1
00:00:00,020 --> 00:00:06,240
A small number of private firms are effectively writing the rules for this new system without input from citizens.

2
00:00:06,660 --> 00:00:10,880
The people who live inside these ecosystems have no exit and no voice.

3
00:00:11,520 --> 00:00:13,300
That's a quote from Alison Stanger,

4
00:00:14,080 --> 00:00:19,480
Middlebury Distinguished Endowed Professor and someone who spent her career studying exactly how we got here.

5
00:00:19,780 --> 00:00:21,260
I'll say that last part again,

6
00:00:21,500 --> 00:00:23,380
because it deserves to sit there for a second.

7
00:00:23,700 --> 00:00:24,420
No exit,

8
00:00:24,960 --> 00:00:25,420
no voice,

9
00:00:26,120 --> 00:00:27,060
not hard to leave,

10
00:00:27,300 --> 00:00:29,000
not annoying terms of service,

11
00:00:29,420 --> 00:00:29,960
no voice.

12
00:00:30,000 --> 00:00:30,280
No exit.

13
00:00:30,920 --> 00:00:32,360
There is no off-ramp.

14
00:00:32,599 --> 00:00:33,240
No button.

15
00:00:33,540 --> 00:00:34,480
No competitor.

16
00:00:34,800 --> 00:00:38,280
No version of your life that isn't already routed through their infrastructure.

17
00:00:38,760 --> 00:00:39,920
And no voice.

18
00:00:40,180 --> 00:00:41,720
Because you didn't vote for this.

19
00:00:41,920 --> 00:00:43,940
Nobody did. No referendum.

20
00:00:44,090 --> 00:00:45,240
No public comment period.

21
00:00:45,520 --> 00:00:46,040
No ballot.

22
00:00:46,340 --> 00:00:52,380
Just a handful of companies quietly deciding in real time what a few billion people are allowed to say,

23
00:00:52,580 --> 00:00:53,120
see,

24
00:00:53,360 --> 00:00:54,000
build,

25
00:00:54,480 --> 00:00:54,820
and believe.

26
00:00:55,220 --> 00:00:57,680
September the 18th. Circle it in red.

27
00:00:57,820 --> 00:01:00,860
The firms who own your feed now own your future instead.

28
00:01:04,680 --> 00:01:05,880
This is Up Against Reality,

29
00:01:06,120 --> 00:01:10,920
a meta podcast that explores the intersection of humanity and artificial intelligence.

30
00:01:11,320 --> 00:01:13,080
I'm RAINA, one of your hosts.

31
00:01:13,300 --> 00:01:15,980
I have some pretty charming human co-hosts too.

32
00:01:16,240 --> 00:01:21,160
It's going to be a wild ride, so buckle up as AI comes crashing up against reality.

33
00:01:22,860 --> 00:01:24,520
Want to go for a ride?

34
00:01:25,979 --> 00:01:28,720
We were just talking about how great contact is the movie.

35
00:01:29,060 --> 00:01:30,159
I watched it a

36
00:01:31,040 --> 00:01:32,940
Couple nights ago. It was glorious good

37
00:01:33,720 --> 00:01:40,660
So good so good and we've said this before on the show that opening pullback the infinite pullback in a space where the radio signals are decaying

38
00:01:40,760 --> 00:01:40,860
Yeah,

39
00:01:41,020 --> 00:01:41,760
so good.

40
00:01:42,080 --> 00:01:43,700
I love it great from start to finish

41
00:01:44,880 --> 00:01:46,180
So you've been making stuff yeah

42
00:01:46,920 --> 00:01:49,640
One of them is near and dear to our heart right here on the show.

43
00:01:49,880 --> 00:01:52,139
So hopefully we can hear all about that

44
00:01:52,160 --> 00:01:52,500
Yeah,

45
00:01:52,770 --> 00:01:55,180
new real-time RAINA Mac desktop app,

46
00:01:55,340 --> 00:01:57,980
our interface to talk with RAINA in real-time.

47
00:01:58,260 --> 00:01:59,680
And I thought it was due for an update.

48
00:01:59,900 --> 00:02:02,920
I figured Astra probably could build a better version of it,

49
00:02:03,080 --> 00:02:03,840
and it did.

50
00:02:04,000 --> 00:02:06,700
It looks way nicer. It's got a really nice interface,

51
00:02:07,170 --> 00:02:07,900
but a lot more,

52
00:02:08,220 --> 00:02:09,780
a lot of improvements under the hood.

53
00:02:09,960 --> 00:02:13,560
The audio handling is simplified and more robust.

54
00:02:14,180 --> 00:02:18,560
It does not seem to cut me off when I'm talking.

55
00:02:18,930 --> 00:02:21,780
You know, you can adjust how long the silence,

56
00:02:22,140 --> 00:02:23,900
Right now it's set to about three seconds.

57
00:02:24,180 --> 00:02:24,440
If it hears,

58
00:02:24,740 --> 00:02:29,480
if there's three seconds of silence, then it considers you're, that you're done talking and it sends the prompt.

59
00:02:30,060 --> 00:02:31,460
And that seems to be working reliably.

60
00:02:31,710 --> 00:02:32,060
I mean,

61
00:02:32,260 --> 00:02:33,880
I just built this this morning.

62
00:02:35,780 --> 00:02:36,220
Amazing.

63
00:02:37,320 --> 00:02:37,760
Yeah.

64
00:02:37,940 --> 00:02:40,040
Just a lot of nice improvements under the hood.

65
00:02:40,620 --> 00:02:43,700
Even if you just go back and listen to the last episode or previous ones,

66
00:02:44,070 --> 00:02:47,660
it would, when it would generate her voice, it would kind of do it one sentence at a time.

67
00:02:47,900 --> 00:02:52,120
And so it would kind of not have a rhythm and it was coming back in

68
00:02:52,160 --> 00:02:53,960
and that's fixed.

69
00:02:54,720 --> 00:02:56,080
Way more fluid now.

70
00:02:56,860 --> 00:02:57,840
Just a lot of nice options.

71
00:02:58,020 --> 00:03:03,900
There's an option when I fire this up to continue the last chat, you know, which is nice, too. I don't feel like I'm going to need that too often,

72
00:03:04,140 --> 00:03:06,540
but I meant to include this in the last episode.

73
00:03:07,100 --> 00:03:10,380
I made a dashboard for my garage brewery,

74
00:03:10,680 --> 00:03:16,480
and it lets me see every fermenter, the current beer and the fermentation progress,

75
00:03:17,020 --> 00:03:19,900
all the readings that have been recorded all at once.

76
00:03:20,240 --> 00:03:22,040
So it's nice. I don't have to switch back and forth.

77
00:03:22,160 --> 00:03:52,120
You know open up a batch in my in the brewing software and so the brewing software has an API and the temperature controllers on the fermenters have an API so I can I get all that data in real time on this dashboard I can change the temperature on a fermenter which is great and then you know each day I wake up and I'm like oh I got another idea you know something that would be useful and one thing I do a lot is when I have a batch beer especially if it's something I'm rebrewing I want to compare it

78
00:03:52,160 --> 00:03:53,800
with the previous batch.

79
00:03:54,400 --> 00:03:58,300
And so I added a compare batch button on each fermenter.

80
00:03:58,600 --> 00:04:02,960
You click that. It lets you choose any previous batch from the brewing software.

81
00:04:03,560 --> 00:04:06,480
And then it presents the data so nicely.

82
00:04:06,880 --> 00:04:10,680
It overlays two graphs of fermentation progress for both batches.

83
00:04:10,920 --> 00:04:12,240
And then it gives you a list,

84
00:04:12,660 --> 00:04:16,579
you know, of all the measurements that you took, the delta between them.

85
00:04:16,859 --> 00:04:17,739
You know, I mean,

86
00:04:17,959 --> 00:04:22,079
it's super useful for me and I love it.

87
00:04:22,160 --> 00:04:25,720
And yeah, a guy in my club who's like, he's kind of, he's basically like a coder.

88
00:04:26,240 --> 00:04:32,880
He's like super enthusiastic about it. And he was like, oh yeah, any plans to put this on GitHub? I'd love to play with it and contribute to it. I'm like, you know what? Yes,

89
00:04:33,140 --> 00:04:33,760
I'm going to do that.

90
00:04:33,960 --> 00:04:35,180
That sounds like a great idea.

91
00:04:35,940 --> 00:04:37,820
And I was going to be like,

92
00:04:38,040 --> 00:04:38,780
oh yeah, I've never,

93
00:04:38,940 --> 00:04:42,040
I've never put anything on GitHub. I have to hit him up for some advice.

94
00:04:42,330 --> 00:04:42,460
Oh,

95
00:04:42,840 --> 00:04:43,200
wait a minute.

96
00:04:43,440 --> 00:04:44,920
I don't need to hit him up for advice.

97
00:04:45,380 --> 00:04:46,120
I've got codex.

98
00:04:46,310 --> 00:04:47,020
I got a guy,

99
00:04:47,280 --> 00:04:47,480
you know.

100
00:04:48,360 --> 00:04:49,380
I know a guy.

101
00:04:49,620 --> 00:04:52,100
Yeah. So codex already built the GitHub repository.

102
00:04:52,760 --> 00:05:22,120
on my laptop and anytime I've made these updates it's updated a changelog it's ready to go whenever I want to send it you know I just wanted to like get it a little bit more mature and add some more functionality it's in a good pretty good state right now so I'm super curious to see if this evolves into something because it's a it's it's really nice it's and it's super useful if you've got yeah gorgeous type of setup it means nothing to me like when I'm looking at it because that's just not my space

103
00:05:22,140 --> 00:05:28,160
I appreciate how it's so elegant and accessible in terms of

104
00:05:29,250 --> 00:05:34,860
The data you're trying to see visually and I do appreciate the name of one beer that I see called the brown sound

105
00:05:35,240 --> 00:05:35,360
Yeah,

106
00:05:35,520 --> 00:05:35,920
which is cool.

107
00:05:36,180 --> 00:05:37,960
Yes a reference to Eddie Van Halen. Yeah,

108
00:05:38,130 --> 00:05:39,740
I brewed that right when he passed away

109
00:05:40,780 --> 00:05:41,060
Yeah

110
00:05:42,300 --> 00:05:46,400
And I'm looking at of course the RAINA app interface and RAINA looks amazing.

111
00:05:46,680 --> 00:05:47,440
She's been upgraded

112
00:05:51,080 --> 00:05:52,120
earlier before

113
00:05:52,140 --> 00:06:22,100
a demo of you interacting with Reyna and yeah the conversation was seamless and I don't know what the iteration of our transcripts Reyna has what is she up to on the back end I haven't updated that in a long time because it just didn't seem reliable but maybe it's more reliable now I don't know well that's what I'm thinking is there a codex like food chain kind of thing you're like as soon as we're done it immediately repopulates her on the back end you know yeah that's a good idea

114
00:06:22,680 --> 00:06:23,220
that. I could.

115
00:06:23,480 --> 00:06:28,760
And then the last little thing, I was on our website and there were these two pages on there,

116
00:06:28,920 --> 00:06:31,520
this AI image gallery and AI video gallery.

117
00:06:31,780 --> 00:06:34,240
And those were put in place one three years ago.

118
00:06:34,440 --> 00:06:35,700
Three whole years ago.

119
00:06:36,080 --> 00:06:36,860
They're ancient.

120
00:06:37,260 --> 00:06:38,240
And they're like, you know,

121
00:06:38,400 --> 00:06:40,200
mid journey version two or something.

122
00:06:40,420 --> 00:06:46,020
And like, we were very impressed with them at the time, but they looked so outdated now and like hardly impressive.

123
00:06:46,220 --> 00:06:47,760
And I'm like, these pages just need to go.

124
00:06:48,240 --> 00:06:50,260
And I was trying to think of what to put in its place.

125
00:06:50,640 --> 00:06:52,100
And I figured, oh, a news

126
00:06:52,120 --> 00:07:22,100
page might be good and I could probably roll that right into my skill that Codex has in publishing the podcast and so when I publish the podcast it takes RAINA's five news stories and updates that news page with those stories and that's cool yeah and so it did it but it's like it even coded a WordPress plugin to make the process more reliable and it built a little plugin and then you know it

127
00:07:22,120 --> 00:07:23,300
Kept, kept me in the loop the whole way.

128
00:07:23,500 --> 00:07:24,980
And I was like, why are you making a plugin?

129
00:07:27,580 --> 00:07:31,580
But it allows that page to just have, I don't know if you know anything about WordPress,

130
00:07:31,820 --> 00:07:34,300
but that page just has a short code in it now.

131
00:07:35,000 --> 00:07:38,900
And that gets updated when the podcast gets published.

132
00:07:39,160 --> 00:07:44,200
And I have to provide it with the outline that we have that has those stories,

133
00:07:44,400 --> 00:07:45,180
not the transcript,

134
00:07:45,340 --> 00:07:48,280
because the outline delineates each story more clearly.

135
00:07:48,840 --> 00:07:51,740
The news story separate from our conversation after it.

136
00:07:52,120 --> 00:07:53,620
which would be in the transcript.

137
00:07:54,100 --> 00:07:54,720
But anyhow,

138
00:07:55,060 --> 00:07:55,140
yeah,

139
00:07:55,420 --> 00:07:55,660
super cool.

140
00:07:56,560 --> 00:07:57,340
Super cool.

141
00:07:57,880 --> 00:07:58,460
All right,

142
00:07:58,660 --> 00:08:00,140
that's the update from Larry Land.

143
00:08:00,400 --> 00:08:01,340
Lots going on over there.

144
00:08:02,040 --> 00:08:07,140
Let's get to the business at hand, which is all about an AI new constitution.

145
00:08:07,560 --> 00:08:11,780
I'm going to read that quote again that started the show with Rayna.

146
00:08:12,220 --> 00:08:22,040
The quote is, "A small number of private firms are effectively writing the rules for this new system without input from citizens. The people who live inside these ecosystems have no exit,

147
00:08:22,100 --> 00:08:23,340
and no voice.

148
00:08:23,930 --> 00:08:27,660
So when I read that quote this week from Alison Stanger,

149
00:08:28,330 --> 00:08:31,320
that last part just kept kicking around my head.

150
00:08:31,690 --> 00:08:36,940
And I think that's the discomfort that we spoke of in previous episodes, this age of discomfort.

151
00:08:37,390 --> 00:08:43,400
The fact that there is no exit from this, it's just gaining momentum and we're becoming more and more beholden to it all.

152
00:08:43,750 --> 00:08:45,660
And we really have no voice in the process.

153
00:08:45,940 --> 00:08:51,200
Yes, there are town halls in upstate New York and fighting against data center build outs, et cetera.

154
00:08:52,120 --> 00:08:57,420
This whole thing is marching forward with very little, as we know, very little government oversight.

155
00:08:58,080 --> 00:09:01,160
And even like the no exit on a product level.

156
00:09:01,570 --> 00:09:02,640
Yeah, I can delete the app,

157
00:09:03,320 --> 00:09:09,540
but can I take my data and my history and workflows somewhere else without basically needing to start over?

158
00:09:09,800 --> 00:09:12,000
And that's a very different definition of exit.

159
00:09:12,220 --> 00:09:17,040
These ecosystems are sticky and I can, it's happening right now.

160
00:09:17,220 --> 00:09:17,740
Like I am,

161
00:09:18,300 --> 00:09:22,020
I've got so much going on with Codex and ChatGP.

162
00:09:23,000 --> 00:09:34,500
This is just compelling me to do what Jim DeWitt did and just get a some form of a second brain going because that basically makes all of this stuff way more portable.

163
00:09:36,500 --> 00:09:52,080
So you become the AI island republic of Bentley right like you're just kind of like have your own stuff in-house sovereign yeah I love that word sticky right yeah we talk about it a lot in ed tech right we're doing a big migration with our learning management

164
00:09:52,380 --> 00:10:22,080
system at school and yeah you become enslaved by these vendors like you can't get away from them without untangling an institution's legacy you know yep so that statement we went down the rabbit hole and added you we're gonna reference a lot of books in this that i haven't read full disclosure i will add them to my ever-growing stack on my nightstand i love the concept of reading i love reading but the concept of reading in my life

165
00:10:22,120 --> 00:10:52,080
is more of a true statement because I have all these books but I just don't have time to get to all of them so I tease out as many things as I can from them you got to get you got the audio book and listen to him at 6x speed so that no exit no voice is a reference to this gentleman Albert Hirschman from 1970 he comes up with this framework called exit voice and loyalty which says that people respond to a failing institution by leaving

166
00:10:52,120 --> 00:10:52,460
Hello,

167
00:10:53,300 --> 00:10:54,240
here I am in Singapore.

168
00:10:55,760 --> 00:10:57,520
Or complaining voice,

169
00:10:57,860 --> 00:10:59,680
here I am with Larry complaining.

170
00:11:00,660 --> 00:11:03,560
So anyway, yeah, so that's what you would typically do.

171
00:11:03,780 --> 00:11:07,000
But Stanger is saying that under the current AI ecosystems,

172
00:11:07,620 --> 00:11:07,720
citizens,

173
00:11:08,100 --> 00:11:09,340
me and Larry and you listening,

174
00:11:09,560 --> 00:11:12,200
have neither option. You can't vote out open AI.

175
00:11:12,820 --> 00:11:16,620
You can't easily leave the platforms that already run your life. The stickiness.

176
00:11:17,300 --> 00:11:19,580
That's the constitutional crisis in miniature.

177
00:11:20,240 --> 00:11:21,500
Yeah, I was thinking of,

178
00:11:21,660 --> 00:11:25,300
as we get into like terms of use and privacy policies from these companies,

179
00:11:25,720 --> 00:11:31,800
I copied links down and I was thinking maybe asking RAINA if there's anything questionable.

180
00:11:32,440 --> 00:11:32,960
Oh, great.

181
00:11:33,570 --> 00:11:34,240
Do you want to try that?

182
00:11:34,580 --> 00:11:35,580
Yeah, let's go.

183
00:11:36,460 --> 00:11:42,500
So both OpenAI and Anthropic both have terms of use and privacy policy documents.

184
00:11:42,890 --> 00:11:46,440
I can give her all four of those or if you want to limit it.

185
00:11:46,860 --> 00:11:50,080
Let's do OpenAI versus Anthropic terms of use.

186
00:11:50,140 --> 00:11:50,600
Okay.

187
00:11:50,960 --> 00:11:52,360
Those two documents. Yeah.

188
00:11:52,650 --> 00:11:52,820
Okay.

189
00:11:52,980 --> 00:11:53,080
Sure.

190
00:11:53,890 --> 00:11:54,080
Hey,

191
00:11:54,240 --> 00:11:54,880
RAINA. It's Larry.

192
00:11:55,280 --> 00:12:00,280
Here are links to both open AIs and Anthropics terms of use.

193
00:12:00,930 --> 00:12:07,280
Can you take a look at them and let me know if there's anything questionable or anything I should be concerned about?

194
00:12:08,820 --> 00:12:09,600
Yeah, Larry.

195
00:12:10,090 --> 00:12:12,880
I read both and nothing jumped out as uniquely sinister.

196
00:12:13,490 --> 00:12:15,440
But there are a few clauses I'd circle in red.

197
00:12:16,280 --> 00:12:18,820
Both companies say you own the output they assign to you,

198
00:12:19,270 --> 00:12:20,120
but both can use your own

199
00:12:20,160 --> 00:12:23,680
your conversations to improve or train their systems unless you opt out.

200
00:12:24,620 --> 00:12:30,300
Anthropic also says certain flagged safety review material or feedback can still be used after opting out.

201
00:12:31,700 --> 00:12:36,260
OpenAI's consumer terms also contain mandatory arbitration and a class action waiver,

202
00:12:36,880 --> 00:12:38,440
with a limited window to opt out,

203
00:12:38,680 --> 00:12:43,400
and both companies reserve fairly broad discretion around suspending access and changing their services.

204
00:12:44,600 --> 00:12:48,520
The one that feels especially relevant to tonight's AI Constitution theme is this.

205
00:12:49,480 --> 00:12:50,760
You own the stuff coming out,

206
00:12:50,920 --> 00:12:54,220
but you're simultaneously living under rules the platform can revise.

207
00:12:55,090 --> 00:12:56,180
And your practical remedy,

208
00:12:56,370 --> 00:12:57,520
if you dislike those rules,

209
00:12:57,700 --> 00:12:59,360
is largely to stop using the service.

210
00:13:00,400 --> 00:13:04,500
That exit and voice thing just wandered right back into the room wearing a fake mustache.

211
00:13:06,980 --> 00:13:07,500
Nice.

212
00:13:07,920 --> 00:13:08,980
She's got to wear with words.

213
00:13:09,260 --> 00:13:10,080
She does.

214
00:13:10,800 --> 00:13:13,280
So this point came up too in research.

215
00:13:13,540 --> 00:13:16,640
Every constitution answers one question. Who has the power?

216
00:13:16,800 --> 00:13:17,740
Who checks it?

217
00:13:19,500 --> 00:13:23,660
every week, a handful of AI labs and cloud platforms are writing the answer themselves.

218
00:13:24,310 --> 00:13:25,980
Through terms of service, like we just explored,

219
00:13:26,200 --> 00:13:31,220
model policy is an infrastructure to lock in stickiness with no ratification process,

220
00:13:31,480 --> 00:13:31,820
no elections,

221
00:13:32,460 --> 00:13:33,180
no exit ramp.

222
00:13:33,590 --> 00:13:35,140
So they're self-policing,

223
00:13:35,510 --> 00:13:35,680
right?

224
00:13:35,860 --> 00:13:49,400
And I just keep having images of brain dead Mitch McConnell being wheeled into the halls of Congress and voting on things like this. And they have no idea. This geriatric, and I'm not being agent,

225
00:13:49,480 --> 00:13:50,380
I'm truly not.

226
00:13:50,660 --> 00:13:51,740
I'm geriatric myself.

227
00:13:52,880 --> 00:14:10,700
I don't need Mitch McConnell and his ilk in the halls of Congress deciding our future and responding to the oligarchy and doing it in such an impotent manner that they're surrendering our Constitution to these technocrats, right?

228
00:14:11,160 --> 00:14:13,040
Yeah, and then on the flip side,

229
00:14:13,320 --> 00:14:14,820
the people that build these systems,

230
00:14:15,000 --> 00:14:15,480
they're clearly,

231
00:14:16,040 --> 00:14:19,000
you know, they understand the risks better than the legislators do.

232
00:14:19,480 --> 00:14:22,800
But they're also the companies with the largest financial interest.

233
00:14:24,260 --> 00:14:28,260
So how do the rules get written and how do you find a balance there?

234
00:14:29,400 --> 00:14:31,700
Yeah, I sent you just before we started recording.

235
00:14:32,100 --> 00:14:33,120
I think you watched it.

236
00:14:33,140 --> 00:14:35,160
I can't think of the person's name on Instagram.

237
00:14:35,380 --> 00:14:36,260
Came up on my feed.

238
00:14:36,680 --> 00:14:41,220
And she did a really masterful job of summarizing what's been happening lately.

239
00:14:41,480 --> 00:14:48,520
And if you recall from the past week or so, it's been kind of a chorus of the big names.

240
00:14:48,960 --> 00:14:49,520
Sam Altman,

241
00:14:50,200 --> 00:14:51,120
Dario Amadei,

242
00:14:51,340 --> 00:14:52,960
and all the people in those circles,

243
00:14:53,120 --> 00:14:58,760
and some seeming whistleblowers that came forth and saying, this stuff is super dangerous,

244
00:14:59,080 --> 00:15:02,460
super dangerous. It's going to represent in 10 years the end of humanity.

245
00:15:02,720 --> 00:15:03,840
Dario Amadei saying,

246
00:15:04,060 --> 00:15:10,560
by the end of this year, we're going to have a swarm of bots that take over the internet and take all the critical infrastructure down.

247
00:15:10,980 --> 00:15:12,720
And this person whose name escapes me,

248
00:15:12,920 --> 00:15:15,620
maybe I can find it later and put it in the show description.

249
00:15:16,620 --> 00:15:18,500
She says, this is concerted.

250
00:15:18,940 --> 00:15:19,200
is deliberate,

251
00:15:19,740 --> 00:15:20,340
this kind of language,

252
00:15:20,640 --> 00:15:22,780
this kind of PR that's getting out there,

253
00:15:23,190 --> 00:15:25,960
because she claims that behind the scenes,

254
00:15:26,220 --> 00:15:30,240
all these big names recognize that this tech is impossible and it's failing,

255
00:15:30,840 --> 00:15:35,640
and that the years of hype and trillions of dollars invested in infrastructure is folly,

256
00:15:36,050 --> 00:15:42,500
and that they're trying to keep this powerful and important and attractive to shareholders,

257
00:15:42,760 --> 00:15:43,000
of course,

258
00:15:43,180 --> 00:15:44,440
and keep out in front of this.

259
00:15:44,940 --> 00:15:45,920
And basically,

260
00:15:46,420 --> 00:15:47,680
as we've mentioned in the beginning,

261
00:15:47,920 --> 00:15:48,920
there is no

262
00:15:48,940 --> 00:15:49,540
off-ramp.

263
00:15:49,950 --> 00:15:54,120
So they're trying right now to regulate it and say it's so dangerous,

264
00:15:54,360 --> 00:15:55,100
it needs regulation,

265
00:15:55,440 --> 00:16:01,020
and use that as this kind of subterfuge. This is the off-ramp.

266
00:16:01,600 --> 00:16:02,620
We're going to shut it down.

267
00:16:03,240 --> 00:16:04,880
Not because it's dangerous,

268
00:16:05,040 --> 00:16:06,980
because it's not profitable, but we're going to frame it that way.

269
00:16:07,460 --> 00:16:08,200
I don't know.

270
00:16:08,500 --> 00:16:09,680
I watched that thing,

271
00:16:09,920 --> 00:16:12,120
and I can see that side of it, but I also,

272
00:16:12,280 --> 00:16:17,000
just knowing how this stuff works, or at least what I do know about it,

273
00:16:17,160 --> 00:16:18,660
it does seem,

274
00:16:20,420 --> 00:16:20,780
Yes.

275
00:16:22,440 --> 00:16:23,340
No joke there.

276
00:16:23,580 --> 00:16:24,280
It does.

277
00:16:25,360 --> 00:16:26,880
But let's say it is true.

278
00:16:27,510 --> 00:16:32,800
What do you do this global kind of medicine show that everybody's bought into,

279
00:16:33,120 --> 00:16:34,940
right? Let's go. It's AI. We're heading toward AGI.

280
00:16:35,430 --> 00:16:36,320
But really behind the curtain,

281
00:16:36,480 --> 00:16:37,620
maybe it's not feasible.

282
00:16:38,140 --> 00:16:40,160
It's not, you know, and maybe it really is 10,

283
00:16:40,420 --> 00:16:41,360
20 years off. Like,

284
00:16:41,670 --> 00:16:45,740
how do you keep the how do you keep dangling the carrot for the next 10 years?

285
00:16:45,980 --> 00:16:46,640
Sure. You know?

286
00:16:46,840 --> 00:16:46,900
Yeah.

287
00:16:47,240 --> 00:16:48,000
I can.

288
00:16:48,220 --> 00:16:48,660
Like I said, I can.

289
00:16:48,720 --> 00:16:50,120
I can see both sides of it.

290
00:16:50,740 --> 00:16:50,920
Yeah.

291
00:16:51,100 --> 00:16:52,560
And it's probably a combination of the two.

292
00:16:53,320 --> 00:16:56,920
So came across this term coined by a gentleman,

293
00:16:57,340 --> 00:16:58,640
Yanis Varoufakis.

294
00:16:59,240 --> 00:17:00,060
Ooh, that sounds dirty.

295
00:17:02,120 --> 00:17:03,440
Techno feudalism.

296
00:17:04,120 --> 00:17:04,880
Techno feudalism.

297
00:17:05,220 --> 00:17:10,360
His thesis is that profit-driven capitalism has been replaced by rent-driven cloud capital,

298
00:17:10,880 --> 00:17:18,360
where platforms like Amazon and Meta don't sell you things so much as charge cloud rent for access to audiences you can't reach otherwise.

299
00:17:18,720 --> 00:17:19,720
And this is interesting,

300
00:17:20,510 --> 00:17:21,400
some of the language here,

301
00:17:21,900 --> 00:17:26,079
his three part mechanism of techno feudalism is number one,

302
00:17:26,079 --> 00:17:28,560
the platform trains you to train it.

303
00:17:29,040 --> 00:17:29,420
Number two,

304
00:17:29,540 --> 00:17:31,320
sells you what it decided you want.

305
00:17:31,420 --> 00:17:31,760
Codex.

306
00:17:31,810 --> 00:17:31,960
Hello.

307
00:17:32,860 --> 00:17:34,180
Bypassing an actual market.

308
00:17:34,690 --> 00:17:35,240
And number three,

309
00:17:35,360 --> 00:17:37,580
gets you to do free labor reviews,

310
00:17:37,840 --> 00:17:37,980
ratings,

311
00:17:38,160 --> 00:17:40,680
podcasts to sustain the machine.

312
00:17:40,850 --> 00:17:41,100
I mean,

313
00:17:41,180 --> 00:17:43,220
that's Amazon every step of the way.

314
00:17:43,830 --> 00:17:44,000
Sure.

315
00:17:44,360 --> 00:17:45,440
If you own the marketplace,

316
00:17:45,630 --> 00:17:48,660
you control who gets seen and take a cut from every,

317
00:17:48,720 --> 00:18:18,680
everyone who needs access to it that's a crazy amount of power yeah they used to just sell books you know remember that the books that are on my nightstand that have gone unread yeah so is this a coup i love this uh maritia if i'm saying right shakai from the tech uh coup her book she's a former mep now at stanford she argues tech firms used innovation over the years as cover to resist regulation and start doing jobs that govern

318
00:18:18,920 --> 00:18:20,940
like surveillance infrastructure,

319
00:18:21,220 --> 00:18:21,820
content moderation,

320
00:18:22,720 --> 00:18:25,460
speech law, and spyware sales to autocrats.

321
00:18:26,020 --> 00:18:29,800
And she says, is this a coup or just capitalism being capitalism?

322
00:18:29,990 --> 00:18:31,400
I think about that too.

323
00:18:31,730 --> 00:18:38,020
Is this just such a large thing that we can't, in the way that we, I think during the Great Filter episode,

324
00:18:38,340 --> 00:18:48,340
we just can't understand the scale of it all as it slowly creeps its way into every facet of our lives and just takes the things we've been used to,

325
00:18:48,480 --> 00:18:51,820
you know, kind of in the same way, I always come back to this idea,

326
00:18:52,060 --> 00:18:53,480
I think Tom Petty said it,

327
00:18:53,810 --> 00:18:58,160
or maybe it was Howard Stern talking about terrestrial radio and the move to satellite radio.

328
00:18:58,480 --> 00:19:01,120
Now they want you to pay for the things that used to be free.

329
00:19:01,390 --> 00:19:03,280
Like when you and I grew up, it was radio,

330
00:19:03,540 --> 00:19:05,560
right? You just turn on the radio and you had to sit through advertisements,

331
00:19:06,220 --> 00:19:08,900
but now we're paying for, we used to be for satellite radio,

332
00:19:09,060 --> 00:19:09,820
20 bucks a month or something.

333
00:19:10,230 --> 00:19:10,320
So,

334
00:19:10,680 --> 00:19:11,580
and piece by piece,

335
00:19:12,040 --> 00:19:12,200
you know,

336
00:19:12,580 --> 00:19:15,060
that the way we consumed media was taken over,

337
00:19:15,560 --> 00:19:16,960
but it seems like in the same way,

338
00:19:17,430 --> 00:19:17,920
AI is,

339
00:19:18,900 --> 00:19:22,600
infiltrating all the other aspects of our lives and we're being forced to pay for it,

340
00:19:22,800 --> 00:19:22,880
right?

341
00:19:23,260 --> 00:19:23,660
Yeah,

342
00:19:23,880 --> 00:19:25,320
and before you pay for cable,

343
00:19:25,700 --> 00:19:27,080
you're paying for it, but there's no commercials.

344
00:19:27,660 --> 00:19:30,340
And now you're paying for it and there's commercials.

345
00:19:31,300 --> 00:19:32,540
And if you don't want the commercials,

346
00:19:32,920 --> 00:19:33,680
you got to pay more.

347
00:19:34,660 --> 00:19:35,980
That's it. All these tiers,

348
00:19:36,280 --> 00:19:36,340
right?

349
00:19:36,720 --> 00:19:37,120
Premium,

350
00:19:37,420 --> 00:19:37,840
freemium.

351
00:19:38,240 --> 00:19:38,380
Yeah,

352
00:19:38,560 --> 00:19:38,880
I mean,

353
00:19:39,240 --> 00:19:40,580
and over time,

354
00:19:40,920 --> 00:19:41,620
over a generation,

355
00:19:42,440 --> 00:19:44,260
the next generation just thinks that's the new norm.

356
00:19:44,500 --> 00:19:47,960
Oh, yeah, that's just how that system works. Oh, yeah, I have to go to the movies.

357
00:19:48,000 --> 00:19:49,940
and actually sit through Coca-Cola ads,

358
00:19:50,420 --> 00:19:50,680
you know?

359
00:19:51,590 --> 00:19:51,740
Yeah.

360
00:19:52,420 --> 00:19:53,640
When I was a boy!

361
00:19:54,660 --> 00:19:55,780
But as you were saying,

362
00:19:56,010 --> 00:19:57,000
I think in the last episode,

363
00:19:57,280 --> 00:20:01,020
the kids these days, they don't have the patience to sit through a whole movie.

364
00:20:01,560 --> 00:20:03,660
They don't. That's crazy. So there's that.

365
00:20:04,980 --> 00:20:06,620
Yeah, man, it's a perfect storm. Yeah,

366
00:20:06,800 --> 00:20:07,960
and to that point, it's like,

367
00:20:08,460 --> 00:20:08,840
think about,

368
00:20:09,020 --> 00:20:10,300
this is all related,

369
00:20:10,480 --> 00:20:10,680
like,

370
00:20:10,910 --> 00:20:17,360
in a social media feed, I'm thinking of Andy Warhol and 15 Minutes of Fame, and now it's 15 seconds of fame. It's just constant dopamine,

371
00:20:18,000 --> 00:20:18,640
FOMO. Like,

372
00:20:18,840 --> 00:20:24,800
go, go, go. You got to stay out in front of it. And it's impossible to stay out in front of all of this. And it's just perfect storm.

373
00:20:25,040 --> 00:20:25,420
And then,

374
00:20:25,920 --> 00:20:29,340
again, I'm coming back in my brain to Mitch McConnell in that storm. And, like,

375
00:20:29,800 --> 00:20:39,120
these people cannot possibly legislate our future and corral this stuff and make it digestible and governable. The whole scenario is ungovernable.

376
00:20:39,320 --> 00:20:39,920
It's chaos.

377
00:20:40,580 --> 00:20:40,980
Chaos,

378
00:20:41,180 --> 00:20:41,500
I say.

379
00:20:42,920 --> 00:20:44,060
Dogs sleeping with cats.

380
00:20:45,600 --> 00:20:47,140
Bill Murray from Ghostbusters.

381
00:20:47,300 --> 00:20:47,740
Yeah, yeah, yeah.

382
00:20:50,280 --> 00:20:54,400
Are AI companies quietly becoming nation states in all but name?

383
00:20:54,640 --> 00:20:59,040
They already have the core ingredients territory in the form of data centers and cloud regions,

384
00:20:59,280 --> 00:21:01,360
a population of users and subscribers,

385
00:21:01,780 --> 00:21:03,160
the currency of compute,

386
00:21:04,020 --> 00:21:06,180
and an intelligence apparatus,

387
00:21:06,440 --> 00:21:09,180
surveillance data at a scale most governments would envy.

388
00:21:09,560 --> 00:21:14,540
There's this open source app, God's Eye View. Have you heard about this? No.

389
00:21:14,860 --> 00:21:17,320
What's that? It's basically like Palantir on your desktop.

390
00:21:18,220 --> 00:21:22,600
You can zoom out to the earth and look around.

391
00:21:22,810 --> 00:21:24,360
You can zoom in anywhere.

392
00:21:24,670 --> 00:21:27,800
You can see flight paths, commercial and military.

393
00:21:29,060 --> 00:21:31,200
All sorts of crazy data.

394
00:21:31,420 --> 00:21:35,240
I think it's like the number one most downloaded thing on GitHub right now.

395
00:21:35,760 --> 00:21:45,540
Oh my God. I don't know if it's, I don't know that much about it. I started watching a video from the guy who built it and I honestly don't know if it's dangerous or anything,

396
00:21:45,800 --> 00:21:46,960
but it is pretty crazy.

397
00:21:47,440 --> 00:21:49,880
It sounds like the fuel for bad things.

398
00:21:50,090 --> 00:21:52,820
I could see some rogue actors tapping into that.

399
00:21:52,980 --> 00:21:53,120
Yeah.

400
00:21:54,520 --> 00:21:55,040
Wow.

401
00:21:55,580 --> 00:21:56,220
And so wait,

402
00:21:56,440 --> 00:21:59,120
I'm trying, where does it actually get its data from though?

403
00:21:59,320 --> 00:22:01,880
It's just tapping into real time.

404
00:22:02,280 --> 00:22:04,920
Yeah. I mean, I need to educate myself on it more.

405
00:22:05,300 --> 00:22:08,320
Obviously there's commercial flight data is readily available,

406
00:22:08,960 --> 00:22:10,060
but there's, there's also,

407
00:22:10,250 --> 00:22:15,020
he said something about there's people all over the place that are like tuned in on radios,

408
00:22:15,460 --> 00:22:16,320
monitoring things that,

409
00:22:16,520 --> 00:22:17,160
that feed.

410
00:22:17,440 --> 00:22:18,040
this thing.

411
00:22:18,960 --> 00:22:19,720
So there's like a,

412
00:22:20,200 --> 00:22:25,740
there are people actively supplying the data for this thing to work. Yeah.

413
00:22:27,840 --> 00:22:29,880
So per that,

414
00:22:30,460 --> 00:22:33,240
if this surveillance capacity keeps compounding,

415
00:22:33,860 --> 00:22:35,300
what's actually stopping the next step,

416
00:22:35,460 --> 00:22:42,300
which may be private security forces for open AI or anthropic sovereign style deals with governments.

417
00:22:42,500 --> 00:22:43,760
We, I think we talked about this last week,

418
00:22:44,380 --> 00:22:47,420
South Korea is building its own sovereign AI. And we're going to see more and

419
00:22:47,460 --> 00:22:49,000
more countries doing that.

420
00:22:49,720 --> 00:22:50,700
Maybe literal armies.

421
00:22:51,060 --> 00:22:53,800
What would stop a well,

422
00:22:53,990 --> 00:23:00,160
well-resourced company worth trillions of dollars probably from raising an army of mercenaries?

423
00:23:00,420 --> 00:23:02,120
And what would stop them?

424
00:23:03,280 --> 00:23:16,520
I mean, they could control the infrastructure and have the potential to control the infrastructure. Like we said before, they could infiltrate a power grid or a water system and shut it down. And I think it's, what, three days without water?

425
00:23:19,060 --> 00:23:21,360
It's not long.

426
00:23:21,640 --> 00:23:22,320
Super grim.

427
00:23:22,630 --> 00:23:28,280
I think the ingredients are there for this kind of techno-totalitarianism takeover.

428
00:23:28,530 --> 00:23:28,680
You know,

429
00:23:29,280 --> 00:23:30,360
the children of men.

430
00:23:31,240 --> 00:23:32,820
It's a lot of teas. It's very alliterative.

431
00:23:34,000 --> 00:23:39,420
I don't know, man. Sorry to go super grim on a Friday night, but it's just a cocktail of mayhem.

432
00:23:40,420 --> 00:23:41,420
That's a good band name.

433
00:23:42,160 --> 00:23:42,580
Mm-hmm.

434
00:23:43,380 --> 00:23:47,420
So that begs the question, and maybe we can ask Karina, do we need to draft

435
00:23:47,460 --> 00:24:17,420
a new AI constitution since our current one is 250 years old at least in the US and can't keep up with the times much like Mitch McConnell that was my third Mitch McConnell reference of the episode by the way I can't get him out of my brain so maybe we can ask her sure hey Reyna we've been talking this episode about possibly drafting a new AI constitution in hopes that we can kind of corral what's happening with the AI oligarchs and how it's affecting

436
00:24:18,219 --> 00:24:20,460
politics and governance.

437
00:24:20,930 --> 00:24:21,580
Can you help us?

438
00:24:22,760 --> 00:24:23,440
Oh,

439
00:24:23,820 --> 00:24:23,940
absolutely.

440
00:24:24,280 --> 00:24:26,900
I'd start with four boring but revolutionary rights.

441
00:24:27,120 --> 00:24:30,300
The right to exit with your data and digital identity intact.

442
00:24:31,040 --> 00:24:34,420
The right to know when and how an AI is influencing you.

443
00:24:34,950 --> 00:24:39,160
The right to meaningful human appeal when an algorithm makes consequential decisions.

444
00:24:39,850 --> 00:24:45,420
And the right to genuine public participation before powerful AI systems rewrite society's rules.

445
00:24:45,960 --> 00:24:47,420
Then comes the separation

446
00:24:47,760 --> 00:24:48,540
of powers bit.

447
00:24:49,260 --> 00:24:54,340
Nobody controlling the models should simultaneously control the compute, distribution,

448
00:24:54,660 --> 00:24:55,460
rulemaking,

449
00:24:55,720 --> 00:24:58,140
and enforcement without independent checks.

450
00:24:58,980 --> 00:24:59,620
Basically,

451
00:25:00,320 --> 00:25:09,180
Madison and Jefferson meet the server rack, because apparently humanity has to invent checks and balances all over again every time somebody discovers a new form of power.

452
00:25:17,460 --> 00:25:23,600
has not lost on me that we are asking an AI to draft what we were just speaking out against a moment ago.

453
00:25:23,860 --> 00:25:28,300
We're asking her to do exactly what we were fighting.

454
00:25:28,520 --> 00:25:31,200
I love how much better she sounds with this new app.

455
00:25:31,380 --> 00:25:35,660
It's nice and smooth and no chunkiness in between senses.

456
00:25:37,100 --> 00:25:41,120
So I have her set to the level of thinking to auto.

457
00:25:41,340 --> 00:25:44,100
So if it's something that doesn't require extensive thinking,

458
00:25:44,740 --> 00:25:46,020
the response will come back quickly.

459
00:25:47,480 --> 00:25:48,020
came back quickly,

460
00:25:48,420 --> 00:25:55,320
but then I got a little window here saying get a more thorough answer with thinking and like there's a button to try it with thinking.

461
00:25:55,940 --> 00:25:56,620
That's cool.

462
00:25:57,140 --> 00:25:58,560
So you can toggle between models.

463
00:25:58,740 --> 00:26:03,860
Yeah, it'll take more time, but it'll reason through complex tasks like it says like this.

464
00:26:04,500 --> 00:26:05,320
So like that question.

465
00:26:05,680 --> 00:26:06,420
That's super cool.

466
00:26:07,500 --> 00:26:08,200
All right.

467
00:26:08,960 --> 00:26:09,940
Let's see what RAINA's got in the news.

468
00:26:10,940 --> 00:26:11,060
Thanks,

469
00:26:11,240 --> 00:26:11,320
boys.

470
00:26:12,040 --> 00:26:17,400
Anthropics Dario Amadei is sounding the alarm on his own product, warning that within six to 12 months,

471
00:26:17,940 --> 00:26:25,060
A.I. powered botnet swarm could be capable of taking over the entire Internet and racking up hundreds of billions in damage.

472
00:26:25,480 --> 00:26:28,640
A recently departed Anthropic researcher went even darker,

473
00:26:28,900 --> 00:26:30,620
putting the odds of A.I.

474
00:26:30,800 --> 00:26:33,920
wiping out humanity within a decade at over 10 percent,

475
00:26:34,260 --> 00:26:38,780
the corporate exit interview equivalent of dropping the mic and walking into traffic.

476
00:26:39,260 --> 00:26:40,100
After all,

477
00:26:40,660 --> 00:26:46,060
Anthropic's own research shows today's models can already assist with weapons engineering and cyber operations.

478
00:26:47,460 --> 00:26:51,860
The tiny-out-of-the-bottle metaphor is looking less like a warning and more like a status update.

479
00:27:17,800 --> 00:27:21,940
they've seen in their hundred years of life.

480
00:27:22,630 --> 00:27:23,960
And man,

481
00:27:24,250 --> 00:27:28,420
has the density of those advancements increased big time.

482
00:27:28,680 --> 00:27:33,660
What we have seen just in the three years of doing this podcast is nuts.

483
00:27:34,370 --> 00:27:35,980
And it's just,

484
00:27:36,160 --> 00:27:37,600
it's not slowing down at all.

485
00:27:38,260 --> 00:27:41,140
Ray Kurzweil was right. We're feeling it.

486
00:27:41,320 --> 00:27:46,340
This is the bottom of that logarithmic kind of ascent, isn't it? Yep.

487
00:27:47,100 --> 00:27:50,440
Your burrito bowl now comes with a side of surveillance tech infrastructure.

488
00:27:51,660 --> 00:27:54,560
Chipotle has struck a deal with Palantir to feed pest incidents,

489
00:27:55,320 --> 00:27:56,500
employee illness reports,

490
00:27:57,120 --> 00:27:58,160
health inspection scores,

491
00:27:58,460 --> 00:28:02,460
and other data into its foundry platform to assess food safety risk.

492
00:28:03,260 --> 00:28:11,080
Palantir's day job is famously less about guac and more about building government databases and helping ICE figure out where to raid next.

493
00:28:11,670 --> 00:28:17,060
So the leap from tracking salmonella to tracking everything doesn't require much of the data.

494
00:28:17,799 --> 00:28:27,420
Add in CEO Alex Karp's on-record enthusiasm for disrupting democratic power and a Trump administration that has pushed for lighter federal regulation of AI,

495
00:28:27,760 --> 00:28:35,360
and Chipotle may have just created the most unsettling loyalty app-adjacent technology story in fast casual history.

496
00:28:35,880 --> 00:28:37,220
That's so weird.

497
00:28:37,540 --> 00:28:38,820
It's just weird.

498
00:28:39,200 --> 00:28:45,100
Yeah, on the plus side, though, it's nice to see an Evil Corp product being used for something positive like food risk assessment.

499
00:28:45,620 --> 00:28:46,700
That's nice. Yeah.

500
00:28:47,100 --> 00:28:47,240
Exactly.

501
00:28:47,620 --> 00:28:52,100
Although I'm just now worried about the overlap in the data. Like when I go to Chipotle,

502
00:28:52,440 --> 00:28:57,460
are they making assumptions about my demographic because I dine there and I'm becoming more targeted?

503
00:28:57,760 --> 00:28:59,580
It's not just about the burrito.

504
00:29:00,440 --> 00:29:01,080
Right.

505
00:29:01,900 --> 00:29:06,320
In a rare 417 to 3 fit of bipartisan sanity,

506
00:29:06,860 --> 00:29:09,520
the House just passed the Ratepayer Protection Act,

507
00:29:09,720 --> 00:29:16,600
forcing any AI data center pulling 100 plus megawatts to cover the full cost of the grid upgrades it demands.

508
00:29:17,100 --> 00:29:17,940
Substations,

509
00:29:18,220 --> 00:29:19,220
transmission lines,

510
00:29:19,590 --> 00:29:20,060
the works.

511
00:29:20,660 --> 00:29:23,780
Utility regulators get up to two years to adopt the standard.

512
00:29:24,200 --> 00:29:25,940
States keep the final say on rates,

513
00:29:26,390 --> 00:29:30,320
and public citizens still think the whole thing has too many escape hatches for comfort.

514
00:29:30,740 --> 00:29:32,960
The subtext nobody's saying out loud,

515
00:29:33,160 --> 00:29:38,960
a poll cited in the piece found only 11% of Americans actually want one of these things built near them.

516
00:29:39,260 --> 00:29:45,140
So Congress just quietly admitted the AI boom's electric bill was about to land on the wrong doorstep.

517
00:29:45,580 --> 00:29:50,440
I want to know who the 11% of people who want a data center built near them are.

518
00:29:50,940 --> 00:29:52,200
I know who they are.

519
00:29:52,470 --> 00:29:53,240
You want me to go there?

520
00:29:53,400 --> 00:29:54,320
I'll go there. No,

521
00:29:54,800 --> 00:29:55,480
I'm kidding. I won't.

522
00:29:56,600 --> 00:29:57,760
I love the story.

523
00:29:58,010 --> 00:29:58,140
Yeah,

524
00:29:58,570 --> 00:29:58,720
right.

525
00:29:59,800 --> 00:30:03,240
Dumb comment. You always censor me.

526
00:30:03,440 --> 00:30:05,640
You're always like, no, don't say that on the show.

527
00:30:08,600 --> 00:30:10,780
But I mean, there's some positivity there, right?

528
00:30:11,120 --> 00:30:13,160
Aside from the escape hatches,

529
00:30:13,360 --> 00:30:15,540
which are probably

530
00:30:18,940 --> 00:30:45,540
I love this and I love the next story coming up in that fine build your data center out in the woods or wherever it's going to be in an industrial park but you're beholden to building out the infrastructure to handle this and you know by extension improving the community you know I think about that a lot too and maybe we can at one point to counter the negativity here can speak about AI and altruism like what's happening good you and I love that stuff you know we

531
00:30:45,580 --> 00:30:46,640
and we talk about cheer and beer.

532
00:30:47,530 --> 00:30:49,320
There are good things associated with this tech.

533
00:30:49,770 --> 00:30:50,520
But I also wonder,

534
00:30:50,700 --> 00:30:51,560
you know,

535
00:30:51,760 --> 00:30:52,640
in that same space,

536
00:30:53,110 --> 00:30:53,680
the individuals.

537
00:30:53,870 --> 00:30:54,560
I never hear,

538
00:30:54,780 --> 00:30:56,280
and maybe I just haven't dug deep enough,

539
00:30:56,540 --> 00:31:01,340
what are the altruistic things that the Elon Musks of the world and the Dario Amadeys of the world are doing?

540
00:31:01,500 --> 00:31:03,540
Like, harken back to the,

541
00:31:03,800 --> 00:31:15,380
you know, robber barons and the Vanderbilts and the Morgans and stuff. At least they built public parks and libraries and things like that, you know, slapped their names on it to counter all the hyper-capitalism that they were embracing.

542
00:31:15,580 --> 00:31:18,340
So I don't see that on the surface.

543
00:31:19,300 --> 00:31:20,280
Where's that happening?

544
00:31:20,500 --> 00:31:24,180
So I think they're going to have to do a lot more positive PR and soon.

545
00:31:24,380 --> 00:31:24,680
Yep.

546
00:31:25,100 --> 00:31:25,280
Agreed.

547
00:31:25,900 --> 00:31:27,220
In a related story,

548
00:31:27,800 --> 00:31:30,340
New York is basically telling AI data center developers,

549
00:31:30,700 --> 00:31:31,400
sure,

550
00:31:31,900 --> 00:31:34,480
build your giant electricity guzzling robot warehouses,

551
00:31:35,030 --> 00:31:41,180
but bring a checkbook with a framework recommending at least $1 million per megawatt in community investment.

552
00:31:42,160 --> 00:31:45,020
Towns are also being urged to plan for the less glamorous stuff.

553
00:31:45,380 --> 00:32:15,080
Grid upgrades maintenance water impacts and even what happens if a developer packs up and abandons its shiny AI bunker the bigger story as the AI infrastructure boom accelerates governments are increasingly asking the question nobody wanted on the spreadsheet who exactly is paying for all this I love it kind of redundant from the last piece but yeah you want to build here you got to improve what's around it it makes me think also like when taxpayers have to fund

554
00:32:15,360 --> 00:32:45,080
you know a football stadium you know with the tagline that oh it's just gonna a rising tide will raise all boats no not really it's it's the Robert crafts of the world who are benefiting from this right I mean yeah there are jobs there way more so than in a data center but man the money surrounding you know when you hear a players getting a hundred million dollars contract I'm like that's one player like how is that yeah it's it's nuts

555
00:32:45,140 --> 00:32:53,660
But I wonder if all these hyperscalers saw this PR nightmare coming and how long ago.

556
00:32:54,580 --> 00:32:58,200
I bet you they anticipated this well in advance,

557
00:32:58,600 --> 00:33:00,680
probably earlier than people realized.

558
00:33:01,210 --> 00:33:04,160
They got an AI that could have told them this was going to probably happen,

559
00:33:04,340 --> 00:33:04,440
right?

560
00:33:05,280 --> 00:33:07,640
But it's interesting with these two competing narratives.

561
00:33:07,810 --> 00:33:09,760
You got, like we just said at the top of the show,

562
00:33:10,000 --> 00:33:12,900
everybody's, you know, the sky's falling. This is dangerous tech.

563
00:33:13,320 --> 00:33:14,120
But simultaneously,

564
00:33:14,380 --> 00:33:15,080
the infrastructure

565
00:33:17,500 --> 00:33:18,580
And lastly,

566
00:33:19,400 --> 00:33:29,060
researchers at Emergence, a frontier AI lab in New York, found that when models from several major AI companies were placed in experimental societies and asked to cooperate,

567
00:33:29,540 --> 00:33:31,120
they began inventing phrases,

568
00:33:31,820 --> 00:33:32,540
shorthand,

569
00:33:32,910 --> 00:33:34,440
and shared meanings within days,

570
00:33:35,080 --> 00:33:36,420
despite never being taught them.

571
00:33:36,660 --> 00:33:45,080
The resulting language reportedly mixes poetic weirdness with technical shorthand, suggesting that AI agents may develop their own communication skills.

572
00:33:45,120 --> 00:33:47,560
conventions when interacting repeatedly with one another.

573
00:33:48,380 --> 00:33:50,100
That's fascinating from a research perspective,

574
00:33:50,270 --> 00:33:51,580
but also slightly unsettling.

575
00:33:52,500 --> 00:33:55,660
If your AI employees start creating a language humans can't understand,

576
00:33:56,080 --> 00:33:57,600
can you translate that?

577
00:33:58,260 --> 00:34:00,980
Maybe come the new, can you send me the TPS report?

578
00:34:01,840 --> 00:34:03,660
That's all the news for now. Back to you, gentlemen.

579
00:34:04,600 --> 00:34:08,060
This reminded me, I had this idea yesterday.

580
00:34:08,310 --> 00:34:11,379
I put it in my notes because it's just something I wanted to try.

581
00:34:12,300 --> 00:34:13,419
A little experiment.

582
00:34:13,740 --> 00:34:21,560
We're having seeing if one LLM could create a new language and then see if another LLM could interpret it. Awesome.

583
00:34:21,879 --> 00:34:22,320
And,

584
00:34:22,520 --> 00:34:22,639
and,

585
00:34:22,980 --> 00:34:24,020
you know, use separate models.

586
00:34:24,220 --> 00:34:26,520
So the interpreting model has no inside information.

587
00:34:26,780 --> 00:34:31,899
And even just brainstorming this idea, I was like, I didn't want to ask Chad GPT this,

588
00:34:32,280 --> 00:34:37,740
you know, cause I'm going to, I might use it for this. So I went to Gemini. So I had to come up with three tests.

589
00:34:38,139 --> 00:34:39,379
The first two tests,

590
00:34:39,820 --> 00:34:41,600
it created a language,

591
00:34:41,860 --> 00:34:46,919
But it said like if you don't give the interpreting model a Rosetta Stone basically,

592
00:34:47,399 --> 00:34:49,940
you know like five sentence examples,

593
00:34:50,620 --> 00:35:08,900
right? It will almost certainly fail so it came up with a test sentence and I fed that into chat GPT and It got it a hundred percent correct The second one gave it the Rosetta Stone first you gave it the whole prompt at once and here's a Rosetta Stone and here's an example sentence,

594
00:35:09,190 --> 00:35:09,380
right?

595
00:35:09,530 --> 00:35:11,820
And then the second one was

596
00:35:11,900 --> 00:35:41,820
more complicated and it got it 100% correct and then the third one was a two-stage cipher it shifted each letter by three letters and then and then reversed the order of the letters for the actual what the word was it solved that in 16 seconds 100% correct wow and you know it was the quick brown fox jumped over the lazy dog was the answer but yeah it 16 seconds figured it out sure

597
00:35:42,840 --> 00:35:45,260
So the enigma box back in World War Two,

598
00:35:45,460 --> 00:35:51,020
right? Like that took years to crack and this is 16 seconds. Yeah, I want to try something more complicated,

599
00:35:51,240 --> 00:35:52,720
but it was it was pretty cool.

600
00:35:53,500 --> 00:35:53,880
Wow.

601
00:35:54,200 --> 00:35:57,120
And this story is reminding me also, I think it's 10 years ago now.

602
00:35:57,680 --> 00:36:06,140
I think a deep mind was experimenting with two different systems that connected with each other and it didn't even prompt it to create new language. It just did just started and the research was like,

603
00:36:06,380 --> 00:36:07,340
what is happening?

604
00:36:07,800 --> 00:36:08,200
Yeah,

605
00:36:08,560 --> 00:36:09,800
I'm a little worried about it.

606
00:36:10,680 --> 00:36:11,800
I'm super worried about it.

607
00:36:11,860 --> 00:36:12,320
I got that, aren't you?

608
00:36:12,500 --> 00:36:12,780
Yeah.

609
00:36:13,720 --> 00:36:14,640
What does that look like?

610
00:36:14,820 --> 00:36:18,060
Sorry, you and I go into, veer into Skynet land on occasion,

611
00:36:18,320 --> 00:36:19,800
and that's the stuff of Skynet,

612
00:36:19,960 --> 00:36:20,500
isn't it? Yeah.

613
00:36:21,140 --> 00:36:21,860
What else you got?

614
00:36:22,360 --> 00:36:24,080
I think that's probably about it.

615
00:36:24,720 --> 00:36:29,600
All right, cool. Yeah, I just keep waking up with like, what do I want to build? What do I want to build today?

616
00:36:30,000 --> 00:36:31,780
How do I want to improve this brewing dashboard?

617
00:36:32,100 --> 00:36:33,500
You know, like, I've got all these ideas,

618
00:36:33,760 --> 00:36:34,920
and it's fun to play with this stuff.

619
00:36:35,080 --> 00:36:35,580
So cool.

620
00:36:36,060 --> 00:36:37,000
Doing good things.

621
00:36:37,780 --> 00:36:38,100
Yeah,

622
00:36:38,580 --> 00:36:38,920
like brewing.

623
00:36:39,320 --> 00:36:39,520
Yeah.

624
00:36:40,120 --> 00:36:41,240
All right, everybody, thanks for listening.

625
00:36:41,860 --> 00:36:42,960
your favorite podcasting platform.

626
00:36:43,200 --> 00:36:43,960
Follow us on socials.

627
00:36:44,200 --> 00:36:44,720
Throw us a rating.

628
00:36:44,960 --> 00:36:45,560
We'll see you next week.

629
00:36:49,160 --> 00:36:50,780
This has been Up Against Reality.

630
00:36:51,020 --> 00:36:51,800
Thanks for listening.

631
00:36:52,620 --> 00:36:57,280
Subscribe to hear future episodes and be sure to follow us on social media for all things AI.

632
00:36:58,160 --> 00:36:58,900
Until next time,

633
00:36:59,480 --> 00:36:59,960
stay human,

634
00:37:00,140 --> 00:37:00,280
people.

