1
00:00:11,840 --> 00:00:13,920
Speaker 1: Read, Taylor, I want to ask you to open this

2
00:00:13,920 --> 00:00:17,640
Speaker 1: week's episode. Would you shop online if you didn't actually

3
00:00:17,720 --> 00:00:21,200
Speaker 1: get anything in the mail. But I'm asking because I've

4
00:00:21,200 --> 00:00:24,320
Speaker 1: recently read about these dopamine sites out of South Korea

5
00:00:24,400 --> 00:00:27,160
Speaker 1: that let you look through non existent products with made

6
00:00:27,200 --> 00:00:29,320
Speaker 1: up reviews and promotions that you can add to your cart,

7
00:00:29,480 --> 00:00:33,280
Speaker 1: click buy, track your packages and never receive a thing,

8
00:00:33,360 --> 00:00:36,600
Speaker 1: No money spent, just the satisfaction of buying something online.

9
00:00:37,159 --> 00:00:39,960
Speaker 2: No you got, Taylor.

10
00:00:40,280 --> 00:00:43,040
Speaker 3: I think I already do this with Zillo every night

11
00:00:43,120 --> 00:00:47,080
Speaker 3: where I'm you know, perusing multimillion dollar houses in Pasadena

12
00:00:47,159 --> 00:00:48,800
Speaker 3: adding them to my favorites.

13
00:00:49,080 --> 00:00:52,240
Speaker 2: Like I don't do the Zilla thing because there's nothing

14
00:00:52,280 --> 00:00:54,920
Speaker 2: on the market here in the day area, but I

15
00:00:55,320 --> 00:00:57,640
Speaker 2: would know. Like what I would do, though, is could

16
00:00:57,680 --> 00:01:00,600
Speaker 2: I order stuff but not have the package just show

17
00:01:00,680 --> 00:01:02,520
Speaker 2: up at my front door, so that, like all my

18
00:01:02,560 --> 00:01:05,160
Speaker 2: neighbors know that we're just like those horrible people who

19
00:01:05,319 --> 00:01:06,959
Speaker 2: order like twenty things on Amazon every.

20
00:01:06,920 --> 00:01:09,520
Speaker 4: Day that the note packages thing.

21
00:01:09,600 --> 00:01:12,080
Speaker 2: For sure, I still want the product stuff.

22
00:01:16,840 --> 00:01:17,880
Speaker 4: Welcome to Tech Stuff.

23
00:01:17,920 --> 00:01:20,400
Speaker 1: I'm as Volocian and this is the week in Tech

24
00:01:20,720 --> 00:01:23,640
Speaker 1: where I'm joined by the world's most clubs and reporters.

25
00:01:23,080 --> 00:01:25,560
Speaker 4: To break down what's really happening in tech right now.

26
00:01:26,080 --> 00:01:28,960
Speaker 1: Today, we're joined by Taylor Lorenzo use a mag and

27
00:01:29,080 --> 00:01:31,399
Speaker 1: Read Albergotti, tech editor at Centophore.

28
00:01:31,440 --> 00:01:34,040
Speaker 2: Welcome both, great to be here, Thanks for having us.

29
00:01:34,040 --> 00:01:37,040
Speaker 1: Great to have you both, As always read, the FCC

30
00:01:37,240 --> 00:01:41,280
Speaker 1: moved to ban Chinese humanoid robots and quadrupeds from entering

31
00:01:41,280 --> 00:01:45,440
Speaker 1: the US, which I can't read without laughing. But I mean,

32
00:01:46,000 --> 00:01:47,319
Speaker 1: what's what's going on here?

33
00:01:47,440 --> 00:01:47,520
Speaker 3: Like?

34
00:01:47,600 --> 00:01:48,760
Speaker 4: How serious is this story?

35
00:01:48,800 --> 00:01:50,480
Speaker 1: What what does it have to do with US robot

36
00:01:50,520 --> 00:01:52,000
Speaker 1: makers lobbying efforts?

37
00:01:52,040 --> 00:01:53,200
Speaker 4: And what's gonna happen next?

38
00:01:53,880 --> 00:01:56,680
Speaker 2: Yeah, I mean this is this is another sort of

39
00:01:56,720 --> 00:02:01,000
Speaker 2: it's its industrial policy protectionism. I guess we shouldn't be surprised.

40
00:02:01,000 --> 00:02:04,240
Speaker 2: I mean, they're already are no buy D Chinese cars

41
00:02:04,320 --> 00:02:07,360
Speaker 2: in the US and it and on its face, you're

42
00:02:07,480 --> 00:02:10,239
Speaker 2: you're you'd sort of put into that box and you say, okay,

43
00:02:10,280 --> 00:02:14,120
Speaker 2: this this makes sense. We want US robot makers to

44
00:02:14,120 --> 00:02:16,280
Speaker 2: to kind of have a clear run at this. But

45
00:02:16,360 --> 00:02:19,240
Speaker 2: I think there's a there and and granted like there

46
00:02:19,280 --> 00:02:21,480
Speaker 2: will be exceptions to this, right, this is not like

47
00:02:21,560 --> 00:02:23,680
Speaker 2: this is not a law. This is you know, this

48
00:02:23,760 --> 00:02:27,040
Speaker 2: is this is policy coming out of the White House essentially,

49
00:02:27,960 --> 00:02:30,240
Speaker 2: but like on its face, you there are these big

50
00:02:30,320 --> 00:02:33,400
Speaker 2: questions like a bunch if you go into any sort

51
00:02:33,400 --> 00:02:36,400
Speaker 2: of like AI company that's doing anything in the physical

52
00:02:36,400 --> 00:02:38,919
Speaker 2: world and in the Bay Area right now, they've got

53
00:02:38,960 --> 00:02:42,440
Speaker 2: these Chinese robots like sitting around from all these different companies,

54
00:02:42,440 --> 00:02:44,560
Speaker 2: and they're you know, some of them are cheap, like

55
00:02:44,639 --> 00:02:48,120
Speaker 2: ten thousand apiece, and they're doing all sorts of testing

56
00:02:48,160 --> 00:02:51,320
Speaker 2: on these things, right They're they're essentially building software and

57
00:02:51,400 --> 00:02:54,360
Speaker 2: the and the hardware is what they use to you know,

58
00:02:54,440 --> 00:02:57,360
Speaker 2: to gather data or just test out their their algorithm.

59
00:02:57,440 --> 00:03:02,720
Speaker 2: So I think it's actually like potentially detrimental to innovation.

60
00:03:02,919 --> 00:03:04,600
Speaker 2: So this is all and it's the same thing with

61
00:03:04,680 --> 00:03:07,760
Speaker 2: like the same issue with these Chinese open source models.

62
00:03:07,800 --> 00:03:09,680
Speaker 1: Right It's like, I want to talk about the open

63
00:03:09,680 --> 00:03:11,720
Speaker 1: source models, but let's stay with the robots for a

64
00:03:11,720 --> 00:03:14,239
Speaker 1: little bit. I mean, the detail about the quadrupeds was

65
00:03:14,280 --> 00:03:16,160
Speaker 1: the one that really really caught my eye because I

66
00:03:16,200 --> 00:03:18,440
Speaker 1: kind of thought the four legged robots were only for

67
00:03:18,480 --> 00:03:21,920
Speaker 1: YouTube videos that I didn't realized they were threat to America.

68
00:03:22,080 --> 00:03:25,040
Speaker 2: I saw I was in Boston a couple of weeks

69
00:03:25,040 --> 00:03:27,880
Speaker 2: ago and I saw some people walking their their robot

70
00:03:28,000 --> 00:03:30,359
Speaker 2: dicers that looked like, yeah, they looked like they were

71
00:03:30,440 --> 00:03:33,960
Speaker 2: sort of like MIT researchers or something. But no, I

72
00:03:33,960 --> 00:03:37,280
Speaker 2: mean it's it's like the same issue though. It's the

73
00:03:37,280 --> 00:03:39,920
Speaker 2: same principle, which is like you try to you try

74
00:03:39,960 --> 00:03:42,400
Speaker 2: to stop the you know, the Chinese from encroaching in

75
00:03:42,440 --> 00:03:45,280
Speaker 2: the US market, and then you sort of like also

76
00:03:45,480 --> 00:03:48,680
Speaker 2: hinder your own innovation. So these things always come at

77
00:03:48,680 --> 00:03:50,440
Speaker 2: a cost. So I think that the trick is going

78
00:03:50,520 --> 00:03:52,360
Speaker 2: to be figuring out how to thread that needle right, Like,

79
00:03:52,880 --> 00:03:55,000
Speaker 2: do what you need to do, you know, if you're

80
00:03:55,040 --> 00:03:57,720
Speaker 2: the White House, to to sort of protect US companies,

81
00:03:58,040 --> 00:04:00,760
Speaker 2: stop the Chinese from i don't know, dumping robots if

82
00:04:00,760 --> 00:04:05,160
Speaker 2: that's even a thing, but then also allow US researchers

83
00:04:05,240 --> 00:04:08,600
Speaker 2: to still, you know, to still be able to innovate

84
00:04:08,720 --> 00:04:10,760
Speaker 2: and come. You know, you don't want like a situation

85
00:04:10,800 --> 00:04:13,800
Speaker 2: where there's like one or two robotics companies in the

86
00:04:13,920 --> 00:04:16,200
Speaker 2: US and if they don't, if they don't just happen

87
00:04:16,240 --> 00:04:18,880
Speaker 2: to win this race outright, you know, the US is

88
00:04:19,560 --> 00:04:20,279
Speaker 2: way behind.

89
00:04:20,480 --> 00:04:22,400
Speaker 1: This will be like the cause in the Soviet Union

90
00:04:22,440 --> 00:04:24,720
Speaker 1: when the when in nineteen eighteen nine, Right, the US

91
00:04:24,760 --> 00:04:28,719
Speaker 1: will have all of these like Dusty Treban style robots

92
00:04:28,800 --> 00:04:32,039
Speaker 1: call the Tita. Have you been following the robot element

93
00:04:32,080 --> 00:04:32,919
Speaker 1: of this story.

94
00:04:32,880 --> 00:04:36,040
Speaker 3: Yes, I've been following. And does this mean that rumbas

95
00:04:36,080 --> 00:04:40,000
Speaker 3: are banned? I saw some headline basically implying.

96
00:04:39,600 --> 00:04:42,839
Speaker 2: That I have a Chinese I have a Chinese robot vacuum.

97
00:04:42,839 --> 00:04:47,880
Speaker 2: That's really really good. I think it's a way outperforms

98
00:04:47,920 --> 00:04:52,520
Speaker 2: the rumba. So I hope, I hope that's not the case. Look,

99
00:04:53,040 --> 00:04:55,599
Speaker 2: I think this is like all of these policies. It's

100
00:04:55,640 --> 00:04:58,039
Speaker 2: just the thing to remember is that you can do

101
00:04:58,120 --> 00:05:00,560
Speaker 2: whatever you want to try to stop China these tech

102
00:05:00,640 --> 00:05:03,360
Speaker 2: and you know, stop them from i don't know, catching

103
00:05:03,440 --> 00:05:07,640
Speaker 2: up or ripping off American technology. The US will not

104
00:05:07,839 --> 00:05:11,160
Speaker 2: win unless it can move faster than China on innovation.

105
00:05:11,560 --> 00:05:14,039
Speaker 2: That is the most important thing about this race, and

106
00:05:14,040 --> 00:05:16,640
Speaker 2: I think it often. I think it often gets sort

107
00:05:16,680 --> 00:05:18,320
Speaker 2: of forgotten or pushed to the side.

108
00:05:18,720 --> 00:05:22,039
Speaker 3: I totally agree with Red and yeah, I mean I

109
00:05:22,080 --> 00:05:25,200
Speaker 3: was just reading this article on the Verge talking about

110
00:05:25,240 --> 00:05:27,479
Speaker 3: like what's actually sort of included in this span and

111
00:05:27,520 --> 00:05:31,120
Speaker 3: it's significantly broader than you would think. I mean, Obviously

112
00:05:31,200 --> 00:05:33,520
Speaker 3: we're talking about the humanoid robots and the quadrupeds, but

113
00:05:33,560 --> 00:05:36,720
Speaker 3: it's not just that, it's it's far broader. This ban

114
00:05:36,880 --> 00:05:40,839
Speaker 3: covers basically any new software controlled robot that travels over

115
00:05:40,880 --> 00:05:44,000
Speaker 3: the ground or weighs more than four point four pounds,

116
00:05:44,000 --> 00:05:47,360
Speaker 3: including the DOC, and that can perceive its own environment

117
00:05:47,480 --> 00:05:50,880
Speaker 3: and has wireless which is basically any robot vacuum like

118
00:05:51,160 --> 00:05:54,320
Speaker 3: it's it's a lot of like household devices. Like I

119
00:05:54,320 --> 00:05:58,120
Speaker 3: think people don't realize as well just how much stuff

120
00:05:58,160 --> 00:06:01,320
Speaker 3: comes from China, Like how like automated I have this

121
00:06:01,400 --> 00:06:04,560
Speaker 3: like automated you know, like watering system thing that I

122
00:06:04,600 --> 00:06:07,640
Speaker 3: got from Amazon. I'm sure it's some Chinese little robot thing.

123
00:06:07,760 --> 00:06:08,120
Speaker 4: Oh you did.

124
00:06:08,160 --> 00:06:10,120
Speaker 1: We were talking about the garden last week and you

125
00:06:10,160 --> 00:06:12,240
Speaker 1: were talking about wanting a gunting robot. And now you've

126
00:06:12,320 --> 00:06:13,560
Speaker 1: now you filled the trigger.

127
00:06:13,960 --> 00:06:19,039
Speaker 3: Exactly and I've ordered a Chinese gardening you know, watering thing.

128
00:06:19,160 --> 00:06:22,599
Speaker 3: Probably that's potentially going to be illegal, legal, but but

129
00:06:22,760 --> 00:06:25,040
Speaker 3: I know I snuck in right under the van. I

130
00:06:25,520 --> 00:06:27,760
Speaker 3: just think, like, I mean, what Reid said is true.

131
00:06:27,960 --> 00:06:30,840
Speaker 3: I think this is ridiculous. Like I understand from a

132
00:06:30,880 --> 00:06:33,719
Speaker 3: consumer perspective, you know, with the electric cars. Okay, protecting

133
00:06:33,760 --> 00:06:36,760
Speaker 3: our car industry in some way, but you know, at

134
00:06:36,839 --> 00:06:42,719
Speaker 3: some point, robotics are becoming so pervasive, and this just

135
00:06:42,760 --> 00:06:45,520
Speaker 3: seems like, you know, we we need to compete harder,

136
00:06:45,680 --> 00:06:48,480
Speaker 3: like we can't. We're I feel like we're the declining

137
00:06:48,520 --> 00:06:51,679
Speaker 3: and we're just trying to like protect this declining industry

138
00:06:51,680 --> 00:06:53,919
Speaker 3: and we need to allow our researchers to compete. And

139
00:06:53,920 --> 00:06:56,680
Speaker 3: and that means that you know, giving technologists access to

140
00:06:56,720 --> 00:06:57,440
Speaker 3: this cutting.

141
00:06:57,160 --> 00:07:00,080
Speaker 1: Head technology on the kind of robotspan Is this like

142
00:07:00,279 --> 00:07:01,960
Speaker 1: kind of a thing that people expect to be in

143
00:07:02,040 --> 00:07:04,279
Speaker 1: forced or is this kind of like a staking of

144
00:07:04,320 --> 00:07:05,640
Speaker 1: some political kind of pulturing.

145
00:07:05,640 --> 00:07:07,760
Speaker 4: Do we do we have any idea?

146
00:07:07,839 --> 00:07:10,040
Speaker 2: Well, I like I was sort of getting at earlier.

147
00:07:10,240 --> 00:07:13,000
Speaker 2: I mean, this is this is just you know, these

148
00:07:13,040 --> 00:07:15,320
Speaker 2: are just orders coming out of the White House. So

149
00:07:15,480 --> 00:07:17,800
Speaker 2: it's not like this is not a law. And I

150
00:07:17,800 --> 00:07:21,400
Speaker 2: think they always they you know, like I think at

151
00:07:21,400 --> 00:07:23,240
Speaker 2: some point they're going to go, oh my god, like

152
00:07:23,760 --> 00:07:25,640
Speaker 2: this is way too broad. We got to find loop.

153
00:07:25,760 --> 00:07:28,560
Speaker 2: You know, there's gonna be loopholes or or carve outs

154
00:07:28,560 --> 00:07:32,200
Speaker 2: for specific areas, like you know, this is what always happens.

155
00:07:32,200 --> 00:07:34,720
Speaker 2: It's like this this shoot from the hip sort of

156
00:07:35,080 --> 00:07:39,680
Speaker 2: you know, proclamation policy, I guess strategy, and then and

157
00:07:39,720 --> 00:07:41,240
Speaker 2: then you go back and make all it. And the

158
00:07:41,280 --> 00:07:43,680
Speaker 2: same thing happened with tariffs, right It's like, oh my god,

159
00:07:43,680 --> 00:07:45,160
Speaker 2: we have all these tariffs. And then it's like, well,

160
00:07:45,200 --> 00:07:47,480
Speaker 2: this thing's carved out and that thing's carved out, and

161
00:07:47,520 --> 00:07:50,480
Speaker 2: you eventually get to like, I don't know someplace where

162
00:07:50,480 --> 00:07:52,760
Speaker 2: it's where it's logical. So I just take all of

163
00:07:52,800 --> 00:07:54,400
Speaker 2: it with a grain of salt and just figure it's

164
00:07:54,400 --> 00:07:55,280
Speaker 2: all going to change.

165
00:07:55,720 --> 00:07:58,720
Speaker 3: I just think in the meantime, the damage is done.

166
00:07:58,920 --> 00:08:01,080
Speaker 3: I mean, the tariffs are good example, that was like

167
00:08:01,120 --> 00:08:05,360
Speaker 3: a huge economic shock. It caused prices to increase for consumers,

168
00:08:05,400 --> 00:08:07,520
Speaker 3: like it did cause a lot of I would argue

169
00:08:07,520 --> 00:08:10,040
Speaker 3: harm to our economy. And I think when we do

170
00:08:10,120 --> 00:08:13,880
Speaker 3: things like this as well, like yeah, maybe you're right

171
00:08:14,000 --> 00:08:16,080
Speaker 3: read and I agree, like it will have to be

172
00:08:16,160 --> 00:08:18,680
Speaker 3: rolled back in some way because it's nonsensical. But in

173
00:08:18,720 --> 00:08:21,080
Speaker 3: the meantime, it's hurt innovation in the meantime, it slowed

174
00:08:21,160 --> 00:08:23,720
Speaker 3: us down. In the meantime, it's created liability for startups

175
00:08:23,760 --> 00:08:26,600
Speaker 3: that don't want to take on any liability. So it

176
00:08:26,640 --> 00:08:27,320
Speaker 3: just seems bad.

177
00:08:27,560 --> 00:08:28,960
Speaker 2: Yeah, I mean, it'd be one thing if they did

178
00:08:28,960 --> 00:08:30,880
Speaker 2: this and they're like, and we're going to put you know,

179
00:08:31,800 --> 00:08:33,640
Speaker 2: more money into academic research.

180
00:08:33,720 --> 00:08:34,679
Speaker 4: That was my next question.

181
00:08:34,760 --> 00:08:37,040
Speaker 1: So is there going to be is there any commensurate

182
00:08:37,120 --> 00:08:40,360
Speaker 1: effort to stimulate the American robotics industry. I so Deep

183
00:08:40,400 --> 00:08:43,120
Speaker 1: Mind had an announcement today about some some new kind

184
00:08:43,120 --> 00:08:45,959
Speaker 1: of robotics program and obviously read you we talked about

185
00:08:46,240 --> 00:08:48,680
Speaker 1: robotics with applied intuition last week on the show. So

186
00:08:48,720 --> 00:08:50,400
Speaker 1: it seems like there is like some action in the

187
00:08:50,480 --> 00:08:52,760
Speaker 1: American robotics sphere.

188
00:08:53,200 --> 00:08:55,760
Speaker 2: Of course, this is a huge industry. There's tons of

189
00:08:55,800 --> 00:08:58,360
Speaker 2: private money going into it. But I think if you're

190
00:08:58,440 --> 00:09:00,480
Speaker 2: the US, like the private sectors going to do what

191
00:09:00,520 --> 00:09:03,800
Speaker 2: it's going to do, the levers that they can pull,

192
00:09:04,040 --> 00:09:07,440
Speaker 2: I think are an academic research. Right. That's where like

193
00:09:07,480 --> 00:09:09,960
Speaker 2: if you think the robotics race is over, like I

194
00:09:10,000 --> 00:09:12,120
Speaker 2: think that's insane. Like I think there's still a lot

195
00:09:12,160 --> 00:09:14,200
Speaker 2: of innovation that's going to happen, and it's maybe not

196
00:09:14,240 --> 00:09:16,080
Speaker 2: all going to come from the private sector. Like if

197
00:09:16,120 --> 00:09:18,720
Speaker 2: you want to do something policy wise, like you know,

198
00:09:19,120 --> 00:09:22,600
Speaker 2: push forward research, right, basic research that's going to you know,

199
00:09:22,679 --> 00:09:26,000
Speaker 2: help advance this industry. There's so many different areas that

200
00:09:26,120 --> 00:09:29,000
Speaker 2: would speed things up. It's it's long term thinking, though,

201
00:09:29,080 --> 00:09:31,800
Speaker 2: right and I think that's what Washington is really short

202
00:09:31,800 --> 00:09:32,360
Speaker 2: on right now.

203
00:09:32,559 --> 00:09:35,040
Speaker 3: I mean, I also just think of like the drama

204
00:09:35,080 --> 00:09:37,560
Speaker 3: over the H one B visas and the way that

205
00:09:37,600 --> 00:09:43,520
Speaker 3: we're treating immigrants. And there are so many talented scientists, engineers,

206
00:09:43,640 --> 00:09:46,920
Speaker 3: roboticists that you know, might be from India, might be

207
00:09:47,000 --> 00:09:50,480
Speaker 3: from other places that we've also pushed out of our country,

208
00:09:50,480 --> 00:09:53,480
Speaker 3: and I think, I mean, I know that's affected startup's

209
00:09:53,559 --> 00:09:56,680
Speaker 3: ability to recruit. I was just talking to somebody last

210
00:09:56,760 --> 00:09:59,319
Speaker 3: night who is living in Canada and what is has

211
00:09:59,360 --> 00:10:00,920
Speaker 3: a job in a Mameric can start up but can't

212
00:10:00,960 --> 00:10:03,840
Speaker 3: start because of this visa process. This is just it's

213
00:10:03,920 --> 00:10:04,920
Speaker 3: really ridiculous.

214
00:10:05,600 --> 00:10:08,160
Speaker 2: Yeah, I mean, it's a it's a this is like

215
00:10:08,240 --> 00:10:11,320
Speaker 2: such a challenge because I mean, what you really want

216
00:10:11,360 --> 00:10:14,320
Speaker 2: to do is like is not sort of bam the

217
00:10:14,400 --> 00:10:17,080
Speaker 2: Chinese robots, but find the people who are really the

218
00:10:17,120 --> 00:10:20,120
Speaker 2: most talented people in China building the robots, bring them

219
00:10:20,120 --> 00:10:23,240
Speaker 2: to the US, have them build the robots here. But

220
00:10:23,360 --> 00:10:26,040
Speaker 2: then there's this worry that they're all spies. Or they

221
00:10:26,040 --> 00:10:28,640
Speaker 2: all will be spies, right, and and there is I mean,

222
00:10:28,679 --> 00:10:31,839
Speaker 2: these are legitimate concerns, but I think there's like a

223
00:10:32,000 --> 00:10:34,439
Speaker 2: I mean this happened in that you mentioned the Soviet

224
00:10:34,480 --> 00:10:36,360
Speaker 2: era too. I mean this happened in the Soviet era

225
00:10:36,480 --> 00:10:38,800
Speaker 2: like the US because of its system, because of its

226
00:10:38,880 --> 00:10:44,160
Speaker 2: open system, you know. There it makes you know, spying easier,

227
00:10:44,240 --> 00:10:46,880
Speaker 2: and so that's just a downside. But then that open

228
00:10:46,920 --> 00:10:50,480
Speaker 2: system leads to like massive innovation, so it's always a

229
00:10:50,520 --> 00:10:51,000
Speaker 2: trade off.

230
00:10:51,480 --> 00:10:53,920
Speaker 1: I got of liked on that word open. Let's talk

231
00:10:53,920 --> 00:10:55,800
Speaker 1: about open source models. I think for the third week

232
00:10:55,800 --> 00:10:58,679
Speaker 1: in a row this is a textuff record, but the

233
00:10:58,720 --> 00:11:03,079
Speaker 1: news keeps rumbling on. So another Chinese export, of course

234
00:11:03,280 --> 00:11:07,240
Speaker 1: is open source models from Chinese companies like Zai and

235
00:11:07,320 --> 00:11:08,199
Speaker 1: moonshot Ai.

236
00:11:09,720 --> 00:11:11,439
Speaker 4: Why is it study in the news for so long?

237
00:11:11,520 --> 00:11:15,319
Speaker 1: And also read you wrote about this concern about censorship

238
00:11:15,480 --> 00:11:18,760
Speaker 1: an Ai models Chinese models that may be unfounded.

239
00:11:19,520 --> 00:11:22,559
Speaker 2: Yeah, there was this research study I wrote about yesterday

240
00:11:23,040 --> 00:11:27,160
Speaker 2: from CTGT which does this interpretability research, like looking at

241
00:11:27,240 --> 00:11:30,680
Speaker 2: how these models actually work so they can build you know,

242
00:11:30,800 --> 00:11:33,400
Speaker 2: products for the private sector that are sort of harnessed

243
00:11:33,440 --> 00:11:36,920
Speaker 2: and don't go off the rails. They did this experiment

244
00:11:36,960 --> 00:11:40,480
Speaker 2: where they took a deep seek model and they fine

245
00:11:40,520 --> 00:11:43,800
Speaker 2: tuned from that model and then tested the before and

246
00:11:43,840 --> 00:11:47,960
Speaker 2: after Chinese censorship and found that the censorship was essentially

247
00:11:48,000 --> 00:11:50,959
Speaker 2: gone once they once they built this new model, the

248
00:11:51,280 --> 00:11:54,800
Speaker 2: base of it was was an open ai OSS like

249
00:11:54,920 --> 00:11:57,719
Speaker 2: open weights model, and then it was fine tuned or

250
00:11:57,800 --> 00:12:01,200
Speaker 2: taught from deep Seek. And so there have been this

251
00:12:01,200 --> 00:12:04,880
Speaker 2: this talk of like subliminal messages being passed on from

252
00:12:04,920 --> 00:12:09,160
Speaker 2: the teacher model to the student model. It gets it

253
00:12:09,200 --> 00:12:11,679
Speaker 2: gets pretty technical and you can get into weeds. But

254
00:12:11,760 --> 00:12:14,160
Speaker 2: like the long story short is like we need more

255
00:12:14,200 --> 00:12:17,120
Speaker 2: research on this. But what it looks like is I

256
00:12:17,160 --> 00:12:21,559
Speaker 2: think these these Chinese models are actually like not as

257
00:12:21,640 --> 00:12:24,400
Speaker 2: much of a propaganda risk as you think because they

258
00:12:24,440 --> 00:12:28,320
Speaker 2: will get changed and filtered through all these different techniques.

259
00:12:28,920 --> 00:12:30,840
Speaker 2: And it also goes to this research point. I mean,

260
00:12:30,880 --> 00:12:34,360
Speaker 2: this is like really interesting research that wouldn't happen unless

261
00:12:34,400 --> 00:12:37,600
Speaker 2: there were these Chinese open source models. So it's another

262
00:12:38,720 --> 00:12:42,319
Speaker 2: I guess point in favor of the pro open source people.

263
00:12:42,960 --> 00:12:47,200
Speaker 3: Yeah. Absolutely, I mean I just think that like it's

264
00:12:47,240 --> 00:12:50,880
Speaker 3: it's really concerning to me that like our response is

265
00:12:50,920 --> 00:12:54,000
Speaker 3: to ban things like I you know, I watched everything

266
00:12:54,040 --> 00:12:57,320
Speaker 3: that happened with TikTok. I covered the musically sale to

267
00:12:57,360 --> 00:13:00,080
Speaker 3: Bite dance, which the government was totally fine with, and

268
00:13:00,080 --> 00:13:02,280
Speaker 3: then they turned around five years later and they were like, actually,

269
00:13:02,320 --> 00:13:06,200
Speaker 3: we're gonna ban TikTok, you know, based on nothing really

270
00:13:06,280 --> 00:13:11,280
Speaker 3: except that they it was keep challenging meta products, you know,

271
00:13:11,440 --> 00:13:13,600
Speaker 3: like there were all these like sort of like made

272
00:13:13,679 --> 00:13:17,280
Speaker 3: up concerns about China that never came to fruition. And

273
00:13:17,320 --> 00:13:19,640
Speaker 3: similar to the Chinese open source models. I mean, we

274
00:13:19,720 --> 00:13:21,679
Speaker 3: did have a lot of research on TikTok and there

275
00:13:21,760 --> 00:13:25,920
Speaker 3: ultimately was just no evidence of anything that the government

276
00:13:26,040 --> 00:13:30,800
Speaker 3: was claiming. And I just it's like if they had

277
00:13:30,920 --> 00:13:33,199
Speaker 3: proof of these things and if they were doing these

278
00:13:33,240 --> 00:13:35,959
Speaker 3: research and showing wow, the Chinese open source models are

279
00:13:36,080 --> 00:13:38,760
Speaker 3: really like biased. So this could be really dangerous for

280
00:13:38,800 --> 00:13:41,560
Speaker 3: American companies because if X Y z all right, then

281
00:13:41,600 --> 00:13:43,240
Speaker 3: I think, you know, we could have a talk about it.

282
00:13:43,280 --> 00:13:46,040
Speaker 3: But in this case, it just it's it seems like

283
00:13:46,120 --> 00:13:48,959
Speaker 3: these speculative fears and as read mentioned earlier, it feels

284
00:13:49,080 --> 00:13:52,440
Speaker 3: very cold war. It feels very like punching at ghosts,

285
00:13:52,480 --> 00:13:55,800
Speaker 3: and I just I really worry about the damage that

286
00:13:55,840 --> 00:13:58,440
Speaker 3: we're doing to our own tech landscape.

287
00:13:58,480 --> 00:14:02,120
Speaker 1: In the meantime you mentioned, I think Markackerberg did two

288
00:14:02,160 --> 00:14:03,960
Speaker 1: interviews this week, with one of the New York Times

289
00:14:03,960 --> 00:14:06,960
Speaker 1: and one with the Journal, his first first media appearances

290
00:14:07,040 --> 00:14:10,560
Speaker 1: since since his uh since his feed me guest lecture

291
00:14:10,640 --> 00:14:11,840
Speaker 1: a few weeks ago.

292
00:14:12,720 --> 00:14:12,800
Speaker 2: And.

293
00:14:14,360 --> 00:14:17,560
Speaker 1: He was basically complaining about the dominance of of open

294
00:14:17,600 --> 00:14:18,599
Speaker 1: a and ionanthropic.

295
00:14:18,720 --> 00:14:18,840
Speaker 2: Right.

296
00:14:19,000 --> 00:14:21,040
Speaker 4: Is that? Is that a fair summary of what the

297
00:14:21,080 --> 00:14:21,960
Speaker 4: media tour was about.

298
00:14:22,480 --> 00:14:25,600
Speaker 2: Yeah, I think there's two minds of this. I mean, one,

299
00:14:26,480 --> 00:14:29,000
Speaker 2: you know, Meta or Facebook has always been sort of

300
00:14:29,000 --> 00:14:32,120
Speaker 2: a pro open source company. They've developed a bunch of

301
00:14:32,320 --> 00:14:35,880
Speaker 2: open source you know, protocols, and but on the other hand,

302
00:14:35,960 --> 00:14:38,400
Speaker 2: I mean you have to sort of take this as like, yeah,

303
00:14:38,400 --> 00:14:40,240
Speaker 2: I mean they're behind in the air raight, so like

304
00:14:40,400 --> 00:14:41,960
Speaker 2: of course, I mean they're going to just be in

305
00:14:42,000 --> 00:14:45,400
Speaker 2: favor of whatever slows their competitors down. So you know,

306
00:14:45,640 --> 00:14:46,080
Speaker 2: there's that.

307
00:14:46,840 --> 00:14:48,360
Speaker 1: There was a headline in the in the New York

308
00:14:48,360 --> 00:14:51,960
Speaker 1: Times Silicon Valley splits over closing the borders to Chinese AI.

309
00:14:52,640 --> 00:14:54,560
Speaker 1: So read is what you're saying that people who want

310
00:14:54,560 --> 00:14:56,360
Speaker 1: to close the borders to Chinese AI are the ones

311
00:14:56,360 --> 00:14:58,120
Speaker 1: who are currently winning and the ones who don't want

312
00:14:58,160 --> 00:14:59,920
Speaker 1: to close the borders of Chinese AI, the ones who

313
00:15:00,120 --> 00:15:01,200
Speaker 1: only losing a no note.

314
00:15:01,280 --> 00:15:03,920
Speaker 3: To me, it's you know, that's a kind of misleading headline.

315
00:15:04,080 --> 00:15:07,320
Speaker 3: To me, it feels very like Anthropic versus everyone else

316
00:15:07,440 --> 00:15:12,560
Speaker 3: in tech. I mean, you had Google, like every single

317
00:15:12,600 --> 00:15:15,440
Speaker 3: tech company kind of signed this open letter right against

318
00:15:15,440 --> 00:15:17,400
Speaker 3: that sort of I felt like was a big subtweet

319
00:15:17,400 --> 00:15:20,240
Speaker 3: of Anthropic. Dario comes out, He's like, Oh no, we

320
00:15:20,280 --> 00:15:22,160
Speaker 3: don't like really want to be an open source models,

321
00:15:22,160 --> 00:15:23,600
Speaker 3: but also we kind of think they're dangerous and we

322
00:15:24,000 --> 00:15:26,680
Speaker 3: kind of I think like dark to me, like I

323
00:15:26,760 --> 00:15:31,200
Speaker 3: see Anthropic running the meta playbook exactly where they have

324
00:15:31,320 --> 00:15:33,440
Speaker 3: this dominance in the market right now and they're trying

325
00:15:33,480 --> 00:15:36,760
Speaker 3: to sort of do regulatory capture to maintain it.

326
00:15:38,200 --> 00:15:40,120
Speaker 4: More about the letter, Taylor, just before we go too far?

327
00:15:40,640 --> 00:15:44,880
Speaker 3: Sure so that So Jensen Wang, CEO of Nvidia, joins

328
00:15:44,920 --> 00:15:47,040
Speaker 3: Twitter big. By the way, I just have to say

329
00:15:47,080 --> 00:15:49,920
Speaker 3: Mark Zuckerberg's on Twitter promoting his Wall Street Journal op

330
00:15:50,040 --> 00:15:52,600
Speaker 3: ed like Jensen's on Twitter. I feel like Twitter has

331
00:15:52,640 --> 00:15:55,920
Speaker 3: gotten peak lately. I love to see all these tech

332
00:15:55,960 --> 00:15:59,880
Speaker 3: CEOs on there. But so Jensen joins Twitter and po

333
00:16:00,360 --> 00:16:03,120
Speaker 3: this open letter, which obviously is you know, shared to

334
00:16:03,160 --> 00:16:07,280
Speaker 3: the website whatever, and it is a defensive open source

335
00:16:07,600 --> 00:16:11,040
Speaker 3: open weight models. Obviously Nvidia has a strong you know,

336
00:16:11,120 --> 00:16:14,440
Speaker 3: business interest in this as well, but it becomes this letter,

337
00:16:14,520 --> 00:16:16,680
Speaker 3: this open letter becomes this sort of like rallying cry,

338
00:16:16,800 --> 00:16:20,000
Speaker 3: and you have just dozens and dozens of tech companies

339
00:16:20,000 --> 00:16:23,080
Speaker 3: sign on to it, you know, by the by the weekend,

340
00:16:23,280 --> 00:16:25,840
Speaker 3: like every almost every major tech company it'd signed on

341
00:16:25,840 --> 00:16:28,520
Speaker 3: to it, including all these startup you know why, y Combinator,

342
00:16:28,520 --> 00:16:31,360
Speaker 3: et cetera, a bunch of startups. So it felt like

343
00:16:32,200 --> 00:16:36,160
Speaker 3: basically the entire tech industry had united to advocate for

344
00:16:36,400 --> 00:16:39,359
Speaker 3: open weight AI models to be accept to remain accessible.

345
00:16:39,920 --> 00:16:43,920
Speaker 3: And then Anthropic came out. Dario issued this this blog

346
00:16:43,960 --> 00:16:48,560
Speaker 3: post saying basically like trying again doing the meta thing

347
00:16:48,600 --> 00:16:51,040
Speaker 3: where it's like, oh, we don't really want to ban TikTok.

348
00:16:51,080 --> 00:16:53,080
Speaker 3: We just think it's a national security thread and it

349
00:16:53,080 --> 00:16:55,600
Speaker 3: should be eliminated and probably we shouldn't be able to

350
00:16:55,600 --> 00:16:57,960
Speaker 3: download the United States and so like That's kind of

351
00:16:57,960 --> 00:16:59,960
Speaker 3: how his letter read to me. I don't know what

352
00:17:00,040 --> 00:17:01,720
Speaker 3: read thought, but it was like, oh no, no, no, we

353
00:17:02,080 --> 00:17:05,080
Speaker 3: don't mind open Wait we just think it's incredibly dangerous

354
00:17:05,119 --> 00:17:07,040
Speaker 3: and you know, I don't like to you know, that's

355
00:17:07,040 --> 00:17:08,359
Speaker 3: probably not worth the resk or whatever.

356
00:17:08,440 --> 00:17:10,600
Speaker 1: So read on Smiley, I want to hear your tape

357
00:17:10,600 --> 00:17:12,560
Speaker 1: of first. Can one of you help us under sound

358
00:17:12,560 --> 00:17:14,200
Speaker 1: difference between open source and open weight?

359
00:17:14,600 --> 00:17:16,879
Speaker 2: Yeah, I mean it's just a this is just a

360
00:17:16,920 --> 00:17:21,399
Speaker 2: semantics thing really because these models don't reveal their underlying

361
00:17:21,480 --> 00:17:24,640
Speaker 2: data sets. And so people in the open source community

362
00:17:24,640 --> 00:17:26,640
Speaker 2: took issue with that term because it's like, well, it's

363
00:17:26,640 --> 00:17:30,480
Speaker 2: not totally open, but you know, you can still for

364
00:17:30,520 --> 00:17:34,199
Speaker 2: all intents and purposes, they're open. I use yeah, I

365
00:17:34,320 --> 00:17:36,840
Speaker 2: use open source all the time because I think it

366
00:17:37,000 --> 00:17:39,080
Speaker 2: just confuses people to a lot of people to say

367
00:17:39,119 --> 00:17:41,840
Speaker 2: open weights. But but you know, there's a there's a

368
00:17:42,000 --> 00:17:44,480
Speaker 2: sort of debate there. There was a big debate about

369
00:17:44,480 --> 00:17:47,840
Speaker 2: this in the open source community. But no, I mean,

370
00:17:47,920 --> 00:17:51,399
Speaker 2: I basically agree with Taylor on this. I think the

371
00:17:51,480 --> 00:17:53,320
Speaker 2: one thing I would sort of take issue with is

372
00:17:53,359 --> 00:17:56,760
Speaker 2: like I actually think that anthropic people in Dario, like

373
00:17:56,800 --> 00:17:59,680
Speaker 2: they're like true believers and they really are scared of

374
00:18:00,640 --> 00:18:03,119
Speaker 2: like AI and what's going to happen if you know,

375
00:18:03,200 --> 00:18:05,760
Speaker 2: you have these open models that can have all the

376
00:18:05,800 --> 00:18:09,800
Speaker 2: guardrails and protection stripped away from them that eventually We're

377
00:18:09,800 --> 00:18:13,560
Speaker 2: not there yet, but eventually, I think pose like a

378
00:18:13,640 --> 00:18:17,800
Speaker 2: large amount of risk in cybersecurity or bioterrorism.

379
00:18:17,080 --> 00:18:20,159
Speaker 3: Doesn't it make us safer? Like, isn't the argument, like

380
00:18:20,200 --> 00:18:22,879
Speaker 3: the argument that I've seen from security people and Alex

381
00:18:22,960 --> 00:18:25,480
Speaker 3: Damos was talking about this sort of a long time

382
00:18:25,480 --> 00:18:30,360
Speaker 3: security guy at Metali. Actually it strengthens everyone to have

383
00:18:30,480 --> 00:18:33,040
Speaker 3: these open source models out there and available because it

384
00:18:33,080 --> 00:18:35,240
Speaker 3: allows I mean the Hugging Face incident is a perfect

385
00:18:35,240 --> 00:18:37,280
Speaker 3: example of this, Like when you have these guard bills

386
00:18:37,320 --> 00:18:39,040
Speaker 3: and you have these restrictions and we're only allowed to

387
00:18:39,119 --> 00:18:41,800
Speaker 3: use anthropic robe and ai basically like it actually made

388
00:18:41,800 --> 00:18:43,400
Speaker 3: it makes it more dangerous.

389
00:18:43,680 --> 00:18:48,320
Speaker 1: Yeah, the hugging Face example was when open Aiy's cyber

390
00:18:48,440 --> 00:18:51,800
Speaker 1: tool kind of went semi rogue in a non properly

391
00:18:51,920 --> 00:18:55,600
Speaker 1: send boxed environment and hecked hugging Face and Hugging Face

392
00:18:55,640 --> 00:18:58,040
Speaker 1: couldn't defend themselves within the US models because of the

393
00:18:58,040 --> 00:19:00,399
Speaker 1: god rails, and so they used a chiny models to

394
00:19:00,400 --> 00:19:02,040
Speaker 1: defend themselves, which we talked about last week.

395
00:19:02,160 --> 00:19:05,000
Speaker 2: Yeah, I mean that's where I would also agree with you,

396
00:19:05,040 --> 00:19:08,000
Speaker 2: like I just my only point was, like, my only

397
00:19:08,000 --> 00:19:10,400
Speaker 2: point of disagreement with you was, like, I actually think

398
00:19:10,440 --> 00:19:13,840
Speaker 2: it's not just some sort of cynical regulatory capture point.

399
00:19:13,920 --> 00:19:16,400
Speaker 2: I think they actually believe in this stuff. I don't

400
00:19:16,400 --> 00:19:19,440
Speaker 2: think their conclusion though, is always right. Like I agree

401
00:19:19,480 --> 00:19:23,040
Speaker 2: with you that, like you can't plug this dam. Like

402
00:19:23,119 --> 00:19:26,520
Speaker 2: these models, these open source models will you know, find

403
00:19:26,560 --> 00:19:29,280
Speaker 2: their way into the country. The Chinese will continue to

404
00:19:29,440 --> 00:19:33,160
Speaker 2: distill from American AI unless there's some new breakthrough that

405
00:19:33,800 --> 00:19:37,600
Speaker 2: you know, allows AI frontier models companies to stop that.

406
00:19:38,200 --> 00:19:40,919
Speaker 2: So yeah, I mean, I think the only answer is, like,

407
00:19:41,400 --> 00:19:45,040
Speaker 2: if these open source models are a threat to cybersecurity,

408
00:19:45,440 --> 00:19:49,800
Speaker 2: we better be plugging as many cybersecurity holes as we can, right, Like,

409
00:19:49,880 --> 00:19:54,080
Speaker 2: we have to protect against this reality rather than try to,

410
00:19:54,520 --> 00:19:57,199
Speaker 2: you know, I don't know, like plug the dam or

411
00:19:57,240 --> 00:19:58,920
Speaker 2: just bury our heads in the sand and hope that

412
00:19:59,040 --> 00:20:02,280
Speaker 2: somehow we can ban this and stop you know, technology

413
00:20:02,280 --> 00:20:03,240
Speaker 2: from marching forward.

414
00:20:03,680 --> 00:20:05,920
Speaker 3: I just want to say, though, like on the true

415
00:20:05,920 --> 00:20:09,280
Speaker 3: believer point, like I totally agree with you. I think

416
00:20:09,359 --> 00:20:11,439
Speaker 3: Dario and a lot of people they're effective ault as

417
00:20:11,480 --> 00:20:15,640
Speaker 3: true believers, but it is also very convenient for their

418
00:20:15,680 --> 00:20:19,639
Speaker 3: business interest And I do think that, like, you know,

419
00:20:19,880 --> 00:20:22,880
Speaker 3: in a lot of ways like Meta, where there were

420
00:20:22,920 --> 00:20:24,840
Speaker 3: true believers in the sense that they thought that it

421
00:20:24,880 --> 00:20:27,480
Speaker 3: was like a dangerous open internet and connecting the world

422
00:20:27,520 --> 00:20:30,240
Speaker 3: would be dangerous, and that's how it was like before Meta,

423
00:20:30,240 --> 00:20:32,440
Speaker 3: it was all sort of like chaotic and the chaos

424
00:20:32,480 --> 00:20:35,600
Speaker 3: of like forums and blogs, there's disinformation. So Meta's whole

425
00:20:35,600 --> 00:20:37,200
Speaker 3: thing was like, we're going to clean it up. We're

426
00:20:37,200 --> 00:20:40,119
Speaker 3: going to provide this standardized experience, like we're going to

427
00:20:40,119 --> 00:20:42,119
Speaker 3: be the one to connect the world and it's going

428
00:20:42,200 --> 00:20:44,520
Speaker 3: to be amazing. And I think part of it is,

429
00:20:44,560 --> 00:20:47,359
Speaker 3: like I mean, and maybe I'm more cynical about all

430
00:20:47,359 --> 00:20:49,560
Speaker 3: of it, but I do think there's this like paternalistic

431
00:20:49,720 --> 00:20:52,800
Speaker 3: power fantasy as well, where it's very much like I'm

432
00:20:52,880 --> 00:20:56,240
Speaker 3: the one that can bring artificial intelligence. I'm the responsible

433
00:20:56,280 --> 00:20:59,720
Speaker 3: one that can oversee it. And I just reject that,

434
00:21:00,440 --> 00:21:01,920
Speaker 3: especially from people like Dario.

435
00:21:02,440 --> 00:21:04,840
Speaker 2: There's a lot of Yeah, there's a lot of paternalistic

436
00:21:04,920 --> 00:21:08,600
Speaker 2: fantasies and tech for sure, I mean, and it is, yeah,

437
00:21:08,640 --> 00:21:10,760
Speaker 2: I mean, it's like trying to boil the ocean. Really,

438
00:21:11,000 --> 00:21:15,040
Speaker 2: it's just like no one person, no one company is

439
00:21:15,119 --> 00:21:18,280
Speaker 2: going to be in charge of safeguarding all of this right,

440
00:21:18,359 --> 00:21:20,400
Speaker 2: And I don't even think I mean, I think even

441
00:21:20,480 --> 00:21:24,760
Speaker 2: government is just like almost moving too fast. And I

442
00:21:24,840 --> 00:21:27,520
Speaker 2: think this partially is because like we don't legislate anymore,

443
00:21:27,560 --> 00:21:31,200
Speaker 2: we just write executive orders. But it's like the people

444
00:21:31,200 --> 00:21:36,600
Speaker 2: are talking about, you know, distillation being and being ip theft, right, Well,

445
00:21:36,720 --> 00:21:40,760
Speaker 2: that essentially means that like when these Chinese models do

446
00:21:40,840 --> 00:21:42,920
Speaker 2: make it to the US and people start using these

447
00:21:42,960 --> 00:21:46,680
Speaker 2: open source models, they won't be allowed to distill them,

448
00:21:46,720 --> 00:21:49,040
Speaker 2: which back to this research study, is how you get

449
00:21:49,119 --> 00:21:52,360
Speaker 2: rid of all this censorship which is real in these models,

450
00:21:52,359 --> 00:21:55,480
Speaker 2: like they are very biased toward you know, the CCP

451
00:21:55,880 --> 00:21:58,840
Speaker 2: point of view. So I mean you're like in all

452
00:21:58,880 --> 00:22:02,240
Speaker 2: these every move that you make, you're like cutting off

453
00:22:02,280 --> 00:22:04,480
Speaker 2: one of your options to deal with this problem, which

454
00:22:04,480 --> 00:22:06,879
Speaker 2: I think is a you know, is a problem.

455
00:22:06,640 --> 00:22:10,280
Speaker 3: Which could be solved by competing. Like I just I'm like,

456
00:22:10,440 --> 00:22:13,600
Speaker 3: why don't we actually try to compete. Why don't we

457
00:22:13,640 --> 00:22:16,480
Speaker 3: invest in research? Why don't we invest Like China has

458
00:22:16,520 --> 00:22:22,040
Speaker 3: this long term vision and has really facilitated innovation and

459
00:22:22,200 --> 00:22:25,879
Speaker 3: prioritize their you know, education system to build you know,

460
00:22:25,920 --> 00:22:29,040
Speaker 3: the skills necessary to build the next generation technology, and

461
00:22:29,119 --> 00:22:31,520
Speaker 3: I just feel like America, it's like, what are we doing?

462
00:22:31,560 --> 00:22:34,439
Speaker 3: We're making gambling apps? And what is Andreas and Horwitz funding?

463
00:22:34,440 --> 00:22:37,320
Speaker 3: Are they funding next generation technology? I don't know, you

464
00:22:37,400 --> 00:22:40,280
Speaker 3: know what I mean. I just like America feels like

465
00:22:41,240 --> 00:22:44,680
Speaker 3: just it's like rickety old cart that's like falling apart,

466
00:22:44,840 --> 00:22:46,639
Speaker 3: and I'm just like, is anyone going to fix it?

467
00:22:47,000 --> 00:22:49,399
Speaker 2: Well? I think that I think the rickety old cart

468
00:22:49,560 --> 00:22:53,320
Speaker 2: is is like actually fine, Like there's there's a America's

469
00:22:53,359 --> 00:22:56,000
Speaker 2: never been perfect, right, and it happens in that like

470
00:22:56,320 --> 00:22:59,720
Speaker 2: innovation happens in a very sort of like haphazard and

471
00:22:59,760 --> 00:23:03,360
Speaker 2: just rebated way. And my point is just get out

472
00:23:03,440 --> 00:23:03,840
Speaker 2: of the way.

473
00:23:04,080 --> 00:23:06,280
Speaker 1: There was a great book called Breakneck, which I'm not

474
00:23:06,320 --> 00:23:08,640
Speaker 1: sure if either you read. A big guy called Dan Wong,

475
00:23:08,720 --> 00:23:13,000
Speaker 1: and he basically counterposed the Chinese Communist Party with the

476
00:23:13,080 --> 00:23:16,760
Speaker 1: history of American presidents and basically as a huge preponderance

477
00:23:16,800 --> 00:23:19,760
Speaker 1: of engineers in the who became you know, supreme leaders

478
00:23:19,840 --> 00:23:23,960
Speaker 1: versus lawyers who became presidents. And you know, his point was,

479
00:23:24,000 --> 00:23:25,359
Speaker 1: if you live in China and you look out the

480
00:23:25,359 --> 00:23:27,280
Speaker 1: window and a year has gone by, you might see

481
00:23:27,280 --> 00:23:29,439
Speaker 1: a new city out of the window and if you

482
00:23:29,480 --> 00:23:31,200
Speaker 1: live in the US, like the best you can hope

483
00:23:31,240 --> 00:23:33,600
Speaker 1: for is a new coffee shop. And I don't know,

484
00:23:33,600 --> 00:23:37,359
Speaker 1: I think there is something. There's something to there's a

485
00:23:37,400 --> 00:23:40,040
Speaker 1: conversation which is bigger than the technology conversation, really about

486
00:23:40,040 --> 00:23:41,919
Speaker 1: like how do you, like, how do you create a

487
00:23:42,160 --> 00:23:46,119
Speaker 1: national culture that promotes innovation without but also without the

488
00:23:46,280 --> 00:23:47,240
Speaker 1: all of the bad things.

489
00:23:47,119 --> 00:23:48,040
Speaker 4: That happen in China.

490
00:23:48,240 --> 00:23:49,600
Speaker 1: Just before we go to the break there, I wanted

491
00:23:49,640 --> 00:23:51,800
Speaker 1: to ask you both. You know, we live in an

492
00:23:51,800 --> 00:23:54,159
Speaker 1: attention economy and we're all you know, all three of

493
00:23:54,240 --> 00:23:58,639
Speaker 1: us work in it. But it's very rare that a

494
00:23:58,720 --> 00:24:01,040
Speaker 1: story rumbles on for three weeks, Like why why is

495
00:24:01,080 --> 00:24:04,480
Speaker 1: everyone still talking about these Chinese open source models and distillations?

496
00:24:04,720 --> 00:24:07,200
Speaker 1: Is just because it's the behind the story is questions

497
00:24:07,240 --> 00:24:10,560
Speaker 1: about like the next open Ai ipo and the anthropic

498
00:24:10,600 --> 00:24:13,480
Speaker 1: ipo or is it about the China US Great Palace struggle?

499
00:24:13,520 --> 00:24:16,000
Speaker 1: Story is always fascinating, like why people still own this?

500
00:24:16,960 --> 00:24:21,240
Speaker 2: Well my one theory and okay, I'm kind I'm kind

501
00:24:21,240 --> 00:24:23,199
Speaker 2: of just going off the cuff here, but like I

502
00:24:23,200 --> 00:24:27,040
Speaker 2: think it's a great question. This AI conversation has sucked

503
00:24:27,119 --> 00:24:29,760
Speaker 2: all of our attention into it, right, It's become everything.

504
00:24:29,880 --> 00:24:33,040
Speaker 2: And so you know, when you have a controversy like

505
00:24:33,040 --> 00:24:37,000
Speaker 2: this and there isn't some major breakthrough that's drawing our

506
00:24:37,000 --> 00:24:39,720
Speaker 2: attention away, like we're going to focus on this stuff

507
00:24:39,760 --> 00:24:42,639
Speaker 2: for longer. But I think as soon as, like, you know,

508
00:24:42,760 --> 00:24:45,879
Speaker 2: we see some new corner turned and there's a shiny

509
00:24:45,960 --> 00:24:48,679
Speaker 2: new thing to go after, we will like that. That

510
00:24:48,720 --> 00:24:52,840
Speaker 2: will that will be the thing everybody talks about. Maybe

511
00:24:52,880 --> 00:24:56,480
Speaker 2: it'll happen later this year, maybe in January. It seems

512
00:24:56,480 --> 00:25:01,280
Speaker 2: like January is it just has some major breakthrough every year, right,

513
00:25:01,400 --> 00:25:03,600
Speaker 2: I wasn't thinking of c yes, but just like every

514
00:25:03,760 --> 00:25:05,880
Speaker 2: like January it was like it was like the reasoning

515
00:25:05,960 --> 00:25:08,359
Speaker 2: models happened, and then like last January it was like

516
00:25:08,480 --> 00:25:10,960
Speaker 2: all of a sudden, the harnesses were just working and

517
00:25:11,040 --> 00:25:13,960
Speaker 2: you could build, you know, build all this technology. I

518
00:25:13,960 --> 00:25:16,439
Speaker 2: think something like that's going to happen, and you know,

519
00:25:16,760 --> 00:25:18,119
Speaker 2: we'll forget about this for a while.

520
00:25:18,320 --> 00:25:21,160
Speaker 3: I hope so. But I feel I'm scared that we're

521
00:25:21,160 --> 00:25:23,320
Speaker 3: going to get distracted by that shiny new object and

522
00:25:24,359 --> 00:25:26,639
Speaker 3: fundamentally not fix all the issues. That sort of reads

523
00:25:26,720 --> 00:25:29,359
Speaker 3: mentioning this idea of not investing in research, Like I

524
00:25:29,400 --> 00:25:31,879
Speaker 3: do think the US has as a major problem with

525
00:25:31,960 --> 00:25:34,399
Speaker 3: long term vision and maybe that's how our political system

526
00:25:34,480 --> 00:25:37,399
Speaker 3: set up, I don't know, but or sort of our

527
00:25:37,440 --> 00:25:51,080
Speaker 3: economic system, but certainly it's it's a problem.

528
00:25:45,440 --> 00:25:46,280
Speaker 4: When we come back.

529
00:25:46,680 --> 00:25:50,040
Speaker 1: Substock has pushed back against a new AI detection feature.

530
00:25:50,440 --> 00:26:06,840
Speaker 1: Stay with us, Welcome back, Taylor. Subset recently launched a

531
00:26:06,920 --> 00:26:09,359
Speaker 1: feature on that app which acts like a kind of

532
00:26:09,400 --> 00:26:12,120
Speaker 1: AI detective fit us in.

533
00:26:12,600 --> 00:26:17,160
Speaker 3: Yeah, So, Substack partnered with this company called Pangram, which

534
00:26:17,200 --> 00:26:20,600
Speaker 3: is a startup that is basically they build themselves as

535
00:26:20,760 --> 00:26:23,600
Speaker 3: like the slop detector of the web, and so they

536
00:26:23,640 --> 00:26:28,800
Speaker 3: have probably the most sophisticated AI detection tools and they've

537
00:26:28,880 --> 00:26:32,080
Speaker 3: just actually they're working on visual AI detection tools in

538
00:26:32,160 --> 00:26:36,040
Speaker 3: terms of images, but they're really good at detecting AI

539
00:26:36,600 --> 00:26:40,760
Speaker 3: use in writing and so basically substack is trying to

540
00:26:40,800 --> 00:26:45,560
Speaker 3: make it easier to click and determine how much of

541
00:26:45,640 --> 00:26:49,640
Speaker 3: an article is written by AI or it's really good

542
00:26:49,640 --> 00:26:52,159
Speaker 3: at determining if one hundred percent of the article is

543
00:26:52,200 --> 00:26:55,680
Speaker 3: AI generated, which I think is valuable because I did

544
00:26:55,680 --> 00:26:58,159
Speaker 3: a partnership with Pangram myself actually where they gave me

545
00:26:58,200 --> 00:27:02,840
Speaker 3: access early access to their API last year or well

546
00:27:02,880 --> 00:27:04,640
Speaker 3: it wasn't even last year, it's like four months ago.

547
00:27:05,000 --> 00:27:08,960
Speaker 3: And tech moves fast, and you know, up to forty

548
00:27:09,040 --> 00:27:11,960
Speaker 3: percent of articles in the tech category, which is what

549
00:27:12,000 --> 00:27:12,800
Speaker 3: my newsletter is in.

550
00:27:13,480 --> 00:27:16,480
Speaker 4: We're fully AI generated, fully AI generated.

551
00:27:16,200 --> 00:27:18,920
Speaker 3: Fully AI generated, So it's really taking over.

552
00:27:19,280 --> 00:27:20,919
Speaker 4: Is this in response to consumer demand?

553
00:27:21,080 --> 00:27:23,560
Speaker 1: Is like this of a substac leadership concern that if

554
00:27:23,560 --> 00:27:26,200
Speaker 1: they let AI proliferate for too long on substack, ultimately

555
00:27:26,240 --> 00:27:28,119
Speaker 1: people will migrate away from substack Like where was it

556
00:27:28,240 --> 00:27:30,000
Speaker 1: was a demand signal kind of driving this?

557
00:27:30,280 --> 00:27:33,800
Speaker 3: You know, I don't know the total thinking behind it.

558
00:27:33,840 --> 00:27:35,480
Speaker 3: I did interview Chris Best, and.

559
00:27:36,080 --> 00:27:38,200
Speaker 4: What he said is like the CEO of subsc right.

560
00:27:38,200 --> 00:27:41,400
Speaker 3: The CEO of Substack, who said, you know, we want

561
00:27:41,440 --> 00:27:45,480
Speaker 3: consumers to basically have more information and know and like

562
00:27:45,520 --> 00:27:48,359
Speaker 3: this is our way of kind of not doing like

563
00:27:48,359 --> 00:27:50,680
Speaker 3: trust and safety, but kind of like making our platform

564
00:27:50,680 --> 00:27:53,719
Speaker 3: more transparent. I think there's a lot of scams and

565
00:27:53,800 --> 00:27:56,199
Speaker 3: spam kind of info. I mean, as I wrote in

566
00:27:56,200 --> 00:27:59,240
Speaker 3: my piece, like one of the most popular subsacts in

567
00:27:59,280 --> 00:28:03,240
Speaker 3: the past year was this like fake conversation between Elon

568
00:28:03,359 --> 00:28:07,200
Speaker 3: Musk and Keanu Reeves. It literally never happened, but it

569
00:28:07,240 --> 00:28:10,680
Speaker 3: was AI generated slop that went megaviral that people still

570
00:28:10,720 --> 00:28:13,440
Speaker 3: share to this day as if it was real. The

571
00:28:13,520 --> 00:28:19,439
Speaker 3: politics category is overrun with AI generated content, so I

572
00:28:19,440 --> 00:28:23,439
Speaker 3: think it's about giving consumers more transparency. I don't know

573
00:28:23,480 --> 00:28:25,280
Speaker 3: that it's totally led to that. I feel like it's

574
00:28:25,280 --> 00:28:29,280
Speaker 3: caused a bit of chaos, but that's certainly the goal.

575
00:28:29,800 --> 00:28:32,919
Speaker 4: So this is not just the detective, It's not X.

576
00:28:33,000 --> 00:28:35,359
Speaker 4: It's that this is more sophisticated.

577
00:28:35,640 --> 00:28:37,280
Speaker 1: Do you think I'm curious for both of your take

578
00:28:37,320 --> 00:28:40,640
Speaker 1: as people who write newsletters, like is all AI writing

579
00:28:40,720 --> 00:28:41,440
Speaker 1: AI slop?

580
00:28:42,920 --> 00:28:45,760
Speaker 3: I get a few AI generated newsletters that I really love.

581
00:28:46,840 --> 00:28:49,120
Speaker 3: There's like a social listening tool that I get the

582
00:28:49,120 --> 00:28:51,680
Speaker 3: newsletter up, and it basically gives me like a roundup

583
00:28:51,720 --> 00:28:56,280
Speaker 3: of trending posts and automatically sort of pulls yeah, like

584
00:28:56,360 --> 00:28:59,840
Speaker 3: trending content in I like Vibe coded my own one

585
00:29:00,120 --> 00:29:03,240
Speaker 3: that like it's mostly it's I trust AI for like

586
00:29:03,280 --> 00:29:06,720
Speaker 3: analysis and scraping, and it's honestly better than like a

587
00:29:06,720 --> 00:29:08,240
Speaker 3: lot of the ones that I subscribe to. I mean,

588
00:29:08,240 --> 00:29:10,200
Speaker 3: I subscribe to a lot of human newsletters as well,

589
00:29:10,200 --> 00:29:13,120
Speaker 3: but I think that like AI can pick up these

590
00:29:13,680 --> 00:29:18,720
Speaker 3: small detections, you know, in sort of trends. I At

591
00:29:18,720 --> 00:29:21,080
Speaker 3: the same time, I think it's all about transparency, right,

592
00:29:21,240 --> 00:29:23,880
Speaker 3: and what does the consumer think they're getting. Do they

593
00:29:23,880 --> 00:29:26,760
Speaker 3: think that they're getting one hundred percent human No, AI

594
00:29:27,120 --> 00:29:29,280
Speaker 3: used it all, but they're actually getting slop. Okay, that's

595
00:29:29,280 --> 00:29:32,600
Speaker 3: going to cause a negative experience. But I think increasingly

596
00:29:33,360 --> 00:29:35,320
Speaker 3: it's somewhere in the middle. And that's kind of where

597
00:29:35,360 --> 00:29:36,040
Speaker 3: things get tricky.

598
00:29:36,840 --> 00:29:38,920
Speaker 2: Yeah, I don't think it's all slop, but I do.

599
00:29:39,800 --> 00:29:41,480
Speaker 2: I don't know about you guys. I mean I can

600
00:29:41,480 --> 00:29:44,440
Speaker 2: tell when something's like AI generated.

601
00:29:44,760 --> 00:29:46,440
Speaker 3: I can to Most people can't.

602
00:29:46,680 --> 00:29:48,840
Speaker 2: Yeah, most people can't, but you get it. There's a

603
00:29:48,920 --> 00:29:52,520
Speaker 2: rhythm to it that sort of like becomes familiar, especially claude,

604
00:29:52,560 --> 00:29:56,040
Speaker 2: like if there's a specially exactly style and you read

605
00:29:56,400 --> 00:29:59,200
Speaker 2: X and sometimes I'm like, well, this is actually an

606
00:29:59,200 --> 00:30:01,960
Speaker 2: interesting point that this person's making. It's also just this

607
00:30:02,120 --> 00:30:04,880
Speaker 2: is just AI generated, and like, I don't know what

608
00:30:04,960 --> 00:30:05,640
Speaker 2: to think about it.

609
00:30:06,000 --> 00:30:08,959
Speaker 3: So one thing that I thought was interesting is this

610
00:30:09,000 --> 00:30:11,040
Speaker 3: person on ax has been doing and I want to

611
00:30:11,040 --> 00:30:15,200
Speaker 3: write about this myself and I haven't. But the big

612
00:30:15,200 --> 00:30:17,920
Speaker 3: thing on YouTube right now is AI generated scripts. So

613
00:30:18,080 --> 00:30:20,880
Speaker 3: Almost every big YouTuber is using AI to auto generate

614
00:30:20,920 --> 00:30:23,040
Speaker 3: their scripts. They're not even tweaking them. They're literally just

615
00:30:23,120 --> 00:30:25,360
Speaker 3: generating the script and reading it. But they're a human

616
00:30:25,840 --> 00:30:30,600
Speaker 3: reading it. So audiences, this person was tweeting out the

617
00:30:30,640 --> 00:30:33,000
Speaker 3: response to your audiences and what I've noticed as well

618
00:30:33,200 --> 00:30:36,320
Speaker 3: one hundred percent positive. They seem to prefer it. Actually

619
00:30:36,800 --> 00:30:42,480
Speaker 3: they are saying incredible analysis, phenomenal. Wow, I can't believe it.

620
00:30:42,720 --> 00:30:44,840
Speaker 2: And how do you know those aren't bots?

621
00:30:45,000 --> 00:30:47,360
Speaker 3: No, no, no, I trust me. I have gone deep

622
00:30:47,400 --> 00:30:49,840
Speaker 3: and I have used AI on one of my own

623
00:30:49,960 --> 00:30:53,240
Speaker 3: multiple of my own scripts, and people praise those AI

624
00:30:53,320 --> 00:30:56,600
Speaker 3: scripts far more than my human scripts. So and I

625
00:30:56,640 --> 00:30:59,040
Speaker 3: didn't use AI like one hundred percent, But like people

626
00:30:59,280 --> 00:31:03,440
Speaker 3: seem there's something about the AI cadence that, like normally

627
00:31:03,520 --> 00:31:06,280
Speaker 3: people online seem to really like. And if you look

628
00:31:06,320 --> 00:31:11,920
Speaker 3: at these like viral articles as well read I generated,

629
00:31:12,000 --> 00:31:14,760
Speaker 3: So there is something And I don't think most most

630
00:31:14,800 --> 00:31:16,920
Speaker 3: of those people do not realize that they're consuming AI.

631
00:31:17,000 --> 00:31:18,920
Speaker 3: And I think that was the issue with substack as well,

632
00:31:18,960 --> 00:31:22,160
Speaker 3: like and I think, but that's the thing they I

633
00:31:22,160 --> 00:31:24,719
Speaker 3: don't know that they want to know that it's they

634
00:31:24,720 --> 00:31:27,200
Speaker 3: get kind of they won't believe it, you know, when

635
00:31:27,240 --> 00:31:29,160
Speaker 3: you say, hey, this is one hundred percent AI generated,

636
00:31:29,200 --> 00:31:31,400
Speaker 3: like they kind of don't want to accept that.

637
00:31:31,840 --> 00:31:35,200
Speaker 1: Did you follow the moment with the Canadian politician this week, Yes,

638
00:31:35,880 --> 00:31:38,400
Speaker 1: here's a more natural, flowing version of that section that

639
00:31:38,480 --> 00:31:40,480
Speaker 1: reads like legislative speech rather than.

640
00:31:40,400 --> 00:31:42,920
Speaker 4: A series of short points. Madam speaker. One of my

641
00:31:43,000 --> 00:31:43,880
Speaker 4: concerns who.

642
00:31:43,680 --> 00:31:47,480
Speaker 1: Went viral reading the instructions from Clode a Loud in

643
00:31:47,480 --> 00:31:48,440
Speaker 1: his political speech.

644
00:31:48,960 --> 00:31:50,160
Speaker 4: I mean, but it's funny.

645
00:31:50,200 --> 00:31:51,600
Speaker 1: I think it's one of those things where it's like,

646
00:31:52,520 --> 00:31:54,480
Speaker 1: oh my god, he was caught with his pants down.

647
00:31:54,800 --> 00:31:57,840
Speaker 1: But on the other hand, like probably ninety nine point

648
00:31:57,880 --> 00:32:01,240
Speaker 1: nine percent of politicians used AI to enhance their speeches

649
00:32:01,520 --> 00:32:03,640
Speaker 1: or even write them, So it's a funny, like funny

650
00:32:03,680 --> 00:32:07,720
Speaker 1: that people were so like panicked or like captivated. I

651
00:32:07,720 --> 00:32:09,960
Speaker 1: guess it's very embarrassing for him, but it was also like.

652
00:32:10,000 --> 00:32:12,680
Speaker 3: Well, it's also like, is AI writing our policy? Like

653
00:32:13,040 --> 00:32:14,880
Speaker 3: this is my stance on all of it. It's like

654
00:32:15,000 --> 00:32:19,400
Speaker 3: I think that by doing like, I think that there's

655
00:32:19,440 --> 00:32:22,440
Speaker 3: not enough transparency. I think there's not enough discussion. I mean,

656
00:32:22,480 --> 00:32:25,000
Speaker 3: I did a Q and A with Substack about this yesterday,

657
00:32:25,080 --> 00:32:28,040
Speaker 3: Like I think we Joe Wisenthal had a good tweet

658
00:32:28,080 --> 00:32:30,000
Speaker 3: a while ago. A while ago he was basically like,

659
00:32:30,480 --> 00:32:32,680
Speaker 3: this is all happening, and we're not going to put

660
00:32:32,760 --> 00:32:34,560
Speaker 3: the genie back in the bottle. We need to accept

661
00:32:34,600 --> 00:32:36,280
Speaker 3: that AI is being used in all these ways and

662
00:32:36,320 --> 00:32:39,040
Speaker 3: we should have discussions about sort of what that means.

663
00:32:39,600 --> 00:32:41,840
Speaker 3: And I think that's much better than the sort of

664
00:32:41,840 --> 00:32:44,160
Speaker 3: witch hunting, which I also see happening and I think

665
00:32:44,160 --> 00:32:46,200
Speaker 3: has sort of happened with Pan Graham now, where it's

666
00:32:46,200 --> 00:32:47,959
Speaker 3: like people just go around like, oh my god, oh

667
00:32:47,960 --> 00:32:49,520
Speaker 3: my god, this AI was used to I was used

668
00:32:49,520 --> 00:32:51,200
Speaker 3: and I was like, okay, but what does it matter

669
00:32:51,240 --> 00:32:53,240
Speaker 3: if AI was used for this like niche thing, you

670
00:32:53,280 --> 00:32:55,280
Speaker 3: know what I mean, Like, let's have a real discussion

671
00:32:55,280 --> 00:32:55,640
Speaker 3: about it.

672
00:32:55,880 --> 00:32:57,760
Speaker 4: How it creators On substeck reacting to.

673
00:32:58,240 --> 00:33:02,800
Speaker 3: Pangram, it runs the gamute. Some people are like so

674
00:33:03,040 --> 00:33:07,080
Speaker 3: mad because again they use AI and they weren't disclosing it.

675
00:33:08,720 --> 00:33:10,520
Speaker 3: I think a lot of consumers, it really depends on

676
00:33:10,560 --> 00:33:13,440
Speaker 3: your audience. You know, ironically, no one really cares about

677
00:33:13,440 --> 00:33:16,800
Speaker 3: the tech, like the tech the tech newsletterspace like tramas

678
00:33:16,880 --> 00:33:20,520
Speaker 3: posted like that full AI generated like you know, article

679
00:33:20,600 --> 00:33:23,000
Speaker 3: basically on X everyone knew it was AI generated and

680
00:33:23,000 --> 00:33:25,120
Speaker 3: Elon Musk was a great article. Like people were like,

681
00:33:25,160 --> 00:33:27,160
Speaker 3: as kind of as read said, people were like, Okay,

682
00:33:27,560 --> 00:33:30,120
Speaker 3: Tamath is colo siding this analysis, and so I find

683
00:33:30,120 --> 00:33:33,280
Speaker 3: it valuable. So but people in the politics category, and

684
00:33:33,320 --> 00:33:36,920
Speaker 3: it's more like the lifestyle things like food and sort

685
00:33:36,960 --> 00:33:39,479
Speaker 3: of lifestyle influencers that have been relying on AI, like

686
00:33:39,680 --> 00:33:42,200
Speaker 3: their audiences are not taking kindly to it. And I

687
00:33:42,600 --> 00:33:46,160
Speaker 3: have a very left leaning audience that rulently hates AI,

688
00:33:46,720 --> 00:33:48,680
Speaker 3: so I've been tried to be very transparent with them.

689
00:33:48,760 --> 00:33:50,760
Speaker 4: Is that a point of tension between you and your audience.

690
00:33:51,240 --> 00:33:54,640
Speaker 3: Yeah, I've lost like a significant amount of subscribers just

691
00:33:54,640 --> 00:33:57,600
Speaker 3: by saying, hey, I'm a tech reporter, so I'm going

692
00:33:57,680 --> 00:34:00,000
Speaker 3: to use these tools for research, and I will alway

693
00:34:00,000 --> 00:34:03,400
Speaker 3: always write my articles like human but I will always

694
00:34:03,440 --> 00:34:05,680
Speaker 3: be transparent. But just so you know, I report on

695
00:34:05,720 --> 00:34:09,000
Speaker 3: technology that means using technology, and this has gotten me

696
00:34:09,040 --> 00:34:10,600
Speaker 3: canceled on Blue Sky multiple times.

697
00:34:11,760 --> 00:34:13,719
Speaker 2: I think in the in the end, though, this is

698
00:34:14,080 --> 00:34:17,520
Speaker 2: in some sense positive because there's always been sort of

699
00:34:17,600 --> 00:34:21,600
Speaker 2: an inauthentic nature to you know, social media to politics,

700
00:34:22,160 --> 00:34:25,040
Speaker 2: and I think this is actually driving people more towards

701
00:34:25,160 --> 00:34:28,799
Speaker 2: like a yearning for authenticity, and that's like part of

702
00:34:28,840 --> 00:34:31,840
Speaker 2: what the anti a AI hate is, right, And we

703
00:34:31,880 --> 00:34:35,040
Speaker 2: saw this with the latest you know, Christopher Nolan movie, right,

704
00:34:35,080 --> 00:34:39,319
Speaker 2: who's sort of rejected CGI and filmed this movie in

705
00:34:39,440 --> 00:34:42,719
Speaker 2: seventy millimeter IMAX like that, and people, I still can't

706
00:34:42,760 --> 00:34:45,920
Speaker 2: get tickets to the seventy millimeter right for months, And

707
00:34:45,960 --> 00:34:49,160
Speaker 2: it's like, you know, I think that's I sort of

708
00:34:49,200 --> 00:34:51,760
Speaker 2: think that's a positive, Like this is this is moving

709
00:34:51,880 --> 00:34:55,440
Speaker 2: us in the right direction for humanity. It's it's and

710
00:34:55,480 --> 00:34:57,799
Speaker 2: it's and it's like I even sort of like the

711
00:34:57,840 --> 00:35:01,640
Speaker 2: fact that even if something's AIG generated, like people can say, well,

712
00:35:01,960 --> 00:35:04,840
Speaker 2: the ideas behind this and the messenger here, I know,

713
00:35:05,480 --> 00:35:07,400
Speaker 2: and I'm gonna take it at face value. I'm not

714
00:35:07,440 --> 00:35:09,719
Speaker 2: going to worry about like what tools were used to

715
00:35:09,760 --> 00:35:11,920
Speaker 2: generate this. I'm gonna actually think about it as long

716
00:35:11,920 --> 00:35:15,440
Speaker 2: as there are people who are thinking critically, like in

717
00:35:15,560 --> 00:35:19,000
Speaker 2: policy and positions of power. That's good read.

718
00:35:19,000 --> 00:35:21,360
Speaker 1: Do you think your sem how would your semaphoe audience

719
00:35:21,480 --> 00:35:25,160
Speaker 1: feel if if you said, Okay, starting next week, I'm

720
00:35:25,160 --> 00:35:27,360
Speaker 1: going to use AI to as like as like a

721
00:35:27,400 --> 00:35:29,120
Speaker 1: full partner to write this newsletter.

722
00:35:30,160 --> 00:35:32,640
Speaker 2: I don't know. I don't think people would really care

723
00:35:32,680 --> 00:35:35,279
Speaker 2: as long as it's ultimately like I'm signing off on

724
00:35:35,480 --> 00:35:39,520
Speaker 2: every part of it, right, the writing of the writing

725
00:35:39,600 --> 00:35:42,200
Speaker 2: part of journalism is like a tiny part of journalism.

726
00:35:42,239 --> 00:35:46,040
Speaker 2: Like I actually would love if AI could just take

727
00:35:46,600 --> 00:35:50,239
Speaker 2: all of my interviews and my notes and my ideas

728
00:35:50,400 --> 00:35:54,279
Speaker 2: or you know, maybe my brains my verbal brainstorming to

729
00:35:54,520 --> 00:35:57,600
Speaker 2: my AI agent and just make like an amazing article

730
00:35:57,680 --> 00:36:00,480
Speaker 2: that's you know, perfectly edited. Like that would be great.

731
00:36:00,520 --> 00:36:03,680
Speaker 2: They'd be huge time saver, right, And I think, you know,

732
00:36:03,800 --> 00:36:05,960
Speaker 2: but you have to think critically about it, like you

733
00:36:05,960 --> 00:36:09,320
Speaker 2: you can't just you know, put stuff out there without

734
00:36:09,600 --> 00:36:13,120
Speaker 2: without like actually going through. It's not good enough yet, right,

735
00:36:13,200 --> 00:36:15,960
Speaker 2: Like I'm sure Taylor agrees. Like I've tried, you know,

736
00:36:16,000 --> 00:36:20,080
Speaker 2: I'm constantly trying. And it's good at some stuff, like

737
00:36:20,120 --> 00:36:22,439
Speaker 2: we use it for copy editing, et cetera. Like it's

738
00:36:22,520 --> 00:36:24,840
Speaker 2: it could be a time saver, but like it's not

739
00:36:24,960 --> 00:36:29,280
Speaker 2: there yet. It can't it can't generate you know, full articles.

740
00:36:29,600 --> 00:36:31,680
Speaker 3: I think AI is also very good at sort of

741
00:36:31,760 --> 00:36:34,399
Speaker 3: optimizing for AI, And I think, like when I talk

742
00:36:34,440 --> 00:36:37,640
Speaker 3: to YouTubers, about why so many YouTubers use it for scripts.

743
00:36:37,880 --> 00:36:40,319
Speaker 3: They're like, well, because the algorithm I'm playing for the

744
00:36:40,320 --> 00:36:44,720
Speaker 3: YouTube algorithm, the algorithm is ingesting my script and making

745
00:36:44,760 --> 00:36:47,359
Speaker 3: a determination about it. And if I say optimize it,

746
00:36:47,400 --> 00:36:51,120
Speaker 3: include these keywords include this cay to like, do include

747
00:36:51,160 --> 00:36:53,480
Speaker 3: these things that perform well for this algorithm, Like I

748
00:36:53,719 --> 00:36:57,320
Speaker 3: I think that's where it, you know, shines.

749
00:36:57,360 --> 00:36:59,040
Speaker 4: I guess it's so interesting.

750
00:36:59,080 --> 00:37:00,759
Speaker 1: I mean it shows up when you look at when

751
00:37:00,760 --> 00:37:03,840
Speaker 1: you look at like the top performing stuff on YouTube,

752
00:37:03,880 --> 00:37:06,440
Speaker 1: by the way the headlines are done, and the and

753
00:37:06,560 --> 00:37:10,040
Speaker 1: the thumbnails and the titling and like the kind of

754
00:37:10,040 --> 00:37:13,280
Speaker 1: like scrunged express like curious expression and stuff like. It's

755
00:37:13,560 --> 00:37:15,799
Speaker 1: it does feel like as I'm not a I have

756
00:37:15,840 --> 00:37:18,560
Speaker 1: to confess to I'm not like a super user review too,

757
00:37:18,600 --> 00:37:20,560
Speaker 1: but like when I look at those thumbnails, I'm like, wow,

758
00:37:20,600 --> 00:37:23,360
Speaker 1: this is like this has been so optimized.

759
00:37:22,719 --> 00:37:24,400
Speaker 4: For something that I am not. It's kind of an

760
00:37:24,400 --> 00:37:25,440
Speaker 4: interesting experience.

761
00:37:26,200 --> 00:37:28,960
Speaker 2: Well, if you think about the way that these models

762
00:37:29,000 --> 00:37:32,600
Speaker 2: were built with reinforcement learning, they had you know, thousands

763
00:37:32,640 --> 00:37:35,520
Speaker 2: of people sitting there just clicking. I like this version

764
00:37:35,640 --> 00:37:38,279
Speaker 2: more than that version. You know, they put a lot

765
00:37:38,320 --> 00:37:41,759
Speaker 2: of money into essentially like finding the common denominator of

766
00:37:41,880 --> 00:37:45,040
Speaker 2: reader listeners. So it sort of makes sense that they

767
00:37:45,080 --> 00:37:48,600
Speaker 2: prefer the AI generated content, which I think it's back

768
00:37:48,640 --> 00:37:51,480
Speaker 2: to this like thing that a lot of journalists and

769
00:37:51,480 --> 00:37:54,960
Speaker 2: and maybe even creators forget, which is like you're you're

770
00:37:55,000 --> 00:37:58,120
Speaker 2: writing this for the reader, not for yourself, right, And

771
00:37:58,160 --> 00:38:01,919
Speaker 2: I think another thing people get really upset about that's

772
00:38:01,920 --> 00:38:05,000
Speaker 2: not AI is when like you just take forever to

773
00:38:05,040 --> 00:38:07,880
Speaker 2: get to the point, and you know, there's certain publications

774
00:38:07,920 --> 00:38:10,560
Speaker 2: that are really bad about this, and like I just

775
00:38:10,680 --> 00:38:13,240
Speaker 2: end up skipping the first three paragraphs of every article

776
00:38:13,280 --> 00:38:16,240
Speaker 2: they write because it's all just you know, basically written

777
00:38:16,280 --> 00:38:19,160
Speaker 2: for the satisfaction of the writer, not for the reader.

778
00:38:19,200 --> 00:38:21,600
Speaker 2: And so we could probably all learn something from me.

779
00:38:21,840 --> 00:38:23,799
Speaker 1: With that in mind, I think we better wrap up

780
00:38:23,840 --> 00:38:25,200
Speaker 1: this episode, but I want to get I want to

781
00:38:25,200 --> 00:38:27,760
Speaker 1: get the last word to Taylor, because I saw you

782
00:38:27,760 --> 00:38:31,160
Speaker 1: you nodding about read saying that writing's only a small

783
00:38:31,200 --> 00:38:34,000
Speaker 1: fraction of journalism, like what's your what's your final word

784
00:38:34,040 --> 00:38:34,239
Speaker 1: on that?

785
00:38:34,880 --> 00:38:39,200
Speaker 3: So I you know, I'm a bad writer, I'm dyslexic.

786
00:38:39,360 --> 00:38:42,200
Speaker 3: I'm a much better writer now than I was. But

787
00:38:42,400 --> 00:38:46,319
Speaker 3: a huge reason, I mean, it was like it was

788
00:38:46,360 --> 00:38:48,720
Speaker 3: really hard for me to even consider being a reporter

789
00:38:48,760 --> 00:38:49,879
Speaker 3: because I thought you had to be a good writer.

790
00:38:49,960 --> 00:38:52,800
Speaker 3: And this really well known editor told me at one point,

791
00:38:53,520 --> 00:38:56,000
Speaker 3: you know, basically there are journalists like the job of

792
00:38:56,000 --> 00:38:59,200
Speaker 3: a journalist is to expose new information, and no one

793
00:38:59,280 --> 00:39:01,880
Speaker 3: will care if you're a bad writer if you can

794
00:39:01,920 --> 00:39:05,120
Speaker 3: get scoops. Like, if you can expose new information, editors

795
00:39:05,120 --> 00:39:07,480
Speaker 3: will hire you no matter what because they'll rewrite your

796
00:39:07,480 --> 00:39:10,680
Speaker 3: stuff and it's fine. And I started to do that

797
00:39:10,840 --> 00:39:13,560
Speaker 3: and that's what got me hired in media, and my

798
00:39:13,600 --> 00:39:15,640
Speaker 3: writing got better over the years, and I think it's

799
00:39:15,680 --> 00:39:17,719
Speaker 3: like good now. But I'm never going to write for

800
00:39:17,760 --> 00:39:20,759
Speaker 3: the like the New Yorker long Form or some you know,

801
00:39:20,800 --> 00:39:24,000
Speaker 3: Like I'm not really that kind of journalist. And there

802
00:39:24,000 --> 00:39:26,359
Speaker 3: are journalists that are amazing writers, right, But I think,

803
00:39:26,440 --> 00:39:28,960
Speaker 3: like what Reid said, I feel the same way. I'm like,

804
00:39:29,000 --> 00:39:30,759
Speaker 3: I wish I could just feed all this stuff in

805
00:39:31,239 --> 00:39:33,640
Speaker 3: have the AI kind of put it together. The goal

806
00:39:33,800 --> 00:39:36,160
Speaker 3: to me is to like get that new information out

807
00:39:36,200 --> 00:39:41,120
Speaker 3: there and kind of let that inform inform the world.

808
00:39:41,560 --> 00:39:44,400
Speaker 1: I feel that's always the deepest cut against the Stannyokagen

809
00:39:44,520 --> 00:39:47,160
Speaker 1: that still hear like rumors that they always hand in

810
00:39:47,200 --> 00:39:51,000
Speaker 1: the shoddiest copy and like the most successful ones to

811
00:39:51,080 --> 00:39:53,040
Speaker 1: do the least good job of handing in good drafts.

812
00:39:53,080 --> 00:39:54,839
Speaker 1: But maybe that's maybe that's what you're getting.

813
00:39:55,000 --> 00:39:57,600
Speaker 2: I am just maybe you can still give Taylor the

814
00:39:57,680 --> 00:39:59,239
Speaker 2: last word, and you can edit this out if you

815
00:39:59,280 --> 00:40:02,359
Speaker 2: want to. But I I wish more journalists would take

816
00:40:02,400 --> 00:40:05,919
Speaker 2: their article before sending it to their editor and run

817
00:40:05,960 --> 00:40:08,399
Speaker 2: it through whatever their favorite chat butt is and say

818
00:40:08,400 --> 00:40:10,360
Speaker 2: what am I missing here? What are my blind spots?

819
00:40:10,400 --> 00:40:13,399
Speaker 2: Where are my biases? And like, actually take that because

820
00:40:13,440 --> 00:40:15,360
Speaker 2: that is I find super helpful.

821
00:40:15,600 --> 00:40:17,920
Speaker 3: Totally. I think, like I mean, I just think like

822
00:40:18,880 --> 00:40:21,440
Speaker 3: we should embrace these tools if they can help us

823
00:40:21,640 --> 00:40:25,279
Speaker 3: improve our jobs. And also we need to understand these

824
00:40:25,320 --> 00:40:28,160
Speaker 3: tools to report on them, especially as technology reporters. And

825
00:40:28,200 --> 00:40:31,319
Speaker 3: I said this in my interview with Substack, but there's

826
00:40:31,360 --> 00:40:34,640
Speaker 3: a lot of people that I see, especially online, that

827
00:40:35,080 --> 00:40:39,440
Speaker 3: are very performative about AI. I will say, and I

828
00:40:39,520 --> 00:40:41,520
Speaker 3: know that they're using it in their personal lives. I

829
00:40:41,560 --> 00:40:43,600
Speaker 3: know that they're using it and they're getting online all

830
00:40:43,680 --> 00:40:46,640
Speaker 3: day and saying all this slop. And I just I

831
00:40:47,080 --> 00:40:49,080
Speaker 3: to me, that bothers me because I'm like let's just

832
00:40:49,120 --> 00:40:51,080
Speaker 3: be on and say that you use it sometimes it's

833
00:40:51,120 --> 00:40:53,600
Speaker 3: not the n to the world, Like let's have a

834
00:40:53,640 --> 00:40:55,719
Speaker 3: discussion about it, but don't just get online all day

835
00:40:55,719 --> 00:40:58,320
Speaker 3: and stoke the sort of fires of the AI hate

836
00:40:58,360 --> 00:41:00,520
Speaker 3: and then like sort of quietly try already lose it

837
00:41:00,560 --> 00:41:03,799
Speaker 3: on the side, like that's silly.

838
00:41:04,400 --> 00:41:07,920
Speaker 4: That's all we have time for this week, Taylor read, thank.

839
00:41:07,760 --> 00:41:10,480
Speaker 2: You great to be here as always, thanks for having us.

840
00:41:20,239 --> 00:41:20,920
Speaker 4: For tech stuff.

841
00:41:21,040 --> 00:41:24,360
Speaker 1: I'm mos Voloshian. This episode was produced by Eliza Dennis.

842
00:41:24,800 --> 00:41:27,520
Speaker 1: It was executive produced by me and Julian Nutter The

843
00:41:27,600 --> 00:41:32,400
Speaker 1: Kaleidoscope and Katrina norvelve iHeart Podcasts. Jack intlument to this

844
00:41:32,520 --> 00:41:35,840
Speaker 1: episode and Kyle Murdoch wrote our theme song. A special

845
00:41:35,880 --> 00:41:39,359
Speaker 1: thank you to Taylor Lorenz and Read Albergotti. Please check

846
00:41:39,360 --> 00:41:41,280
Speaker 1: out all the work be put out into the world.

847
00:41:41,560 --> 00:41:43,400
Speaker 1: We're lucky to call them friends of the Pod.