
from none other than cheapu, Josh's favorite word, cheapu.

417
00:24:26,715 --> 00:24:30,789
Ejaaz:
And it is an extremely capable model. But the one thing that it is very capable

418
00:24:30,789 --> 00:24:32,746
Ejaaz:
at is cybersecurity technology.

419
00:24:33,257 --> 00:24:36,878
Ejaaz:
Attacks, which, you know, isn't the greatest thing, but it's a very capable

420
00:24:36,878 --> 00:24:40,918
Ejaaz:
cybersecurity model. Now, it is held under its own model weights,

421
00:24:40,918 --> 00:24:45,412
Ejaaz:
which was just open-weighted or open-sourced about, I think, two weeks ago.

422
00:24:45,661 --> 00:24:48,773
Ejaaz:
And in those two weeks, this company, this rogue random team,

423
00:24:49,022 --> 00:24:53,518
Ejaaz:
has decided to stress test this model and remove all the safeguards that this

424
00:24:53,518 --> 00:24:57,718
Ejaaz:
model has. So, typically, if you get access to GLM 5.3, it has safeguards on,

425
00:24:57,718 --> 00:24:58,862
Ejaaz:
it won't do the bad thing.

426
00:24:59,222 --> 00:25:02,478
Ejaaz:
This team kind of took that open weights model, tweaked the model weights,

427
00:25:02,478 --> 00:25:06,578
Ejaaz:
and fine-tuned it to create a version of it that basically has no safeguards.

428
00:25:06,578 --> 00:25:09,378
Ejaaz:
Now, of course, that's the kind of side effect of open source.

429
00:25:09,378 --> 00:25:12,778
Ejaaz:
You are able to kind of experiment with these things. And the intention that

430
00:25:12,778 --> 00:25:17,038
Ejaaz:
this company says is you can use it to help defend your own company and your

431
00:25:17,038 --> 00:25:21,298
Ejaaz:
own software, which is, you know, an admirable goal. But I can imagine that

432
00:25:21,298 --> 00:25:24,212
Ejaaz:
some people might use this for other nefarious purposes.

433
00:25:24,572 --> 00:25:28,218
Ejaaz:
But the second thing I'll say is, this is accessible via API.

434
00:25:28,218 --> 00:25:32,368
Ejaaz:
Now, typically, with open source models, like you could take it and make it

435
00:25:32,368 --> 00:25:35,268
Ejaaz:
dangerous, but you would have to download and run it yourself.

436
00:25:35,268 --> 00:25:39,154
Ejaaz:
And GLM 5.3 models, this incredibly capable cybersecurity threat,

437
00:25:39,500 --> 00:25:42,967
Ejaaz:
is quite large. You would need to spend hundreds of thousands of dollars to be able to do that.

438
00:25:43,112 --> 00:25:46,758
Ejaaz:
No more. With this company, you can get access to it via API and wreak havoc.

439
00:25:46,758 --> 00:25:47,860
Ejaaz:
It's pretty insane to see.

440
00:25:48,226 --> 00:25:52,748
Josh:
And next up we have a narrative violation this does not fall in line with the

441
00:25:52,748 --> 00:25:53,748
Josh:
public sentiment talk to

442
00:25:53,748 --> 00:25:54,298
Ejaaz:
Me about the new.

443
00:25:54,298 --> 00:25:59,018
Josh:
School turns out turns out data centers aren't actually ruining the world who

444
00:25:59,018 --> 00:26:02,888
Josh:
would have thought we talked about water first then the almonds are way worse

445
00:26:02,888 --> 00:26:06,019
Josh:
hate the almond suppliers don't hate the ai people second

446
00:26:06,639 --> 00:26:10,082
Josh:
You want to remove property taxes? Get a data center in town.

447
00:26:10,296 --> 00:26:11,834
Josh:
Columbia County, Georgia.

448
00:26:12,066 --> 00:26:15,928
Josh:
It's nearing a deal with Google that will nearly double its $25 million property

449
00:26:15,928 --> 00:26:20,058
Josh:
tax revenues to $40 million, allowing the county to end its residential property

450
00:26:20,058 --> 00:26:22,075
Josh:
tax thanks to massive data centers.

451
00:26:22,273 --> 00:26:25,946
Josh:
Also, another fun side quest item that I came across this week.

452
00:26:26,155 --> 00:26:30,748
Josh:
You know the most wealthy country or the most wealthy county in the country

453
00:26:31,268 --> 00:26:35,458
Josh:
is actually the same county, which in Virginia, that has the most amount of

454
00:26:35,458 --> 00:26:36,540
Josh:
data centers per capita.

455
00:26:36,749 --> 00:26:40,347
Josh:
In it uh turns out like data centers actually make a ton of money

456
00:26:40,655 --> 00:26:44,178
Josh:
and all of the money that we talk about being spent on this goes through data

457
00:26:44,178 --> 00:26:47,928
Josh:
centers those data centers have to pay a lot of money and they have to incentivize

458
00:26:47,928 --> 00:26:49,084
Josh:
the people to want to have them there

459
00:26:49,385 --> 00:26:53,388
Josh:
and this is another example of like hey you have a data center town it's kind

460
00:26:53,388 --> 00:26:57,318
Josh:
of quiet but it provides a lot of jobs and it provides a lot of subsidies and

461
00:26:57,318 --> 00:26:58,308
Josh:
you don't pay property tax

462
00:26:58,610 --> 00:27:01,358
Josh:
and you're responsible for generating the tokens that will hopefully make the

463
00:27:01,358 --> 00:27:04,108
Josh:
world a much better place so this was a fun feel-good story i was like oh all

464
00:27:04,108 --> 00:27:05,134
Josh:
right nice another data point

465
00:27:05,506 --> 00:27:07,828
Josh:
where the data centers aren't quite ending the world. Pretty cool.

466
00:27:08,147 --> 00:27:11,178
Ejaaz:
Those are blue collar jobs, by the way. Like I think the whole narrative around

467
00:27:11,178 --> 00:27:15,048
Ejaaz:
like AI taking away jobs, they're actually one, not taking away software engineering

468
00:27:15,048 --> 00:27:18,728
Ejaaz:
jobs, they're adding more. And two, they're adding even more blue collar jobs,

469
00:27:18,728 --> 00:27:22,660
Ejaaz:
which is actually really beneficial for the states that these data centers are being set up.

470
00:27:22,944 --> 00:27:26,118
Ejaaz:
And then the other thing I'll say is like, listen, this is great for Georgia

471
00:27:26,118 --> 00:27:29,518
Ejaaz:
getting, you know, whatever, $25 million worth of property tax revenues and

472
00:27:29,518 --> 00:27:32,969
Ejaaz:
like, you know, great for the residents over there. But it's happening in every single state.

473
00:27:33,178 --> 00:27:37,026
Ejaaz:
And if it's not coming in the form of property tax reimbursement.

474
00:27:37,153 --> 00:27:41,458
Ejaaz:
It's coming in the form of new schools, new hospitals, brand new equipment that

475
00:27:41,458 --> 00:27:42,558
Ejaaz:
makes your lives better.

476
00:27:42,558 --> 00:27:46,378
Ejaaz:
So I don't want to sound like a propagandist as I'm listening to myself right

477
00:27:46,378 --> 00:27:50,148
Ejaaz:
now, but like data centers aren't all bad. They aren't all great either.

478
00:27:50,148 --> 00:27:53,193
Ejaaz:
And I'm addressing that they are issues that need to be solved, but

479
00:27:53,558 --> 00:27:57,278
Ejaaz:
The hatred is kind of like overwhelming right now and it's unjustified.

480
00:27:57,278 --> 00:28:00,608
Josh:
You know what issue needs to be solved? Like getting more energy in these freaking data centers.

481
00:28:00,608 --> 00:28:03,868
Ejaaz:
That's the problem. Slant PowerShell, LPS. Let's figure out,

482
00:28:03,868 --> 00:28:06,190
Ejaaz:
like let's upgrade our transformer system. My God.

483
00:28:06,405 --> 00:28:10,438
Ejaaz:
Anyway, moving away from power and transformers, can we talk about electrics

484
00:28:10,438 --> 00:28:11,988
Ejaaz:
for a bit? Because there is an update.

485
00:28:12,708 --> 00:28:14,588
Josh:
There's an update from RoboTaxi.

486
00:28:14,588 --> 00:28:20,923
Ejaaz:
I know, there's too many. Listen, I am a simple man. I look on my screen. I see two things.

487
00:28:21,219 --> 00:28:25,108
Ejaaz:
Really cheap prices, which is going to cost me kind of like a fraction of an

488
00:28:25,108 --> 00:28:26,200
Ejaaz:
Uber cost, which is amazing.

489
00:28:26,565 --> 00:28:29,631
Ejaaz:
And then number two, Terry's Black's, Terry Black's Barbecue.

490
00:28:29,776 --> 00:28:32,632
Ejaaz:
A phenomenal spot. If you're in Austin, Texas, fantastic.

491
00:28:32,812 --> 00:28:35,440
Ejaaz:
Go and visit. Josh, what on earth is going on here, please?

492
00:28:35,661 --> 00:28:40,108
Josh:
So by the time you're watching this, the Cyber Cab will have officially launched

493
00:28:40,108 --> 00:28:42,841
Josh:
in Austin, Texas. They will have had the launch event. They are going to,

494
00:28:43,114 --> 00:28:47,665
Josh:
for the first time ever, distribute the Cyber Cabs across the entire county or city of Austin.

495
00:28:47,944 --> 00:28:50,388
Josh:
This is the first time they're doing this with the Cyber Cab.

496
00:28:50,388 --> 00:28:52,774
Josh:
They've been using robo taxis that are serving as Model Ys.

497
00:28:53,058 --> 00:28:55,206
Josh:
Now we get those little gold things that are rolling around.

498
00:28:55,491 --> 00:28:58,888
Josh:
Luke, who everyone knows who helps out with the show, he lives in Austin and

499
00:28:58,888 --> 00:29:01,848
Josh:
he was describing what they look like in person. He's been seeing these all

500
00:29:01,848 --> 00:29:05,876
Josh:
over town and he described it as these small little cars with the mullet on the back.

501
00:29:06,212 --> 00:29:10,498
Josh:
They look adorable. But the reason why is because the efficiency of these vehicles

502
00:29:10,498 --> 00:29:13,678
Josh:
exceeds that of any other car ever made in history.

503
00:29:13,979 --> 00:29:17,948
Josh:
It is a two-seater. It has a very aerodynamically efficient design.

504
00:29:17,948 --> 00:29:22,888
Josh:
And for the nerds who drive Teslas, you know, generally your tesla drives excuse

505
00:29:22,888 --> 00:29:28,503
Josh:
me at around 250 to 350 kilowatt hours per mile the cyber cab is 150 it's built for

506
00:29:29,154 --> 00:29:33,423
Josh:
Efficiency. It's built for low cost. Therefore, the cost per mile will be lower

507
00:29:33,423 --> 00:29:34,882
Josh:
because the cost of the vehicle will be lower.

508
00:29:35,120 --> 00:29:39,263
Josh:
The cost per mile will be lower. And I find it impossibly challenging for anybody

509
00:29:39,263 --> 00:29:41,772
Josh:
to compete once these cybercabs are rolled out at scale.

510
00:29:42,010 --> 00:29:45,760
Josh:
This is monumental because it's the first time it's rolled out in a city fully autonomously.

511
00:29:46,010 --> 00:29:49,093
Josh:
As soon as the legislation passes to make it more broadly available,

512
00:29:49,342 --> 00:29:50,399
Josh:
they will be able to do so.

513
00:29:50,642 --> 00:29:54,713
Josh:
And this is the beginning of the cybercab takeover. And I think this is a one-way

514
00:29:54,713 --> 00:29:56,513
Josh:
street in which cybercabs start

515
00:29:56,513 --> 00:29:59,687
Josh:
entering more and more and more cities and nobody can compete on cost.

516
00:30:00,337 --> 00:30:03,367
Josh:
This is almost half the price of Uber and you don't have to tip.

517
00:30:03,977 --> 00:30:07,503
Josh:
And the cars are cleaner. They're more futuristic. They're far safer than a

518
00:30:07,503 --> 00:30:09,369
Josh:
human driving. It is a win, win, win, win, win.

519
00:30:09,636 --> 00:30:14,033
Josh:
And this is the beginning of a very long duration kind of tidal wave that I

520
00:30:14,033 --> 00:30:16,237
Josh:
think a lot of the world isn't expecting to come as soon as it will.

521
00:30:16,684 --> 00:30:22,053
Ejaaz:
I wish this was present in New York City and I wish that state regulations would

522
00:30:22,053 --> 00:30:25,683
Ejaaz:
stop banning cool stuff like this. I mean, this is a car that doesn't have a

523
00:30:25,683 --> 00:30:27,203
Ejaaz:
steering wheel, for goodness sake.

524
00:30:27,583 --> 00:30:30,543
Ejaaz:
I would love to jump in that thing and see what this technology can do.

525
00:30:30,543 --> 00:30:32,512
Ejaaz:
And it's well for a fraction of the cost.

526
00:30:32,808 --> 00:30:35,943
Ejaaz:
I'm looking forward to these things kind of becoming way more prevalent in every

527
00:30:35,943 --> 00:30:38,079
Ejaaz:
single state in the US, first at least.

528
00:30:38,703 --> 00:30:41,068
Ejaaz:
But moving on, we have officially, folks,

529
00:30:41,486 --> 00:30:47,657
Ejaaz:
reached peak AI bubble. We now have AI in your toothbrush.

530
00:30:47,877 --> 00:30:52,260
Ejaaz:
James Dyson, founder of none other than Dyson itself, 79 years old,

531
00:30:52,556 --> 00:30:56,562
Ejaaz:
had a keynote where he pitched his brand new product. It wasn't a GPU.

532
00:30:56,846 --> 00:31:01,101
Ejaaz:
It wasn't a new data center. It wasn't even an AI model. It was a toothbrush

533
00:31:01,357 --> 00:31:06,790
Ejaaz:
that has an AI powered camera jet and it only costs $500. But let me tell you why it does.

534
00:31:07,005 --> 00:31:10,536
Ejaaz:
Now, the problem that people are trying to fix here is you brush your teeth,

535
00:31:10,536 --> 00:31:13,186
Ejaaz:
people take around, whatever, 45 seconds, they should take two minutes,

536
00:31:13,186 --> 00:31:15,936
Ejaaz:
but they don't. And then flossing, well, that's not even a thing.

537
00:31:15,936 --> 00:31:18,847
Ejaaz:
Like a second thing to do after brushing your teeth, I'm not going to be doing that.

538
00:31:19,097 --> 00:31:26,186
Ejaaz:
But what if you had a toothbrush with a camera and a jet inside of it that scans,

539
00:31:26,186 --> 00:31:29,076
Ejaaz:
it takes around 500 pictures of your entire mouth whilst you're brushing your

540
00:31:29,076 --> 00:31:33,463
Ejaaz:
teeth, identifies where the plaque is, and then shoots water whilst you're brushing your teeth

541
00:31:33,782 --> 00:31:40,586
Ejaaz:
to clean the plaque away and save you the trips to the dentist for infinite amount of times.

542
00:31:40,586 --> 00:31:44,606
Ejaaz:
I just think this is a crazy, crazy gizmo device. And the more I read about

543
00:31:44,606 --> 00:31:46,236
Ejaaz:
it, the more I want to buy one.

544
00:31:46,236 --> 00:31:49,766
Josh:
I am obsessed with this product. And that is by no means a recommendation to

545
00:31:49,766 --> 00:31:52,004
Josh:
spend $500 on a toothbrush because that is insane.

546
00:31:52,312 --> 00:31:56,396
Josh:
But I haven't seen this level of care and obsession over design since the old

547
00:31:56,396 --> 00:31:59,452
Josh:
school Steve Jobs era. And that's what gets me really excited.

548
00:31:59,667 --> 00:32:02,418
Josh:
There's an eight minute video online of James Dyson talking about this.

549
00:32:02,627 --> 00:32:04,206
Josh:
James Dyson now is 79 years old.

550
00:32:04,421 --> 00:32:07,886
Josh:
He is one of the most incredible designers alive right now. And he describes

551
00:32:07,886 --> 00:32:13,646
Josh:
the kind of vitriol, the tribes and tribulations of what it takes to make the

552
00:32:13,646 --> 00:32:14,713
Josh:
best toothbrush in the world.

553
00:32:14,905 --> 00:32:19,576
Josh:
And he describes the sensors and the vacuum sealed chamber and how difficult

554
00:32:19,576 --> 00:32:22,428
Josh:
it was to accomplish all these amazing things. And it's so overkill.

555
00:32:22,805 --> 00:32:26,306
Josh:
Nobody needs this, but it is such a beautiful device. It is such a thoughtful

556
00:32:26,306 --> 00:32:30,336
Josh:
device. It's really cool to see people who still care about the craft of industrial

557
00:32:30,336 --> 00:32:32,237
Josh:
design, like James Dyson does.

558
00:32:32,551 --> 00:32:35,686
Josh:
It's like whenever this guy stands on stage, I know whatever he's going to announce.

559
00:32:35,686 --> 00:32:37,467
Josh:
He's spent years thinking about it. They've spent

560
00:32:37,845 --> 00:32:40,400
Josh:
hundreds of millions of dollars researching this thing and they are going to

561
00:32:40,400 --> 00:32:44,040
Josh:
produce the best of it in the world and that is what we got with this toothbrush

562
00:32:44,040 --> 00:32:47,530
Josh:
so i mean i'm probably not going to spend 500 on a toothbrush but man it is

563
00:32:47,530 --> 00:32:48,770
Josh:
so beautiful i just want to look at

564
00:32:48,770 --> 00:32:52,560
Ejaaz:
It you and i both know you are going to spend 500 bucks and you also know that

565
00:32:52,560 --> 00:32:56,910
Ejaaz:
i'm also going to spend 500 bucks on it so let's just kind of like put our hands up there i mean.

566
00:32:56,910 --> 00:32:59,695
Josh:
Man if you're thinking of a christmas present for me i'm not going to say no

567
00:33:00,020 --> 00:33:06,405
Ejaaz:
Okay okay i'll heavily consider it that's for sure um moving on last two things.

568
00:33:06,719 --> 00:33:08,588
Josh:
Um tim cook good stuff dude

569
00:33:08,808 --> 00:33:16,050
Ejaaz:
Tim cook the goat uh cp ceo former ceo of uh of apple has stepped down officially

570
00:33:16,050 --> 00:33:19,160
Ejaaz:
uh this is not new news we announced this about a couple months ago whenever

571
00:33:19,160 --> 00:33:23,083
Ejaaz:
the news broke uh but his successor john turnus who is

572
00:33:23,396 --> 00:33:27,180
Ejaaz:
a hardware expert he's been at apple for around 27 years and he's been uh,

573
00:33:27,930 --> 00:33:32,750
Ejaaz:
very uh present in the design and manufacturing of most of Apple's hardware

574
00:33:32,750 --> 00:33:36,903
Ejaaz:
devices, including the iPad, the iWatch, and a number of the phone devices,

575
00:33:37,159 --> 00:33:41,100
Ejaaz:
for the last 18 years at least, has officially taken the helm.

576
00:33:41,373 --> 00:33:47,070
Ejaaz:
And he introduces himself in one single word, hello, which got around 85 million

577
00:33:47,070 --> 00:33:48,693
Ejaaz:
views and impressions on X.

578
00:33:48,914 --> 00:33:52,230
Josh:
Talk to me about this. There's New King in charge. He is coming in hot with

579
00:33:52,230 --> 00:33:53,660
Josh:
the unveiling event next week.

580
00:33:53,660 --> 00:33:57,090
Josh:
Next week, Wednesday, 1 p.m. Eastern Standard Time, is the new

581
00:33:57,430 --> 00:34:00,550
Josh:
Apple hardware event where John turnis will come on stage reveal himself as

582
00:34:00,550 --> 00:34:04,738
Josh:
new ceo and then share the new hardware that is going to be released we're expecting

583
00:34:04,907 --> 00:34:08,700
Josh:
a foldable iphone which i'm very excited about a new 18 pro there's a whole

584
00:34:08,700 --> 00:34:11,971
Josh:
bunch of stuff going on one of the things though that i saw that i was most excited about is

585
00:34:12,743 --> 00:34:16,312
Josh:
iOS 27 has a handoff feature. Because one of the things that I've been deciding

586
00:34:16,312 --> 00:34:18,782
Josh:
in my head, and I don't know about the toothbrush, but one thing I will be getting

587
00:34:18,782 --> 00:34:19,785
Josh:
is both of these iPhones.

588
00:34:19,988 --> 00:34:23,442
Josh:
And how do you decide between a foldable phone with two cameras or a professional

589
00:34:23,442 --> 00:34:24,311
Josh:
phone with three cameras?

590
00:34:24,642 --> 00:34:28,342
Josh:
You don't. You get both. And there is a new feature built into iOS 27 that allows

591
00:34:28,342 --> 00:34:32,102
Josh:
you to seamlessly hand off your phone number between devices so you can dual-wheel

592
00:34:32,102 --> 00:34:33,901
Josh:
devices on a singular plan.

593
00:34:34,249 --> 00:34:37,022
Josh:
That is very exciting to me. Quick shout out to Tim Cook also.

594
00:34:37,022 --> 00:34:39,636
Josh:
Like, oh my God, what an unbelievable run this man had.

595
00:34:39,752 --> 00:34:45,172
Josh:
Just a really decent guy who took an impossible challenge of following off Steve

596
00:34:45,172 --> 00:34:50,892
Josh:
Jobs and did a really remarkable job of stewarding the company over 15 years

597
00:34:50,892 --> 00:34:54,027
Josh:
into what was just the most valuable company on earth.

598
00:34:54,148 --> 00:34:56,837
Josh:
So I'm excited for this new era. I'm grateful for the previous one.

599
00:34:57,110 --> 00:34:58,822
Josh:
And you know, hopefully we'll get some more hardware innovation.

600
00:34:58,822 --> 00:35:02,292
Josh:
We'll get Apple intelligence. We'll get new hardware. We'll get the glasses

601
00:35:02,292 --> 00:35:04,969
Josh:
we've always wanted. I'm excited for this new era of hardware at Apple.

602
00:35:05,556 --> 00:35:09,660
Ejaaz:
So to summarize this week, We have around four to five new models.

603
00:35:09,799 --> 00:35:12,207
Ejaaz:
A lot of them are cheap, but super, super intelligent.

604
00:35:12,457 --> 00:35:16,662
Ejaaz:
We have a brand new world model. We have a new toothbrush, which costs around

605
00:35:16,662 --> 00:35:19,070
Ejaaz:
500 bucks, which will solve all your dental problems.

606
00:35:19,353 --> 00:35:23,502
Ejaaz:
And Josh is going to buy one, two, three, four, maybe five new devices in the

607
00:35:23,502 --> 00:35:27,392
Ejaaz:
space of between now and the end of the year, which will maybe put him in bankruptcy.

608
00:35:27,392 --> 00:35:29,697
Ejaaz:
We don't know, but he'll be able to switch between different phones.

609
00:35:30,016 --> 00:35:33,182
Ejaaz:
Apple is going to release so much cool hardware.

610
00:35:33,182 --> 00:35:37,102
Ejaaz:
OpenAI is still going to tease us continually for the foreseeable future until

611
00:35:37,102 --> 00:35:39,242
Ejaaz:
they release this damn thing in 2027.

612
00:35:39,242 --> 00:35:43,662
Ejaaz:
And I can't wait for any and all of it. If you have listened to this, this is the roundup.

613
00:35:43,662 --> 00:35:46,922
Ejaaz:
You're done. You have caught up on anything and everything. And if you've missed

614
00:35:46,922 --> 00:35:49,522
Ejaaz:
some of our earlier episodes this week, you should probably go check that out.

615
00:35:49,522 --> 00:35:52,667
Ejaaz:
The Hugging Face incident was pretty insane. And we did an in-depth coverage

616
00:35:52,812 --> 00:35:55,494
Ejaaz:
of that episode, within that episode, rather.

617
00:35:56,144 --> 00:35:58,922
Ejaaz:
And if you haven't shared this with a friend yet, please do.

618
00:35:58,922 --> 00:36:01,241
Ejaaz:
So if you aren't subscribed to us, if you haven't left us a comment,

619
00:36:01,438 --> 00:36:03,203
Ejaaz:
we love hearing from you guys. please let us know.

620
00:36:03,752 --> 00:36:07,682
Ejaaz:
Turn on notifications. It helps us out massively. And I, Josh, is there anything else?

621
00:36:07,682 --> 00:36:09,699
Josh:
Can we take like a collective deep breath? We've made it through like,

622
00:36:11,429 --> 00:36:15,938
Josh:
ah you have made it through the week the crazy week in which all these models

623
00:36:15,938 --> 00:36:18,658
Josh:
have released all this stuff has happened you could go enjoy your weekend now

624
00:36:18,658 --> 00:36:22,388
Josh:
enjoy go it's beautiful outside it's the end of summer fall is coming get some

625
00:36:22,388 --> 00:36:26,288
Josh:
time touching grass and then right back in this seat next week lock in

626
00:36:26,840 --> 00:36:29,868
Josh:
it's go time the train doesn't stop but for now it does you're fully caught

627
00:36:29,868 --> 00:36:32,488
Josh:
up thank you so much for watching and yeah we'll see you guys in the next one