1
00:00:09,840 --> 00:00:12,880
Speaker 1: Yo, what's going on? It's Texter. So technology does not

2
00:00:13,039 --> 00:00:15,920
Speaker 1: slow down for anybody. And sometimes I wish that I

3
00:00:15,920 --> 00:00:18,520
Speaker 1: could just talk to y'all live about what's happening right now.

4
00:00:19,160 --> 00:00:21,120
Speaker 1: But then we were talking about it, we realized, wait

5
00:00:21,120 --> 00:00:24,000
Speaker 1: a second, we can talk to you live. So we're

6
00:00:24,000 --> 00:00:26,319
Speaker 1: gonna try something new here. Every once in a while,

7
00:00:26,320 --> 00:00:28,639
Speaker 1: we're gonna talk about something that's happening right now and

8
00:00:28,720 --> 00:00:31,400
Speaker 1: let people jump into the audience to ask questions, and

9
00:00:31,440 --> 00:00:33,720
Speaker 1: we'll bring a guest along. And so what you're about

10
00:00:33,760 --> 00:00:36,240
Speaker 1: to listen to is our first episode that we live

11
00:00:36,280 --> 00:00:39,239
Speaker 1: streamed on Wednesday, March third. So if you want to

12
00:00:39,240 --> 00:00:41,440
Speaker 1: catch the next one as we're recording it and maybe

13
00:00:41,440 --> 00:00:44,199
Speaker 1: even ask some questions, you can subscribe to our YouTube

14
00:00:44,240 --> 00:00:46,280
Speaker 1: page and you can catch the links to that in

15
00:00:46,360 --> 00:00:57,200
Speaker 1: the description. So everyone, this is we're doing something different here.

16
00:00:57,240 --> 00:01:01,319
Speaker 1: We're doing the first live episode of kill Switch. We're

17
00:01:01,360 --> 00:01:03,720
Speaker 1: gonna probably try to do these maybe once a month,

18
00:01:04,200 --> 00:01:06,480
Speaker 1: just to give people opportunity to actually jump into the

19
00:01:06,560 --> 00:01:10,560
Speaker 1: chat and talk about stuff that is happening right now.

20
00:01:10,640 --> 00:01:13,360
Speaker 1: So thanks for bearing with us while we experiment a

21
00:01:13,400 --> 00:01:17,160
Speaker 1: little bit. But today what I'd like to talk about

22
00:01:17,520 --> 00:01:21,160
Speaker 1: is what is happening with Anthropic, what's going on with

23
00:01:21,200 --> 00:01:23,760
Speaker 1: open Ai, what's going on with the US government. To

24
00:01:23,800 --> 00:01:26,440
Speaker 1: talk about it today, our guest is Will Knight, who's

25
00:01:26,480 --> 00:01:29,360
Speaker 1: a senior writer at Wired. He's been writing about AI

26
00:01:29,480 --> 00:01:32,360
Speaker 1: for a while. He's got a weekly AI newsletter called

27
00:01:32,400 --> 00:01:35,040
Speaker 1: AI Lab. Will Welcome to kill Switch.

28
00:01:35,080 --> 00:01:37,280
Speaker 2: Welcome in, Thank you for having me. I'm wanting to

29
00:01:37,280 --> 00:01:40,120
Speaker 2: be on the first experimental live stream.

30
00:01:40,480 --> 00:01:43,679
Speaker 1: Yeah, hey, look, we'll see. Everything's in experiment at this point,

31
00:01:43,720 --> 00:01:48,080
Speaker 1: for better or for worse. Sure, So I want to

32
00:01:48,160 --> 00:01:51,520
Speaker 1: start this off with a timeline, Okay. On the twenty seventh,

33
00:01:52,080 --> 00:01:55,280
Speaker 1: Donald Trump gets on truth Social and says that Claude

34
00:01:55,320 --> 00:01:59,480
Speaker 1: is a radical left woke company. Okay, yeah, out of nowhere,

35
00:02:00,040 --> 00:02:02,720
Speaker 1: People like, what why is he saying this? Pete heg

36
00:02:02,800 --> 00:02:06,840
Speaker 1: Saith gets on Twitter and says that Claude Anthropic, the

37
00:02:06,840 --> 00:02:10,200
Speaker 1: company that makes Claude, is a supply chain risk, which

38
00:02:10,200 --> 00:02:12,320
Speaker 1: most people have never thought about what that even means.

39
00:02:13,000 --> 00:02:17,040
Speaker 1: The next day, Open Ai, who makes chat GPT, announces

40
00:02:17,080 --> 00:02:19,680
Speaker 1: that they're working with the Department of War, and then

41
00:02:19,800 --> 00:02:23,880
Speaker 1: Donald Trump announces that we're bombing Iran. As I say

42
00:02:23,919 --> 00:02:27,480
Speaker 1: that that sounds like a completely unhinged sequence of events.

43
00:02:28,360 --> 00:02:30,880
Speaker 1: Please tell me where I'm wrong, because I must be

44
00:02:30,919 --> 00:02:31,640
Speaker 1: missing something.

45
00:02:32,120 --> 00:02:34,320
Speaker 2: I mean, I think I think there's a lot that's

46
00:02:34,960 --> 00:02:37,840
Speaker 2: that's right in there and a lot that's unhinged. But

47
00:02:38,639 --> 00:02:40,639
Speaker 2: I mean to sort of set the scene a little bit,

48
00:02:41,040 --> 00:02:43,920
Speaker 2: you know. So this goes back. So the use of

49
00:02:44,240 --> 00:02:49,440
Speaker 2: Claude in classified military systems has its origins last year,

50
00:02:49,520 --> 00:02:52,799
Speaker 2: last summer, Anthropic was the first to get involved in

51
00:02:53,160 --> 00:02:55,480
Speaker 2: working with the military. You know, they've pursued a very

52
00:02:55,880 --> 00:02:58,519
Speaker 2: business focus, and I think that sort of led them

53
00:02:58,520 --> 00:03:00,440
Speaker 2: to focus on the d D, which is like one

54
00:03:00,480 --> 00:03:02,840
Speaker 2: of the biggest businesses in the US. So they ended

55
00:03:02,919 --> 00:03:05,880
Speaker 2: up supplying Claude their model, a version of that which

56
00:03:05,960 --> 00:03:11,400
Speaker 2: runs on the Pentagon's own systems for military use on

57
00:03:11,520 --> 00:03:15,480
Speaker 2: classified systems. In many ways, this is totally remarkable because

58
00:03:16,160 --> 00:03:18,320
Speaker 2: just a few years ago, everybody in the AI industry

59
00:03:18,440 --> 00:03:20,480
Speaker 2: was saying, one thing we don't want to do is

60
00:03:20,520 --> 00:03:22,800
Speaker 2: work with the military, and people were staging, you know,

61
00:03:22,880 --> 00:03:26,120
Speaker 2: walkouts over this, right, So it shows you how the

62
00:03:26,160 --> 00:03:29,000
Speaker 2: tech world has really pivoted to that. What happened more

63
00:03:29,040 --> 00:03:33,800
Speaker 2: recently was that its themes and you know, this is

64
00:03:33,840 --> 00:03:38,360
Speaker 2: somewhat disputed but there's been reports that when President Maduro

65
00:03:38,520 --> 00:03:42,280
Speaker 2: was captured, Claude was used, and somebody at Anthropic may

66
00:03:42,320 --> 00:03:46,520
Speaker 2: have raised questions about that. This seems like it snowballed.

67
00:03:46,840 --> 00:03:49,760
Speaker 2: And one of the carve outs in the original contract

68
00:03:49,840 --> 00:03:53,680
Speaker 2: signed last year there were two. One was against mass

69
00:03:53,680 --> 00:03:56,800
Speaker 2: surveillance of US citizen and the others was that the

70
00:03:56,840 --> 00:04:00,160
Speaker 2: model shouldn't be used to build fully autonomous weapons. And

71
00:04:01,280 --> 00:04:06,200
Speaker 2: the DoD or W said that they wanted to change

72
00:04:06,240 --> 00:04:08,520
Speaker 2: the contract essentially, which is an unusual thing to do,

73
00:04:08,760 --> 00:04:11,760
Speaker 2: but they wanted to allow all your lawful use. And

74
00:04:12,080 --> 00:04:15,200
Speaker 2: to some degree, this dispute seems like it's a little

75
00:04:15,200 --> 00:04:17,520
Speaker 2: bit like people, you know, each side talking past each

76
00:04:17,520 --> 00:04:19,920
Speaker 2: other to some degree. And I think that the DOW

77
00:04:20,080 --> 00:04:24,160
Speaker 2: the Department of War became very concerned about a tech

78
00:04:24,200 --> 00:04:27,200
Speaker 2: company saying how it could and couldn't use the technology.

79
00:04:27,400 --> 00:04:30,360
Speaker 2: And to be fair, that is quite an unusual idea, right.

80
00:04:30,400 --> 00:04:33,680
Speaker 2: It's like somebody selling a missile or a jet fighter

81
00:04:33,720 --> 00:04:35,760
Speaker 2: to the Pentagon and then later on saying we don't

82
00:04:35,760 --> 00:04:37,320
Speaker 2: want you to use it in this way or that way.

83
00:04:38,080 --> 00:04:40,680
Speaker 2: But then again, ai is is a really different technology,

84
00:04:40,760 --> 00:04:43,760
Speaker 2: is very new, it's very experimental and I believe anthropic,

85
00:04:44,200 --> 00:04:46,719
Speaker 2: especially among AI companies, it's very focused on safety and

86
00:04:46,720 --> 00:04:49,920
Speaker 2: they have concerns just about how it could end up

87
00:04:49,960 --> 00:04:50,640
Speaker 2: being used.

88
00:04:50,960 --> 00:04:54,240
Speaker 1: Yeah, well, so let's jump into this so anthropic you know,

89
00:04:54,640 --> 00:04:57,680
Speaker 1: this is a statement that is put on anthropic dot com.

90
00:04:57,760 --> 00:05:00,400
Speaker 1: So this is their own, you know, pr statement here

91
00:05:00,920 --> 00:05:03,000
Speaker 1: and the title is statement from Daria Ama Day on

92
00:05:03,040 --> 00:05:06,320
Speaker 1: our discussions with the Department of War, and it starts

93
00:05:06,360 --> 00:05:09,720
Speaker 1: out the first line, I believe deeply in the existential

94
00:05:09,720 --> 00:05:12,800
Speaker 1: importance of using AI to defend the United States and

95
00:05:12,839 --> 00:05:17,840
Speaker 1: other democracies and to defeat our autocratic adversaries. So already

96
00:05:17,839 --> 00:05:20,240
Speaker 1: out of the gate, that's kind of an interesting thing

97
00:05:20,440 --> 00:05:23,320
Speaker 1: for again a company in Silicon Valley to be saying,

98
00:05:23,480 --> 00:05:27,560
Speaker 1: even just in the past couple of years, really, but

99
00:05:27,680 --> 00:05:30,360
Speaker 1: it seems like the two big points of contention here,

100
00:05:30,440 --> 00:05:34,600
Speaker 1: there's two. He says, using these systems for mass domestic

101
00:05:34,800 --> 00:05:39,560
Speaker 1: surveillance is incompatible with democratic values. So spying out other people,

102
00:05:39,880 --> 00:05:42,640
Speaker 1: we're okay with that. Spying on people in the US,

103
00:05:42,960 --> 00:05:45,320
Speaker 1: We're not okay with that. Right, And then the next

104
00:05:45,320 --> 00:05:49,400
Speaker 1: thing is, quote frontier AI systems are simply not reliable

105
00:05:49,520 --> 00:05:54,400
Speaker 1: enough to power fully autonomous weapons both of those sound

106
00:05:54,760 --> 00:05:59,239
Speaker 1: totally okay and reasonable. Not only do they sound okay

107
00:05:59,279 --> 00:06:02,120
Speaker 1: and reasonable, but I mean, look, I've spent a whole

108
00:06:02,120 --> 00:06:05,600
Speaker 1: lot of time around people on the pretty far rate

109
00:06:06,680 --> 00:06:09,200
Speaker 1: and you know, for work, right, and they will tell

110
00:06:09,240 --> 00:06:12,919
Speaker 1: you I don't want no government spying on me, and

111
00:06:13,000 --> 00:06:16,680
Speaker 1: I don't trust these computers. Right, where's the beef? What's

112
00:06:16,720 --> 00:06:17,520
Speaker 1: the problem here?

113
00:06:18,200 --> 00:06:20,640
Speaker 2: That's a great question. You know. I've spent a lot

114
00:06:20,680 --> 00:06:24,919
Speaker 2: of time talking to people in the Pentagon and military

115
00:06:25,520 --> 00:06:29,520
Speaker 2: defense companies, and nobody believes that those frontier models are

116
00:06:29,520 --> 00:06:32,679
Speaker 2: ready were even remotely near ready being used in autonomous weapons,

117
00:06:33,360 --> 00:06:35,760
Speaker 2: And consistently the government said we have no desire to

118
00:06:35,839 --> 00:06:39,160
Speaker 2: do either of those things. I think the way what

119
00:06:39,240 --> 00:06:42,200
Speaker 2: it really came down to was that they didn't just

120
00:06:43,120 --> 00:06:46,280
Speaker 2: you know, take a knee and agree to change the

121
00:06:46,360 --> 00:06:48,960
Speaker 2: terms of their contract. And as you say, you know,

122
00:06:49,720 --> 00:06:54,120
Speaker 2: it culminated in Trump and Hegxeth and others putting out

123
00:06:54,120 --> 00:06:58,080
Speaker 2: these very very you know, as often happens, they were

124
00:06:58,080 --> 00:06:59,680
Speaker 2: sort of classifying them all of a sudden as the

125
00:06:59,720 --> 00:07:03,359
Speaker 2: radical left, which is it's pretty incredible for company supplying

126
00:07:03,360 --> 00:07:06,400
Speaker 2: technology that's working on classified US systems.

127
00:07:06,520 --> 00:07:10,080
Speaker 1: Right, So this is the post that Donald Trump made

128
00:07:10,120 --> 00:07:13,680
Speaker 1: on his social media network Truth Social and this I'm

129
00:07:13,720 --> 00:07:16,600
Speaker 1: reading from the top, and I'm reading verbatim. The United

130
00:07:16,600 --> 00:07:20,120
Speaker 1: States of America will never allow a radical left woke

131
00:07:20,280 --> 00:07:25,280
Speaker 1: company to dictate how a great military fights and wins wars.

132
00:07:25,480 --> 00:07:28,120
Speaker 1: That decision belongs to your commander in chief and the

133
00:07:28,120 --> 00:07:31,400
Speaker 1: tremendous leaders I appoint to run are military. And then

134
00:07:31,480 --> 00:07:33,200
Speaker 1: he finally gets to what he's talking about. He says,

135
00:07:33,200 --> 00:07:37,760
Speaker 1: the left wing nut jobs Anthropic have made a disastrous

136
00:07:37,800 --> 00:07:41,160
Speaker 1: mistake trying to strong arm the Department of War and

137
00:07:41,200 --> 00:07:43,520
Speaker 1: force them to obey their terms of service instead of

138
00:07:43,520 --> 00:07:44,320
Speaker 1: our constitution.

139
00:07:44,760 --> 00:07:45,320
Speaker 3: Yeah.

140
00:07:45,400 --> 00:07:47,400
Speaker 2: I mean, like a lot of that doesn't really even

141
00:07:47,920 --> 00:07:51,120
Speaker 2: make sense. I think Anthropic is not a radical left

142
00:07:51,720 --> 00:07:56,160
Speaker 2: organization by a stretch. It's since come out that that

143
00:07:56,200 --> 00:07:59,240
Speaker 2: should The DoD was negotiating with that for they were

144
00:07:59,240 --> 00:08:01,880
Speaker 2: trying to get this war odding kind of massalage into

145
00:08:01,880 --> 00:08:04,040
Speaker 2: a format that everybody was happy with. And it doesn't

146
00:08:04,080 --> 00:08:06,840
Speaker 2: seem like that would have been impossible. It feels like

147
00:08:07,560 --> 00:08:10,880
Speaker 2: some people within the administration lost patients, And I think

148
00:08:10,880 --> 00:08:12,960
Speaker 2: a lot of it came down to just power and

149
00:08:14,360 --> 00:08:18,560
Speaker 2: the Department of Defense and the leadership not being able

150
00:08:18,640 --> 00:08:20,960
Speaker 2: to handle being denied that by a tech company that

151
00:08:21,000 --> 00:08:23,720
Speaker 2: they could very easily see as as you know, part

152
00:08:23,760 --> 00:08:26,800
Speaker 2: of this sort of you know, coastal elite and in

153
00:08:26,840 --> 00:08:30,160
Speaker 2: the case of Anthropic, woke and liberal. I mean, it's

154
00:08:30,200 --> 00:08:32,880
Speaker 2: another reflection of how AI is just is it sort

155
00:08:32,920 --> 00:08:36,640
Speaker 2: of also a cultural lightning rod, right, it becomes suddenly

156
00:08:36,640 --> 00:08:41,559
Speaker 2: it becomes this kind of left versus right. So, I mean,

157
00:08:41,760 --> 00:08:46,199
Speaker 2: you know that that post kind of does unfortunately sum

158
00:08:46,280 --> 00:08:49,480
Speaker 2: up the entire situation, which is slightly bewildering if you

159
00:08:49,920 --> 00:08:52,800
Speaker 2: know that now they've they've taken the action to try

160
00:08:52,800 --> 00:08:56,440
Speaker 2: and to classify, as you said, Anthropic a supply chain risk,

161
00:08:56,880 --> 00:08:59,040
Speaker 2: which is something that's only ever been to foreign nations.

162
00:08:59,040 --> 00:09:02,319
Speaker 2: And we're doing this to one of the foremost AI

163
00:09:02,440 --> 00:09:05,280
Speaker 2: companies in the United States, one of them I supposedly,

164
00:09:05,360 --> 00:09:09,200
Speaker 2: you know, should be our national champions. It it doesn't

165
00:09:09,240 --> 00:09:13,560
Speaker 2: sound like to me they're really promoting American technology. If

166
00:09:13,559 --> 00:09:15,319
Speaker 2: you if you're too keen to do that, I think

167
00:09:15,360 --> 00:09:16,920
Speaker 2: you don't. You know, it's a real danger you're going

168
00:09:17,000 --> 00:09:21,160
Speaker 2: to harm some of you know, this this nascent, very

169
00:09:21,160 --> 00:09:24,439
Speaker 2: important American company also maybe dissuade other companies from working

170
00:09:24,480 --> 00:09:26,880
Speaker 2: with the government or the Pentagon if they feel that

171
00:09:26,880 --> 00:09:27,839
Speaker 2: that's going to happen.

172
00:09:27,640 --> 00:09:32,480
Speaker 1: Right, Yeah, So the supply chain risk thing again that

173
00:09:32,480 --> 00:09:35,720
Speaker 1: that's the language that Pete Hegseth used. What does it

174
00:09:35,800 --> 00:09:40,319
Speaker 1: mean to call anthropic a supply chain risk? What does

175
00:09:40,360 --> 00:09:41,960
Speaker 1: that even mean? Yeah?

176
00:09:42,000 --> 00:09:43,599
Speaker 2: I mean, you know, well, the first thing is to

177
00:09:43,640 --> 00:09:49,760
Speaker 2: say that this supply chain risk designation is incredibly unusual, extraordinary,

178
00:09:49,880 --> 00:09:53,040
Speaker 2: and just the kind of thing that I think the

179
00:09:53,080 --> 00:09:56,480
Speaker 2: sort of left field thinking that we're become accustomed to

180
00:09:56,600 --> 00:09:59,440
Speaker 2: with the administration because it's usually applied to, you know,

181
00:10:00,040 --> 00:10:02,400
Speaker 2: a company that you want to remove from the US

182
00:10:02,520 --> 00:10:05,439
Speaker 2: supply chain because you're worried that their software or their

183
00:10:05,440 --> 00:10:09,800
Speaker 2: hardware might contain backdoors, might be a threat to national

184
00:10:09,800 --> 00:10:13,000
Speaker 2: security because it could be used by an adversary. So

185
00:10:13,120 --> 00:10:14,839
Speaker 2: this is the very sort of all or nothing. You

186
00:10:15,280 --> 00:10:18,160
Speaker 2: either agree with everything we say, or you are the enemy.

187
00:10:18,920 --> 00:10:21,880
Speaker 2: I mean, that's the kind of tenor which sounds familiar

188
00:10:21,880 --> 00:10:26,000
Speaker 2: to me generally. Often the most extreme version of that designation,

189
00:10:26,120 --> 00:10:27,760
Speaker 2: and we haven't seen the full details of it because

190
00:10:27,760 --> 00:10:30,000
Speaker 2: they haven't produced it yet, but would mean that any

191
00:10:30,360 --> 00:10:33,240
Speaker 2: company who's working with the d id couldn't use your technology,

192
00:10:33,240 --> 00:10:36,280
Speaker 2: which is a vast I mean, Amazon, which has invested

193
00:10:36,320 --> 00:10:39,920
Speaker 2: billions in Anthropy, has massive contracts with the DoD so

194
00:10:39,960 --> 00:10:43,000
Speaker 2: it's an enormous thing to designate them. It's also doesn't

195
00:10:43,000 --> 00:10:45,400
Speaker 2: really make sense because you know, they were also saying

196
00:10:45,440 --> 00:10:47,880
Speaker 2: we may invoke the Defense Production Act, which has been

197
00:10:48,000 --> 00:10:51,000
Speaker 2: used in wartimes or in actually also during the Cold

198
00:10:51,080 --> 00:10:56,559
Speaker 2: War to force companies to prioritize production and supply because

199
00:10:56,760 --> 00:11:01,240
Speaker 2: what they were making is so important. So it's it's wait,

200
00:11:01,600 --> 00:11:03,400
Speaker 2: I think this may be what, Yeah, this may be

201
00:11:03,400 --> 00:11:06,480
Speaker 2: a legal problem that the d ID has, and I

202
00:11:06,520 --> 00:11:09,160
Speaker 2: think you know Anthropic is going to as a signal

203
00:11:09,200 --> 00:11:11,760
Speaker 2: that's going to take some legal action, So you can't

204
00:11:11,800 --> 00:11:14,800
Speaker 2: it's difficult to argue. On one hand, it's so vital

205
00:11:14,840 --> 00:11:17,520
Speaker 2: we may invoke the Defense Production Act, but then we

206
00:11:17,559 --> 00:11:19,880
Speaker 2: may also designate you a supply chain risk.

207
00:11:20,400 --> 00:11:24,600
Speaker 1: Yeah, like you're too dangerous for us to use, but

208
00:11:25,200 --> 00:11:28,840
Speaker 1: you're so useful that we want to use your product.

209
00:11:28,559 --> 00:11:31,640
Speaker 2: But we want to force you to supply it big one.

210
00:11:32,280 --> 00:11:35,600
Speaker 1: Yeah, like nobody can have it, but we want it. Okay,

211
00:11:35,880 --> 00:11:49,480
Speaker 1: what are we doing here? So basically the next day

212
00:11:50,480 --> 00:11:54,640
Speaker 1: open Ai signs something with the government. How does that

213
00:11:54,720 --> 00:11:57,120
Speaker 1: sequence of events? Like, what kind of sense are you

214
00:11:57,160 --> 00:11:57,840
Speaker 1: able to make of that?

215
00:11:58,160 --> 00:12:02,720
Speaker 2: Yeah, it is very interesting that anthropics chief rival. And

216
00:12:03,000 --> 00:12:06,040
Speaker 2: quick reminder that Anthropic was founded by people who left

217
00:12:06,400 --> 00:12:10,079
Speaker 2: OpenAI because they said that company was not developing AI

218
00:12:10,559 --> 00:12:15,600
Speaker 2: with enough focus on safety. And they're really really bitter rivals.

219
00:12:15,640 --> 00:12:18,360
Speaker 2: And as it was going on, Sam al Mun was

220
00:12:18,400 --> 00:12:21,960
Speaker 2: working through an alternative deal. The deal that they signed

221
00:12:22,000 --> 00:12:25,760
Speaker 2: seems extremely similar to the one that Anthropic we're trying

222
00:12:25,760 --> 00:12:28,520
Speaker 2: to hash out, has similar carve outs, so it's sort

223
00:12:28,520 --> 00:12:31,360
Speaker 2: of doesn't actually carve them out, but it specifies that

224
00:12:31,360 --> 00:12:36,000
Speaker 2: those things that massurbillance is unlawful and that nobody wants

225
00:12:36,040 --> 00:12:37,240
Speaker 2: to build autonomous weapons.

226
00:12:37,559 --> 00:12:41,880
Speaker 1: So are the details or the terms of the contract

227
00:12:41,880 --> 00:12:45,880
Speaker 1: that OpenAI signed with the Department of War, are they

228
00:12:45,920 --> 00:12:51,199
Speaker 1: materially different from what Anthropic did? Because this is from CNBC,

229
00:12:51,679 --> 00:12:55,040
Speaker 1: Altman said, and I want to read this here quote.

230
00:12:55,120 --> 00:12:57,959
Speaker 1: I believe that we will hopefully have the best models

231
00:12:57,960 --> 00:13:00,440
Speaker 1: that were encouraged the government to be willing to work

232
00:13:00,480 --> 00:13:03,679
Speaker 1: with us, even if our safety stack annoys them. But

233
00:13:03,679 --> 00:13:05,800
Speaker 1: there will be at least one other actor, which I

234
00:13:05,840 --> 00:13:09,520
Speaker 1: assume will be x Ai, which effectively will say, will

235
00:13:09,559 --> 00:13:14,040
Speaker 1: do whatever you want. So the vibe here seems to

236
00:13:14,080 --> 00:13:17,280
Speaker 1: be like, okay, well listen, at least worthy adults in

237
00:13:17,280 --> 00:13:19,880
Speaker 1: the room. They're gonna let Grop do whatever Groc wants

238
00:13:19,880 --> 00:13:22,280
Speaker 1: to do, whatever, whatever Trump wants to do. They're just

239
00:13:22,320 --> 00:13:24,719
Speaker 1: gonna let it do it, drop bombs wherever. At least

240
00:13:24,760 --> 00:13:26,240
Speaker 1: we're going to be some adults in the room. But

241
00:13:26,960 --> 00:13:30,360
Speaker 1: is Opening Eye doing something fundamentally different. Did they seed

242
00:13:30,400 --> 00:13:33,480
Speaker 1: more grounded than Anthropic did or is it the same thing?

243
00:13:35,280 --> 00:13:36,920
Speaker 2: You know, Yeah, there's a lot going on there. I

244
00:13:37,200 --> 00:13:41,120
Speaker 2: think that they they did see slightly more ground in

245
00:13:41,160 --> 00:13:45,200
Speaker 2: that they basically agreed to a contract that specified all

246
00:13:45,280 --> 00:13:49,560
Speaker 2: lawful use. They changed the contractors make Oh but you know,

247
00:13:49,640 --> 00:13:51,800
Speaker 2: to be clear, here's what's what is lawful and what

248
00:13:51,840 --> 00:13:54,880
Speaker 2: the deity doesn't want to do. But I think there

249
00:13:54,880 --> 00:13:56,760
Speaker 2: isn't a huge amount of difference. I mean, I think

250
00:13:56,800 --> 00:14:00,000
Speaker 2: what you're seeing with that that internal statement was really,

251
00:14:01,200 --> 00:14:04,600
Speaker 2: you know, Sam open Ai willing to take a bit

252
00:14:04,640 --> 00:14:09,720
Speaker 2: of flat because it's just incredibly important to get a

253
00:14:09,720 --> 00:14:11,920
Speaker 2: win over Anthropic, to be seen to be very on

254
00:14:12,000 --> 00:14:14,040
Speaker 2: board with the US government because you want more and

255
00:14:14,120 --> 00:14:17,840
Speaker 2: more involvement, more and more contracts and they've since you know,

256
00:14:17,880 --> 00:14:20,000
Speaker 2: open a I since got quite a lot of blowback.

257
00:14:20,640 --> 00:14:22,920
Speaker 1: Yeah, well, let's talk about the blowback here. I mean

258
00:14:23,120 --> 00:14:27,360
Speaker 1: this is on tech crunch us app. Uninstalls of chat

259
00:14:27,400 --> 00:14:31,440
Speaker 1: gbt's mobile app went up almost three hundred per so

260
00:14:31,600 --> 00:14:40,640
Speaker 1: people were uninstalling chat gbt and started installing Claude Anthropics Claude.

261
00:14:40,720 --> 00:14:44,120
Speaker 1: And so there's a weird way in which there's this

262
00:14:45,040 --> 00:14:47,280
Speaker 1: and I really hesitate to call this a game, but

263
00:14:47,280 --> 00:14:49,800
Speaker 1: there's there's a weird, really weird way in which there's

264
00:14:49,840 --> 00:14:53,520
Speaker 1: this game being played over who gets to be the

265
00:14:53,600 --> 00:14:57,160
Speaker 1: contractor for the Department of War. But then also we're

266
00:14:57,240 --> 00:14:59,720
Speaker 1: still fighting for consumers here.

267
00:15:00,720 --> 00:15:03,800
Speaker 2: Definitely, And it is interesting. It's the sort of perception

268
00:15:03,840 --> 00:15:06,120
Speaker 2: of who's the good guy and you had, you know,

269
00:15:06,200 --> 00:15:10,920
Speaker 2: Katie Perry was subscribing to claud pro and posting about

270
00:15:10,920 --> 00:15:14,120
Speaker 2: it and this, you know, the sort of interesting cultural moment.

271
00:15:14,160 --> 00:15:17,040
Speaker 2: And I think the thing that was that struck me

272
00:15:17,160 --> 00:15:20,560
Speaker 2: was like, the people really know how Claude is being

273
00:15:20,640 --> 00:15:23,640
Speaker 2: used by the military, Do they know what Anthropic was

274
00:15:23,680 --> 00:15:26,920
Speaker 2: willing to do? I think it's gotten a little oversimplified

275
00:15:26,960 --> 00:15:29,040
Speaker 2: to my mind, and I think that one of the

276
00:15:29,160 --> 00:15:32,680
Speaker 2: risks there is that we ought to be rather necessarily

277
00:15:32,760 --> 00:15:36,320
Speaker 2: uninstalling something or posting about who you think is best,

278
00:15:37,120 --> 00:15:39,080
Speaker 2: really pushing for more of an explanation about how the

279
00:15:39,160 --> 00:15:42,400
Speaker 2: technology is being used and developed and tested, like what

280
00:15:42,520 --> 00:15:45,240
Speaker 2: is it mean that it's used on classified systems? There

281
00:15:45,240 --> 00:15:48,200
Speaker 2: should be more accountability there. And the fact of the

282
00:15:48,200 --> 00:15:51,160
Speaker 2: matter is there is nothing legally to putting the issue

283
00:15:51,160 --> 00:15:53,440
Speaker 2: of the bonus to once nothing prohibiting the development of

284
00:15:53,480 --> 00:15:57,640
Speaker 2: autonomous weapons, and there isn't really a legal framework around

285
00:15:57,720 --> 00:16:00,520
Speaker 2: that sort. If they, you know, civilians were affected by

286
00:16:00,520 --> 00:16:03,600
Speaker 2: fully autonomous weapons, you suddenly have a situation where nobody

287
00:16:03,680 --> 00:16:06,640
Speaker 2: is legally accountable. And so these are things that really

288
00:16:06,640 --> 00:16:09,560
Speaker 2: need to be figured out. And that doesn't mean a

289
00:16:09,600 --> 00:16:12,720
Speaker 2: fully autonomous web that could be your automated part of

290
00:16:12,760 --> 00:16:15,280
Speaker 2: the so called kill chain and so decisions were kind

291
00:16:15,280 --> 00:16:17,960
Speaker 2: of automated away. I think that is really important to

292
00:16:18,000 --> 00:16:20,800
Speaker 2: have more focus on this and to have more discussion

293
00:16:20,840 --> 00:16:24,120
Speaker 2: of it. And you know, one of the reasons, honestly,

294
00:16:24,440 --> 00:16:27,680
Speaker 2: as someone who's looked at the DoD and the work

295
00:16:27,680 --> 00:16:31,080
Speaker 2: going on around autonomy, is that unfortunately it is the

296
00:16:31,120 --> 00:16:33,040
Speaker 2: case that we're just going to have more and more autonomy,

297
00:16:33,400 --> 00:16:36,160
Speaker 2: and most analysts expect that they're going to be swarms

298
00:16:36,200 --> 00:16:39,440
Speaker 2: of drones attacking systems, and the truth is you just

299
00:16:39,480 --> 00:16:41,600
Speaker 2: won't be able to have humans in the loop always.

300
00:16:41,760 --> 00:16:43,520
Speaker 2: So this is going to become a really big issue.

301
00:16:43,520 --> 00:16:47,080
Speaker 2: And it's not necessarily about anthropics models in those situations,

302
00:16:47,080 --> 00:16:50,120
Speaker 2: but the question of who's accountable, how do we know

303
00:16:50,160 --> 00:16:52,600
Speaker 2: that the systems are reliable? That is going to become

304
00:16:52,680 --> 00:16:54,200
Speaker 2: more and more important. So I think it's good that

305
00:16:54,200 --> 00:16:57,000
Speaker 2: the situation is getting some attention. I do think that

306
00:16:57,040 --> 00:17:00,320
Speaker 2: the positive thing is that it is important for people

307
00:17:00,320 --> 00:17:03,520
Speaker 2: who understand this technology really really well. Do you maybe

308
00:17:03,560 --> 00:17:06,680
Speaker 2: have a role in at least influencing how it's used.

309
00:17:06,840 --> 00:17:08,560
Speaker 2: And that doesn't mean, you know, like you get to

310
00:17:08,600 --> 00:17:10,920
Speaker 2: say yes or no to a mission. It just means

311
00:17:10,920 --> 00:17:15,119
Speaker 2: perhaps that you say, have some input on how you

312
00:17:15,160 --> 00:17:17,640
Speaker 2: ought to try and deploy it within the Pentagon I.

313
00:17:17,600 --> 00:17:20,720
Speaker 1: Think, I mean, how does that functionally work? Though maybe

314
00:17:20,720 --> 00:17:23,240
Speaker 1: that's the question here, That is the question, Yeah, can

315
00:17:23,280 --> 00:17:28,240
Speaker 1: you have a contract with the government and tell them no,

316
00:17:28,400 --> 00:17:31,320
Speaker 1: you can't do this? And I'm asking that partially because

317
00:17:31,720 --> 00:17:34,320
Speaker 1: certainly there is a lot of talk that will wait

318
00:17:34,320 --> 00:17:37,520
Speaker 1: a second, The iron Strake. Was Claude used there.

319
00:17:38,000 --> 00:17:41,119
Speaker 2: Yeah. To answer the first question, I think that is

320
00:17:41,200 --> 00:17:43,919
Speaker 2: a really good question. How do companies, how does the

321
00:17:43,920 --> 00:17:47,719
Speaker 2: public understand how things are developed, what the rules are

322
00:17:47,720 --> 00:17:51,920
Speaker 2: around them. And there are specified rules around autonomous weapons,

323
00:17:51,960 --> 00:17:55,439
Speaker 2: around surveillance, but this is also a technology that's different,

324
00:17:55,840 --> 00:17:58,600
Speaker 2: that can misbehave in surprising ways and can be kind

325
00:17:58,640 --> 00:18:01,080
Speaker 2: vulnerable in surprising ways. So I mean that it needs

326
00:18:01,119 --> 00:18:05,000
Speaker 2: a bit more scrutiny. The Washington Posters has reported. I

327
00:18:05,000 --> 00:18:08,440
Speaker 2: think it's that it seems very likely true that Claude

328
00:18:08,520 --> 00:18:12,080
Speaker 2: was absolutely used in the Iran operation as part of

329
00:18:12,080 --> 00:18:14,960
Speaker 2: this system called Maize and which was built by Palanteer,

330
00:18:15,080 --> 00:18:18,160
Speaker 2: which includes a bunch of things, not just a language model.

331
00:18:18,200 --> 00:18:21,119
Speaker 2: It can do you know, analyzed maps and images and

332
00:18:21,760 --> 00:18:24,679
Speaker 2: mission plan and so on. But yeah, as part of that.

333
00:18:24,760 --> 00:18:26,840
Speaker 2: And so now, one of the things that was kind

334
00:18:26,840 --> 00:18:29,280
Speaker 2: of crazy with Trump's initial statement was that he said,

335
00:18:29,720 --> 00:18:33,560
Speaker 2: I'm ordering everybody to stop working with Anthropic because they're

336
00:18:33,760 --> 00:18:36,000
Speaker 2: all these terrible people. But there'll be a six month

337
00:18:36,040 --> 00:18:38,560
Speaker 2: period of what we carry on doing. Right, it can't

338
00:18:38,600 --> 00:18:41,960
Speaker 2: be that bad, and we're going to use We're about

339
00:18:42,000 --> 00:18:45,040
Speaker 2: to use them in this you know, incredible operation in

340
00:18:45,800 --> 00:18:46,520
Speaker 2: the Middle East.

341
00:18:47,000 --> 00:18:51,479
Speaker 1: So yeah, I mean we're laughing. It's terrible. I mean

342
00:18:51,520 --> 00:18:54,760
Speaker 1: it's ridiculous and also terrible. I suppose it can be

343
00:18:54,760 --> 00:18:59,080
Speaker 1: two things at the same time. And because we're doing

344
00:18:59,119 --> 00:19:02,520
Speaker 1: this live, I wanted to take a few questions. And

345
00:19:02,840 --> 00:19:06,040
Speaker 1: one of the questions that I got actually previously I

346
00:19:06,080 --> 00:19:09,399
Speaker 1: posted on Instagram does anybody have any questions? And somebody posted,

347
00:19:09,480 --> 00:19:12,160
Speaker 1: and I think they were kind of joking, but they said, hey,

348
00:19:12,160 --> 00:19:15,520
Speaker 1: can we just make all the country's Ais fight against

349
00:19:15,520 --> 00:19:17,600
Speaker 1: each other? Can we just have them fight instead of

350
00:19:17,640 --> 00:19:21,080
Speaker 1: the real war. But it's kind of funny that they

351
00:19:21,119 --> 00:19:24,200
Speaker 1: say that because there's I don't know if you've seen

352
00:19:24,200 --> 00:19:29,960
Speaker 1: this article in the New Scientist, but basically different AIS,

353
00:19:30,000 --> 00:19:32,960
Speaker 1: and the headline here kind of says it, all AIS

354
00:19:33,040 --> 00:19:37,359
Speaker 1: can't stop recommending nuclear strikes in wargames simulations, and the

355
00:19:37,400 --> 00:19:42,320
Speaker 1: subhead leading AIS from Open Ai, Anthropic and Google opted

356
00:19:42,359 --> 00:19:46,040
Speaker 1: to use nuclear weapons in simulated wargames in ninety five

357
00:19:46,080 --> 00:19:47,440
Speaker 1: percent of cases.

358
00:19:48,520 --> 00:19:52,000
Speaker 2: Right, Yeah, So I read that paper. It's important to

359
00:19:52,080 --> 00:19:55,840
Speaker 2: know that this person was just trying to understand what

360
00:19:55,880 --> 00:19:57,720
Speaker 2: the kind of behavior of models would be. Well, I'm

361
00:19:57,720 --> 00:19:59,959
Speaker 2: not advocating to do that, and I don't think anybody.

362
00:20:00,240 --> 00:20:02,520
Speaker 2: I think actually this is one of the like not

363
00:20:02,680 --> 00:20:06,520
Speaker 2: letting AI have a role in nuclear weapon decision making

364
00:20:06,640 --> 00:20:08,960
Speaker 2: is one thing that the US and China have actually

365
00:20:09,000 --> 00:20:11,480
Speaker 2: agreed on, which is extremely unusual these days.

366
00:20:11,520 --> 00:20:11,680
Speaker 1: Right.

367
00:20:11,720 --> 00:20:14,840
Speaker 2: So, but it's unsurprising to me that if you have

368
00:20:14,880 --> 00:20:17,879
Speaker 2: a model trade on the entire Internet, it's going to

369
00:20:18,400 --> 00:20:23,760
Speaker 2: tend toward escalation and not be very very expert in

370
00:20:24,440 --> 00:20:27,440
Speaker 2: how to avoid you know, it's not going to be

371
00:20:27,840 --> 00:20:30,040
Speaker 2: someone who spent their life worrying about that sort of thing.

372
00:20:30,080 --> 00:20:33,160
Speaker 2: And on the question of whether ais could fight each other,

373
00:20:33,359 --> 00:20:37,679
Speaker 2: the truth is, like you are already seeing in certain conflicts,

374
00:20:38,080 --> 00:20:41,560
Speaker 2: you know, autonomous systems, semi autonomous systems are remotely operated

375
00:20:42,160 --> 00:20:43,960
Speaker 2: fighting other semi autonomous systems.

376
00:20:44,000 --> 00:20:44,119
Speaker 1: Right.

377
00:20:44,119 --> 00:20:47,040
Speaker 2: And some people may argue that that would remove people

378
00:20:47,080 --> 00:20:50,800
Speaker 2: from harm's way. That is not the reality of a conflict.

379
00:20:50,840 --> 00:20:52,640
Speaker 2: There's going to be plenty of people in harm's way.

380
00:20:53,240 --> 00:20:55,639
Speaker 2: And the question is, you know, how do you do

381
00:20:55,720 --> 00:20:59,200
Speaker 2: that in ways that are reliable and don't make mistakes.

382
00:20:59,280 --> 00:21:02,040
Speaker 2: And there has been talking in recent years of these

383
00:21:02,080 --> 00:21:04,679
Speaker 2: sort of things like existential risks, but long before that,

384
00:21:04,720 --> 00:21:08,200
Speaker 2: there are lots of systemic dangers of using this technology

385
00:21:08,200 --> 00:21:12,040
Speaker 2: in any critical system. So you know, just as you

386
00:21:12,080 --> 00:21:14,719
Speaker 2: have hallucinations if you're using a chat, or just as

387
00:21:14,760 --> 00:21:17,840
Speaker 2: it gets simple reasoning wrong in a very convincing way,

388
00:21:18,240 --> 00:21:19,880
Speaker 2: that is what we should be worried about. I think

389
00:21:19,920 --> 00:21:23,439
Speaker 2: in how this and other models language models are used

390
00:21:23,480 --> 00:21:24,800
Speaker 2: by the military, by the Pentagon.

391
00:21:25,200 --> 00:21:25,520
Speaker 1: You know what.

392
00:21:25,640 --> 00:21:29,320
Speaker 2: I once talked to the excuse and naval officer in

393
00:21:29,320 --> 00:21:32,120
Speaker 2: the US Navy whose job it was to, you know,

394
00:21:32,440 --> 00:21:34,920
Speaker 2: be in command of these nuclear submarines, and he said

395
00:21:35,200 --> 00:21:37,120
Speaker 2: one of the first things he did when he got

396
00:21:37,119 --> 00:21:38,960
Speaker 2: the job was get permission to go and talk to

397
00:21:38,960 --> 00:21:42,400
Speaker 2: his counterpart in China, go meet him and spend time there.

398
00:21:42,560 --> 00:21:45,760
Speaker 2: The reason being because if something went wrong, if one

399
00:21:45,760 --> 00:21:48,320
Speaker 2: of the systems went wrong with some a person made

400
00:21:48,359 --> 00:21:52,159
Speaker 2: a mistake, you know, or say there's a collision between

401
00:21:52,200 --> 00:21:54,719
Speaker 2: two vessels, and you know, you've a risk of escalation.

402
00:21:55,200 --> 00:21:56,560
Speaker 2: He wants to be able to pick up the phone

403
00:21:56,640 --> 00:21:58,000
Speaker 2: and talk to someone he knows. He doesn't want that

404
00:21:58,040 --> 00:22:00,080
Speaker 2: to be the first time he's spoken to them, and

405
00:22:00,119 --> 00:22:02,919
Speaker 2: so that sort of shows the importance of the human element.

406
00:22:03,000 --> 00:22:06,720
Speaker 2: And that's that's not something that those we talk about

407
00:22:07,119 --> 00:22:09,440
Speaker 2: a g I and intelligence being sold. But that is

408
00:22:09,480 --> 00:22:12,160
Speaker 2: a very human thing that no language want to really,

409
00:22:12,800 --> 00:22:17,080
Speaker 2: they don't have that kind of human human intelligence, right.

410
00:22:17,240 --> 00:22:19,960
Speaker 1: Yeah, chat GBT can't take groc out for coffee.

411
00:22:20,400 --> 00:22:23,320
Speaker 2: Yeah No, well no not yet. I mean, that's a

412
00:22:23,359 --> 00:22:25,600
Speaker 2: frightening thought, it is. I don't think anybody would want

413
00:22:25,640 --> 00:22:26,280
Speaker 2: to take rock count.

414
00:22:26,400 --> 00:22:38,560
Speaker 3: Yeah.

415
00:22:39,160 --> 00:22:42,040
Speaker 1: So there's a question also here in the chat that

416
00:22:42,119 --> 00:22:45,439
Speaker 1: I want to get to, so b masters asking saying

417
00:22:46,080 --> 00:22:49,640
Speaker 1: they read some sources that say that Anthropic was actually

418
00:22:49,680 --> 00:22:52,600
Speaker 1: willing to give in to the Department of Wars demands,

419
00:22:52,800 --> 00:22:56,120
Speaker 1: but the communication broke down. The CEO just doubled down

420
00:22:56,160 --> 00:22:58,640
Speaker 1: on the ethical argument, just as a pr move. I mean,

421
00:22:58,640 --> 00:23:01,040
Speaker 1: what your reporting have you? Have you seen anything that

422
00:23:01,080 --> 00:23:02,000
Speaker 1: would substantiate that.

423
00:23:02,680 --> 00:23:06,639
Speaker 2: I couldn't tell you definitively if they were about to

424
00:23:06,720 --> 00:23:10,440
Speaker 2: cave to the d d's terms. I have heard that

425
00:23:10,480 --> 00:23:12,840
Speaker 2: they were very close to reaching an agreement which seemed

426
00:23:12,840 --> 00:23:15,960
Speaker 2: like sounded to me like a compromise between the two

427
00:23:16,040 --> 00:23:18,440
Speaker 2: and it wasn't very far off. I think it's gonna

428
00:23:18,440 --> 00:23:20,080
Speaker 2: it's gonna be too easy to get caught up in

429
00:23:21,520 --> 00:23:24,560
Speaker 2: the drama of these companies in there and missed the

430
00:23:24,560 --> 00:23:27,879
Speaker 2: bigger picture. Which is we'll hang on. Maybe we should

431
00:23:27,920 --> 00:23:30,240
Speaker 2: have more of a discussion about how those things are

432
00:23:30,280 --> 00:23:32,880
Speaker 2: you what are the benefits, what are the It's important

433
00:23:32,920 --> 00:23:34,760
Speaker 2: to say that I don't think you should not use

434
00:23:35,160 --> 00:23:38,280
Speaker 2: AI in defense. It's all about how reliably you can

435
00:23:38,320 --> 00:23:40,680
Speaker 2: do it, What are the limits? What accountability is that?

436
00:23:41,000 --> 00:23:43,720
Speaker 2: How how does citizens sort of know how that's being

437
00:23:43,800 --> 00:23:44,520
Speaker 2: used in their name?

438
00:23:44,640 --> 00:23:44,840
Speaker 3: Right?

439
00:23:44,880 --> 00:23:47,400
Speaker 2: And it you know, Trump wrote in that that message

440
00:23:47,400 --> 00:23:51,160
Speaker 2: that he decides how wars are fought, and in some way,

441
00:23:51,240 --> 00:23:53,800
Speaker 2: right he decides when we invadeer on, but he doesn't

442
00:23:53,800 --> 00:23:57,439
Speaker 2: decide the terms of war, like you know, what is

443
00:23:57,880 --> 00:24:01,160
Speaker 2: a war crime? That is decided by society at large.

444
00:24:01,160 --> 00:24:02,760
Speaker 2: And that's the sort of a bigger thing that we

445
00:24:02,960 --> 00:24:05,119
Speaker 2: to be asking these questions when it comes to the

446
00:24:05,160 --> 00:24:08,960
Speaker 2: new capabilities that the AI is going to give governments,

447
00:24:09,040 --> 00:24:11,680
Speaker 2: which is they would be able to do new kinds

448
00:24:11,680 --> 00:24:14,359
Speaker 2: of massurveilance you could never you know, never imagine before. Right,

449
00:24:14,400 --> 00:24:16,760
Speaker 2: you can be able to find patterns you couldn't think

450
00:24:16,760 --> 00:24:18,880
Speaker 2: of it, or you'll be able to sort of automate

451
00:24:18,920 --> 00:24:22,639
Speaker 2: that much more so, I think, yeah, that's the bigger

452
00:24:22,680 --> 00:24:24,600
Speaker 2: thing for me, and I think I should remember that

453
00:24:24,800 --> 00:24:27,800
Speaker 2: you know, real life, real human beings at stake here

454
00:24:27,880 --> 00:24:30,680
Speaker 2: and that war as hell, and it is something you

455
00:24:30,760 --> 00:24:31,960
Speaker 2: want to try and avoid.

456
00:24:31,760 --> 00:24:35,000
Speaker 1: Right, Yeah, you were talking about the bigger picture, and

457
00:24:35,040 --> 00:24:37,919
Speaker 1: also you mentioned palents here, which, as it happens, I

458
00:24:38,320 --> 00:24:40,720
Speaker 1: just a little bit ago talk to mechanic Kelly, your

459
00:24:40,880 --> 00:24:44,719
Speaker 1: colleague about talents here, and we actually have an episode

460
00:24:44,720 --> 00:24:48,240
Speaker 1: about tech drop in next week. But this kind of

461
00:24:48,280 --> 00:24:51,960
Speaker 1: seems to be part of a larger trend, like you said,

462
00:24:52,560 --> 00:24:56,919
Speaker 1: big tech companies who were very interested in working with

463
00:24:57,240 --> 00:25:02,560
Speaker 1: the government, but specifically working with the government in military capacities,

464
00:25:03,680 --> 00:25:07,240
Speaker 1: which feels new and in some ways it's not. Of course,

465
00:25:07,400 --> 00:25:09,680
Speaker 1: in some ways it's not. Some of the origins of

466
00:25:09,720 --> 00:25:12,160
Speaker 1: things like the Internet, you know, does come from the military.

467
00:25:12,520 --> 00:25:15,159
Speaker 1: But in terms of the way that we think about

468
00:25:15,359 --> 00:25:18,800
Speaker 1: the culture of Silicon Valley, I think this turn has

469
00:25:18,800 --> 00:25:22,359
Speaker 1: been a little bit unsettling but also surprising for a

470
00:25:22,400 --> 00:25:24,679
Speaker 1: lot of people. But it does to be a trend

471
00:25:24,920 --> 00:25:28,920
Speaker 1: right now. Where do you see this going, say, even

472
00:25:28,960 --> 00:25:29,879
Speaker 1: over the next year or so.

473
00:25:31,080 --> 00:25:33,040
Speaker 2: I mean, that's a great question. That is a really

474
00:25:33,040 --> 00:25:39,560
Speaker 2: interesting question. It's hard to imagine ALI companies becoming disentangled

475
00:25:39,760 --> 00:25:43,440
Speaker 2: because they are is so strategically important, and the incentive

476
00:25:43,440 --> 00:25:46,719
Speaker 2: for working with the government is enormous. Right, The amount

477
00:25:46,720 --> 00:25:48,919
Speaker 2: of data, the amount of compute, the amount of money

478
00:25:49,520 --> 00:25:51,840
Speaker 2: that the government has to throw out those companies is

479
00:25:51,880 --> 00:25:54,600
Speaker 2: so huge that it would be surprising to me if

480
00:25:54,640 --> 00:25:58,080
Speaker 2: everybody said, our conscience is not clear, so we're going

481
00:25:58,160 --> 00:26:01,280
Speaker 2: to rein that in So if you imagine that we're

482
00:26:01,280 --> 00:26:04,040
Speaker 2: going to have more and more tension, more more conflict,

483
00:26:04,880 --> 00:26:08,080
Speaker 2: there will be pressures on them too, for everybody to

484
00:26:08,640 --> 00:26:11,639
Speaker 2: use that AI and new ways that might be concerning,

485
00:26:12,280 --> 00:26:17,840
Speaker 2: besides the use in military situations, which I think we

486
00:26:17,840 --> 00:26:20,240
Speaker 2: should have try and have transparency on. I think it

487
00:26:20,320 --> 00:26:23,760
Speaker 2: is really telling that Anthropic had this concern about mass

488
00:26:23,760 --> 00:26:29,840
Speaker 2: surveillance because I think language models are preternaturally brilliant at

489
00:26:30,359 --> 00:26:35,560
Speaker 2: passing large amounts of text and audio and making sense

490
00:26:35,600 --> 00:26:38,520
Speaker 2: of it, finding patterns and making predictions like we haven't

491
00:26:38,520 --> 00:26:41,040
Speaker 2: done anything like that, and that's exactly what you might

492
00:26:41,080 --> 00:26:43,840
Speaker 2: want to do with in surveillance. So the temptation is

493
00:26:43,880 --> 00:26:47,600
Speaker 2: going to be huge, and yeah, I think it will.

494
00:26:47,640 --> 00:26:50,280
Speaker 2: I think it will be very interesting to see if

495
00:26:50,320 --> 00:26:53,359
Speaker 2: this turns into more of a pushback if more people.

496
00:26:54,840 --> 00:26:57,159
Speaker 2: One thing is that tech workers have a ridiculous I

497
00:26:57,160 --> 00:27:00,000
Speaker 2: mean especially AI engineers, right has an incredible amount of

498
00:27:00,119 --> 00:27:02,880
Speaker 2: cloud these days because they're so valuable. And I think

499
00:27:02,880 --> 00:27:06,080
Speaker 2: this is you know, one reason why Sam Alman and

500
00:27:06,119 --> 00:27:10,520
Speaker 2: others are going to extraordinary lengths to try and explain

501
00:27:10,600 --> 00:27:13,960
Speaker 2: and to a piece their their staff and maybe change

502
00:27:14,000 --> 00:27:15,600
Speaker 2: even change tax because of that.

503
00:27:15,840 --> 00:27:17,720
Speaker 1: Yeah, well, there were there were a bunch of workers,

504
00:27:17,760 --> 00:27:20,639
Speaker 1: I mean hundreds of workers from open Ai and Google

505
00:27:20,640 --> 00:27:25,760
Speaker 1: who signed an open letter basically supporting anthropic. There is

506
00:27:26,760 --> 00:27:30,520
Speaker 1: something of a little bit of pushback, if I can

507
00:27:30,560 --> 00:27:33,800
Speaker 1: call it that, from inside of these companies, and even

508
00:27:33,840 --> 00:27:37,520
Speaker 1: you know, at the consumer angle, people uninstalling chat ubt

509
00:27:37,680 --> 00:27:42,600
Speaker 1: and installing Claude. The question, though, becomes, are both of

510
00:27:42,600 --> 00:27:45,240
Speaker 1: those things enough? Is the pushback from inside the company?

511
00:27:45,359 --> 00:27:47,640
Speaker 1: Enough is a bunch of people deciding that, Hey, I'm

512
00:27:47,640 --> 00:27:51,680
Speaker 1: gonna uninstall open Ai and use the woke claud which

513
00:27:51,680 --> 00:27:53,600
Speaker 1: I think we've established is is maybe not that maybe

514
00:27:53,680 --> 00:27:56,600
Speaker 1: not as woke as some people might have be given

515
00:27:56,680 --> 00:28:00,440
Speaker 1: it credit for. What does that do? Does that anything

516
00:28:00,840 --> 00:28:01,440
Speaker 1: move the needle?

517
00:28:02,400 --> 00:28:03,920
Speaker 2: I don't think it does at this stage. I think

518
00:28:04,040 --> 00:28:06,720
Speaker 2: the question is does it go from hundreds of workers

519
00:28:06,720 --> 00:28:11,600
Speaker 2: to thousands and does the pushback from the public, which

520
00:28:11,600 --> 00:28:15,080
Speaker 2: already seems to be fading somewhat, does that turn into

521
00:28:15,080 --> 00:28:18,160
Speaker 2: something maybe with further incidents that has a meaningful impact

522
00:28:18,240 --> 00:28:21,560
Speaker 2: on bottom line, you know. And it is sort of

523
00:28:21,600 --> 00:28:24,359
Speaker 2: hard these days, right for people. It's harder for people

524
00:28:24,400 --> 00:28:26,919
Speaker 2: to push back and draw lines when it comes to

525
00:28:26,960 --> 00:28:29,879
Speaker 2: the government. It seems like everybody's a bit more curtailed

526
00:28:29,920 --> 00:28:32,679
Speaker 2: and scared, right, But at some point that you know,

527
00:28:32,720 --> 00:28:35,400
Speaker 2: as we've seen with some of the protests, that there

528
00:28:35,720 --> 00:28:37,320
Speaker 2: is a sort of breaking point, and I think it

529
00:28:37,359 --> 00:28:39,600
Speaker 2: needs to be a bigger pushback. But I think it

530
00:28:39,640 --> 00:28:45,560
Speaker 2: also we probably should have more nuance around what that

531
00:28:45,600 --> 00:28:48,040
Speaker 2: pushback here it really is, you know, just saying anthropic

532
00:28:48,200 --> 00:28:52,120
Speaker 2: is better. It's actually maybe like maybe everybody should get

533
00:28:52,120 --> 00:28:54,720
Speaker 2: together and say all these companies should be saying, hey,

534
00:28:54,760 --> 00:28:57,560
Speaker 2: we're not totally sure this stuff is ready even for

535
00:28:58,680 --> 00:29:01,560
Speaker 2: half the use cases whatever it might be like, and

536
00:29:01,720 --> 00:29:04,479
Speaker 2: then have the government explain to people how it's being mean.

537
00:29:04,520 --> 00:29:07,480
Speaker 2: It's one of the things that America can do, you know,

538
00:29:07,640 --> 00:29:10,760
Speaker 2: is pushes political leaders in theory to do that. So

539
00:29:10,920 --> 00:29:12,280
Speaker 2: I would rewind hope.

540
00:29:12,120 --> 00:29:16,040
Speaker 1: For well, thank you so much for coming through and

541
00:29:16,080 --> 00:29:17,960
Speaker 1: talking about this. It's heavy stuff, but we got to

542
00:29:18,000 --> 00:29:19,280
Speaker 1: talk about it. Thank you for real.

543
00:29:19,440 --> 00:29:21,120
Speaker 2: You're very welcome, ye, thanks for having me.

544
00:29:24,280 --> 00:29:26,880
Speaker 1: Thank you so much for listening to another episode of

545
00:29:26,960 --> 00:29:29,000
Speaker 1: kill Switch. You can email us if you want to

546
00:29:29,040 --> 00:29:32,840
Speaker 1: talk at kill Switch at Kaleidoscope dot NYC or on

547
00:29:32,880 --> 00:29:35,840
Speaker 1: Instagram or at kill switch Pod, and if you like

548
00:29:35,840 --> 00:29:38,320
Speaker 1: what you're hearing, maybe leave us a review. It helps

549
00:29:38,360 --> 00:29:40,840
Speaker 1: other people find the show, which helps us keep doing

550
00:29:40,840 --> 00:29:43,040
Speaker 1: our thing. And as I said at the beginning of

551
00:29:43,080 --> 00:29:45,920
Speaker 1: the show, kill Switch is on YouTube, so if you

552
00:29:45,960 --> 00:29:48,040
Speaker 1: want to catch the next one live, the link for

553
00:29:48,080 --> 00:29:51,880
Speaker 1: that and everything else is in the show notes. Kill

554
00:29:51,920 --> 00:29:54,920
Speaker 1: Switch is hosted by Me Dexter Thomas. It's produced by

555
00:29:54,960 --> 00:29:59,000
Speaker 1: Sena Ozaki, Darluck Potts, and Julia Nutter. Our theme song

556
00:29:59,120 --> 00:30:02,680
Speaker 1: is by me and Kyle Murdoch from Kaleidoscope. Our executive

557
00:30:02,680 --> 00:30:06,880
Speaker 1: producers are Ozma Lashin, mungesh Hati Kadur, and Kate Osborne

558
00:30:06,960 --> 00:30:11,000
Speaker 1: from iHeart. Our executive producers are Katrina Norvil and Nikki Ecord.

559
00:30:11,360 --> 00:30:11,880
Speaker 1: Catch on the

560
00:30:11,920 --> 00:30:24,000
Speaker 3: Next one, Goodbye,