1
00:00:00,160 --> 00:00:02,960
Speaker 1: Leave zero some quant finance and come build the machine.

2
00:00:02,960 --> 00:00:05,480
Gud uh. They're going to have a rude awakening when

3
00:00:05,519 --> 00:00:07,000
it turns out they have to work on ads to

4
00:00:07,000 --> 00:00:07,440
pay for it.

5
00:00:07,480 --> 00:00:11,119
Speaker 2: All That one lands hard, right.

6
00:00:11,240 --> 00:00:13,439
Speaker 1: I mean, that's a tweet from seven months ago, just

7
00:00:13,480 --> 00:00:15,880
seven months, just seven months, and you know, in the

8
00:00:15,880 --> 00:00:20,199
normal world that's nothing, but in Ai time, that feels

9
00:00:20,239 --> 00:00:22,839
like a decade ago, and reading it today, it just

10
00:00:23,320 --> 00:00:25,800
it feels less like a tweet and more like a prophecy.

11
00:00:25,879 --> 00:00:28,120
Speaker 2: It is absolutely prophetic. It's like someone saw the entire

12
00:00:28,160 --> 00:00:29,679
movie and just tweeted out the ending.

13
00:00:29,760 --> 00:00:34,920
Speaker 1: It's the whole story arc, the grand ambition, the noble crusade,

14
00:00:34,960 --> 00:00:36,520
and then the gritty.

15
00:00:36,159 --> 00:00:38,399
Speaker 2: Reality check, the crash landing.

16
00:00:38,240 --> 00:00:41,799
Speaker 1: Exactly, and that rude awakening he mentioned. It's not some

17
00:00:42,000 --> 00:00:45,399
far off theoretical thing anymore. It's here. We're watching it

18
00:00:45,439 --> 00:00:48,479
happen in real time. We're literally watching this collision between

19
00:00:48,479 --> 00:00:52,079
these huge grand dreams of building a digital deity, a

20
00:00:52,200 --> 00:00:56,240
super intelligence, yeah, and the you know, the boring, gritty,

21
00:00:56,640 --> 00:00:59,479
capitalistic reality of just paying the server bills. And who

22
00:00:59,520 --> 00:01:05,480
is standing there pointing a finger. It's Google's AI CEO Demissabis.

23
00:01:05,000 --> 00:01:07,680
Speaker 2: And he's not just pointing, he's he's laughing a little bit.

24
00:01:07,680 --> 00:01:09,480
It's getting incredibly spicy out there.

25
00:01:09,560 --> 00:01:12,799
Speaker 1: It is so spicy. Welcome to thrilling threads. I am

26
00:01:12,920 --> 00:01:15,280
so ready to pull on this particular thread because it

27
00:01:15,319 --> 00:01:19,200
feels like the whole AI world is, I don't know,

28
00:01:19,280 --> 00:01:20,599
unraveling a bit right now.

29
00:01:20,719 --> 00:01:23,120
Speaker 2: It really does. And you know, usually you see these

30
00:01:23,159 --> 00:01:25,480
tech feuds and they're all kind of buried.

31
00:01:25,159 --> 00:01:27,400
Speaker 1: In subtext, right and like a press release or something.

32
00:01:27,239 --> 00:01:30,040
Speaker 2: Yeah, or some subtle dig in a shareholder meeting that

33
00:01:30,079 --> 00:01:33,400
you have to decode. But this is just it's out

34
00:01:33,519 --> 00:01:35,959
in the open. This isn't just a business disagreement. It

35
00:01:36,000 --> 00:01:38,719
feels like a battle for the soul of what AI

36
00:01:38,840 --> 00:01:39,640
is supposed.

37
00:01:39,319 --> 00:01:42,920
Speaker 1: To be exactly. So our mission today is to untangle

38
00:01:43,000 --> 00:01:45,439
this whole drama. We're looking at Google, We're looking at

39
00:01:45,439 --> 00:01:50,480
open AI, and specifically this friction between the promise of AGI,

40
00:01:51,599 --> 00:01:54,640
you know, this artificial general intelligence that's supposed to change.

41
00:01:54,439 --> 00:01:56,319
Speaker 2: Everything, it's supposed to save the world, right.

42
00:01:56,319 --> 00:01:58,840
Speaker 1: And the reality of well, pop up ads.

43
00:01:58,959 --> 00:02:01,640
Speaker 2: It is the oldest story in the book. It's idealism

44
00:02:01,719 --> 00:02:04,959
versus capitalism. And what's so fascinating here is that. Yeah,

45
00:02:05,000 --> 00:02:06,840
it's a little petty, which is always fun for us,

46
00:02:06,879 --> 00:02:10,919
oh for sure, But it actually reveals this fundamental economic

47
00:02:11,039 --> 00:02:14,280
truth about the AI revolution that I think nobody really

48
00:02:14,360 --> 00:02:15,840
wanted to talk about until now.

49
00:02:15,960 --> 00:02:18,400
Speaker 1: Okay, so let's get right into it. The Google burn

50
00:02:18,560 --> 00:02:21,560
let's do it. Demis Hasabis, he's the head of Google

51
00:02:21,599 --> 00:02:24,280
Deep Mind. He basically just came out and publicly torched

52
00:02:24,360 --> 00:02:27,199
Open Ai. He didn't just you know, say I disagree

53
00:02:27,240 --> 00:02:29,800
with their strategy. He went straight for the jagular.

54
00:02:29,919 --> 00:02:32,719
Speaker 2: He really did. And it's important to understand who Demis is.

55
00:02:32,840 --> 00:02:34,960
You know, he's not a hype guy. He's not a

56
00:02:35,000 --> 00:02:38,439
marketing ceo. He's a scientist. He was a chess prodigy.

57
00:02:38,800 --> 00:02:41,919
He created DeepMind, which is, you know, arguably the most

58
00:02:41,960 --> 00:02:45,319
serious AI research lab on the planet. They're the ones

59
00:02:45,360 --> 00:02:49,360
who cracked protein folding with alpha fold. He usually stays very.

60
00:02:49,319 --> 00:02:51,840
Speaker 1: Very quiet, so when he decides to speak up.

61
00:02:51,800 --> 00:02:55,000
Speaker 2: Everybody listens. And what he said, and I'm paraphrasing the

62
00:02:55,039 --> 00:02:57,520
sentiment here, but this is the core of it. He said,

63
00:02:57,599 --> 00:02:59,800
if AGI was around the corner, why do you need

64
00:02:59,800 --> 00:03:01,520
to sell ads inside of your product?

65
00:03:01,560 --> 00:03:05,120
Speaker 1: That is just it's the billion dollar question. Isn't it.

66
00:03:05,120 --> 00:03:06,800
It's so simple, but it's devastating.

67
00:03:06,800 --> 00:03:07,840
Speaker 2: It's devastatingly simple.

68
00:03:07,879 --> 00:03:10,960
Speaker 1: Let's actually unpack that logic because open AI's whole narrative,

69
00:03:11,000 --> 00:03:12,960
the story they've been telling us in their investors for

70
00:03:13,080 --> 00:03:16,039
years is that AGI is imminent. It's almost here.

71
00:03:16,319 --> 00:03:19,000
Speaker 2: Yeah. Sam Altman talks about the singularity like it's a

72
00:03:19,039 --> 00:03:21,400
flight on a departure's board, now boarding, Just.

73
00:03:21,360 --> 00:03:25,240
Speaker 1: Hold on, We're almost there. And Demis's logic is, Okay,

74
00:03:25,280 --> 00:03:28,159
if you really have a technology that is that powerful,

75
00:03:28,199 --> 00:03:31,599
that world changing, you don't need to interrupt it with

76
00:03:31,639 --> 00:03:33,800
a commercial for car insurance to make money.

77
00:03:35,080 --> 00:03:36,840
Speaker 2: The idea would be laughable.

78
00:03:37,120 --> 00:03:39,400
Speaker 1: But why explain that a bit more? Why wouldn't you

79
00:03:39,439 --> 00:03:42,199
just slap an ad on a machine god?

80
00:03:42,360 --> 00:03:44,439
Speaker 2: Okay, So think about it in terms of value capture.

81
00:03:44,560 --> 00:03:47,080
If you actually have AGI, you have an entity that

82
00:03:47,159 --> 00:03:50,919
can I don't know, cure cancer, or solve cold fusion,

83
00:03:51,240 --> 00:03:55,680
or completely restructure global logistics to end world hunger. The

84
00:03:55,759 --> 00:03:58,919
economic value of that isn't just big, it's almost infinite.

85
00:03:59,000 --> 00:03:59,919
Speaker 1: So you wouldn't be selling BAM.

86
00:04:00,800 --> 00:04:05,159
Speaker 2: You'd be licensing the core technology to governments, to massive

87
00:04:05,159 --> 00:04:08,719
pharmaceutical companies, to investment banks for you know, millions or

88
00:04:08,759 --> 00:04:11,840
billions of dollars per transaction. You'd be selling the cure

89
00:04:11,879 --> 00:04:14,960
for cancer, not the eyeballs watching a video about it.

90
00:04:15,039 --> 00:04:17,639
Speaker 1: I see, So you're selling the solution itself, not the

91
00:04:17,680 --> 00:04:19,399
attention of the people asking for the solution.

92
00:04:19,879 --> 00:04:24,079
Speaker 2: The idea of monetizing AGI through like a cost per

93
00:04:24,120 --> 00:04:26,240
click model where you get paid a few cents because

94
00:04:26,279 --> 00:04:28,519
someone clicks a link for a pair of sneakers. It's

95
00:04:28,560 --> 00:04:29,680
a complete mismatch of such.

96
00:04:30,040 --> 00:04:32,480
Speaker 1: It's like inventing a faster than light starship and then

97
00:04:32,560 --> 00:04:34,639
deciding the best way to make money is to, I

98
00:04:34,639 --> 00:04:36,240
don't know, use it for pizza delivery.

99
00:04:36,319 --> 00:04:38,959
Speaker 2: That's a great analogy. Or painting a giant Coca Cola

100
00:04:39,000 --> 00:04:41,120
logo on the side and just flying it past the beach.

101
00:04:41,160 --> 00:04:41,959
It makes no sense.

102
00:04:42,160 --> 00:04:45,800
Speaker 1: It's inventing cold fusion to power a single neon open

103
00:04:45,879 --> 00:04:48,360
sign yes perfect.

104
00:04:48,920 --> 00:04:50,680
Speaker 2: If you have the machine God, you don't need to

105
00:04:50,680 --> 00:04:54,040
put a sponsored by Pepsi sticker on its forehead. So

106
00:04:54,240 --> 00:04:57,279
by just pivoting to ads demis is basically saying open

107
00:04:57,319 --> 00:05:01,639
AI is admitting publicly that they're tech isn't as magical

108
00:05:01,759 --> 00:05:03,959
or as close to perfection as they claim it is.

109
00:05:04,160 --> 00:05:07,560
Speaker 1: It's a huge reality check. And this source material points

110
00:05:07,600 --> 00:05:11,560
out this really interesting contrast in strategy right now because

111
00:05:11,600 --> 00:05:14,560
Google Google is kind of taking the high road here.

112
00:05:14,439 --> 00:05:16,839
Speaker 2: Which is a strategy in itself. I mean, they're playing

113
00:05:16,879 --> 00:05:19,639
forty E chss while everyone else is playing checkers.

114
00:05:19,279 --> 00:05:22,800
Speaker 1: Right demonstated very clearly, Google has no plans at the

115
00:05:22,839 --> 00:05:25,240
moment to put ads in the Gemini app.

116
00:05:25,319 --> 00:05:27,800
Speaker 2: And okay, let's be real, that phrase at the moment

117
00:05:28,000 --> 00:05:30,639
is doing a lot of heavy lifting. So sure, Google

118
00:05:30,680 --> 00:05:32,600
is an ad company, it's what they do, but for

119
00:05:32,680 --> 00:05:34,040
now they're holding back.

120
00:05:34,360 --> 00:05:38,279
Speaker 1: But just compare the products right now today. The source

121
00:05:38,319 --> 00:05:42,480
describes Google Gemini as a simple core offering. It's clean,

122
00:05:42,759 --> 00:05:45,040
you open it, you talk to the AI, that's it,

123
00:05:45,560 --> 00:05:47,879
and then you have open AI, who this source says,

124
00:05:48,000 --> 00:05:49,120
is fumbling around.

125
00:05:49,240 --> 00:05:52,199
Speaker 2: And that word fumbling is so important. It implies a

126
00:05:52,319 --> 00:05:54,759
lack of a clear plan, a little bit of desperation.

127
00:05:54,879 --> 00:05:55,600
Speaker 1: Maybe it does.

128
00:05:55,959 --> 00:05:58,759
Speaker 2: And so Google is winning this perception battle right now

129
00:05:58,839 --> 00:06:01,839
just by doing nothing. They're not cluttering their user experience.

130
00:06:01,839 --> 00:06:04,319
They're letting the text speak for itself. And you have

131
00:06:04,319 --> 00:06:07,399
to remember the resources they have. Google is the king

132
00:06:07,600 --> 00:06:08,959
of online advertising.

133
00:06:09,360 --> 00:06:11,560
Speaker 1: That's the wild irony. The one company that could do

134
00:06:11,600 --> 00:06:12,800
it best is choosing not to.

135
00:06:13,519 --> 00:06:17,519
Speaker 2: Exactly, if anyone knows how to perfectly integrate ADS, it's them.

136
00:06:17,800 --> 00:06:21,120
They have the whole global infrastructure, the auction systems, the

137
00:06:21,199 --> 00:06:25,319
tracking pixels on what half the Internet, But they're deliberately

138
00:06:25,360 --> 00:06:26,439
holding off on Gemini.

139
00:06:26,879 --> 00:06:27,600
Speaker 1: Why tell me?

140
00:06:27,920 --> 00:06:30,480
Speaker 2: Because they know it degrades the experience. They've learned this,

141
00:06:30,759 --> 00:06:32,920
they know that when you're trying to build trust with

142
00:06:33,000 --> 00:06:36,839
this new form of intelligence, an AD is this jarring

143
00:06:37,199 --> 00:06:41,160
commercial reminder that you're not a collaborator, you're a customer,

144
00:06:41,240 --> 00:06:41,959
you're a target.

145
00:06:42,279 --> 00:06:43,040
Speaker 1: It breaks the spell.

146
00:06:43,120 --> 00:06:46,959
Speaker 2: It completely breaks the spell, the illusion of this pure intelligence.

147
00:06:46,959 --> 00:06:49,480
I mean, imagine you're asking an AI for help writing

148
00:06:49,480 --> 00:06:51,800
a eulogy for a loved one. Oh wow, and in

149
00:06:51,839 --> 00:06:55,680
the middle of this very personal emotional moment, it pauses

150
00:06:55,720 --> 00:06:58,360
to tell you about a twenty percent discount on mattresses.

151
00:06:59,319 --> 00:07:00,240
Speaker 1: The magic is on.

152
00:07:00,360 --> 00:07:03,240
Speaker 2: It's gone instantly. You're no longer talking to a genius

153
00:07:03,240 --> 00:07:05,439
in a box, You're talking to a digital salesman.

154
00:07:05,560 --> 00:07:09,399
Speaker 1: And Demis mentioned this specifically. He said AD implementation has

155
00:07:09,399 --> 00:07:11,519
to be handled very carefully.

156
00:07:11,160 --> 00:07:14,720
Speaker 2: And what's the reason he gave privacy. See that's the

157
00:07:14,759 --> 00:07:17,720
other angle. Google is just sitting back and watching open

158
00:07:17,759 --> 00:07:21,399
ai walk into a minefield. They're letting open ai be

159
00:07:21,519 --> 00:07:22,800
the canary in the coal mine.

160
00:07:22,879 --> 00:07:24,879
Speaker 1: You guys, go ahead, you try to figure out how

161
00:07:24,920 --> 00:07:27,600
to put ads in a chatbot without completely creeping everyone out.

162
00:07:27,680 --> 00:07:30,800
Speaker 2: We'll just watch from over here and see if you explode.

163
00:07:30,839 --> 00:07:32,279
Speaker 1: That is such a power move.

164
00:07:32,360 --> 00:07:35,480
Speaker 2: It's a total power move. Go ahead, touch the electric

165
00:07:35,519 --> 00:07:37,519
fence for us, tell us how it feels, and.

166
00:07:37,519 --> 00:07:42,399
Speaker 1: It positions Google, at least for now, as the safer, cleaner,

167
00:07:42,759 --> 00:07:46,079
more serious alternative. They look like the adults in the room,

168
00:07:46,360 --> 00:07:49,319
while open Ai looks a little desperate, like they're scrambling

169
00:07:49,360 --> 00:07:49,959
for cash.

170
00:07:50,000 --> 00:07:52,759
Speaker 2: And honestly, if you look at Google's recent history with

171
00:07:52,800 --> 00:07:56,319
their own AI launches, They've learned the hard way that

172
00:07:56,439 --> 00:07:59,839
rushing things and getting it wrong breaks user trust. Remember

173
00:07:59,839 --> 00:08:03,040
some of the early weird answers from their search overviews.

174
00:08:03,079 --> 00:08:05,519
Speaker 1: So, yeah, the put glue on pizza thing exactly.

175
00:08:05,560 --> 00:08:09,680
Speaker 2: They know the spotlight is incredibly hot, so letting open

176
00:08:09,720 --> 00:08:12,399
ai take all the heat for this ad experiment. It's

177
00:08:12,519 --> 00:08:15,000
just a brilliant patient strategy.

178
00:08:15,279 --> 00:08:17,800
Speaker 1: Speaking of taking the heat and scrambling this brings us

179
00:08:17,839 --> 00:08:20,199
to what I think is maybe the most damaging part

180
00:08:20,240 --> 00:08:23,319
of this whole story for open Ai, the flip flop,

181
00:08:23,399 --> 00:08:25,720
the flip flop, the walk back. We have to talk

182
00:08:25,720 --> 00:08:30,399
about Sam Moltman because the Internet, the Internet never.

183
00:08:30,279 --> 00:08:33,519
Speaker 2: Forgets, No, it does not. The digital footprint is forever,

184
00:08:33,840 --> 00:08:36,399
and for Sam it's pretty glaring right now.

185
00:08:36,879 --> 00:08:39,120
Speaker 1: So okay, recently he's been warming up to the idea

186
00:08:39,120 --> 00:08:41,799
of ads. But let's just rewind the clock to twenty

187
00:08:41,799 --> 00:08:43,320
twenty four, not even that long ago.

188
00:08:43,519 --> 00:08:46,480
Speaker 2: We are talking very recent history. This isn't some tweet

189
00:08:46,480 --> 00:08:47,960
from twenty ten. We're digging up.

190
00:08:48,039 --> 00:08:51,480
Speaker 1: No, he made a statement where he called ads a

191
00:08:51,600 --> 00:08:55,600
last resort and he explicitly said, and this is a quote,

192
00:08:55,720 --> 00:08:58,639
that moving to ads would mean the company is kind

193
00:08:58,679 --> 00:08:59,279
of in trouble.

194
00:08:59,360 --> 00:09:01,960
Speaker 2: He literally at his own trap. He created the metric

195
00:09:02,039 --> 00:09:04,759
for his own failure. We mean, well, think about it

196
00:09:04,759 --> 00:09:08,120
from a purely logical standpoint. He the CEO, defined the

197
00:09:08,120 --> 00:09:11,200
success condition for his own company. He said, in essence,

198
00:09:11,480 --> 00:09:13,799
if we're doing well, we won't need ads. If we're

199
00:09:13,799 --> 00:09:16,000
failing and in trouble, we will have ads.

200
00:09:16,200 --> 00:09:18,279
Speaker 1: It's a binary state he created himself.

201
00:09:18,480 --> 00:09:22,759
Speaker 2: It's a binary a then B and now they are

202
00:09:23,039 --> 00:09:27,320
very publicly exploring ADS. So by his own definition or trouble,

203
00:09:27,440 --> 00:09:30,360
it's an inescapable conclusion. And you can't even argue with

204
00:09:30,399 --> 00:09:33,720
it because the premise came directly from his own mouth.

205
00:09:34,000 --> 00:09:37,320
He signaled to the entire market that adds equal failure.

206
00:09:38,000 --> 00:09:40,000
So what does the market see now?

207
00:09:40,279 --> 00:09:40,720
Speaker 1: Failure?

208
00:09:40,840 --> 00:09:44,120
Speaker 2: It's unbelievable, but it gets so much.

209
00:09:43,919 --> 00:09:46,000
Speaker 1: Worse it does. The Lex Friedman.

210
00:09:45,720 --> 00:09:47,759
Speaker 2: Interview, Oh, the Lex Fridman interview.

211
00:09:47,879 --> 00:09:49,679
Speaker 1: If you haven't seen this clip, you have to find

212
00:09:49,679 --> 00:09:52,639
it because it's not just the words, it's his body language,

213
00:09:52,679 --> 00:09:55,440
his tone. He was so adamant. He called ADS and

214
00:09:55,559 --> 00:09:59,240
Chat GPT a dystopian idea dystopian.

215
00:09:59,399 --> 00:10:02,600
Speaker 2: That is such a word. It's not suboptimal or annoying.

216
00:10:02,840 --> 00:10:07,080
It's dystopian that invokes George Orwell, that invokes Blade Runner.

217
00:10:07,279 --> 00:10:09,919
That is a society collapsing, nightmarish future.

218
00:10:10,080 --> 00:10:12,879
Speaker 1: And he painted this very specific picture, didn't there He

219
00:10:12,960 --> 00:10:15,919
visualized the dystopia force. He said, you know, imagine you

220
00:10:15,960 --> 00:10:18,320
go to the AI for advice. Maybe it's something personal,

221
00:10:18,399 --> 00:10:20,440
or maybe it's just hey, where should I go on vacation,

222
00:10:20,919 --> 00:10:26,120
and instead of a genuine objective answer, the AI replies, well,

223
00:10:26,200 --> 00:10:28,639
you should really think about buying this product, or you

224
00:10:28,639 --> 00:10:31,799
should go to this specific resort for your vacation.

225
00:10:32,200 --> 00:10:36,000
Speaker 2: He described with perfect clarity the exact user fear he

226
00:10:36,120 --> 00:10:39,919
was trying to avoid. He validated everyone's worst nightmare about

227
00:10:39,919 --> 00:10:43,440
this technology. He admitted that the moment you introduce ads

228
00:10:43,559 --> 00:10:46,879
you change the nature of the truth you receive. And

229
00:10:46,919 --> 00:10:50,080
this is the critical point. This is the difference between

230
00:10:50,120 --> 00:10:51,759
a search engine and an AI.

231
00:10:52,000 --> 00:10:54,559
Speaker 1: Okay, break that down for me. Why is it so different.

232
00:10:54,600 --> 00:10:55,840
It's just text on a screen.

233
00:10:56,000 --> 00:10:59,240
Speaker 2: No, No, it's fundamentally different. On Google Search, you type

234
00:10:59,240 --> 00:11:02,720
a query and you get a list of ten blue links, right,

235
00:11:02,799 --> 00:11:05,440
list of options, and you, the human, are still the

236
00:11:05,480 --> 00:11:08,000
final arbiter of truth. You do the sorting. You see

237
00:11:08,039 --> 00:11:11,000
the little sponsored tag, You consciously or subconsciously ignore it.

238
00:11:11,039 --> 00:11:13,240
You click the Wikipedia link, you click a news article.

239
00:11:13,440 --> 00:11:16,320
You synthesize the answer yourself. But with an LM, a

240
00:11:16,440 --> 00:11:19,399
large language model, it gives you one answer, right.

241
00:11:19,440 --> 00:11:20,559
Speaker 1: It does the thinking for you.

242
00:11:20,799 --> 00:11:23,480
Speaker 2: It does the synthesis for you. It presents itself as

243
00:11:23,480 --> 00:11:28,080
the definitive truth. And if that one single authoritative answer

244
00:11:28,360 --> 00:11:32,279
has been secretly influenced by money, the entire concept of

245
00:11:32,320 --> 00:11:34,120
truth has been corrupted at the source.

246
00:11:34,279 --> 00:11:35,399
Speaker 1: You can't trust it anymore.

247
00:11:35,440 --> 00:11:38,840
Speaker 2: You can't If the AI recommends a specific hotel, is

248
00:11:38,879 --> 00:11:41,799
it actually the best hotel for me based on my needs?

249
00:11:41,919 --> 00:11:43,720
Or is it just the one that paid the most

250
00:11:43,720 --> 00:11:46,840
to be in that answer? The objectivity is completely gone.

251
00:11:46,919 --> 00:11:50,559
And Altman knew this. He said he hates ads as

252
00:11:50,559 --> 00:11:54,559
an esthetic choice. He wanted the paid subscription model specifically

253
00:11:54,720 --> 00:11:56,759
so that the answers wouldn't be bought and sold.

254
00:11:56,799 --> 00:11:59,279
Speaker 1: He the one line that really stuck with me, He said,

255
00:11:59,799 --> 00:12:01,200
I like knowing I am not.

256
00:12:01,200 --> 00:12:04,360
Speaker 2: The product, and that resonates so deeply with people today.

257
00:12:04,399 --> 00:12:07,120
We are so so tired of being the product on Facebook,

258
00:12:07,159 --> 00:12:10,279
on Instagram, on TikTok chat GPT. You felt different. It

259
00:12:10,320 --> 00:12:13,960
felt like a clean utility, like a calculator exactly. Calculators

260
00:12:13,960 --> 00:12:16,399
don't have ads. Can you imagine if you typed two

261
00:12:16,559 --> 00:12:19,759
plus two into your calculator and it flashed back it's four.

262
00:12:19,919 --> 00:12:22,159
But have you considered that for just five dollars you

263
00:12:22,159 --> 00:12:24,159
can upgrade your life with this new soda.

264
00:12:24,240 --> 00:12:26,679
Speaker 1: I would throw that calculator out the window. It would

265
00:12:26,720 --> 00:12:27,559
be infuriating.

266
00:12:27,679 --> 00:12:30,320
Speaker 2: It completely destroys the utility of the tool, and now

267
00:12:30,440 --> 00:12:33,200
that certainty, that feeling of having a clean tool, it's

268
00:12:33,240 --> 00:12:34,679
just eroding before our eyes.

269
00:12:35,080 --> 00:12:38,679
Speaker 1: And that erosion leads directly to the trust gap. The

270
00:12:38,720 --> 00:12:41,240
source material talks about this, and I think it's crucial

271
00:12:41,919 --> 00:12:45,879
trust in. Sam Altman was already let's be generous and

272
00:12:45,919 --> 00:12:46,600
call it shaky.

273
00:12:46,799 --> 00:12:49,240
Speaker 2: It's been a turbulent few years for him to say

274
00:12:49,240 --> 00:12:49,679
the least.

275
00:12:49,759 --> 00:12:51,879
Speaker 1: I mean, you've got the whole history with Elon Musk.

276
00:12:52,159 --> 00:12:54,919
The founding of the company is a nonprofit. Then this

277
00:12:55,000 --> 00:12:58,639
weird pivot to a capped profit model that nobody really understands,

278
00:12:59,240 --> 00:13:02,799
the lawsuits, and of course the insane firing and then

279
00:13:02,879 --> 00:13:04,799
rehiring Debaggle in November twenty three.

280
00:13:04,919 --> 00:13:06,159
Speaker 2: It's been a total soap opera.

281
00:13:06,600 --> 00:13:09,360
Speaker 1: But this reversal on ADS feels different. It feels more

282
00:13:09,399 --> 00:13:12,639
personal because it hits the user directly in the product experience.

283
00:13:12,919 --> 00:13:16,200
Speaker 2: It's a betrayal of the product's promise. It's one thing

284
00:13:16,240 --> 00:13:19,600
to have boardroom drama, it's another thing to fundamentally change

285
00:13:19,600 --> 00:13:21,879
the nature of the tool you've given to millions of people.

286
00:13:22,519 --> 00:13:25,000
When you tell your users this is a dystopian idea

287
00:13:25,039 --> 00:13:26,559
and we will never do it. And then a year

288
00:13:26,679 --> 00:13:29,080
later you say, actually, you know what we're doing it.

289
00:13:29,759 --> 00:13:31,879
You're not just changing a business model.

290
00:13:31,960 --> 00:13:32,559
Speaker 1: What are you doing?

291
00:13:32,720 --> 00:13:36,320
Speaker 2: You are, in a way gaslighting your user base. You

292
00:13:36,399 --> 00:13:39,159
are telling them that their definition of dystopian was wrong

293
00:13:39,279 --> 00:13:42,480
and now it's just necessary. It makes you question everything

294
00:13:42,480 --> 00:13:43,840
else they've said, right.

295
00:13:43,679 --> 00:13:45,840
Speaker 1: If they're willing to flip flop on this, what else

296
00:13:45,919 --> 00:13:46,559
is on the table?

297
00:13:46,720 --> 00:13:49,960
Speaker 2: Exactly? If the last resort is happy? Now, what other

298
00:13:50,039 --> 00:13:53,840
last resorts are waiting in the wings? Selling user training data,

299
00:13:54,039 --> 00:13:58,480
compromising on privacy promises, giving back doors to government agencies.

300
00:13:58,720 --> 00:14:01,879
It just opens this Pandora box of skepticism. If your

301
00:14:01,960 --> 00:14:05,120
big maral redline was no ads and you've just casually

302
00:14:05,159 --> 00:14:08,240
stepped over it, where's the next red line? Or is

303
00:14:08,279 --> 00:14:08,639
there one?

304
00:14:08,679 --> 00:14:11,120
Speaker 1: And that skepticism leads us right into the belly of

305
00:14:11,159 --> 00:14:15,360
the beast. The money, the incentives. Why are they doing this?

306
00:14:15,519 --> 00:14:17,279
Why risk all this backlash?

307
00:14:17,399 --> 00:14:19,639
Speaker 2: Show me the incentive and I'll show you the outcome.

308
00:14:19,759 --> 00:14:23,039
Speaker 1: The source quotes that famous principle from Charlie Munger and

309
00:14:23,320 --> 00:14:25,919
man does it ever apply here? Let's look at the

310
00:14:25,919 --> 00:14:27,240
incentive for open.

311
00:14:27,000 --> 00:14:29,679
Speaker 2: AI because the pot of gold at the end of

312
00:14:29,720 --> 00:14:33,360
this particular rainbow is bigger than anything the Internet has

313
00:14:33,399 --> 00:14:36,919
ever seen before. If they can build the best advertising

314
00:14:36,919 --> 00:14:41,159
business on the planet, they don't just become a successful company,

315
00:14:41,799 --> 00:14:44,480
they become a monster, a titan.

316
00:14:44,639 --> 00:14:48,480
Speaker 1: The source makes a really compelling comparison to Meta, to Facebook,

317
00:14:48,799 --> 00:14:51,399
and this part, I have to admit it gave me chills.

318
00:14:51,480 --> 00:14:54,720
Speaker 2: It should. It reveals the raw mechanics of the Internet economy.

319
00:14:54,960 --> 00:14:58,360
Speaker 1: So the narrator of our source material admits, and I

320
00:14:58,360 --> 00:15:00,159
think a lot of us can relate to this, that

321
00:15:00,200 --> 00:15:02,120
they actually click on most of the ads they see

322
00:15:02,159 --> 00:15:03,240
on Instagram or Facebook.

323
00:15:03,279 --> 00:15:04,240
Speaker 2: I do too, embarrassingly.

324
00:15:04,360 --> 00:15:07,080
Speaker 1: Why though, yea, And the reason given is it's like

325
00:15:07,159 --> 00:15:08,360
someone's reading my brain.

326
00:15:08,840 --> 00:15:11,039
Speaker 2: We have all had that moment. You're having a spoken

327
00:15:11,080 --> 00:15:14,360
conversation with your spouse about a specific brand of coffee maker,

328
00:15:14,799 --> 00:15:17,639
and an hour later, an ad for that exact coffee

329
00:15:17,639 --> 00:15:20,120
maker is at the top of your Instagram feed. It

330
00:15:20,200 --> 00:15:21,399
feels like telepathy.

331
00:15:21,639 --> 00:15:26,080
Speaker 1: It's spooky, and Meta prints literally billions of dollars a

332
00:15:26,159 --> 00:15:30,200
year from this spooky brain reading ability to predict what

333
00:15:30,240 --> 00:15:32,320
we want before we even know we want it.

334
00:15:32,399 --> 00:15:34,759
Speaker 2: Okay, but let's look at the technical reason why Meta

335
00:15:34,799 --> 00:15:35,240
is so good.

336
00:15:35,240 --> 00:15:35,480
Speaker 1: At that.

337
00:15:36,039 --> 00:15:38,720
Speaker 2: They have what's called the social graph. They know who

338
00:15:38,759 --> 00:15:41,320
your friends are, what pages you like, what photos you

339
00:15:41,360 --> 00:15:43,440
stop scrolling to look at for an extra half second.

340
00:15:43,759 --> 00:15:47,639
They infer your interests based on your behavior. It's basically

341
00:15:47,879 --> 00:15:49,360
very very sophisticated guesswork.

342
00:15:49,480 --> 00:15:51,120
Speaker 1: Okay, so they're guessing what.

343
00:15:51,120 --> 00:15:54,360
Speaker 2: I want right now. Take that predictive power and apply

344
00:15:54,440 --> 00:15:57,440
it to an LM like chat GPT. Meta knows what

345
00:15:57,480 --> 00:15:59,919
you like based on what you click, but chat GPT

346
00:16:00,399 --> 00:16:01,759
chat GPT knows what you think.

347
00:16:01,879 --> 00:16:04,320
Speaker 1: Oh wow. Put that into perspective for me.

348
00:16:04,399 --> 00:16:06,679
Speaker 2: When you type a query into chat GPT, you are

349
00:16:06,759 --> 00:16:10,200
articulating your intent with perfect explicit clarity. You're not just

350
00:16:10,200 --> 00:16:13,360
searching for shoes on Google. You're typing into chat GPT.

351
00:16:14,200 --> 00:16:16,000
I need a new pair of running shoes that are

352
00:16:16,039 --> 00:16:18,320
good for someone with flat feet cost under one hundred

353
00:16:18,360 --> 00:16:21,120
dollars because I'm training for my first marathon next month

354
00:16:21,120 --> 00:16:22,159
and my knees have been heard it.

355
00:16:22,279 --> 00:16:24,000
Speaker 1: That's that is so much more data.

356
00:16:24,360 --> 00:16:27,240
Speaker 2: It's not just more data, it's the context. It's the

357
00:16:27,279 --> 00:16:30,840
why Meta has to guess, the why. Open ai is

358
00:16:30,919 --> 00:16:34,639
just told the why. It knows, the nuance, the insecurity,

359
00:16:34,679 --> 00:16:38,840
the motivation behind every question. If OpenAI can leverage that

360
00:16:39,000 --> 00:16:42,399
depth of conversation data to target ads, they won't just

361
00:16:42,440 --> 00:16:44,919
beat meta at this game. They will create the most

362
00:16:45,039 --> 00:16:47,679
powerful advertising machine in human history.

363
00:16:47,759 --> 00:16:49,720
Speaker 1: They will just be sitting on a pile of cash.

364
00:16:49,759 --> 00:16:52,120
Speaker 2: You'll be sitting on the entire global economy.

365
00:16:52,240 --> 00:16:53,960
Speaker 1: It's the infinite money printer button.

366
00:16:54,080 --> 00:16:56,000
Speaker 2: That is exactly how the source describes it. And so

367
00:16:56,039 --> 00:16:59,559
the real question is how long can a company that

368
00:16:59,720 --> 00:17:03,799
is all accounts hemorrhaging cash? How long can they resist

369
00:17:03,799 --> 00:17:06,279
pressing that big, red, shiny button.

370
00:17:06,279 --> 00:17:08,599
Speaker 1: We should probably touch on that hamorrhaging cash part. I

371
00:17:08,599 --> 00:17:10,720
think a lot of people assume that because chat GPT

372
00:17:10,839 --> 00:17:13,079
is so popular, it must be insanely profits.

373
00:17:13,240 --> 00:17:17,599
Speaker 2: The exact opposite. It is astonishingly mind bogglingly expensive to run.

374
00:17:17,920 --> 00:17:20,559
Every single time you type a query and hit enter,

375
00:17:20,960 --> 00:17:23,720
a massive bank of GPUs somewhere has to spin up.

376
00:17:23,759 --> 00:17:26,240
Speaker 1: These are like super powerful graphics.

377
00:17:25,839 --> 00:17:31,000
Speaker 2: Cards, super powerful, very expensive graphics cards. They consume enormous

378
00:17:31,000 --> 00:17:33,279
amounts of electricity, they generate a ton of heat that

379
00:17:33,319 --> 00:17:37,319
needs to be cooled, and they perform trillions of calculations

380
00:17:37,400 --> 00:17:39,440
just to figure out the next word to generate in

381
00:17:39,480 --> 00:17:42,720
your sentence. Some public estimates put the operational cost of

382
00:17:42,799 --> 00:17:46,640
chat GPT at seven hundred thousand dollars a day, or even.

383
00:17:46,440 --> 00:17:49,599
Speaker 1: More seven hundred thousand dollars a day just so I

384
00:17:49,599 --> 00:17:52,119
can ask it to write a poem about my dog exactly.

385
00:17:52,359 --> 00:17:54,720
Speaker 2: And the subscriptions, you know, the twenty dollars a month

386
00:17:54,799 --> 00:17:57,480
for Chat GPT plus. That helps it brings in some revenue,

387
00:17:57,880 --> 00:18:00,319
but does it cover the insane R and d P cost.

388
00:18:00,559 --> 00:18:02,839
Does it cover the hundreds of millions of dollars it

389
00:18:02,960 --> 00:18:06,200
costs just to train the next model like GPT five?

390
00:18:06,400 --> 00:18:09,480
Probably not, almost certainly not. The math just doesn't balance

391
00:18:09,480 --> 00:18:11,799
out with subscriptions alone, not at this scale.

392
00:18:11,839 --> 00:18:14,279
Speaker 1: So you have the investors, right Microsoft is a big one.

393
00:18:14,400 --> 00:18:16,279
They're looking at this massive burn rate.

394
00:18:16,559 --> 00:18:19,680
Speaker 2: They've corn in what thirteen billion dollars now, and at

395
00:18:19,680 --> 00:18:21,680
some point they're going to come knocking and say, okay,

396
00:18:21,759 --> 00:18:25,240
where is our return on investment. Software is supposed to

397
00:18:25,240 --> 00:18:27,240
have high margins. You build it once, you sell it

398
00:18:27,279 --> 00:18:31,559
a million times. But Ai right now has terrible margins

399
00:18:31,599 --> 00:18:35,200
because the cost of running it, the compute cost is

400
00:18:35,319 --> 00:18:38,599
so high for every single user query and ads.

401
00:18:38,720 --> 00:18:41,400
Speaker 1: Ads are the one business model on the Internet that

402
00:18:41,480 --> 00:18:43,839
scales almost infinitely with very.

403
00:18:43,720 --> 00:18:45,680
Speaker 2: Low marginal costs. That's why it's so tempting.

404
00:18:45,880 --> 00:18:49,319
Speaker 1: Okay, But to be fair, open AI claims, and I

405
00:18:49,319 --> 00:18:51,680
have to quote them here, we're not going to accept

406
00:18:51,720 --> 00:18:53,960
money to influence the chat answers, right.

407
00:18:54,000 --> 00:18:55,599
Speaker 2: The official line is trust us.

408
00:18:55,720 --> 00:18:57,400
Speaker 1: But the source argues, and I think this is a

409
00:18:57,400 --> 00:19:01,640
powerful point that the incentive is just too high. If

410
00:19:01,640 --> 00:19:03,880
you are bleeding cash every single day and there's a

411
00:19:03,880 --> 00:19:06,960
button on your desk that says press for billions of dollars,

412
00:19:07,200 --> 00:19:10,759
are we really supposed to believe they won't eventually push it.

413
00:19:10,759 --> 00:19:14,960
Speaker 2: It creates this massive, unavoidable conflict of interest. Even if

414
00:19:15,000 --> 00:19:18,559
they don't explicitly let an advertiser pay to insert their

415
00:19:18,559 --> 00:19:21,759
product into an answer today, the algorithms that run the

416
00:19:21,799 --> 00:19:24,920
ad system will naturally start to optimize for things that

417
00:19:25,000 --> 00:19:26,160
make more money.

418
00:19:26,279 --> 00:19:28,359
Speaker 1: How does that happen? What does that drift look like?

419
00:19:28,720 --> 00:19:32,200
Speaker 2: Well, let's say the AI's internal metrics realize that when

420
00:19:32,240 --> 00:19:35,400
it gives shorter punch your answers, users are more likely

421
00:19:35,440 --> 00:19:37,920
to see and click on the ad in the sidebar. Okay,

422
00:19:38,079 --> 00:19:42,079
the system might start subtly optimizing or shorter answers. Even

423
00:19:42,079 --> 00:19:44,920
if a longer, more nuanced answer would be better. Or

424
00:19:45,200 --> 00:19:47,680
maybe it realizes that when users are in a buying

425
00:19:47,720 --> 00:19:50,920
mood asking about products, the AD rates are higher, so

426
00:19:51,000 --> 00:19:54,559
it might start to gently steer conversations towards shopping topics.

427
00:19:54,559 --> 00:19:55,200
Speaker 1: It nudges you.

428
00:19:55,319 --> 00:19:57,240
Speaker 2: It nudges you. The AI might not tell you an

429
00:19:57,240 --> 00:19:59,720
outright lie, but it will nudge you. And a subtle

430
00:19:59,799 --> 00:20:03,440
nude from a super intelligent AI that sounds incredibly confident

431
00:20:03,480 --> 00:20:07,279
and authoritative, that is an incredibly powerful form of persuasion.

432
00:20:07,279 --> 00:20:08,559
It's the ultimate influencer.

433
00:20:08,799 --> 00:20:10,880
Speaker 1: It is, and that brings us to the part of

434
00:20:10,920 --> 00:20:14,440
this that actually terrifies me. We've talked about the annoyance

435
00:20:14,480 --> 00:20:17,000
of ads, we've talked about the money, but we have

436
00:20:17,079 --> 00:20:20,119
to talk about privacy because this isn't just about seeing

437
00:20:20,160 --> 00:20:23,279
a banner ad for sneakers. This is what the Force

438
00:20:23,359 --> 00:20:25,079
calls the privacy nightmare.

439
00:20:25,400 --> 00:20:28,799
Speaker 2: This is where we get into the really dark timeline stuff.

440
00:20:29,000 --> 00:20:30,200
This is where it gets personal.

441
00:20:30,440 --> 00:20:33,480
Speaker 1: The source brings up this comment from a community member,

442
00:20:33,519 --> 00:20:37,000
a guy named Nate Hurrah, and Nate asks the one

443
00:20:37,119 --> 00:20:39,359
question that I think everybody should be asking.

444
00:20:39,160 --> 00:20:40,440
Speaker 2: Right now, and what was the question?

445
00:20:40,759 --> 00:20:44,640
Speaker 1: He said, I don't care for ads but how will

446
00:20:44,680 --> 00:20:47,119
they know what to show me? Will they use my

447
00:20:47,240 --> 00:20:48,039
memory against me?

448
00:20:48,440 --> 00:20:52,240
Speaker 2: Will they use my memory against me? That phrasing is

449
00:20:52,440 --> 00:20:54,599
it's haunting. It sounds like something out of a psychological

450
00:20:54,680 --> 00:20:55,400
thriller movie.

451
00:20:55,440 --> 00:20:57,720
Speaker 1: It really does. And the source goes deeper into this.

452
00:20:57,720 --> 00:21:00,000
This is completely different from seeing an ad on an

453
00:21:00,119 --> 00:21:03,400
news website because you once searched for flights to Hawaii.

454
00:21:03,480 --> 00:21:05,880
Speaker 2: Oh, it's a different universe of personal data.

455
00:21:06,000 --> 00:21:10,119
Speaker 1: This implies that the AI is scanning your intimate journal

456
00:21:10,240 --> 00:21:14,119
entries that you've pasted into it, your confidential work notes,

457
00:21:14,480 --> 00:21:17,000
your deepest fears that you've shared with it in a

458
00:21:17,079 --> 00:21:17,960
late night chat.

459
00:21:18,079 --> 00:21:21,680
Speaker 2: Let's play this out with a real example. Millions of

460
00:21:21,720 --> 00:21:24,599
people right now are using chad GPT as a sort

461
00:21:24,640 --> 00:21:25,880
of makeshift therapist.

462
00:21:26,079 --> 00:21:29,920
Speaker 1: Right. It's available twenty four to seven. It's non judgmental, exactly.

463
00:21:30,279 --> 00:21:32,920
Speaker 2: So you talk to it about your anxiety, your struggles,

464
00:21:32,960 --> 00:21:36,359
and your marriage. Your insecurity is about your job performance.

465
00:21:36,599 --> 00:21:38,519
You pour your heart out to this machine.

466
00:21:38,680 --> 00:21:41,519
Speaker 1: Okay, so I'm being extremely vulnerable.

467
00:21:41,160 --> 00:21:44,480
Speaker 2: Right, And the AI has that feature called memory, right,

468
00:21:44,559 --> 00:21:46,680
so it stores these facts to be more helpful later.

469
00:21:46,759 --> 00:21:50,319
It logs away user is anxious about public speaking user

470
00:21:50,359 --> 00:21:53,599
is currently going through a divorce, user feels insecure about

471
00:21:53,599 --> 00:21:54,039
their weight.

472
00:21:54,319 --> 00:21:55,920
Speaker 1: Yeah, so it doesn't have to ask me the same

473
00:21:56,000 --> 00:21:57,680
questions over and over. That's a helpful feature.

474
00:21:57,759 --> 00:22:00,599
Speaker 2: It's a helpful feature for you. But what happens if

475
00:22:00,640 --> 00:22:04,079
that memory database is accessible to the advertising engine. Then

476
00:22:04,119 --> 00:22:06,359
the very next day you open your browser and you

477
00:22:06,400 --> 00:22:09,759
see a targeted ad for a local divorce lawyer, or

478
00:22:09,839 --> 00:22:13,359
for a specific antidepressant, or for a weight loss program

479
00:22:13,359 --> 00:22:16,720
that targets the exact insecurity you confess to the machine

480
00:22:16,759 --> 00:22:17,920
in confidence.

481
00:22:17,480 --> 00:22:20,000
Speaker 1: The night before. That feels predatory.

482
00:22:20,119 --> 00:22:23,680
Speaker 2: It feels predatory because it is. It's using human vulnerability

483
00:22:23,720 --> 00:22:26,440
as a targeting metric. It's one thing for Amazon to

484
00:22:26,480 --> 00:22:29,079
know I bought a toaster and recommend another kitchen appliance.

485
00:22:29,359 --> 00:22:31,720
It's a completely different thing for an intelligence to know

486
00:22:31,759 --> 00:22:35,000
that I feel profoundly lonely and then use that knowledge

487
00:22:35,079 --> 00:22:37,319
to sell me a subscription to a dating app. That

488
00:22:37,440 --> 00:22:41,519
is literally using your memory, your own private thoughts against

489
00:22:41,559 --> 00:22:43,319
you to extract money.

490
00:22:43,640 --> 00:22:47,359
Speaker 1: And we know this isn't just paranoia. Someone actually asked Grock,

491
00:22:47,680 --> 00:22:51,400
which is Elon Musk's AI, about Sam Altman's quotes on this,

492
00:22:51,839 --> 00:22:55,079
and Grock confirmed that Altman himself previously called this exact

493
00:22:55,079 --> 00:22:59,240
scenario dystopian. Even the other ais are keeping the receipts on.

494
00:22:59,240 --> 00:23:02,279
Speaker 2: Him are fascinating because they're all fact checking each other

495
00:23:02,319 --> 00:23:05,039
in public. But really, looking at the track record of

496
00:23:05,039 --> 00:23:07,680
big tech over the last decade, why on earth should

497
00:23:07,720 --> 00:23:09,519
we trust open ai to be any different.

498
00:23:09,759 --> 00:23:12,000
Speaker 1: That's the exact point the source makes every right look

499
00:23:12,039 --> 00:23:16,440
at Facebook, Cabridge Analytica, all of the data scandals, and what.

500
00:23:16,319 --> 00:23:19,000
Speaker 2: Was the end result? Did Facebook get shut down? Did

501
00:23:19,000 --> 00:23:22,440
they stop showing hyper targeted ads? No, they paid a fine,

502
00:23:22,599 --> 00:23:24,319
a big one, but just a fine.

503
00:23:24,160 --> 00:23:25,480
Speaker 1: A cost of doing business.

504
00:23:25,519 --> 00:23:29,200
Speaker 2: Exactly. The source mentions that Facebook makes hundreds of millions

505
00:23:29,240 --> 00:23:32,240
of dollars just from letting fraudulent scam ads run on

506
00:23:32,279 --> 00:23:35,079
its platform, and they largely turn a blind eye because

507
00:23:35,119 --> 00:23:37,799
the revenue is so massive. The fines are a rounding

508
00:23:37,880 --> 00:23:40,559
error for them. A five billion dollar fine sounds like

509
00:23:40,559 --> 00:23:42,519
a lot to us, but if you make one hundred

510
00:23:42,519 --> 00:23:45,799
and twenty billion dollars a year, it's just a business expense.

511
00:23:45,839 --> 00:23:46,480
It's attacks.

512
00:23:46,640 --> 00:23:48,720
Speaker 1: So the fears that open ai looks at that model

513
00:23:48,759 --> 00:23:50,359
and just says, hey, we can afford to pay a

514
00:23:50,400 --> 00:23:52,720
parking ticket every few years if.

515
00:23:52,559 --> 00:23:55,400
Speaker 2: The choices go bankrupt because we can't afford our servers

516
00:23:55,680 --> 00:23:58,839
or violate some privacy norms, make billions of dollars and

517
00:23:58,880 --> 00:24:01,119
then pay a small fine and issue an apology later.

518
00:24:02,079 --> 00:24:05,720
Which path do you think a for profit corporation chooses,

519
00:24:06,240 --> 00:24:09,000
especially one under immense pressure to deliver a return to

520
00:24:09,079 --> 00:24:12,920
investors like Microsoft who have thirteen billion dollars on the line.

521
00:24:13,000 --> 00:24:16,039
Speaker 1: It's a really sobering thought. We are essentially just relying

522
00:24:16,119 --> 00:24:19,720
on their moral compass, which, as we've already established with

523
00:24:19,720 --> 00:24:23,480
the whole last resort flip flop, seems to be spinning

524
00:24:23,480 --> 00:24:24,839
a bit wildly right now.

525
00:24:25,079 --> 00:24:27,960
Speaker 2: And this is where it gets technically tricky too. They'll

526
00:24:28,039 --> 00:24:32,160
use language like anonymized data to reassure.

527
00:24:31,640 --> 00:24:34,119
Speaker 1: Everyone right, don't worry, your name isn't attached to it.

528
00:24:34,480 --> 00:24:38,359
Speaker 2: But with natural language, it is incredibly difficult to truly

529
00:24:38,400 --> 00:24:42,359
anonymize data. If I write a long, specific story about

530
00:24:42,400 --> 00:24:45,279
my day at work, you can remove my name, but

531
00:24:45,359 --> 00:24:48,000
the context might be so unique that it could still

532
00:24:48,039 --> 00:24:48,920
be traced back to me.

533
00:24:49,440 --> 00:24:52,240
Speaker 1: If I tell the AI about a specific problem I'm

534
00:24:52,279 --> 00:24:54,640
having with a specific project at a specific type.

535
00:24:54,519 --> 00:24:57,839
Speaker 2: Of company, exactly the data is the person. So this

536
00:24:57,960 --> 00:25:01,160
idea that they can build some perfect impati wall between

537
00:25:01,400 --> 00:25:04,079
what you tell the AI and what the ad engine knows.

538
00:25:04,480 --> 00:25:07,920
It's technically very very difficult, if not impossible, especially if

539
00:25:07,960 --> 00:25:10,279
you want the ads to be highly relevant, and if the.

540
00:25:10,279 --> 00:25:12,279
Speaker 1: Ads aren't relevant, they don't make as much money.

541
00:25:12,319 --> 00:25:15,200
Speaker 2: Precisely, to make the big money, they have to invade

542
00:25:15,200 --> 00:25:17,759
the privacy. There really isn't a middle ground.

543
00:25:17,599 --> 00:25:22,240
Speaker 1: And users are starting to notice this potential betrayal, which

544
00:25:22,319 --> 00:25:24,440
leads us to the final part of this the root

545
00:25:24,480 --> 00:25:25,519
awakening in the market.

546
00:25:25,640 --> 00:25:26,599
Speaker 2: The backlash.

547
00:25:26,720 --> 00:25:29,359
Speaker 1: Yeah, let's talk about the fallout, because people are not

548
00:25:29,400 --> 00:25:33,119
just sitting back and taking this. The source highlights a

549
00:25:33,160 --> 00:25:34,720
potential user exodus.

550
00:25:34,920 --> 00:25:39,240
Speaker 2: People are on Twitter on Reddit posting screenshots of them

551
00:25:39,279 --> 00:25:44,079
canceling their subscriptions. The hashtag goodbye chat GPT has been trending.

552
00:25:44,240 --> 00:25:44,920
Speaker 1: Where are they going?

553
00:25:45,200 --> 00:25:48,200
Speaker 2: Well, in the ultimate irony, they might be going to Google,

554
00:25:48,319 --> 00:25:50,319
to the ad giant, to the ad giant to get

555
00:25:50,319 --> 00:25:53,680
away from ads. It sounds completely insane when you say

556
00:25:53,720 --> 00:25:55,480
it out loud, it does, but think about it from

557
00:25:55,519 --> 00:25:58,720
the user's perspective. Right now, Google Gemini is free and

558
00:25:58,759 --> 00:26:02,119
it is ad free. Course calls this the friction problem.

559
00:26:02,279 --> 00:26:05,119
If using chat GPT suddenly becomes annoying, if you have

560
00:26:05,160 --> 00:26:07,160
to close a pop up AD, or you have to

561
00:26:07,200 --> 00:26:10,960
constantly wonder if the answer you're getting is sponsored, that

562
00:26:11,119 --> 00:26:11,599
is friction.

563
00:26:11,880 --> 00:26:13,799
Speaker 1: And we are lazy creatures by nature.

564
00:26:14,160 --> 00:26:16,599
Speaker 2: We are we always follow the path of least resistance.

565
00:26:17,039 --> 00:26:18,640
If I have to click an X to close a

566
00:26:18,680 --> 00:26:20,839
banner AD before I get my answer, that's like a

567
00:26:20,880 --> 00:26:24,640
tiny microaggression from the software. If Google Gemini can offer

568
00:26:24,680 --> 00:26:27,839
a comparable level of intelligence, and let's be honest, Gemini

569
00:26:27,839 --> 00:26:31,519
one point five pro is very very good. It's arguably

570
00:26:31,559 --> 00:26:34,400
on par with GPT four for many tasks. If the

571
00:26:34,480 --> 00:26:37,440
quality is the same, but one is cluttered with ads

572
00:26:37,440 --> 00:26:40,759
and the other is clean, the migration could be huge.

573
00:26:40,839 --> 00:26:42,920
Speaker 1: It's smooth sailing versus a bumpy.

574
00:26:42,680 --> 00:26:46,559
Speaker 2: Road, exactly. And Google can afford to play this long game.

575
00:26:46,599 --> 00:26:48,759
They have YouTube money, they have Google Search money. They

576
00:26:48,759 --> 00:26:51,200
can run Gemini at a loss for years and years

577
00:26:51,359 --> 00:26:54,359
just to starve open AI of users and revenue.

578
00:26:54,440 --> 00:26:57,839
Speaker 1: So you think Google could intentionally keep Gemini ad free,

579
00:26:58,079 --> 00:27:01,200
not because they're morally superior, but just as a competitive

580
00:27:01,240 --> 00:27:02,559
weapon to kill tat GPT.

581
00:27:02,799 --> 00:27:06,480
Speaker 2: Absolutely, it's a classic business strategy. It's like predatory pricing.

582
00:27:06,839 --> 00:27:09,119
We have basically the same technology, but ours is pure

583
00:27:09,160 --> 00:27:11,519
and clean. There's has sold out. If you're a user,

584
00:27:11,640 --> 00:27:13,359
why would you stay with the one that sold out?

585
00:27:13,599 --> 00:27:16,200
Speaker 1: It's the old clean web argument. It's like Netflix versus

586
00:27:16,240 --> 00:27:18,920
cable tv back in the day. The killer feature of

587
00:27:18,960 --> 00:27:20,680
early Netflix was no commercials.

588
00:27:20,960 --> 00:27:23,559
Speaker 2: And now open ai is becoming cable TV.

589
00:27:23,720 --> 00:27:26,279
Speaker 1: Well that is a burn. That's a good one. Open

590
00:27:26,319 --> 00:27:27,799
ai is becoming cable TV.

591
00:27:28,160 --> 00:27:31,400
Speaker 2: It fits perfectly. You pay a monthly subscription fee, but

592
00:27:31,480 --> 00:27:33,559
you still have to watch ads. That is the cable

593
00:27:33,559 --> 00:27:34,279
TV model.

594
00:27:34,319 --> 00:27:37,240
Speaker 1: That is the worst model. That's the model everyone has.

595
00:27:37,359 --> 00:27:40,279
Speaker 2: It's the model that an entire generation cut the cord

596
00:27:40,480 --> 00:27:44,000
to escape from, and here we are reinventing it for

597
00:27:44,240 --> 00:27:46,279
twenty first century artificial intelligence.

598
00:27:46,319 --> 00:27:48,599
Speaker 1: It's wild and it just brings us right back around

599
00:27:48,640 --> 00:27:53,000
to Demossopus's original point. If the magic isn't paying the bills,

600
00:27:53,880 --> 00:27:55,839
how real is the magic?

601
00:27:56,079 --> 00:27:58,400
Speaker 2: That's the final verdict. The source seems to land on.

602
00:27:58,559 --> 00:28:00,920
The claims of AGI being just around the corner are

603
00:28:00,960 --> 00:28:03,519
starting to look really far fetched. If the company leading

604
00:28:03,559 --> 00:28:06,119
the charge has to pivot to ads just to keep

605
00:28:06,160 --> 00:28:07,079
the lights on if you.

606
00:28:07,160 --> 00:28:10,039
Speaker 1: Really had a technology that could create unlimited energy, or

607
00:28:10,119 --> 00:28:14,079
cure cancer or a true superintelligence, you would not be

608
00:28:14,200 --> 00:28:14,920
selling ad.

609
00:28:14,759 --> 00:28:18,079
Speaker 2: Space now, you wouldn't. This pivot to ads suggests that

610
00:28:18,119 --> 00:28:22,200
AGI is hard. It's really, really, really hard. Recreating the

611
00:28:22,200 --> 00:28:25,200
functionality of the human brain is proving to be incredibly

612
00:28:25,200 --> 00:28:28,720
difficult and expensive, and the magic that we see today

613
00:28:29,079 --> 00:28:33,000
isn't generating enough direct revenue to sustain the massive operation

614
00:28:33,160 --> 00:28:36,759
behind it. We're not in the post scarcity AGI future,

615
00:28:36,839 --> 00:28:39,920
yet we're still very much in the scarcity driven present,

616
00:28:40,079 --> 00:28:40,839
which brings.

617
00:28:40,680 --> 00:28:43,039
Speaker 1: Us full circle right back to that tweet from the beginning,

618
00:28:43,519 --> 00:28:46,039
the one about the machine God, the machine God, leave

619
00:28:46,160 --> 00:28:49,000
zero some quant finance and come build the machine God.

620
00:28:49,599 --> 00:28:52,680
That was the recruitment pitch. That was the dream. Come

621
00:28:52,720 --> 00:28:56,759
here and build something divine, something that will change humanity forever.

622
00:28:57,079 --> 00:28:59,480
Speaker 2: And the reality, the root awakening is come here and

623
00:28:59,519 --> 00:29:01,920
help us optimizer ad placement algorithms.

624
00:29:02,319 --> 00:29:05,720
Speaker 1: Irony is just it's so thick. You have these brilliant

625
00:29:05,759 --> 00:29:09,720
people who love jobs, where they made millions optimizing financial.

626
00:29:09,279 --> 00:29:11,119
Speaker 2: Algorithms at zero sum game, a.

627
00:29:11,200 --> 00:29:14,759
Speaker 1: Zero sum game, right, They left that to do something meaningful,

628
00:29:14,960 --> 00:29:17,759
something that would create value for the whole world, and

629
00:29:17,839 --> 00:29:20,759
now they're being told their new job is to figure

630
00:29:20,759 --> 00:29:23,319
out the best way to sell toothpaste to people who

631
00:29:23,359 --> 00:29:25,240
are asking for philosophy advice.

632
00:29:25,440 --> 00:29:28,400
Speaker 2: Imagine the morale inside that company right now. You thought

633
00:29:28,400 --> 00:29:30,319
you were working on the computer from Star Trek, and

634
00:29:30,359 --> 00:29:33,519
suddenly your manager walks in and says, hey, great work

635
00:29:33,559 --> 00:29:36,160
on that emergent reasoning, But can you make sure the

636
00:29:36,200 --> 00:29:39,880
model mentions coca cola naturally every fifteen queries.

637
00:29:39,559 --> 00:29:41,519
Speaker 1: Or so naturally that's the key word.

638
00:29:41,759 --> 00:29:44,240
Speaker 2: And you know what, that technical challenge might actually be

639
00:29:44,359 --> 00:29:48,000
harder than building agi. How do you make a paid

640
00:29:48,119 --> 00:29:52,680
advertisement sound like the genuine, trusted advice of a wise friend.

641
00:29:52,559 --> 00:29:55,519
Speaker 1: That goes right back to the medic comparifon reading my brain?

642
00:29:55,880 --> 00:29:58,759
Speaker 2: Exactly if they actually succeed at this, and these are

643
00:29:58,759 --> 00:30:00,680
some of the smartest people on the plant, they might.

644
00:30:00,839 --> 00:30:03,960
If they succeed in making ads that are perfectly conversational,

645
00:30:04,279 --> 00:30:07,599
that are woven seamlessly into the dialogue, that's not just

646
00:30:07,640 --> 00:30:10,599
annoying anymore. That's psychologically potent in a way we have

647
00:30:10,680 --> 00:30:13,759
never seen before. It's a level of personalized persuasion that

648
00:30:13,880 --> 00:30:15,400
is unprecedented.

649
00:30:15,759 --> 00:30:19,039
Speaker 1: It blurs the line between reality and advertising.

650
00:30:19,359 --> 00:30:22,599
Speaker 2: It raises it. If I ask my AI, what's the

651
00:30:22,640 --> 00:30:25,359
best way to clean this weinstain, and it says, oh,

652
00:30:25,400 --> 00:30:28,759
you should use this specific brand of stain remover. I'm

653
00:30:28,799 --> 00:30:31,480
inclined to trust it because I see it as an objective,

654
00:30:31,640 --> 00:30:32,640
all knowing authority.

655
00:30:32,720 --> 00:30:33,559
Speaker 1: The encyclopedia.

656
00:30:33,720 --> 00:30:37,240
Speaker 2: Right, we don't assume the encyclopedia is sponsored, but now

657
00:30:38,000 --> 00:30:42,079
it is. It exploits our natural human tendency the authority

658
00:30:42,079 --> 00:30:46,039
bias on a massive scale. We're wired to trust the

659
00:30:46,119 --> 00:30:48,880
expert in the room, and they're positioning the AI as

660
00:30:48,920 --> 00:30:50,039
the ultimate expert.

661
00:30:50,279 --> 00:30:53,000
Speaker 1: It just highlights that no matter how futuristic and sci

662
00:30:53,039 --> 00:30:57,519
fi the technology gets, the underlying economics are stubbornly old school.

663
00:30:57,759 --> 00:30:59,960
Someone has to pay for the electricity, someone has to

664
00:31:00,039 --> 00:31:02,559
pay for the servers, and on the Internet, for twenty

665
00:31:02,640 --> 00:31:06,160
five years, that someone has usually been the advertiser.

666
00:31:06,240 --> 00:31:08,400
Speaker 2: We tried to build a god and we ended up

667
00:31:08,400 --> 00:31:09,559
with a billboard.

668
00:31:09,160 --> 00:31:10,799
Speaker 1: A billboard that knows all of our secrets.

669
00:31:10,920 --> 00:31:12,480
Speaker 2: Yeah, that's the part to remember.

670
00:31:12,599 --> 00:31:14,519
Speaker 1: Okay, let's try to wrap this up. Let' summarize what

671
00:31:14,559 --> 00:31:18,680
we've untangled here today on thrilling threads. Let's do it right. So, first,

672
00:31:18,680 --> 00:31:22,720
we have Google led by Demesisavus calling out the AGI bluff.

673
00:31:23,200 --> 00:31:26,480
He's basically pointing out that if the tech was as

674
00:31:26,559 --> 00:31:29,920
world changing as claimed, the AD model wouldn't even be

675
00:31:30,039 --> 00:31:30,559
on the table.

676
00:31:30,680 --> 00:31:33,519
Speaker 2: Then you have the trust issue with Sam Altman, his

677
00:31:33,680 --> 00:31:36,400
complete one hundred and eighty degree pivot from calling ads

678
00:31:36,480 --> 00:31:40,200
dystopian to embracing them as a necessary evil, which has

679
00:31:40,240 --> 00:31:42,759
seriously damaged his in the company's credibility.

680
00:31:42,920 --> 00:31:45,920
Speaker 1: And finally, and maybe most importantly, we have the very

681
00:31:45,960 --> 00:31:50,160
real privacy nightmare, the terrifying prospect of an AI that

682
00:31:50,240 --> 00:31:53,880
has access to your memories, your fears, your private conversations

683
00:31:54,440 --> 00:31:57,599
and using that intimate knowledge to target you with sales pitches.

684
00:31:57,799 --> 00:32:00,720
Speaker 2: It's really a trifecta of disappointment, isn't it. It's economic

685
00:32:00,759 --> 00:32:03,920
reality crashing into broken promises, all built on a foundation

686
00:32:03,960 --> 00:32:05,400
of potential privacy erosion.

687
00:32:05,599 --> 00:32:07,319
Speaker 1: So here's the question that I want to leave you

688
00:32:07,400 --> 00:32:10,519
our listeners with today. It's the one the source really drives.

689
00:32:10,240 --> 00:32:11,400
Speaker 2: Home, a provocative one.

690
00:32:11,519 --> 00:32:14,119
Speaker 1: If an AI knows your secrets, your fears, and your

691
00:32:14,200 --> 00:32:17,440
dreams and it uses that deeply personal information to sell

692
00:32:17,480 --> 00:32:21,279
you a product, is that just a helpful personalized service?

693
00:32:21,720 --> 00:32:24,400
Is that just good marketing in the twenty first century,

694
00:32:24,720 --> 00:32:27,960
or is that the literal definition of the dystopia that

695
00:32:28,000 --> 00:32:30,000
Sam Altman himself warned us about.

696
00:32:30,240 --> 00:32:34,359
Speaker 2: It's such a fine line between helpful and manipulative, and

697
00:32:34,400 --> 00:32:36,640
I think we are standing right on that line about

698
00:32:36,680 --> 00:32:39,480
to fall over to the wrong side. The technology allows

699
00:32:39,519 --> 00:32:42,839
for a level of psychological manipulation that our laws and

700
00:32:42,880 --> 00:32:45,480
maybe even our own minds just aren't prepared to handle.

701
00:32:45,559 --> 00:32:47,200
Speaker 1: Yet. I really want to hear from you on this.

702
00:32:47,400 --> 00:32:49,279
Go to the comment section wherever you're listening and tell

703
00:32:49,319 --> 00:32:51,680
us would you switch to Google Gemini right now just

704
00:32:51,720 --> 00:32:55,039
to avoid ads? Or is chat GPT's product just too good,

705
00:32:55,119 --> 00:32:57,480
too integrated into your life to quit no matter how

706
00:32:57,480 --> 00:32:58,960
many commercials they shove in your face.

707
00:32:59,119 --> 00:33:01,559
Speaker 2: I'm genuinely here for us to see the answers. Is

708
00:33:01,680 --> 00:33:04,599
brand loyalty to chat gpt strong enough to withstand the

709
00:33:04,640 --> 00:33:07,400
friction of ads? Or will people just move to the cleaner,

710
00:33:07,559 --> 00:33:09,880
easier alternative, That is the question.

711
00:33:12,000 --> 00:33:14,519
Speaker 1: Thank you for joining us on this episode of thrilling Threads.

712
00:33:14,759 --> 00:33:17,160
We're going to keep pulling on these loose threads until

713
00:33:17,200 --> 00:33:18,680
we find the truth, or until the.

714
00:33:18,640 --> 00:33:21,200
Speaker 2: AI gently suggests we talk about something else.

715
00:33:21,119 --> 00:33:24,680
Speaker 1: Don't even say that. Keep questioning the narrative. Everyone, We'll

716
00:33:24,720 --> 00:33:25,400
see you next time.

717
00:33:25,480 --> 00:33:26,200
Speaker 2: Stay curious,