1
00:00:02,400 --> 00:00:06,840
Speaker 1: Also media, Hello and welcome to Better Offline. I'm your

2
00:00:06,840 --> 00:00:20,840
Speaker 1: host ed zich Tron. As ever, remember you can buy

3
00:00:20,840 --> 00:00:23,840
Speaker 1: Better Offline merchandise linkers in the episode notes. Today, I

4
00:00:23,840 --> 00:00:26,040
Speaker 1: am joined by Karen Howe, the author of the upcoming

5
00:00:26,079 --> 00:00:28,680
Speaker 1: book Empire of Ai, which tells the story of Open

6
00:00:28,720 --> 00:00:31,440
Speaker 1: Ai and the arms rais surrounding large language models. Karen,

7
00:00:31,480 --> 00:00:32,280
Speaker 1: thank you for joining me.

8
00:00:32,840 --> 00:00:34,120
Speaker 2: Thank you so much for having me ed.

9
00:00:35,000 --> 00:00:39,279
Speaker 1: So you describe the progress of these models in these

10
00:00:39,280 --> 00:00:41,600
Speaker 1: companies as a kind of colonialism, Can you get into

11
00:00:41,640 --> 00:00:42,120
Speaker 1: that for me?

12
00:00:43,520 --> 00:00:46,239
Speaker 2: Yeah, So, if you think about the way that empires

13
00:00:46,240 --> 00:00:49,960
Speaker 2: of old operated during the very long history of European colonialism,

14
00:00:50,280 --> 00:00:53,480
Speaker 2: they were essentially taking resources that were not their own,

15
00:00:54,480 --> 00:00:57,880
Speaker 2: exploiting massive amounts of labor as in not paying them

16
00:00:58,160 --> 00:01:01,680
Speaker 2: or paying them extremely small amounts of money, and they

17
00:01:01,680 --> 00:01:05,320
Speaker 2: were doing this all under a civilizing mission, this idea

18
00:01:05,480 --> 00:01:09,560
Speaker 2: that they were bringing modernity and progress to all of humanity,

19
00:01:09,600 --> 00:01:11,600
Speaker 2: when in fact what was actually happening was they were

20
00:01:11,600 --> 00:01:14,920
Speaker 2: just fortifying themselves and the empire, and the people at

21
00:01:14,959 --> 00:01:17,640
Speaker 2: the top of the empire, and everyone else that kind

22
00:01:17,640 --> 00:01:20,160
Speaker 2: of lived in the world had to live in the

23
00:01:20,160 --> 00:01:22,559
Speaker 2: thrash of what the people at the top decided based

24
00:01:22,600 --> 00:01:25,000
Speaker 2: on their whims for what was part of their self

25
00:01:25,040 --> 00:01:28,680
Speaker 2: serving agenda. And that's essentially what we're seeing with Empires

26
00:01:28,680 --> 00:01:32,000
Speaker 2: of AI today, where they are taking data that is

27
00:01:32,040 --> 00:01:34,480
Speaker 2: not their own, their laying claim to it. They're taking land,

28
00:01:34,640 --> 00:01:38,720
Speaker 2: they're taking energy, they're taking water. They are exploiting massive

29
00:01:38,720 --> 00:01:41,679
Speaker 2: amounts of labor, both labor that goes into the inputs

30
00:01:41,959 --> 00:01:45,080
Speaker 2: for developing these am models, but also exploiting labor in

31
00:01:45,080 --> 00:01:48,440
Speaker 2: the sense that they are ultimately creating labor automating technologies

32
00:01:49,200 --> 00:01:52,200
Speaker 2: that is eroding away people's labor rights as we speak,

33
00:01:53,240 --> 00:01:55,760
Speaker 2: and they're doing it under this civilizing mission of they

34
00:01:55,800 --> 00:01:58,600
Speaker 2: are doing it for the benefit of all of humanity.

35
00:01:59,800 --> 00:02:02,120
Speaker 2: And what I say in the book is Empires of

36
00:02:02,160 --> 00:02:05,040
Speaker 2: AI they're not as overtly violent as empires of old.

37
00:02:05,240 --> 00:02:09,799
Speaker 2: And so maybe that can become confusing and people think, oh, well,

38
00:02:09,800 --> 00:02:12,359
Speaker 2: it can't be that bad. But the thing is, we've

39
00:02:12,360 --> 00:02:15,399
Speaker 2: had one hundred and fifty years of social and moral progress,

40
00:02:15,480 --> 00:02:18,280
Speaker 2: and so empires of modern day are going to look

41
00:02:18,320 --> 00:02:22,280
Speaker 2: different in the way that empires of old operated. And

42
00:02:22,520 --> 00:02:26,480
Speaker 2: when you look just at like the actual parallels, there

43
00:02:26,480 --> 00:02:31,480
Speaker 2: are just so many extraordinary parallels between the kind of

44
00:02:31,600 --> 00:02:35,120
Speaker 2: basis of empire building back then and now that I

45
00:02:35,160 --> 00:02:39,080
Speaker 2: think it is fundamentally the only frame that I have

46
00:02:39,160 --> 00:02:42,120
Speaker 2: found to really help understand and grapple with the sheer

47
00:02:42,160 --> 00:02:46,280
Speaker 2: scope and scale and the actual like what is actually

48
00:02:46,360 --> 00:02:48,280
Speaker 2: happening here within the AI industry.

49
00:02:49,440 --> 00:02:51,799
Speaker 1: One theme from the book I also noticed was that

50
00:02:52,320 --> 00:02:55,160
Speaker 1: despite all of the backs and forths between all the people,

51
00:02:55,760 --> 00:02:58,680
Speaker 1: very rarely product came out. Though, Like it was interesting,

52
00:02:58,720 --> 00:03:01,360
Speaker 1: there seemed to be all that these sations about research

53
00:03:01,440 --> 00:03:04,360
Speaker 1: and all of these things they were saying, but it

54
00:03:04,440 --> 00:03:06,800
Speaker 1: usually just ended with like some sort of release and

55
00:03:06,840 --> 00:03:10,560
Speaker 1: then kind of just moved on. Yeah, it almost it

56
00:03:10,639 --> 00:03:12,960
Speaker 1: almost makes me wonder what they're always what they're working

57
00:03:13,000 --> 00:03:13,840
Speaker 1: on half the time.

58
00:03:15,000 --> 00:03:18,000
Speaker 2: Yeah, you know, it's I think it's a product of

59
00:03:18,080 --> 00:03:21,120
Speaker 2: two different things that that you noticed that in the book.

60
00:03:21,200 --> 00:03:24,160
Speaker 2: One is that I finished writing the book before a

61
00:03:24,200 --> 00:03:28,080
Speaker 2: lot of the most recent product releases came out, Right,

62
00:03:28,440 --> 00:03:31,680
Speaker 2: That's just the nature of writing things on the timescale of.

63
00:03:31,639 --> 00:03:34,160
Speaker 1: Books, is it's not fun.

64
00:03:34,680 --> 00:03:39,920
Speaker 2: Yeah. I froze the manuscript in like the early days

65
00:03:39,920 --> 00:03:43,520
Speaker 2: of January, right before Deep Seek, right before Stargate, right

66
00:03:43,520 --> 00:03:48,760
Speaker 2: before you know, a string of other releases. So that's

67
00:03:48,800 --> 00:03:53,080
Speaker 2: one is that through most of open AI's history, it

68
00:03:53,160 --> 00:03:56,240
Speaker 2: is really it was really more focused on research conversations,

69
00:03:56,280 --> 00:03:58,680
Speaker 2: and it's only been in the last year or so

70
00:03:58,800 --> 00:04:02,520
Speaker 2: that it's really dramatically shifted much more to talking about product.

71
00:04:03,080 --> 00:04:06,480
Speaker 2: But the second reason is that I personally like that

72
00:04:06,600 --> 00:04:11,480
Speaker 2: is my expertise. I came up in AI reporting covering

73
00:04:11,560 --> 00:04:15,160
Speaker 2: the research, and so I wanted to focus on that

74
00:04:15,200 --> 00:04:18,280
Speaker 2: in the book and really unpack it, especially because there's

75
00:04:18,360 --> 00:04:22,320
Speaker 2: not as much reporting on the research these days, and

76
00:04:22,400 --> 00:04:24,640
Speaker 2: I wanted to kind of track that history and the

77
00:04:24,640 --> 00:04:27,880
Speaker 2: internal conversations that happened when people say that they're developing

78
00:04:28,200 --> 00:04:30,360
Speaker 2: so called AGI and.

79
00:04:30,360 --> 00:04:32,560
Speaker 1: You talk about in twenty nineteen in the book the

80
00:04:32,560 --> 00:04:35,480
Speaker 1: your rose colored Glasses got knocked off by a story?

81
00:04:35,760 --> 00:04:39,040
Speaker 1: What was it that really made you start being suspicious

82
00:04:39,080 --> 00:04:40,040
Speaker 1: of these companies?

83
00:04:41,520 --> 00:04:45,360
Speaker 2: Yeah, So in twenty nineteen was when I started covering

84
00:04:45,440 --> 00:04:48,000
Speaker 2: Opening Eye, and I embedded within the company for three

85
00:04:48,080 --> 00:04:52,040
Speaker 2: days in August of twenty nineteen to profile what had

86
00:04:52,080 --> 00:04:57,040
Speaker 2: then become a newly minted capped profit nested in a nonprofit.

87
00:04:58,680 --> 00:05:00,880
Speaker 2: And I think the thing that but that really started

88
00:05:00,880 --> 00:05:05,360
Speaker 2: tipping me off was it was actually really small things. Initially,

89
00:05:05,640 --> 00:05:09,839
Speaker 2: the first thing was they publicly professed to be this

90
00:05:10,000 --> 00:05:12,760
Speaker 2: bastion of transparency and they were going to share all

91
00:05:12,839 --> 00:05:14,880
Speaker 2: their research to the world, and they had accumulated a

92
00:05:14,880 --> 00:05:18,839
Speaker 2: significant amount of good will on the basis of this idea,

93
00:05:19,320 --> 00:05:23,080
Speaker 2: and they were raising not literally fundraising, but they had

94
00:05:23,120 --> 00:05:26,240
Speaker 2: amassed a lot of capital on the basis of this idea.

95
00:05:26,560 --> 00:05:29,960
Speaker 2: And when I started embedding within the company, I realized

96
00:05:29,960 --> 00:05:33,320
Speaker 2: that they were incredibly secretive, like they were not. They

97
00:05:33,320 --> 00:05:37,479
Speaker 2: wouldn't allow me to see anything or time to anyone

98
00:05:37,680 --> 00:05:42,720
Speaker 2: beyond very strictly sanctioned conversations. And even in those conversations,

99
00:05:42,760 --> 00:05:45,719
Speaker 2: I would notice that researchers were giving side eye to

100
00:05:45,760 --> 00:05:49,839
Speaker 2: the communications officer every other sentence because they were worried

101
00:05:49,880 --> 00:05:53,520
Speaker 2: about stepping into a lane that was considered proprietary. And

102
00:05:53,560 --> 00:05:55,800
Speaker 2: I was like, wait a minute, why are there things

103
00:05:55,839 --> 00:05:58,280
Speaker 2: that are proprietary, and why are the people being secretive

104
00:05:58,279 --> 00:06:02,200
Speaker 2: if all this is supposed to ultimately be shared with

105
00:06:02,240 --> 00:06:05,800
Speaker 2: the public. But the other thing was when I was

106
00:06:05,880 --> 00:06:09,680
Speaker 2: talking with executives, Like the very first interview that I

107
00:06:09,760 --> 00:06:12,520
Speaker 2: had was with Greg Brockman and Ilia Setskav, the CTO

108
00:06:12,520 --> 00:06:16,880
Speaker 2: and she's scientist, And I just asked them very basic questions,

109
00:06:16,920 --> 00:06:19,160
Speaker 2: like why do you think we should spend this much

110
00:06:19,200 --> 00:06:22,720
Speaker 2: money on Agi and not on something else? And can

111
00:06:22,760 --> 00:06:25,599
Speaker 2: you articulate for me what does AGI look like? What

112
00:06:25,640 --> 00:06:29,560
Speaker 2: would you even want Agi to do? And can you

113
00:06:29,640 --> 00:06:32,559
Speaker 2: articulate for me? You know, part of their origin story

114
00:06:32,600 --> 00:06:36,080
Speaker 2: as a company was they want to build agi good

115
00:06:36,120 --> 00:06:39,640
Speaker 2: agi first before the bad people built bad Agi. So

116
00:06:39,680 --> 00:06:42,159
Speaker 2: I was like, well, what would bad agi look like

117
00:06:42,200 --> 00:06:45,280
Speaker 2: as well? Or like what are the harms that are

118
00:06:45,279 --> 00:06:48,359
Speaker 2: coming out of some of this rapid AI progress? And

119
00:06:48,920 --> 00:06:53,200
Speaker 2: they weren't able to answer any of those questions. And

120
00:06:53,640 --> 00:06:56,359
Speaker 2: that was when I thought, hold on a second, Like

121
00:06:56,480 --> 00:07:00,000
Speaker 2: I thought that this was a nonprofit meant to care

122
00:07:00,080 --> 00:07:04,320
Speaker 2: counter some of the ills of Silicon Valley, one of

123
00:07:04,360 --> 00:07:08,719
Speaker 2: the ills being that the most companies end up being

124
00:07:08,760 --> 00:07:12,640
Speaker 2: thrown boatloads of cash without like clear articulated ideas about

125
00:07:12,640 --> 00:07:15,520
Speaker 2: what they're going to do with that cash. And here

126
00:07:15,560 --> 00:07:17,920
Speaker 2: I am in this meeting room trying to just ask

127
00:07:18,000 --> 00:07:21,640
Speaker 2: the most basic question, like the most boiler plate stuff

128
00:07:21,680 --> 00:07:25,920
Speaker 2: that there should be some kind of answer too, uh,

129
00:07:25,960 --> 00:07:28,840
Speaker 2: And they can't even answer that. So it seems like

130
00:07:29,040 --> 00:07:33,320
Speaker 2: it is actually very much just an animal of Silicon Valley.

131
00:07:33,680 --> 00:07:36,640
Speaker 2: This is not actually something different from what we're seeing

132
00:07:36,680 --> 00:07:37,880
Speaker 2: with the rest of the tech industry.

133
00:07:38,840 --> 00:07:41,480
Speaker 1: It felt as well, that was a comment and forgive

134
00:07:41,480 --> 00:07:43,640
Speaker 1: me for forgetting it exactly where where it was like

135
00:07:43,720 --> 00:07:45,600
Speaker 1: the all secrets could be written on a grain of

136
00:07:45,680 --> 00:07:50,800
Speaker 1: rice or something like that. Yeah, and I have to admit,

137
00:07:51,000 --> 00:07:53,120
Speaker 1: as I read it, I got this weird feeling like

138
00:07:53,160 --> 00:07:56,720
Speaker 1: does anyone actually have any ip because when you actually

139
00:07:56,720 --> 00:07:59,480
Speaker 1: look at the conversations they're having, and you likely privy

140
00:07:59,520 --> 00:08:02,040
Speaker 1: to more hair, it felt like they wouldn't talk about

141
00:08:02,040 --> 00:08:05,080
Speaker 1: what they were doing at all, and not I say,

142
00:08:05,120 --> 00:08:07,960
Speaker 1: this is run apof written a lot about the valley.

143
00:08:08,520 --> 00:08:11,760
Speaker 1: It feels like they'd say more, but no one wanted

144
00:08:11,800 --> 00:08:14,679
Speaker 1: to say anything, not even secret. It's like nobody really knew.

145
00:08:15,080 --> 00:08:17,800
Speaker 1: And you even described some of the managerial stuff in there,

146
00:08:17,800 --> 00:08:19,760
Speaker 1: like no one really knew what was going on anyway.

147
00:08:19,960 --> 00:08:23,880
Speaker 1: It just feels like a remarkably disorganized company considering the scale.

148
00:08:25,160 --> 00:08:28,000
Speaker 2: Yeah, so I think early on at Opening Eye was

149
00:08:28,040 --> 00:08:31,360
Speaker 2: completely disorganized in the sense that they had no idea.

150
00:08:31,920 --> 00:08:34,680
Speaker 2: You know, they decide, Okay, we're going to build this

151
00:08:34,720 --> 00:08:36,559
Speaker 2: ag I think, but then they were like, what does

152
00:08:36,600 --> 00:08:40,000
Speaker 2: that even mean? We have no idea, And there was

153
00:08:40,240 --> 00:08:42,760
Speaker 2: a lot of there weren't real managers at the company either,

154
00:08:42,800 --> 00:08:46,080
Speaker 2: because they had just gathered up a bunch of researchers

155
00:08:46,120 --> 00:08:49,560
Speaker 2: from academia, and they didn't really have much of a

156
00:08:49,600 --> 00:08:52,360
Speaker 2: sense of how to organize themselves other than a traditional

157
00:08:52,400 --> 00:08:56,440
Speaker 2: academic lab where there's a professor and grad students, and

158
00:08:57,679 --> 00:09:02,199
Speaker 2: I mean academia, you know, as it's has its function,

159
00:09:02,600 --> 00:09:06,040
Speaker 2: But that ultimately wasn't the right structure for trying to

160
00:09:06,559 --> 00:09:10,720
Speaker 2: move a group of people towards a similar goal. And

161
00:09:10,760 --> 00:09:14,199
Speaker 2: over time, Opening Eye did start cleaning itself up a

162
00:09:14,240 --> 00:09:16,800
Speaker 2: little bit, It did start restructuring itself. It started focusing

163
00:09:16,840 --> 00:09:20,199
Speaker 2: more on GPT models because they hit on that in

164
00:09:20,320 --> 00:09:26,080
Speaker 2: around twenty eighteen twenty nineteen. But similarly, there's still just

165
00:09:26,240 --> 00:09:31,160
Speaker 2: because there's no clarity about its mission and ultimately what

166
00:09:31,240 --> 00:09:34,680
Speaker 2: it is trying to build, you end up with just

167
00:09:34,920 --> 00:09:38,800
Speaker 2: a lot of riffs within the organization over this very

168
00:09:38,840 --> 00:09:42,320
Speaker 2: fundamental question. People fundamentally disagree about what Opening I is.

169
00:09:43,000 --> 00:09:46,400
Speaker 2: They disagree about what AGI is, They disagree about what

170
00:09:46,440 --> 00:09:49,520
Speaker 2: it means to ensure that something should benefit all of humanity.

171
00:09:51,240 --> 00:09:56,040
Speaker 2: And I think because there was all this confusion or

172
00:09:56,120 --> 00:09:59,520
Speaker 2: there were all these different interpretations ultimately of these like

173
00:09:59,600 --> 00:10:04,200
Speaker 2: basic tenants of the organization. I think people also just

174
00:10:04,920 --> 00:10:08,120
Speaker 2: there was they wouldn't quite clearly articulate to one another

175
00:10:08,400 --> 00:10:10,679
Speaker 2: what they were doing. It wasn't necessarily that they were

176
00:10:10,720 --> 00:10:13,600
Speaker 2: trying to be secretive to one another. It was more

177
00:10:13,760 --> 00:10:16,880
Speaker 2: just that they weren't really on the same page. And

178
00:10:18,840 --> 00:10:23,080
Speaker 2: this eventually became sort of less and less true in

179
00:10:23,080 --> 00:10:26,480
Speaker 2: the sense that as sam Oltman's installed himself as CEO

180
00:10:26,520 --> 00:10:31,760
Speaker 2: and started really exerting a particular type of path for

181
00:10:31,920 --> 00:10:35,160
Speaker 2: AI progress, then they started having, you know, research documents

182
00:10:35,200 --> 00:10:38,480
Speaker 2: that explicitly articulated we are a scaling lab, like we

183
00:10:38,520 --> 00:10:41,160
Speaker 2: are going after scale way.

184
00:10:41,280 --> 00:10:44,480
Speaker 1: How long did it take them to put those documents

185
00:10:44,480 --> 00:10:46,360
Speaker 1: together though? What year about?

186
00:10:47,520 --> 00:10:50,360
Speaker 2: I think their first research roadmap was in twenty seventeen,

187
00:10:50,400 --> 00:10:52,480
Speaker 2: so it was one and a half years to them

188
00:10:53,000 --> 00:10:57,520
Speaker 2: so bad into the non profit Yeah, yeah.

189
00:10:56,640 --> 00:10:59,600
Speaker 1: So I will admit there is another colonial thing that

190
00:10:59,640 --> 00:11:04,320
Speaker 1: stood out well. Two specifically, One, it definitely feels that

191
00:11:04,320 --> 00:11:07,240
Speaker 1: there are a lot of unintelligent cousin types who were

192
00:11:07,240 --> 00:11:09,840
Speaker 1: put in there because their mate was there throughout the company.

193
00:11:10,240 --> 00:11:13,800
Speaker 1: But two it's this kind of religious view around agi,

194
00:11:14,280 --> 00:11:18,200
Speaker 1: this kind of nebulous justification for just about every I

195
00:11:18,240 --> 00:11:20,240
Speaker 1: was disappointed, and I understand what you mentioned it like

196
00:11:20,640 --> 00:11:23,480
Speaker 1: did Yadowski was in there? I think the less wrong people.

197
00:11:23,640 --> 00:11:26,640
Speaker 1: This is a personal belief. Just no need to mention

198
00:11:26,720 --> 00:11:29,679
Speaker 1: them again for anyone. I think that Yadowski, anyone who

199
00:11:29,679 --> 00:11:31,840
Speaker 1: writes a six hundred thousand word Harry Potter book should

200
00:11:31,880 --> 00:11:35,760
Speaker 1: be put in prison, including JK. Rowling. But it feels

201
00:11:35,840 --> 00:11:38,960
Speaker 1: like there really is this belief system that's pushed throughout

202
00:11:38,960 --> 00:11:43,080
Speaker 1: this industry which mirrors colonialism. Mirror is the very judao

203
00:11:43,200 --> 00:11:46,880
Speaker 1: Christian push of the British and many other colonial entities.

204
00:11:47,760 --> 00:11:50,320
Speaker 2: Yeah. Absolutely so. One of the things that I was

205
00:11:50,480 --> 00:11:54,760
Speaker 2: most surprised by when reporting the book is I had

206
00:11:54,840 --> 00:11:58,320
Speaker 2: seen all the divisions around boomers and boomers people saying hey,

207
00:11:58,400 --> 00:12:00,760
Speaker 2: I can bring us to Utopiera, people saying I can

208
00:12:00,840 --> 00:12:04,080
Speaker 2: kill us all. I really did think initially that it

209
00:12:04,160 --> 00:12:07,200
Speaker 2: was just rhetoric and that it was just a tool

210
00:12:07,360 --> 00:12:11,120
Speaker 2: for accruing more power. And the thing that surprised me

211
00:12:11,280 --> 00:12:16,319
Speaker 2: most was how many people I met that genuinely deeply

212
00:12:16,320 --> 00:12:19,240
Speaker 2: believed in both, especially the Dumer ideology like I was

213
00:12:19,240 --> 00:12:23,200
Speaker 2: interviewing people whose voices were quivering because they were talking

214
00:12:23,240 --> 00:12:27,320
Speaker 2: about their anxiety around the potential end of the world,

215
00:12:27,400 --> 00:12:31,880
Speaker 2: and that was a very sincere reaction. And I think

216
00:12:31,920 --> 00:12:34,000
Speaker 2: that is part You're exactly right that it is a

217
00:12:34,080 --> 00:12:37,720
Speaker 2: huge parallel with empire building in the past, is that

218
00:12:38,280 --> 00:12:43,680
Speaker 2: empires need to have an ideology that convinces themselves why

219
00:12:43,720 --> 00:12:46,800
Speaker 2: they are ultimately doing something that is for the benefit

220
00:12:46,840 --> 00:12:49,680
Speaker 2: of the world. So in the past, when they had

221
00:12:49,679 --> 00:12:52,280
Speaker 2: the civilizing mission, we're bringing this to the world. It

222
00:12:52,360 --> 00:12:57,320
Speaker 2: also wasn't rhetoric. It was also a deeply seated religious

223
00:12:57,840 --> 00:13:02,600
Speaker 2: and spiritual and scientific belief that they were doing something

224
00:13:02,640 --> 00:13:04,440
Speaker 2: that was better off for everyone.

225
00:13:04,920 --> 00:13:07,600
Speaker 1: I mean, the origins of the BBC in England were

226
00:13:07,880 --> 00:13:12,600
Speaker 1: religious indoctrination on some level. It kind of it's I

227
00:13:12,679 --> 00:13:16,800
Speaker 1: admit I'm surprised to hear the quivering voices stuff I

228
00:13:16,840 --> 00:13:21,360
Speaker 1: think because I think that again, this personal opinion, Ydowski,

229
00:13:21,480 --> 00:13:23,000
Speaker 1: I think is full. I think a lot of those

230
00:13:23,040 --> 00:13:24,840
Speaker 1: less wrong guys are full of shit. I think they're

231
00:13:24,880 --> 00:13:27,440
Speaker 1: doing it for not for the bit. But it's the

232
00:13:27,480 --> 00:13:31,800
Speaker 1: same kind of horse trading shit that people do around anything.

233
00:13:31,840 --> 00:13:33,480
Speaker 1: It's like, we don't have anything to believe in, so

234
00:13:33,559 --> 00:13:36,680
Speaker 1: let's all agree on this. But it's interesting to hear

235
00:13:36,720 --> 00:13:40,320
Speaker 1: that people are I don't know how to put this,

236
00:13:40,480 --> 00:13:44,360
Speaker 1: actually believing this crap even though it doesn't feel like

237
00:13:44,520 --> 00:13:46,560
Speaker 1: there's any real evidence. You know.

238
00:13:46,880 --> 00:13:50,280
Speaker 2: Yeah, well, I think the analogy that I've started using

239
00:13:50,400 --> 00:13:54,520
Speaker 2: is I really feel like opening Eyes done, where you know,

240
00:13:54,679 --> 00:13:57,920
Speaker 2: in Dune, there is a mythology that is created by

241
00:13:57,960 --> 00:14:00,960
Speaker 2: a certain group of people with four understanding that they're

242
00:14:00,960 --> 00:14:06,120
Speaker 2: creating a mythology, right right, But then as they start

243
00:14:06,160 --> 00:14:10,640
Speaker 2: to embody and act out this mythology, not only do

244
00:14:11,160 --> 00:14:13,360
Speaker 2: many many people who didn't know that it was originally

245
00:14:13,400 --> 00:14:16,400
Speaker 2: created come to believe it, also the people who created

246
00:14:16,400 --> 00:14:20,000
Speaker 2: it come to believe it themselves. And I think this

247
00:14:20,240 --> 00:14:26,960
Speaker 2: is essentially exactly what is happening within AI with the ideologies,

248
00:14:27,040 --> 00:14:30,520
Speaker 2: is that maybe there was at some point someone who

249
00:14:30,960 --> 00:14:36,520
Speaker 2: was more aware that there was some kind of rhetorical

250
00:14:36,600 --> 00:14:39,880
Speaker 2: trick that they were playing around, really propagating this kind

251
00:14:39,920 --> 00:14:43,040
Speaker 2: of belief, but it is not We're not at that

252
00:14:43,120 --> 00:14:46,160
Speaker 2: point anymore. Like, there are lots and lots of people

253
00:14:46,160 --> 00:14:49,840
Speaker 2: who genuinely believe these things, and I think it's self

254
00:14:49,880 --> 00:14:53,640
Speaker 2: perpetuating because when you believe it, you look for signs

255
00:14:53,680 --> 00:14:58,320
Speaker 2: of it, and you research things that would suggest more

256
00:14:58,400 --> 00:15:01,800
Speaker 2: evidence for your belief and so they're kind of continuing

257
00:15:01,840 --> 00:15:04,960
Speaker 2: to reinforce their beliefs. And the more these A models

258
00:15:04,960 --> 00:15:09,320
Speaker 2: have progressed, the stronger these beliefs have become. Because whether

259
00:15:09,400 --> 00:15:12,840
Speaker 2: you believe AI will bring utopia or dystopia, there is

260
00:15:12,880 --> 00:15:16,320
Speaker 2: an abundance of evidence that you can point to now

261
00:15:16,800 --> 00:15:21,000
Speaker 2: to reinforce your own yeah, exactly, to reinforce your own

262
00:15:21,800 --> 00:15:24,160
Speaker 2: starting point. And so it's sort of like a microcosm

263
00:15:25,120 --> 00:15:29,520
Speaker 2: of society today where you know, most the average person

264
00:15:29,720 --> 00:15:32,880
Speaker 2: no longer encounters information that can change their minds. It

265
00:15:32,960 --> 00:15:36,600
Speaker 2: just continues to entrench whatever they already believed before.

266
00:15:37,120 --> 00:15:40,840
Speaker 1: Do you believe Sam Moultman believes this shite? Do you

267
00:15:40,840 --> 00:15:44,240
Speaker 1: think he believes in the AGI? Is he part of it?

268
00:15:44,240 --> 00:15:49,360
Speaker 2: It's really It's really interesting because I think the no

269
00:15:49,440 --> 00:15:52,560
Speaker 2: matter who I interviewed, and no matter how long they

270
00:15:52,600 --> 00:15:55,880
Speaker 2: worked with Sam Aultman or how closely they worked with

271
00:15:55,960 --> 00:15:58,360
Speaker 2: Sam Altman, and not a single person was able to

272
00:15:58,560 --> 00:16:02,760
Speaker 2: fully articulate what his beliefs are. And I think that

273
00:16:02,840 --> 00:16:03,920
Speaker 2: is very much by design.

274
00:16:04,240 --> 00:16:08,000
Speaker 1: Is that it's beautiful that's.

275
00:16:07,720 --> 00:16:11,360
Speaker 2: Yeah, and and and and and it wasn't. And they

276
00:16:11,360 --> 00:16:14,800
Speaker 2: would explicitly say this too. They would they would call out,

277
00:16:15,000 --> 00:16:19,720
Speaker 2: I'm not actually sure what he believes. And this was

278
00:16:19,800 --> 00:16:34,120
Speaker 2: the most consistent thing that people said about him.

279
00:16:34,400 --> 00:16:37,840
Speaker 1: I really noticed as well. If you read your book

280
00:16:38,120 --> 00:16:40,800
Speaker 1: and you really look, you actually can't get much of

281
00:16:40,800 --> 00:16:43,040
Speaker 1: an idea of who Sam Mortman is at all, And

282
00:16:43,080 --> 00:16:46,200
Speaker 1: in fact, you can't work out why he's brilliant at all.

283
00:16:46,200 --> 00:16:47,880
Speaker 1: And I've read a lot of stuff about Sam Moultman.

284
00:16:48,120 --> 00:16:50,120
Speaker 1: The long and short of it I can understand is

285
00:16:50,200 --> 00:16:54,400
Speaker 1: that he's got good psychology and he's really charming. Everyone

286
00:16:54,440 --> 00:16:57,360
Speaker 1: talks about the psychology and the charming, and it's just

287
00:16:57,520 --> 00:17:01,040
Speaker 1: really it. It is so busy. He's so a bizarre

288
00:17:01,120 --> 00:17:04,160
Speaker 1: man like everything about him is so like. The way

289
00:17:04,200 --> 00:17:07,680
Speaker 1: that people talk about him is so strange.

290
00:17:08,000 --> 00:17:10,560
Speaker 2: Yeah, So this is what I sort of concluded about

291
00:17:10,600 --> 00:17:15,160
Speaker 2: why he's able to pull off what he does. He

292
00:17:15,320 --> 00:17:17,600
Speaker 2: is a once in a generation talent when it comes

293
00:17:17,600 --> 00:17:23,720
Speaker 2: to storytelling, and he has a loose relationship with the truth, yes,

294
00:17:23,800 --> 00:17:27,600
Speaker 2: which is a really powerful combination. And so when he's

295
00:17:27,640 --> 00:17:31,960
Speaker 2: talking to someone, he shines most when he's talking to

296
00:17:32,040 --> 00:17:35,639
Speaker 2: small groups of people in one on one meetings, and

297
00:17:35,720 --> 00:17:40,280
Speaker 2: what he says is more tightly correlated with what that

298
00:17:40,440 --> 00:17:45,080
Speaker 2: person needs to hear rather than what he believes, which

299
00:17:45,119 --> 00:17:47,040
Speaker 2: is part of the reason why people say, ultimately like

300
00:17:47,080 --> 00:17:50,000
Speaker 2: they don't really know what he believes because he doesn't

301
00:17:50,119 --> 00:17:54,439
Speaker 2: really indicate it. And so I think that is what

302
00:17:54,520 --> 00:17:58,600
Speaker 2: makes him incredibly persuasive. And he is really good at

303
00:17:58,640 --> 00:18:01,960
Speaker 2: understanding people and what they and what they want, and

304
00:18:02,080 --> 00:18:04,160
Speaker 2: you know, he's well resourced, so he's able to then

305
00:18:04,200 --> 00:18:07,000
Speaker 2: deliver to them what they need and want. And what

306
00:18:07,080 --> 00:18:13,679
Speaker 2: I realized is with that kind of talent, you would

307
00:18:13,720 --> 00:18:19,280
Speaker 2: inherently be incredibly polarizing as a figure, because if the

308
00:18:19,320 --> 00:18:23,479
Speaker 2: person agrees with you, you're the best asset in the

309
00:18:23,520 --> 00:18:27,200
Speaker 2: world for what they want to achieve. You're incredibly persuasive,

310
00:18:27,440 --> 00:18:31,200
Speaker 2: You're able to get the resources you can do exactly

311
00:18:31,840 --> 00:18:34,680
Speaker 2: what that person can do for you, exactly what you

312
00:18:34,720 --> 00:18:37,879
Speaker 2: want them to do. But if you disagree with this person,

313
00:18:38,119 --> 00:18:42,639
Speaker 2: that person becomes the greatest threat ever because they are

314
00:18:42,680 --> 00:18:46,159
Speaker 2: so persuasive. You have fear that they're going to be

315
00:18:46,200 --> 00:18:48,399
Speaker 2: able to carry out exactly what you don't want them

316
00:18:48,440 --> 00:18:51,400
Speaker 2: to carry out. And so that kind of boils down

317
00:18:51,440 --> 00:18:56,639
Speaker 2: to why he's just such an enigmatic and extremely polarizing person.

318
00:18:57,440 --> 00:18:59,520
Speaker 2: Is it really depends on whether or not someone agrees

319
00:18:59,600 --> 00:19:00,440
Speaker 2: or disagree with him.

320
00:19:00,680 --> 00:19:02,680
Speaker 1: He also doesn't seem that smart. I don't know, he

321
00:19:02,720 --> 00:19:04,800
Speaker 1: seems quite good at talking to people, but when I

322
00:19:04,840 --> 00:19:07,760
Speaker 1: hear him talk, he doesn't seem that eloquent. And it

323
00:19:07,800 --> 00:19:10,800
Speaker 1: makes me wonder if perhaps Sam Mortman is a symptom

324
00:19:10,840 --> 00:19:12,840
Speaker 1: of a greater problem that so much of our power,

325
00:19:12,880 --> 00:19:16,200
Speaker 1: structure and money is based on someone making decisions based

326
00:19:16,240 --> 00:19:19,399
Speaker 1: on the last intelligent person or intelligent seeming person they

327
00:19:19,480 --> 00:19:19,840
Speaker 1: talk to.

328
00:19:21,200 --> 00:19:24,639
Speaker 2: Yeah, I mean, I think our society is also just

329
00:19:24,800 --> 00:19:31,360
Speaker 2: we still have such a have we worship people that

330
00:19:31,480 --> 00:19:36,199
Speaker 2: are wealthy. Yeah, like, And so even if he's not

331
00:19:36,480 --> 00:19:40,199
Speaker 2: saying something that is convincing you in real time, he

332
00:19:40,320 --> 00:19:43,040
Speaker 2: has all of the kind of indicators that this person

333
00:19:43,119 --> 00:19:46,679
Speaker 2: has been remarkably successful and you should listen to what

334
00:19:46,720 --> 00:19:49,200
Speaker 2: he says because then that will make you successful too, right,

335
00:19:50,280 --> 00:19:53,040
Speaker 2: And so I think that is part of the part

336
00:19:53,200 --> 00:19:56,960
Speaker 2: of the kind of mythos around him, is that if

337
00:19:56,960 --> 00:20:00,240
Speaker 2: you can join up with him, it will greatly enrich you.

338
00:20:00,800 --> 00:20:02,439
Speaker 2: And you know, like, there's a lot of evidence to

339
00:20:02,440 --> 00:20:05,520
Speaker 2: suggest that too, that like, there have been plenty of

340
00:20:05,520 --> 00:20:08,560
Speaker 2: people that I've allied themselves with Sam Aulman and that

341
00:20:08,600 --> 00:20:12,879
Speaker 2: have become much richer for it. And so whether or

342
00:20:12,920 --> 00:20:16,879
Speaker 2: not people are joining up with him because they necessarily

343
00:20:16,920 --> 00:20:19,320
Speaker 2: one hundred percent agree with like his ideology or his

344
00:20:19,400 --> 00:20:22,720
Speaker 2: actions or anything like that, or if it's if it's

345
00:20:22,760 --> 00:20:26,280
Speaker 2: more because ultimately they get to benefit from that alliance,

346
00:20:27,200 --> 00:20:29,080
Speaker 2: I think is Yeah, what.

347
00:20:29,000 --> 00:20:31,720
Speaker 1: Most feels like, it's people connecting with other people to

348
00:20:31,760 --> 00:20:34,560
Speaker 1: see how far they can get, far more than AI.

349
00:20:34,800 --> 00:20:37,639
Speaker 1: Because one of the other things I really noticed when

350
00:20:37,680 --> 00:20:41,160
Speaker 1: you were telling the story of the firing Sam Wulman

351
00:20:41,160 --> 00:20:44,080
Speaker 1: getting fired in No. Twenty twenty three, As much as

352
00:20:44,080 --> 00:20:47,440
Speaker 1: people wanted to pretend, they kept bringing up the tender

353
00:20:48,000 --> 00:20:50,080
Speaker 1: and to explain for the listeners the tender was the

354
00:20:50,119 --> 00:20:52,960
Speaker 1: open AI hadn't had a plans to let have plans

355
00:20:52,960 --> 00:20:55,479
Speaker 1: to let people sell their stock. It really felt like

356
00:20:55,560 --> 00:20:59,680
Speaker 1: that was more the primary concern than any loyalty to Altman.

357
00:21:00,800 --> 00:21:02,960
Speaker 2: It was. I wouldn't say it was the primary concern,

358
00:21:03,000 --> 00:21:06,080
Speaker 2: but I mean, yeah, it really depended on.

359
00:21:06,080 --> 00:21:06,639
Speaker 1: Who I was.

360
00:21:07,320 --> 00:21:11,679
Speaker 2: It's taught, Yeah, exactly like every employee sort of had

361
00:21:11,680 --> 00:21:15,639
Speaker 2: a different calculus that ultimately led them to revolt against

362
00:21:15,640 --> 00:21:18,720
Speaker 2: the board and want Altman back, And there were different

363
00:21:18,720 --> 00:21:21,760
Speaker 2: calculuses among Microsoft and investors. But one of the key

364
00:21:21,800 --> 00:21:25,679
Speaker 2: things that I think is uh necessary to understand just

365
00:21:25,800 --> 00:21:29,040
Speaker 2: why there is so much seeming loyalty around Altman in general,

366
00:21:29,200 --> 00:21:33,960
Speaker 2: is he is very very good at at establishing relationships

367
00:21:34,840 --> 00:21:38,159
Speaker 2: with money involved, where he is the lynchpin to the

368
00:21:38,200 --> 00:21:42,160
Speaker 2: other person accessing that much yes, and so the tender

369
00:21:42,200 --> 00:21:44,720
Speaker 2: offer is a perfect example of this, and that employees

370
00:21:45,000 --> 00:21:48,680
Speaker 2: ultimately they realized that Altman is just really he's really

371
00:21:48,680 --> 00:21:52,320
Speaker 2: good at fundraising, and whether or not an employee believes

372
00:21:52,359 --> 00:21:56,000
Speaker 2: in the AGI thing, they all agreed that open AA

373
00:21:56,040 --> 00:21:59,080
Speaker 2: ultimately needs an enormous amount of capital, and also many

374
00:21:59,160 --> 00:22:01,160
Speaker 2: of them are doing it in part because they can

375
00:22:01,200 --> 00:22:05,239
Speaker 2: then like guarantee their own financial future. And so with

376
00:22:05,960 --> 00:22:09,679
Speaker 2: Altman gone, it became increasingly clear that open a I

377
00:22:09,720 --> 00:22:12,760
Speaker 2: wouldn't survive, and so that's not something that a lot

378
00:22:12,760 --> 00:22:16,440
Speaker 2: of employees wanted. It became clear that even if Opening

379
00:22:16,480 --> 00:22:18,960
Speaker 2: I did survive, they would be a lot more short

380
00:22:19,080 --> 00:22:21,240
Speaker 2: changed in terms of the amount of capital that they

381
00:22:21,280 --> 00:22:23,840
Speaker 2: would be able to get because he would no longer

382
00:22:23,880 --> 00:22:27,960
Speaker 2: be their champion for that, and also the tender offer

383
00:22:28,000 --> 00:22:30,480
Speaker 2: could potentially go away and they would not be personally

384
00:22:30,920 --> 00:22:35,000
Speaker 2: enriched as well, and you know, many of them. The

385
00:22:35,040 --> 00:22:39,000
Speaker 2: thing to also understand is like it's very expensive to

386
00:22:39,000 --> 00:22:42,040
Speaker 2: live in the Bay Area, and so for the worker,

387
00:22:42,240 --> 00:22:45,240
Speaker 2: for the employees, in the moment, losing the tender offer

388
00:22:45,359 --> 00:22:48,480
Speaker 2: wasn't like oh no, I'm going to lose like my retirement.

389
00:22:48,640 --> 00:22:51,560
Speaker 2: It was also this sense of like, I'm literally going

390
00:22:51,640 --> 00:22:55,760
Speaker 2: to lose my financial security right now, like I already

391
00:22:56,280 --> 00:22:59,320
Speaker 2: tried to, I already bought, you know, a house based

392
00:22:59,320 --> 00:23:02,600
Speaker 2: on the fact. I feel like you mentioned that yeah, yeah.

393
00:23:02,640 --> 00:23:04,400
Speaker 1: That was someone who put money.

394
00:23:04,680 --> 00:23:07,960
Speaker 2: Yeah, who put plenty of money down, and that the

395
00:23:09,040 --> 00:23:13,440
Speaker 2: tender offer dissolving was a real financial stress. It was

396
00:23:13,760 --> 00:23:16,040
Speaker 2: a threat to their financial existence in the life.

397
00:23:16,080 --> 00:23:19,359
Speaker 1: I imagine so, But I'm just the way it was

398
00:23:19,440 --> 00:23:22,800
Speaker 1: framed in public was that this was some big loyalty

399
00:23:22,880 --> 00:23:25,480
Speaker 1: thing where everyone was like, I love open Ai and

400
00:23:25,560 --> 00:23:29,879
Speaker 1: Sam Altman, and that just didn't feel like it was

401
00:23:29,920 --> 00:23:32,760
Speaker 1: what was happening. That people seemed angry at Ilia, but

402
00:23:32,800 --> 00:23:36,280
Speaker 1: they just seemed angry because something changed, rather than.

403
00:23:36,520 --> 00:23:39,480
Speaker 2: Like, yeah, yeah, I mean I think I think there

404
00:23:39,480 --> 00:23:43,280
Speaker 2: were certainly people within the company that did feel loyalty

405
00:23:43,600 --> 00:23:46,760
Speaker 2: to Altman and and and that was one of their

406
00:23:46,800 --> 00:23:49,639
Speaker 2: primary motivating things. But by and Leche, when I was

407
00:23:49,640 --> 00:23:54,640
Speaker 2: interviewing lots of employees for understanding like what ultimately led

408
00:23:54,640 --> 00:24:00,280
Speaker 2: them to rally around Sam, there was actually more more

409
00:24:01,080 --> 00:24:04,760
Speaker 2: practical concerns than just personal loyalty that was driving the thing,

410
00:24:04,960 --> 00:24:07,600
Speaker 2: whether it was financial or whether it was just I

411
00:24:07,640 --> 00:24:09,919
Speaker 2: really believe in Agi and I don't want open Aiye

412
00:24:09,960 --> 00:24:13,160
Speaker 2: to go away because it'll scrap all of the work

413
00:24:13,200 --> 00:24:17,720
Speaker 2: that we've done. And of course, you know, of course

414
00:24:17,960 --> 00:24:20,840
Speaker 2: the narrative would I mean, the open ai themselves have

415
00:24:20,920 --> 00:24:24,240
Speaker 2: been pushing again and again and again this idea that

416
00:24:25,040 --> 00:24:28,000
Speaker 2: all of the employee or whatever, more than ninety percent

417
00:24:28,080 --> 00:24:31,520
Speaker 2: of the employees ended up signing the petition, and they

418
00:24:31,720 --> 00:24:34,520
Speaker 2: show they cite this number as just a show of

419
00:24:34,520 --> 00:24:37,399
Speaker 2: solidarity and loyalty to Altman. But then, of course, if

420
00:24:37,440 --> 00:24:39,720
Speaker 2: you look at the track record after the Boyd crisis

421
00:24:39,720 --> 00:24:43,760
Speaker 2: of how many people have subsequently left the company once

422
00:24:45,480 --> 00:24:48,800
Speaker 2: once things have sort of stabilized and there isn't a

423
00:24:48,880 --> 00:24:52,840
Speaker 2: crisis situation, that is I think much more revealing of

424
00:24:52,960 --> 00:24:55,119
Speaker 2: how much loyalty people have to him.

425
00:24:55,920 --> 00:24:59,560
Speaker 1: So tell me about Jack Clock. So Jack Clock is

426
00:24:59,640 --> 00:25:01,640
Speaker 1: the what is he? And he's one of the co

427
00:25:01,720 --> 00:25:05,280
Speaker 1: founders of Anthropic. Now, yeah, I, without putting you on

428
00:25:05,320 --> 00:25:07,640
Speaker 1: the spot, kind of feels like Jack Clark has got

429
00:25:07,680 --> 00:25:10,320
Speaker 1: off a little easy with everyone not even saying you.

430
00:25:10,320 --> 00:25:12,159
Speaker 1: You're one of the few people. Jack Clark worked at

431
00:25:12,160 --> 00:25:15,800
Speaker 1: The Register, which is an extremely critical IT publication, and

432
00:25:15,840 --> 00:25:18,840
Speaker 1: now he's out in conferences saying that AI agents will

433
00:25:18,880 --> 00:25:22,280
Speaker 1: control everything. He just feels like one of the weirdest

434
00:25:22,359 --> 00:25:23,920
Speaker 1: characters in this whole story.

435
00:25:26,359 --> 00:25:29,520
Speaker 2: Yeah, Yeah, it's interesting. Like when I went to profile

436
00:25:29,520 --> 00:25:33,240
Speaker 2: Opening Eye in twenty nineteen, I actually the first person

437
00:25:33,280 --> 00:25:36,240
Speaker 2: I reached out to was Jack because I had spoken

438
00:25:36,280 --> 00:25:39,320
Speaker 2: to him before and he had until then, until recently

439
00:25:39,359 --> 00:25:43,280
Speaker 2: been playing communications for head for Open Eye, and then

440
00:25:43,680 --> 00:25:46,080
Speaker 2: he had shifted into a policy role. And I remember

441
00:25:46,119 --> 00:25:50,240
Speaker 2: when I when I was at the company, I was like, hey, Jack, like,

442
00:25:50,280 --> 00:25:53,200
Speaker 2: do you think you can actually give me more access

443
00:25:53,240 --> 00:25:57,280
Speaker 2: to seeing the things that I like, stuff that I

444
00:25:57,400 --> 00:25:58,800
Speaker 2: like to say? Yeah, Like I was literally I was

445
00:25:58,800 --> 00:26:01,040
Speaker 2: literally asking so they wouldn't let me go beyond the

446
00:26:01,040 --> 00:26:04,480
Speaker 2: first floor. There were there's there were three floors second.

447
00:26:04,200 --> 00:26:06,880
Speaker 1: And I'm so sorry there's computers there. There's not It's

448
00:26:06,920 --> 00:26:08,800
Speaker 1: not like they have an AI machine. Come on.

449
00:26:09,280 --> 00:26:11,800
Speaker 2: Yeah, And I was like, hey, like, can I just

450
00:26:12,040 --> 00:26:14,080
Speaker 2: literally just go up to the second? Can you like

451
00:26:14,119 --> 00:26:15,920
Speaker 2: take me up and just like let me walk around?

452
00:26:16,080 --> 00:26:20,119
Speaker 2: And he looked at me with this like like deep

453
00:26:20,200 --> 00:26:24,160
Speaker 2: deep side eye of like no, Karen, like you absolutely cannot.

454
00:26:24,680 --> 00:26:27,960
Speaker 2: And I was like, so, Fawny, you're a former journalist.

455
00:26:28,000 --> 00:26:30,560
Speaker 2: You know how this works. Like the.

456
00:26:32,000 --> 00:26:35,200
Speaker 1: Jack Frow in twenty fourteen for The Register was shock

457
00:26:35,240 --> 00:26:37,840
Speaker 1: an a w West the fall of Amazon's deflationary cloud,

458
00:26:38,040 --> 00:26:39,719
Speaker 1: just as Jeff Bezos did the books and see these

459
00:26:39,720 --> 00:26:42,160
Speaker 1: Amazon's rivals are now doing it. He used to write

460
00:26:42,240 --> 00:26:47,199
Speaker 1: these like very grouchy l reg style pig It's just

461
00:26:47,320 --> 00:26:48,320
Speaker 1: so weird.

462
00:26:49,240 --> 00:26:51,760
Speaker 2: Yeah, I mean I think this is like I've.

463
00:26:52,040 --> 00:26:53,400
Speaker 1: But it gets back to the thing you were saying

464
00:26:53,400 --> 00:26:54,800
Speaker 1: about the kind of the doctrine.

465
00:26:55,359 --> 00:27:00,480
Speaker 2: Yeah. So, like, because I started covering this company intoing nineteen,

466
00:27:01,480 --> 00:27:03,560
Speaker 2: I talked with people then that I then talked to

467
00:27:03,640 --> 00:27:05,520
Speaker 2: for the book, and I was able to sort of

468
00:27:06,359 --> 00:27:10,399
Speaker 2: have this unique opportunity to track how people's individual beliefs

469
00:27:10,400 --> 00:27:15,400
Speaker 2: evolve when they are seeped in this world, and there

470
00:27:15,400 --> 00:27:18,000
Speaker 2: were people that I was talking to back then that

471
00:27:18,080 --> 00:27:20,239
Speaker 2: were like, I don't really believe in this AGI thing,

472
00:27:20,400 --> 00:27:22,280
Speaker 2: that by the time I was talking to them for

473
00:27:22,320 --> 00:27:24,680
Speaker 2: the book were like AGI all the way, like that

474
00:27:24,920 --> 00:27:28,359
Speaker 2: this is a genuine, true belief. And I think there's

475
00:27:28,400 --> 00:27:31,480
Speaker 2: a lot of reasons for this transformation, Like one is

476
00:27:31,480 --> 00:27:35,520
Speaker 2: that you are only talking to people who believe this,

477
00:27:35,680 --> 00:27:39,360
Speaker 2: so you're just constantly in this environment where you're not

478
00:27:39,480 --> 00:27:43,600
Speaker 2: talking to people who are challenging or testing that belief

479
00:27:43,680 --> 00:27:49,000
Speaker 2: and instead just like continuously being reinforced in this echo chamber.

480
00:27:50,119 --> 00:27:53,159
Speaker 2: But I think there's another thing that I kind of

481
00:27:53,200 --> 00:27:55,720
Speaker 2: came to realize while reporting on the book is like

482
00:27:55,760 --> 00:27:59,000
Speaker 2: people who really really believe that AGI is possible, that

483
00:27:59,040 --> 00:28:01,840
Speaker 2: we will actually be able to replicate human intelligence. It's

484
00:28:01,880 --> 00:28:05,360
Speaker 2: not a belief about what AI is capable of. It

485
00:28:05,400 --> 00:28:11,639
Speaker 2: is a belief about what human intelligence is. And a

486
00:28:11,720 --> 00:28:14,120
Speaker 2: lot of people in the AI world today have this

487
00:28:14,320 --> 00:28:17,359
Speaker 2: belief that human intelligence or everything in the world is

488
00:28:17,480 --> 00:28:21,360
Speaker 2: inherently computable and all you have to do is just

489
00:28:21,480 --> 00:28:23,840
Speaker 2: a mass more data and more compute and eventually you

490
00:28:23,920 --> 00:28:26,919
Speaker 2: will get to that thing. You will be able to

491
00:28:26,920 --> 00:28:29,920
Speaker 2: replicate that thing. And when you are in this kind

492
00:28:29,960 --> 00:28:34,560
Speaker 2: of environment where you have people constantly arguing to you

493
00:28:34,920 --> 00:28:37,480
Speaker 2: that this is why AGA is possible, because everything is computable,

494
00:28:37,520 --> 00:28:41,480
Speaker 2: and then you see the rapid clip of your models

495
00:28:41,760 --> 00:28:44,080
Speaker 2: being able to do more and more functions that you

496
00:28:44,120 --> 00:28:48,240
Speaker 2: know other people outside in society previously would have suggested

497
00:28:48,760 --> 00:28:52,520
Speaker 2: were not possible. It's sort of the self reinforcing belief machine,

498
00:28:52,960 --> 00:29:00,120
Speaker 2: like it just manufacturers gives you yeah, exactly, And so

499
00:29:00,200 --> 00:29:04,480
Speaker 2: I think. And one of the things that I also

500
00:29:05,120 --> 00:29:09,200
Speaker 2: have just has a general realization not just with opening

501
00:29:09,200 --> 00:29:12,360
Speaker 2: eye but in general. When I'm covering tech companies, I

502
00:29:12,440 --> 00:29:14,040
Speaker 2: kind of have a policy for myself to do a

503
00:29:14,040 --> 00:29:15,800
Speaker 2: little bit of a detox after I spend a lot

504
00:29:15,800 --> 00:29:18,560
Speaker 2: of time talking with them, because it is really like,

505
00:29:18,600 --> 00:29:20,640
Speaker 2: when you're talking with all of these people that exist

506
00:29:20,640 --> 00:29:24,560
Speaker 2: in this world, you do adopt their worldview, and you

507
00:29:24,600 --> 00:29:27,120
Speaker 2: do adopt their talking points, and you do see things

508
00:29:27,440 --> 00:29:31,880
Speaker 2: through their eyes. And usually I then have to like

509
00:29:32,440 --> 00:29:36,000
Speaker 2: let myself just be in the actual world for a

510
00:29:36,040 --> 00:29:39,320
Speaker 2: little bit and remind myself of what the average person

511
00:29:39,360 --> 00:29:43,480
Speaker 2: thinks and what the average person values, and remind myself

512
00:29:43,480 --> 00:29:46,360
Speaker 2: that there are you know, there are problems beyond the

513
00:29:46,400 --> 00:29:50,840
Speaker 2: Silicon Valley's borders that just look fundamentally different from what

514
00:29:51,040 --> 00:29:54,640
Speaker 2: they conceive the world to be. And so I did

515
00:29:54,720 --> 00:29:57,000
Speaker 2: that with the Opening Eye profile. I did that with

516
00:29:57,760 --> 00:30:00,760
Speaker 2: I profiled Facebook years ago, and I that with Facebook.

517
00:30:00,840 --> 00:30:03,240
Speaker 2: I did that with the book, where I would interview

518
00:30:03,280 --> 00:30:05,800
Speaker 2: lots of people in like these big batches and kind

519
00:30:05,840 --> 00:30:08,400
Speaker 2: of really do my best to try and occupy their

520
00:30:08,440 --> 00:30:12,800
Speaker 2: shoes for a couple of weeks a month, and then

521
00:30:12,960 --> 00:30:17,840
Speaker 2: I would spend my time explicitly not interviewing open ani people,

522
00:30:17,960 --> 00:30:21,080
Speaker 2: just interviewing other people that were out in the world

523
00:30:21,120 --> 00:30:23,800
Speaker 2: to just like reset my brain chemistry a little bit,

524
00:30:24,320 --> 00:30:26,200
Speaker 2: because it really does feel that way. It really does

525
00:30:26,280 --> 00:30:31,280
Speaker 2: feel like you kind of get absorbed into this singular

526
00:30:31,520 --> 00:30:33,440
Speaker 2: world view and then you have to kind of remind

527
00:30:33,440 --> 00:30:36,760
Speaker 2: yourself of the greater reality.

528
00:30:49,880 --> 00:30:51,840
Speaker 1: I'm gonna ask this question without getting you in too

529
00:30:51,920 --> 00:30:54,120
Speaker 1: much trouble. You think that's what happened with Kevin Russe,

530
00:30:56,360 --> 00:31:00,360
Speaker 1: because it's really I know, I don't want to put

531
00:31:00,360 --> 00:31:02,240
Speaker 1: you in a situation we have to talk kill of someone,

532
00:31:02,520 --> 00:31:07,120
Speaker 1: but that interview was bizarre and hard fork I So.

533
00:31:08,720 --> 00:31:13,400
Speaker 2: I've been I think really lucky in that I've covered

534
00:31:13,560 --> 00:31:17,040
Speaker 2: the tech industry almost always not living in SF.

535
00:31:18,120 --> 00:31:19,280
Speaker 1: I agree, that's a great thing.

536
00:31:19,720 --> 00:31:23,600
Speaker 2: And and you know, like I've been able to figure

537
00:31:23,640 --> 00:31:25,320
Speaker 2: that out in my career, and that was an explicit

538
00:31:25,320 --> 00:31:27,560
Speaker 2: dis like I did not want to live in SF

539
00:31:27,640 --> 00:31:30,920
Speaker 2: anymore like I had lived in SF. I wanted to

540
00:31:31,200 --> 00:31:35,520
Speaker 2: get out. And I think this is a really it's

541
00:31:35,320 --> 00:31:40,280
Speaker 2: it's a really hard balance for any journalist. Is you

542
00:31:40,400 --> 00:31:42,360
Speaker 2: need to decide whether you're close to your subject and

543
00:31:42,400 --> 00:31:44,920
Speaker 2: immerse in their world and therefore might be co opted

544
00:31:45,040 --> 00:31:49,719
Speaker 2: by their world, or whether you exist outside of that

545
00:31:49,800 --> 00:31:52,800
Speaker 2: world and therefore you don't have as much access. You

546
00:31:53,080 --> 00:31:55,080
Speaker 2: don't get to go to the parties where you hear

547
00:31:55,160 --> 00:31:58,880
Speaker 2: tips all the time. And and that's one just like

548
00:31:59,040 --> 00:32:01,280
Speaker 2: it's been attention in my career as well as like

549
00:32:01,360 --> 00:32:03,440
Speaker 2: I constantly feel like I'm missing things because I'm not

550
00:32:03,440 --> 00:32:05,120
Speaker 2: an SF. But the thing that I think I have

551
00:32:05,480 --> 00:32:09,120
Speaker 2: gained from not being an SF is just a continued

552
00:32:09,200 --> 00:32:16,040
Speaker 2: connection to non SF world, you know, Like I I

553
00:32:16,240 --> 00:32:19,560
Speaker 2: notice when I spend too much time with s F

554
00:32:19,640 --> 00:32:24,040
Speaker 2: people that I start in my vocabulary changes, like how

555
00:32:24,480 --> 00:32:27,120
Speaker 2: I talk about things changes, because because people in SF

556
00:32:27,160 --> 00:32:31,120
Speaker 2: talk about things in a very particular way, you know,

557
00:32:31,240 --> 00:32:36,560
Speaker 2: like they are talking about like optimization hacking, and like

558
00:32:37,400 --> 00:32:42,840
Speaker 2: they have a particular utilitarian maximization mindset around how they

559
00:32:42,880 --> 00:32:47,480
Speaker 2: do things and why, and I have to then kind

560
00:32:47,480 --> 00:32:52,000
Speaker 2: of step away from that and reseet my language, even

561
00:32:52,560 --> 00:32:55,520
Speaker 2: when when then I sit down to write a story

562
00:32:55,600 --> 00:32:58,920
Speaker 2: that's for the greater public. And so yeah, so I

563
00:32:58,960 --> 00:33:01,440
Speaker 2: think this is something that's just challenging in general. Is

564
00:33:01,520 --> 00:33:05,080
Speaker 2: like it's really hard to not get too close to

565
00:33:05,160 --> 00:33:09,920
Speaker 2: your sources and to not start adopting everything that they

566
00:33:10,200 --> 00:33:15,640
Speaker 2: say as as your own, especially if you are literally

567
00:33:15,680 --> 00:33:16,240
Speaker 2: living with them.

568
00:33:16,800 --> 00:33:21,000
Speaker 1: And yes, in some cases right could be anyone. But

569
00:33:21,360 --> 00:33:23,640
Speaker 1: it does feel like there is a kind of almost

570
00:33:23,680 --> 00:33:27,360
Speaker 1: word containent or thought contagion with this stuff with AGI

571
00:33:27,600 --> 00:33:30,560
Speaker 1: that it pickles certain people. They hear about the idea

572
00:33:30,600 --> 00:33:34,200
Speaker 1: of the autonomous computer and it drives them mad and everything.

573
00:33:34,280 --> 00:33:38,160
Speaker 1: To your point, they start chasing it even though there's

574
00:33:38,240 --> 00:33:40,240
Speaker 1: not really any evidence that we can do it.

575
00:33:41,960 --> 00:33:44,959
Speaker 2: Yeah, I mean, I mean, like when I first started

576
00:33:45,280 --> 00:33:50,080
Speaker 2: covering AI, I also was so enthralled by the premise,

577
00:33:50,760 --> 00:33:53,840
Speaker 2: Like when I when I so before I feel AI,

578
00:33:56,000 --> 00:33:58,400
Speaker 2: I didn't really like that when I first started covering

579
00:33:58,400 --> 00:34:03,200
Speaker 2: it at my two technology review. I did not realize

580
00:34:03,240 --> 00:34:07,800
Speaker 2: before then that AI was actually trying to recreate human intelligence.

581
00:34:07,800 --> 00:34:10,319
Speaker 2: I thought it was just you know, I mean, it

582
00:34:10,360 --> 00:34:12,719
Speaker 2: is a marketing it is a marketing term.

583
00:34:12,760 --> 00:34:14,799
Speaker 1: But even then, this sounds like it might be a

584
00:34:14,840 --> 00:34:16,520
Speaker 1: definition that people argue.

585
00:34:16,280 --> 00:34:19,880
Speaker 2: Over right, right right, But I mean, like in the original,

586
00:34:19,960 --> 00:34:22,920
Speaker 2: like when AI was coined as a term in nineteen

587
00:34:22,960 --> 00:34:27,279
Speaker 2: fifty six, like John McCarthy, he did explicitly coin the

588
00:34:27,400 --> 00:34:30,360
Speaker 2: term both to be to attract funders, so as a

589
00:34:30,400 --> 00:34:34,400
Speaker 2: marketing term, and because he was trying to describe what

590
00:34:34,560 --> 00:34:37,440
Speaker 2: he wanted the field to do, which was to recreate

591
00:34:37,520 --> 00:34:40,959
Speaker 2: human intelligence. And that is just it's it's it's such

592
00:34:41,040 --> 00:34:47,040
Speaker 2: an evocative thing, like to think, wait a minute, could

593
00:34:47,040 --> 00:34:49,560
Speaker 2: we actually do that? And what would that mean? And

594
00:34:49,600 --> 00:34:53,840
Speaker 2: there's so much philosophical Uh, it's just a philosophical mindfield.

595
00:34:53,880 --> 00:34:58,400
Speaker 2: Like and if you are someone that loves philosophizing, you

596
00:34:58,440 --> 00:35:00,440
Speaker 2: can just sit there for like days in days and

597
00:35:00,520 --> 00:35:02,920
Speaker 2: days and think, holy crap, like what would that be?

598
00:35:03,000 --> 00:35:06,640
Speaker 1: What would that look like? How can we do times columns?

599
00:35:07,640 --> 00:35:10,360
Speaker 2: And and so I really got I got pulled into

600
00:35:10,760 --> 00:35:18,600
Speaker 2: just the kind of sheer enigma of that and yeah,

601
00:35:18,719 --> 00:35:21,960
Speaker 2: also the power of that of oh wow, if we

602
00:35:21,960 --> 00:35:24,720
Speaker 2: could do this, like if if you know, if I

603
00:35:24,719 --> 00:35:29,200
Speaker 2: imagine being in the shoes of someone who's actually doing

604
00:35:29,239 --> 00:35:33,239
Speaker 2: the AI research and thinking to yourself, I might be

605
00:35:33,320 --> 00:35:37,279
Speaker 2: contributing to the recreation of my own intelligence sort of

606
00:35:37,280 --> 00:35:40,279
Speaker 2: of our collective intelligence. Like that's intoxicating, you.

607
00:35:40,239 --> 00:35:45,279
Speaker 1: Know, feels like philosophy marketing though, because I I just

608
00:35:45,360 --> 00:35:47,480
Speaker 1: look at this stuff and I hear about this stuff,

609
00:35:47,480 --> 00:35:49,800
Speaker 1: and I always think, Okay, but what you're doing today,

610
00:35:50,640 --> 00:35:52,520
Speaker 1: And I look at what they're doing today, and I say,

611
00:35:52,520 --> 00:35:54,960
Speaker 1: that doesn't seem anything like that, And I understand. I

612
00:35:55,000 --> 00:35:57,880
Speaker 1: actually don't think that there's anything harmful in discussing AGI.

613
00:35:58,120 --> 00:36:00,000
Speaker 1: What pisses me off is how many people don't seem

614
00:36:00,000 --> 00:36:04,400
Speaker 1: to be discussing AGI. They discuss the ramifications on the edges.

615
00:36:04,560 --> 00:36:07,839
Speaker 1: Because something that and Casey Kawa, friend of the show,

616
00:36:07,840 --> 00:36:09,480
Speaker 1: has brought up a number of times with me, is like,

617
00:36:09,560 --> 00:36:12,719
Speaker 1: no one seems to be discussing personage, Like if we

618
00:36:12,800 --> 00:36:15,360
Speaker 1: make a conscious computer, do we give it a Social

619
00:36:15,360 --> 00:36:16,240
Speaker 1: Security number?

620
00:36:16,440 --> 00:36:18,279
Speaker 2: That's actually really funny because I think there are too

621
00:36:18,320 --> 00:36:20,120
Speaker 2: many people discussing personage.

622
00:36:21,160 --> 00:36:23,120
Speaker 1: I don't see them in the black. Well, perhaps they're

623
00:36:23,120 --> 00:36:25,200
Speaker 1: not doing it in the media because AGI gets brought

624
00:36:25,239 --> 00:36:28,040
Speaker 1: up as this vague term and then they go, ah,

625
00:36:28,239 --> 00:36:31,080
Speaker 1: what do you think, Yes, could be good, could be bad,

626
00:36:31,200 --> 00:36:35,280
Speaker 1: millions brillion sound fucking And it's just it's so bizarre

627
00:36:35,400 --> 00:36:39,640
Speaker 1: because I look at I've been covering I personally with

628
00:36:39,680 --> 00:36:42,399
Speaker 1: AI really only started looking at it hard in twenty

629
00:36:42,440 --> 00:36:45,760
Speaker 1: twenty three, which is my own fault. And I've looked

630
00:36:45,760 --> 00:36:48,400
Speaker 1: and perhaps that has also colored my belief system because

631
00:36:49,480 --> 00:36:52,719
Speaker 1: I ken looking for the thing, like the stuff, the

632
00:36:52,800 --> 00:36:54,719
Speaker 1: thing that everyone was freaking out about. And you look

633
00:36:54,719 --> 00:36:58,160
Speaker 1: and it's like we've extrapolated from large language models that

634
00:36:58,960 --> 00:37:01,520
Speaker 1: AGI will come out. But actually that kind of leads

635
00:37:01,560 --> 00:37:05,440
Speaker 1: me to another question. You Sam Morton's a confusing person.

636
00:37:05,480 --> 00:37:09,799
Speaker 1: What about Dario Amma Day, Darry Ama Day. What do

637
00:37:09,840 --> 00:37:12,319
Speaker 1: you think does he believe in AGI? You think you

638
00:37:12,360 --> 00:37:13,560
Speaker 1: think he's a true believer.

639
00:37:14,280 --> 00:37:16,319
Speaker 2: I do think Dario is a true believer, yeah, And

640
00:37:16,360 --> 00:37:18,200
Speaker 2: I do think that he's a true duomer as well,

641
00:37:18,239 --> 00:37:22,920
Speaker 2: like he genuinely has a lot of anxiety around the

642
00:37:23,480 --> 00:37:27,600
Speaker 2: AGI creating the end of the world, whether or not,

643
00:37:28,640 --> 00:37:30,480
Speaker 2: and also like what does it mean to be a

644
00:37:30,480 --> 00:37:31,080
Speaker 2: true believer?

645
00:37:31,200 --> 00:37:34,440
Speaker 1: You know, like, does he believe the bollocks he's saying

646
00:37:34,719 --> 00:37:37,000
Speaker 1: because he claims that AGI will be here but twenty

647
00:37:37,080 --> 00:37:38,360
Speaker 1: twenty seven or quicker.

648
00:37:38,760 --> 00:37:41,719
Speaker 2: Yeah, So that then is when he's just wearing his

649
00:37:41,800 --> 00:37:43,560
Speaker 2: CEO hat and he needs to say something.

650
00:37:43,960 --> 00:37:45,880
Speaker 1: When you say wearing the CEO, can you be a

651
00:37:45,880 --> 00:37:46,719
Speaker 1: bit more specific?

652
00:37:47,800 --> 00:37:52,640
Speaker 2: I think Dario is an interesting case in that he

653
00:37:52,960 --> 00:37:55,759
Speaker 2: originally he has a different background than Sam. You know,

654
00:37:56,040 --> 00:37:59,040
Speaker 2: Sam is a VC that or an investor that then

655
00:37:59,520 --> 00:38:02,239
Speaker 2: became the CEO of an AI company, and his skill

656
00:38:02,280 --> 00:38:06,719
Speaker 2: is storytelling, right, That's what all investors do. Dario was,

657
00:38:07,520 --> 00:38:11,640
Speaker 2: he was a scientist. He studied I think computational neuroscience,

658
00:38:11,840 --> 00:38:14,600
Speaker 2: and he had a kind of deep fascination with this

659
00:38:14,719 --> 00:38:16,920
Speaker 2: idea of how do you how do you figure out

660
00:38:16,920 --> 00:38:18,360
Speaker 2: how the brain works and how do you replicate it

661
00:38:18,400 --> 00:38:20,960
Speaker 2: like it was? It was. He didn't initially call himself

662
00:38:20,960 --> 00:38:23,120
Speaker 2: an AI researcher, and I think the early days of

663
00:38:23,160 --> 00:38:27,440
Speaker 2: his academic career, but like he was essentially studying a

664
00:38:27,480 --> 00:38:31,360
Speaker 2: lot of the things that hardcore AI researchers study, the brain,

665
00:38:32,120 --> 00:38:34,920
Speaker 2: computer science, like all all of these things. And so

666
00:38:35,040 --> 00:38:38,880
Speaker 2: I think he has this fascination and I don't know

667
00:38:38,960 --> 00:38:41,160
Speaker 2: this for sure, but I would guess that he is

668
00:38:41,600 --> 00:38:44,880
Speaker 2: of the category of people that I described that believes

669
00:38:44,880 --> 00:38:47,640
Speaker 2: that everything is fundamentally computable in the world, and human

670
00:38:47,640 --> 00:38:52,319
Speaker 2: intelligence is computable, and so he does really believe that

671
00:38:52,440 --> 00:38:55,640
Speaker 2: if he can figure it out like AGI will happen.

672
00:38:57,080 --> 00:39:00,600
Speaker 2: But then he has to run a company, and a

673
00:39:00,640 --> 00:39:02,600
Speaker 2: company can't just do science. And actually, one of the

674
00:39:02,640 --> 00:39:08,920
Speaker 2: things that people mentioned to me about their criticisms of

675
00:39:09,000 --> 00:39:12,560
Speaker 2: Daria when he initially ran Anthropic was that he didn't

676
00:39:12,600 --> 00:39:15,000
Speaker 2: care about the business at all, Like he seemed to

677
00:39:15,040 --> 00:39:19,120
Speaker 2: have no interest in anything other than the science. And

678
00:39:19,200 --> 00:39:21,640
Speaker 2: there were people within the company that were like, this

679
00:39:21,760 --> 00:39:23,680
Speaker 2: is not going to work as a company if you

680
00:39:23,920 --> 00:39:28,319
Speaker 2: cannot literally do business, if you cannot raise money, And

681
00:39:28,360 --> 00:39:33,759
Speaker 2: so I think what happened I didn't actually report this out,

682
00:39:33,800 --> 00:39:36,080
Speaker 2: but my guess is what happened is Dario then had

683
00:39:36,120 --> 00:39:39,720
Speaker 2: to shift to not just being a scientist but also

684
00:39:39,960 --> 00:39:43,360
Speaker 2: being a businessman, and he had to learn how to storytell,

685
00:39:43,440 --> 00:39:48,400
Speaker 2: and is he and I think honestly he tries to.

686
00:39:49,120 --> 00:39:51,719
Speaker 2: You know, Sam Altman is a really successful storyteller and

687
00:39:52,200 --> 00:39:54,760
Speaker 2: able to accreu a out of capital. I think Dario

688
00:39:55,080 --> 00:39:59,960
Speaker 2: tries to match the stories that Altman tells in order

689
00:40:00,160 --> 00:40:03,040
Speaker 2: to try and accrue the same amount of capital and

690
00:40:03,280 --> 00:40:07,000
Speaker 2: try to take capital away. Maybe because ultimately they are

691
00:40:07,320 --> 00:40:12,120
Speaker 2: personal archite nemesis and anthropic and open AI are competiti.

692
00:40:11,880 --> 00:40:13,719
Speaker 1: Hate each other so much? Is it just because Sam

693
00:40:13,760 --> 00:40:16,399
Speaker 1: Wlman doesn't like the Warrio walked off?

694
00:40:19,239 --> 00:40:24,120
Speaker 2: I don't know that Sam. I. I can't figure out

695
00:40:24,160 --> 00:40:30,200
Speaker 2: whether Sam genuinely ever hates anyone, but people certainly hate him,

696
00:40:30,280 --> 00:40:35,640
Speaker 2: and Dario hates him for sure. I think it goes

697
00:40:35,680 --> 00:40:38,920
Speaker 2: back to this idea of do you agree or do

698
00:40:38,960 --> 00:40:41,759
Speaker 2: you disagree with Sam about something fundamental and therefore do

699
00:40:41,800 --> 00:40:43,880
Speaker 2: you perceive him as the greatest asset ever or the

700
00:40:43,880 --> 00:40:47,840
Speaker 2: greatest threat ever? And from in Dario's case, he fundamentally

701
00:40:47,920 --> 00:40:57,000
Speaker 2: disagreed with Sam about certain key decisions around safety AAI safety,

702
00:40:57,080 --> 00:41:01,000
Speaker 2: the Dumer, the Dumer brand of AI safety, where they

703
00:41:01,280 --> 00:41:06,000
Speaker 2: Daria was the one that decided to blow up the

704
00:41:06,040 --> 00:41:08,000
Speaker 2: amount of computer chips that were being used to train

705
00:41:08,040 --> 00:41:10,719
Speaker 2: a single model. So he did that. From GPT two

706
00:41:10,760 --> 00:41:12,840
Speaker 2: to GPT three, they went from a couple dozen chips

707
00:41:13,280 --> 00:41:16,440
Speaker 2: to ten thousand chips all at once to train GPT

708
00:41:16,560 --> 00:41:20,240
Speaker 2: three and Daria wanted to do this because he wanted

709
00:41:20,360 --> 00:41:24,160
Speaker 2: to create an internal lead in order to then have

710
00:41:24,480 --> 00:41:28,080
Speaker 2: some time to do research on this model that would

711
00:41:28,120 --> 00:41:31,920
Speaker 2: emerge from ten thousand chips. And Allmann does this thing

712
00:41:31,960 --> 00:41:34,920
Speaker 2: where he will convince, he will ally with people, so

713
00:41:35,520 --> 00:41:37,560
Speaker 2: he was like, oh, ten thousand chips, that just a

714
00:41:37,600 --> 00:41:40,919
Speaker 2: brilliant idea. We should totally do that. But then once

715
00:41:40,960 --> 00:41:45,080
Speaker 2: it was done, he sort of shifted to, Okay, now

716
00:41:45,080 --> 00:41:46,759
Speaker 2: we should release it, or now we should give it

717
00:41:46,800 --> 00:41:49,360
Speaker 2: to Microsoft because we have this deal with Microsoft. We

718
00:41:49,400 --> 00:41:51,799
Speaker 2: need to make them happy. We need to give them

719
00:41:51,840 --> 00:41:56,080
Speaker 2: some kind of really exciting thing deliverable to justify the

720
00:41:56,160 --> 00:41:58,080
Speaker 2: first one billion dollars they gave us so that they

721
00:41:58,080 --> 00:42:01,640
Speaker 2: can then give us more money. And so it was

722
00:42:01,680 --> 00:42:04,160
Speaker 2: actually it was like the two It was both Altman

723
00:42:04,360 --> 00:42:09,439
Speaker 2: and Amoday together that I would credit as being responsible

724
00:42:09,520 --> 00:42:14,360
Speaker 2: for dramatically accelerating the AI race, because Amoday was the

725
00:42:14,360 --> 00:42:16,680
Speaker 2: one that decided we need to blow it up to

726
00:42:16,680 --> 00:42:19,160
Speaker 2: ten thousand chips, and then Altman was the one that

727
00:42:19,200 --> 00:42:22,120
Speaker 2: persuaded him, yes, you should do it because I agree

728
00:42:22,120 --> 00:42:24,279
Speaker 2: with you, and then kind of flipped to, Okay, now

729
00:42:24,280 --> 00:42:25,480
Speaker 2: we need to get this out in the world as

730
00:42:25,560 --> 00:42:34,319
Speaker 2: quickly as possible, and amoday I think feels like his

731
00:42:34,600 --> 00:42:39,960
Speaker 2: intelligent Like Altman, as a politically savvy person, was able

732
00:42:40,000 --> 00:42:45,080
Speaker 2: to use his intelligence against him to achieve exactly the

733
00:42:45,120 --> 00:42:50,600
Speaker 2: opposite of what he ultimately wanted, which was to slow Yeah,

734
00:42:50,640 --> 00:42:52,360
Speaker 2: to slow things down rather than accelerate it.

735
00:42:53,000 --> 00:42:56,399
Speaker 1: This sounds like colonial This sounds like colonial Britain. It's

736
00:42:56,480 --> 00:42:59,360
Speaker 1: just white guys getting angry at each other over tiny

737
00:42:59,480 --> 00:43:04,000
Speaker 1: grievance is from years ago. Here's a weird question. Well,

738
00:43:04,000 --> 00:43:06,040
Speaker 1: first of all, do any of them seem happy in

739
00:43:06,080 --> 00:43:08,800
Speaker 1: any way? Do any of them seem to enjoy anything?

740
00:43:08,880 --> 00:43:12,759
Speaker 1: I asked this seriously, I genuinely do they seem miserable?

741
00:43:12,880 --> 00:43:15,880
Speaker 1: That is the consistent theme from all of Jack Clark included.

742
00:43:16,600 --> 00:43:21,440
Speaker 1: They all seem pissed off, scared, paranoid. Weird. It's that

743
00:43:21,520 --> 00:43:23,160
Speaker 1: they're being driven mad by this.

744
00:43:23,280 --> 00:43:28,440
Speaker 2: Yeah, yeah, yeah, I think that is an entirely accurate description.

745
00:43:28,760 --> 00:43:33,879
Speaker 2: I think you cannot be not driven mad in this

746
00:43:34,000 --> 00:43:38,080
Speaker 2: world where you have convinced yourself that the stakes are

747
00:43:38,120 --> 00:43:41,880
Speaker 2: the future of humanity. You know, like, how do you

748
00:43:41,920 --> 00:43:43,280
Speaker 2: not buckle under that pressure?

749
00:43:44,160 --> 00:43:46,960
Speaker 1: I mean, skill issue? I think I'd be fine give

750
00:43:47,000 --> 00:43:50,439
Speaker 1: me one billion dollars. But it does make me think

751
00:43:50,480 --> 00:43:53,319
Speaker 1: that right now as and Bloomberg came out with the

752
00:43:53,320 --> 00:43:56,960
Speaker 1: headline just as we're recording this, that Stargate soft Banks,

753
00:43:56,960 --> 00:44:01,239
Speaker 1: Stargate is hitting snags over pariff is they can't seem

754
00:44:01,280 --> 00:44:04,920
Speaker 1: to raise the money. I wonder if we're going to

755
00:44:04,960 --> 00:44:09,080
Speaker 1: see new levels of paranoia anxiety with all of these

756
00:44:09,080 --> 00:44:12,080
Speaker 1: people as the AI trade starts to collapse a bit.

757
00:44:13,920 --> 00:44:19,120
Speaker 2: Yeah, this has been an interesting theme that I've picked

758
00:44:19,200 --> 00:44:22,080
Speaker 2: up on with the way that Alman operates is when

759
00:44:22,640 --> 00:44:28,920
Speaker 2: he starts sounding incredibly optimistic in public about the future

760
00:44:29,040 --> 00:44:31,200
Speaker 2: of open AI, the future of AGI, the future of

761
00:44:31,200 --> 00:44:34,520
Speaker 2: all these things, it means that something is going wrong.

762
00:44:35,160 --> 00:44:40,759
Speaker 2: Like it's become the opposite signal, because he will roll

763
00:44:40,800 --> 00:44:44,040
Speaker 2: out the most grandiose language when he needs to cover

764
00:44:44,200 --> 00:44:51,160
Speaker 2: up something that is really stressing him out. And so

765
00:44:51,560 --> 00:44:55,000
Speaker 2: we're seeing, you know, like this happening again more recently

766
00:44:55,040 --> 00:44:57,520
Speaker 2: where I mean, in the beginning of the year, he

767
00:44:57,600 --> 00:45:00,239
Speaker 2: had this post where he was like, we we are

768
00:45:00,239 --> 00:45:04,480
Speaker 2: no longer just building AGI, we are now on our

769
00:45:04,520 --> 00:45:06,960
Speaker 2: path to building superintelligence. Like he was sort of like

770
00:45:07,040 --> 00:45:11,640
Speaker 2: upping the ante saying okay, could continue to hold on

771
00:45:12,120 --> 00:45:14,760
Speaker 2: could continue to stay with the program because we're about

772
00:45:14,800 --> 00:45:21,600
Speaker 2: to uh supercharge turbocharged this like ten times more. And

773
00:45:22,000 --> 00:45:24,160
Speaker 2: it was like at the time when Opening Eye was

774
00:45:24,960 --> 00:45:28,360
Speaker 2: starting to really feel weak because it had just lost

775
00:45:28,400 --> 00:45:31,720
Speaker 2: a string of executives, including some of its most important

776
00:45:31,760 --> 00:45:35,839
Speaker 2: ones Elias ask Govern and Miramurati, and it was under

777
00:45:36,000 --> 00:45:38,080
Speaker 2: just a massive amount of scrutiny and it wasn't making

778
00:45:38,120 --> 00:45:40,279
Speaker 2: the clip of research progress that it needed to kind

779
00:45:40,320 --> 00:45:45,080
Speaker 2: of solve what it itself defined as the key challenges

780
00:45:45,160 --> 00:45:49,400
Speaker 2: to reaching AGI and so yeah, so I think the

781
00:45:49,400 --> 00:45:53,440
Speaker 2: more that that that it sort of becomes clear that

782
00:45:53,520 --> 00:45:57,560
Speaker 2: people are no longer really buying into this AI future

783
00:45:57,600 --> 00:46:00,680
Speaker 2: that they've painted. The stronger they're going to painted, the

784
00:46:00,680 --> 00:46:02,800
Speaker 2: more they're going to roll out this rhetoric.

785
00:46:03,840 --> 00:46:06,000
Speaker 1: You mentioned this because there was a tweet from April

786
00:46:06,080 --> 00:46:09,200
Speaker 1: fifteenth where he said, the Open Ai team is executing

787
00:46:09,280 --> 00:46:11,520
Speaker 1: just ridiculously well at so many things right now. The

788
00:46:11,560 --> 00:46:13,400
Speaker 1: coming months and years should be amazing. So I'm going

789
00:46:13,440 --> 00:46:16,080
Speaker 1: to guess things were bad. Yeah, Yeah, I mean, like

790
00:46:16,200 --> 00:46:18,799
Speaker 1: cool read Sam's tweets.

791
00:46:19,120 --> 00:46:22,640
Speaker 2: Yeah, this was like a thing that I just consistently,

792
00:46:22,719 --> 00:46:25,560
Speaker 2: consistently every time I was reporting on things that were

793
00:46:25,560 --> 00:46:29,120
Speaker 2: going really bad, sure enough Altman would roll out some

794
00:46:29,200 --> 00:46:33,960
Speaker 2: kind of like really crazy, some really crazy yeah statement

795
00:46:34,040 --> 00:46:37,359
Speaker 2: out in public. So that was actually that tweet's actually

796
00:46:37,400 --> 00:46:40,960
Speaker 2: a perfect example because he says things will be awesome

797
00:46:40,960 --> 00:46:44,640
Speaker 2: in the coming months and years. It's always like hold on,

798
00:46:44,960 --> 00:46:48,440
Speaker 2: stick with me, Yeah, stick with me. Things might look

799
00:46:48,440 --> 00:46:51,279
Speaker 2: a little bit weird now, but oh boy, like just

800
00:46:51,400 --> 00:46:54,600
Speaker 2: you wait for what I'm seeing inside that like you

801
00:46:54,719 --> 00:46:57,759
Speaker 2: need to just have patience for you know. It's it's always.

802
00:46:57,400 --> 00:47:00,719
Speaker 1: That kind of May seventh picture, It's great to see

803
00:47:00,719 --> 00:47:02,960
Speaker 1: progress on the first Star Gay in Abilene with our

804
00:47:03,000 --> 00:47:04,880
Speaker 1: partners at Oracle. Today will be the biggest day I

805
00:47:05,000 --> 00:47:07,319
Speaker 1: training facility in the world. The scale, speed and skill

806
00:47:07,560 --> 00:47:09,839
Speaker 1: of the people building this is awesome. And then this

807
00:47:09,880 --> 00:47:13,279
Speaker 1: story comes out a week later, bloody hell. This does

808
00:47:13,960 --> 00:47:16,400
Speaker 1: so final question, how do you feel? What do you

809
00:47:16,440 --> 00:47:19,680
Speaker 1: think this fijisimo? Forgive me if I messed up the

810
00:47:19,760 --> 00:47:22,120
Speaker 1: name there. What do you think about her becoming CEO

811
00:47:22,200 --> 00:47:26,640
Speaker 1: of Applications and Sam Altman doing something else? Yeah?

812
00:47:26,680 --> 00:47:29,960
Speaker 2: So I haven't actually I haven't actually done reporting on

813
00:47:29,960 --> 00:47:34,120
Speaker 2: this myself, but my sense of what happening, what's happening

814
00:47:34,280 --> 00:47:37,880
Speaker 2: is Altmand's not a good manager. He's not actually like

815
00:47:37,960 --> 00:47:40,319
Speaker 2: he's a he's a fundraising CEO. He's not someone that

816
00:47:40,360 --> 00:47:44,920
Speaker 2: can run the company. And I think probably what happened

817
00:47:45,560 --> 00:47:49,520
Speaker 2: is that after Miramrati left, she was the one that

818
00:47:49,640 --> 00:47:51,960
Speaker 2: was actually doing the actually the day to day operations

819
00:47:52,000 --> 00:47:54,640
Speaker 2: and the running of the show. After she left, he

820
00:47:54,719 --> 00:47:57,480
Speaker 2: then made a big show of I'm going to be

821
00:47:57,640 --> 00:48:00,480
Speaker 2: much more close to the the work. Wow, I'm going

822
00:48:00,520 --> 00:48:03,840
Speaker 2: to do the day to day running. And probably his

823
00:48:03,960 --> 00:48:08,720
Speaker 2: time is up in doing that because what in my book,

824
00:48:08,719 --> 00:48:12,520
Speaker 2: I like talk a lot about how like he's not

825
00:48:12,640 --> 00:48:15,680
Speaker 2: good at that, like he is he will he's not

826
00:48:15,719 --> 00:48:18,560
Speaker 2: good at making decisions. He's very conflict averse. So what

827
00:48:18,640 --> 00:48:21,120
Speaker 2: he does is he'll just say he'll agree with every

828
00:48:21,120 --> 00:48:23,440
Speaker 2: single team even when they're disagreeing with one another. And

829
00:48:23,520 --> 00:48:27,200
Speaker 2: it causes chaos, and it causes riffs where you don't

830
00:48:27,239 --> 00:48:29,919
Speaker 2: the person at the top is not able to make

831
00:48:29,960 --> 00:48:31,759
Speaker 2: a decision and say we are all going to go

832
00:48:31,840 --> 00:48:33,200
Speaker 2: this way now and some of you are going to

833
00:48:33,200 --> 00:48:35,360
Speaker 2: be unhappy. Like he does not do that, and so

834
00:48:35,960 --> 00:48:38,440
Speaker 2: it just leads to a lot of tumult chaos. Part

835
00:48:38,520 --> 00:48:41,040
Speaker 2: of the reason why Opening Eye has had so many

836
00:48:41,080 --> 00:48:44,560
Speaker 2: product releases and features and things like that. I think

837
00:48:44,640 --> 00:48:47,719
Speaker 2: is actually also a product of this in that he

838
00:48:47,760 --> 00:48:50,560
Speaker 2: doesn't want to tell any team, like all of these

839
00:48:50,719 --> 00:48:53,600
Speaker 2: product releases and features are different teams working on these things,

840
00:48:53,880 --> 00:48:56,359
Speaker 2: and he doesn't want to tell any team like we're

841
00:48:56,360 --> 00:48:58,680
Speaker 2: going to have this person release first and have their

842
00:48:58,719 --> 00:49:00,919
Speaker 2: moment in the sun, and then we're gonna like work

843
00:49:00,960 --> 00:49:02,759
Speaker 2: a little bit more and then you get your moment

844
00:49:02,840 --> 00:49:05,000
Speaker 2: in the sun, you know, a year later. He's like

845
00:49:05,239 --> 00:49:07,560
Speaker 2: everyone gets their moment in the sun. Like we're gonna

846
00:49:07,600 --> 00:49:11,080
Speaker 2: do releases. We're gonna do like twelve days of ship Miss.

847
00:49:11,520 --> 00:49:15,520
Speaker 2: We're gonna just release. That was insane case in twelve

848
00:49:15,640 --> 00:49:17,240
Speaker 2: days twelve.

849
00:49:17,080 --> 00:49:20,160
Speaker 1: Days of ship Miss for the listeners that don't remember,

850
00:49:20,200 --> 00:49:21,839
Speaker 1: that was when they claimed they were going to release

851
00:49:21,880 --> 00:49:24,000
Speaker 1: twelve new twelve new products.

852
00:49:23,640 --> 00:49:25,200
Speaker 2: Twelve new products over the twelve.

853
00:49:25,000 --> 00:49:27,680
Speaker 1: And it wasn't twelve new. It was like four new products,

854
00:49:28,040 --> 00:49:30,040
Speaker 1: and like some of them were like an API for

855
00:49:30,080 --> 00:49:33,319
Speaker 1: an API. It's just so strange. It feels like, while

856
00:49:33,360 --> 00:49:38,439
Speaker 1: you're also describing an empire, you're also describing this kind

857
00:49:38,440 --> 00:49:43,000
Speaker 1: of very petty underpa. It really does mirror British colonialisten right,

858
00:49:43,840 --> 00:49:45,719
Speaker 1: You've got a guy who doesn't want to rule, who

859
00:49:45,800 --> 00:49:48,440
Speaker 1: wants the power of a ruler and all the assets,

860
00:49:48,680 --> 00:49:53,759
Speaker 1: but someone else, ideally in another country, should take responsibility. Yeah,

861
00:49:54,239 --> 00:49:55,120
Speaker 1: truly awful.

862
00:49:55,960 --> 00:49:58,080
Speaker 2: I mean this is a paradox of empire. Is like

863
00:49:58,360 --> 00:50:02,160
Speaker 2: it feels inevitable, well because it feels so strong and

864
00:50:02,360 --> 00:50:04,720
Speaker 2: it also feels so weak when you start to look

865
00:50:05,000 --> 00:50:06,040
Speaker 2: at an end of the surface.

866
00:50:07,440 --> 00:50:10,280
Speaker 1: It's it was a really great book, and I really

867
00:50:10,280 --> 00:50:12,279
Speaker 1: appreciate your time. Where can people find you?

868
00:50:13,440 --> 00:50:16,160
Speaker 2: I am on LinkedIn and blue sky these days, and

869
00:50:16,320 --> 00:50:20,719
Speaker 2: also on my website karindhow dot com and Yeah reach out.

870
00:50:20,880 --> 00:50:22,279
Speaker 2: I have a contact for him there and I try

871
00:50:22,280 --> 00:50:24,120
Speaker 2: to respond to as many people as possible.

872
00:50:24,800 --> 00:50:26,800
Speaker 1: Wonderful. Thank you so much for joining us. I'm of

873
00:50:26,880 --> 00:50:28,840
Speaker 1: course aid zechro on. You'll now get a thing I

874
00:50:28,880 --> 00:50:31,479
Speaker 1: recorded over a year ago that people still complain about about.

875
00:50:31,480 --> 00:50:42,920
Speaker 1: Where you can find stuff. Thank you for listening. Thank

876
00:50:43,000 --> 00:50:45,719
Speaker 1: you for listening to Better Offline. The editor and composer

877
00:50:45,760 --> 00:50:48,560
Speaker 1: of the Better Offline theme song is Matasowski. You can

878
00:50:48,640 --> 00:50:51,000
Speaker 1: check out more of his music and audio projects at

879
00:50:51,000 --> 00:50:54,480
Speaker 1: Matasowski dot com. M A T T O s O

880
00:50:55,160 --> 00:50:58,640
Speaker 1: W s KI dot com. You can email me an

881
00:50:58,640 --> 00:51:01,480
Speaker 1: easy at better offline dot com or visit better Offline

882
00:51:01,520 --> 00:51:03,600
Speaker 1: dot com to find more podcast links and of course

883
00:51:03,640 --> 00:51:06,759
Speaker 1: my newsletter. I also really recommend you go to chat

884
00:51:06,760 --> 00:51:09,440
Speaker 1: dot Where's youread dot at to visit the discord, and

885
00:51:09,440 --> 00:51:12,160
Speaker 1: go to our slash Better Offline to check out our reddit.

886
00:51:12,920 --> 00:51:14,280
Speaker 1: Thank you so much for listening.

887
00:51:15,120 --> 00:51:17,840
Speaker 3: Better Offline is a production of cool Zone Media. For

888
00:51:17,920 --> 00:51:21,080
Speaker 3: more from cool Zone Media, visit our website cool Zonemedia

889
00:51:21,160 --> 00:51:24,000
Speaker 3: dot com, or check us out on the iHeartRadio app,

890
00:51:24,040 --> 00:51:26,760
Speaker 3: Apple Podcasts, or wherever you get your podcasts.