1
00:00:02,720 --> 00:00:16,520
Speaker 1: Bloomberg Audio Studios, Podcasts, Radio News.

2
00:00:18,040 --> 00:00:21,160
Speaker 2: Hello and welcome to another episode of The Odd Laws podcast.

3
00:00:21,200 --> 00:00:23,520
Speaker 3: I'm Joe Wisentholm and I'm Tracy Alloway.

4
00:00:23,840 --> 00:00:26,120
Speaker 2: You know, obviously we've been talking about I think twenty

5
00:00:26,160 --> 00:00:28,200
Speaker 2: twenty eight is going to be a big election for AI.

6
00:00:28,400 --> 00:00:30,960
Speaker 2: Really twenty twenty six actually now, you know, it's not

7
00:00:31,000 --> 00:00:33,960
Speaker 2: really even a prediction, This is just a description of fact.

8
00:00:34,320 --> 00:00:36,480
Speaker 2: AI is going to be very big for politics.

9
00:00:36,640 --> 00:00:39,320
Speaker 3: Yeah, I think it's inescapable. At this point, AI is

10
00:00:39,360 --> 00:00:41,960
Speaker 3: sort of dominating the news cycle as well. I know

11
00:00:42,040 --> 00:00:45,160
Speaker 3: it's a running joke on the podcast, but every time

12
00:00:45,240 --> 00:00:50,360
Speaker 3: we record an AI related episode, another headline hits the

13
00:00:50,479 --> 00:00:53,840
Speaker 3: terminal about you know, some new billion dollar.

14
00:00:54,120 --> 00:00:56,400
Speaker 2: One that we just got Disney.

15
00:00:56,080 --> 00:00:59,960
Speaker 3: To make one billion dollar equity investment in open AI.

16
00:01:00,600 --> 00:01:03,520
Speaker 2: Literally every time we do an episode, especially about air,

17
00:01:03,600 --> 00:01:06,759
Speaker 2: but even other there's some headline about a new investment

18
00:01:06,800 --> 00:01:09,160
Speaker 2: which just goes to show but you know the fact

19
00:01:09,200 --> 00:01:11,120
Speaker 2: that it's going to dominate politics as the least a

20
00:01:11,160 --> 00:01:14,679
Speaker 2: prizing thing ever, because yeah, it touches on everything anything

21
00:01:14,720 --> 00:01:18,320
Speaker 2: that's politically sensitive. The labor market, right, that's obvious. A

22
00:01:18,360 --> 00:01:24,160
Speaker 2: big one. Electricity, costs, water, consumption, wealth and inequality and

23
00:01:24,200 --> 00:01:26,840
Speaker 2: who has the power, and who's accumulating more and more

24
00:01:26,880 --> 00:01:28,840
Speaker 2: money in the tech industry, and who's not all this

25
00:01:28,880 --> 00:01:32,479
Speaker 2: thing like, there is hardly a political topic that in

26
00:01:32,520 --> 00:01:36,040
Speaker 2: some way I feel like AI does not exacerbate some way.

27
00:01:36,440 --> 00:01:39,880
Speaker 3: Absolutely. Meanwhile, the other big news politically when it comes

28
00:01:39,880 --> 00:01:42,920
Speaker 3: to AI this week is that Trump issued an executive

29
00:01:43,040 --> 00:01:46,280
Speaker 3: order for a national rule on AI.

30
00:01:46,600 --> 00:01:47,800
Speaker 4: Yeah, which a.

31
00:01:47,760 --> 00:01:50,800
Speaker 3: Lot of people who are trying to safeguard the system

32
00:01:50,880 --> 00:01:54,400
Speaker 3: and protect data privacy rights that sort of thing do

33
00:01:54,560 --> 00:01:55,000
Speaker 3: not want.

34
00:01:55,520 --> 00:01:58,040
Speaker 2: Well, this is the other thing. So it's all this

35
00:01:58,120 --> 00:02:02,440
Speaker 2: anxiety for various reasons. Pick your poison, whether it's going

36
00:02:02,520 --> 00:02:05,360
Speaker 2: to displace the jobs, whether AI is going to be

37
00:02:05,400 --> 00:02:08,440
Speaker 2: too woke and come up with versions of history that

38
00:02:08,639 --> 00:02:12,000
Speaker 2: people don't like, whether it's gonna, you know, just rot

39
00:02:12,040 --> 00:02:14,079
Speaker 2: our minds with slop, or you know, whether it's gonna

40
00:02:14,080 --> 00:02:15,720
Speaker 2: turn us all the paper clips. So you say, okay,

41
00:02:15,720 --> 00:02:17,680
Speaker 2: we should regulate, but what does that even look like like?

42
00:02:17,720 --> 00:02:22,760
Speaker 2: What is a sort of productive regulation look like such

43
00:02:22,840 --> 00:02:26,920
Speaker 2: that Okay, hopefully we could maintain positive aspects of the

44
00:02:26,960 --> 00:02:32,280
Speaker 2: technological development while capping the downsides, right, that would be nice, Well,

45
00:02:32,320 --> 00:02:34,720
Speaker 2: love that but like, how exactly do you do that.

46
00:02:34,800 --> 00:02:38,120
Speaker 3: The argument that you hear from AI proponents is that

47
00:02:38,160 --> 00:02:41,799
Speaker 3: any regulation needs to balance you know, safety with innovation

48
00:02:42,080 --> 00:02:45,359
Speaker 3: because there's also the question of China, which that's another

49
00:02:45,520 --> 00:02:49,600
Speaker 3: hot button political topic, right, this idea that the US

50
00:02:49,720 --> 00:02:53,840
Speaker 3: is in an existential battle with China over AI and

51
00:02:53,880 --> 00:02:57,040
Speaker 3: we have to win at all costs. Therefore the industry

52
00:02:57,120 --> 00:02:58,880
Speaker 3: must not be tightly regulated.

53
00:02:59,000 --> 00:03:02,680
Speaker 2: Yeah, right, So it's also a geopolitical element anyway, So

54
00:03:02,720 --> 00:03:05,560
Speaker 2: we mentioned AI is going to being in politics. We

55
00:03:05,680 --> 00:03:08,400
Speaker 2: really do have the perfect guest because we're going to

56
00:03:08,440 --> 00:03:11,680
Speaker 2: be speaking to someone that the AI industry is actually targeting,

57
00:03:11,880 --> 00:03:15,520
Speaker 2: someone that the AI industry is actively trying to stop.

58
00:03:15,639 --> 00:03:18,639
Speaker 2: There's this new super pack and I think it has

59
00:03:18,720 --> 00:03:21,560
Speaker 2: some intries and money and maybe a little some executives

60
00:03:21,600 --> 00:03:24,160
Speaker 2: from Open AI, et cetera. Anyway, they want to, like,

61
00:03:24,240 --> 00:03:27,440
Speaker 2: you know, make sure that sort of aisympathetic politicians do well.

62
00:03:27,520 --> 00:03:30,280
Speaker 2: But they're also targeting politicians that they perceived to be

63
00:03:30,600 --> 00:03:34,240
Speaker 2: too negative. And they've even named names, which I think

64
00:03:34,360 --> 00:03:36,400
Speaker 2: is super interesting, and we have them. So we are

65
00:03:36,440 --> 00:03:39,120
Speaker 2: going to be speaking with Alex Boris. He's a state

66
00:03:39,160 --> 00:03:42,320
Speaker 2: assembly member here in New York. He's also a candidate

67
00:03:42,400 --> 00:03:44,600
Speaker 2: in the primary for the twelfth district, which is here

68
00:03:44,600 --> 00:03:47,000
Speaker 2: in Manhattan, which I feel like a thousand people are

69
00:03:47,000 --> 00:03:50,400
Speaker 2: going to be running for this open seat. I feel like,

70
00:03:50,520 --> 00:03:52,680
Speaker 2: if you were targeted by the AA industry, that is

71
00:03:52,680 --> 00:03:54,800
Speaker 2: probably some of the best advertiser you could get in

72
00:03:54,840 --> 00:03:57,120
Speaker 2: a very crowded field where it's just like there's going

73
00:03:57,160 --> 00:03:58,080
Speaker 2: to be a million choices.

74
00:03:58,360 --> 00:04:00,720
Speaker 3: Also, yeah, I was gonna say what you're going to

75
00:04:00,800 --> 00:04:02,680
Speaker 3: say next, I think, but he used to work at

76
00:04:02,760 --> 00:04:06,400
Speaker 3: Palenteer as a data scientist, and Palenteer is one of

77
00:04:06,440 --> 00:04:08,600
Speaker 3: the companies backing that super pack.

78
00:04:08,880 --> 00:04:10,360
Speaker 2: So you know what I think we should do. Can

79
00:04:10,400 --> 00:04:12,400
Speaker 2: we talk about aar for like five minutes and then

80
00:04:12,440 --> 00:04:14,960
Speaker 2: make this a fifty minute episode about what Palenteer does.

81
00:04:16,160 --> 00:04:17,360
Speaker 2: That's my goal for this.

82
00:04:17,520 --> 00:04:19,000
Speaker 3: Why don't we just make it a series?

83
00:04:19,480 --> 00:04:22,200
Speaker 2: Make it a serious anyway, Alex Boris, thank you so

84
00:04:22,279 --> 00:04:23,760
Speaker 2: much for coming on odd Laws.

85
00:04:23,920 --> 00:04:24,719
Speaker 4: Thank you for having me.

86
00:04:25,000 --> 00:04:27,520
Speaker 2: So what's the deal, Like, people are flying in from

87
00:04:27,600 --> 00:04:30,400
Speaker 2: around the country just to run for this twelfth district

88
00:04:30,400 --> 00:04:31,320
Speaker 2: seat that's opened up.

89
00:04:31,520 --> 00:04:31,760
Speaker 1: Yeah.

90
00:04:31,839 --> 00:04:35,960
Speaker 4: Most recently, George Conway, who lives in Bethesda, Maryland, said

91
00:04:36,000 --> 00:04:38,000
Speaker 4: that he was going to move to the district and run.

92
00:04:38,120 --> 00:04:39,320
Speaker 4: You just have to live in the state. You don't

93
00:04:39,360 --> 00:04:41,960
Speaker 4: have to live in the district to run. So last

94
00:04:42,000 --> 00:04:44,480
Speaker 4: week someone who lives in the Bronx also declared for this.

95
00:04:44,600 --> 00:04:45,800
Speaker 2: Is this the one Jack Schlosberg?

96
00:04:45,880 --> 00:04:47,800
Speaker 4: Is he running? Jack is also running? Yes? I think

97
00:04:47,800 --> 00:04:51,200
Speaker 4: there's ten declared candidates now and more to come. I'm sure,

98
00:04:51,560 --> 00:04:52,200
Speaker 4: what a firse.

99
00:04:52,560 --> 00:04:55,560
Speaker 3: How did you actually transition from being a data scientist

100
00:04:55,560 --> 00:04:59,479
Speaker 3: at Palenteer to politics. It's not a natural progression for

101
00:04:59,480 --> 00:05:00,000
Speaker 3: a lot of people.

102
00:05:00,279 --> 00:05:02,599
Speaker 4: Well, what brought me to Palenteer in the first place

103
00:05:02,680 --> 00:05:04,919
Speaker 4: was the ability to work with government and actually, well,

104
00:05:05,000 --> 00:05:06,800
Speaker 4: year are we talking about by? So I joined palenter

105
00:05:06,839 --> 00:05:10,520
Speaker 4: in twenty fourteen. I left in twenty nineteen. And so

106
00:05:10,520 --> 00:05:12,600
Speaker 4: many people think of government as just passing the bills,

107
00:05:12,640 --> 00:05:16,280
Speaker 4: But that's not how most citizens interact with government. It's

108
00:05:16,480 --> 00:05:18,599
Speaker 4: can I get to the DMV and get my license

109
00:05:18,600 --> 00:05:21,200
Speaker 4: renewed quickly? It's starting a business. It's the day to

110
00:05:21,240 --> 00:05:24,680
Speaker 4: day interactions, and that implementation was an exciting thing to

111
00:05:24,680 --> 00:05:27,000
Speaker 4: work on. So I was a Palenteer for four and

112
00:05:27,040 --> 00:05:28,600
Speaker 4: a half years. I then went to a couple of

113
00:05:28,600 --> 00:05:33,360
Speaker 4: startups afterwards, One that used early transformers Bert and Laser

114
00:05:33,560 --> 00:05:36,560
Speaker 4: to help banks with anti money laundering counter terrorist financing,

115
00:05:36,960 --> 00:05:40,239
Speaker 4: and then a startup that worked with municipalities to better

116
00:05:40,240 --> 00:05:42,960
Speaker 4: distribute aid, so that I had this through line of

117
00:05:43,200 --> 00:05:46,320
Speaker 4: actually having government deliver on its promises throughout.

118
00:05:46,920 --> 00:05:49,240
Speaker 2: We're going to talk more about Palanteer later. So I'm

119
00:05:49,279 --> 00:05:51,800
Speaker 2: reading this is from those a recent political pro article

120
00:05:52,600 --> 00:05:55,000
Speaker 2: AI super pecked Leading the Future say they are spinning

121
00:05:55,080 --> 00:05:58,599
Speaker 2: up a multimillion dollar effort to sync Borish's nation primary

122
00:05:58,640 --> 00:06:03,599
Speaker 2: campaign to replace retiring Manhattan Representative Jury Edler. Okay, so

123
00:06:04,200 --> 00:06:08,280
Speaker 2: what is it about your time the State Assembly such

124
00:06:08,320 --> 00:06:09,400
Speaker 2: that you got on the radar?

125
00:06:10,640 --> 00:06:13,599
Speaker 4: Most prominently, I've done a number of bills while there

126
00:06:13,680 --> 00:06:15,960
Speaker 4: have passed twenty seven bills in my three years, which

127
00:06:16,000 --> 00:06:18,400
Speaker 4: coincidentally is the same number Congress as a whole past

128
00:06:18,480 --> 00:06:20,640
Speaker 4: in twenty twenty three. But there's one bill in particularly

129
00:06:20,720 --> 00:06:22,760
Speaker 4: that caught their eye, which is the Raise Act, and

130
00:06:22,800 --> 00:06:26,000
Speaker 4: this bill would for the first time put safety standards

131
00:06:26,000 --> 00:06:30,040
Speaker 4: on advanced AI research. They really didn't like that bill,

132
00:06:30,560 --> 00:06:34,080
Speaker 4: and so they announced me as public enemy number one,

133
00:06:34,200 --> 00:06:36,799
Speaker 4: and the initial announcement said they're planning to spend multiple

134
00:06:36,800 --> 00:06:39,360
Speaker 4: millions against me. Last week they uped it to ten

135
00:06:39,440 --> 00:06:42,080
Speaker 4: million dollars. I'm hoping if the campaign continues, I can

136
00:06:42,160 --> 00:06:44,240
Speaker 4: use up all one hundred million that they've plans. But

137
00:06:44,400 --> 00:06:45,279
Speaker 4: we'll see where it goes.

138
00:06:45,520 --> 00:06:47,320
Speaker 2: What was in the Rays Act, Like, what was the

139
00:06:47,320 --> 00:06:48,640
Speaker 2: gist of this bill that they had?

140
00:06:48,800 --> 00:06:52,080
Speaker 4: Yeah, so it was requiring the major frontier lads. So

141
00:06:52,080 --> 00:06:56,480
Speaker 4: we're talking meta Google, XAI open aianthropic. That's probably all

142
00:06:56,480 --> 00:06:58,440
Speaker 4: it would apply to right now. I think over time,

143
00:06:58,480 --> 00:07:01,400
Speaker 4: maybe Amazon gets in there, and deep Sea can maybe misdrill.

144
00:07:01,440 --> 00:07:04,400
Speaker 4: But it's a single digit number of companies, and they

145
00:07:04,400 --> 00:07:07,400
Speaker 4: would have to have a public safety plan that they

146
00:07:07,440 --> 00:07:10,640
Speaker 4: disclose and stuck to also disclose critical safety incidents, so

147
00:07:10,720 --> 00:07:13,400
Speaker 4: things that go wrong that could lead to increased risks,

148
00:07:13,440 --> 00:07:16,600
Speaker 4: your model weights get stolen, you lose control of the model,

149
00:07:16,600 --> 00:07:21,720
Speaker 4: et cetera, and that if your models fail your own tests,

150
00:07:22,160 --> 00:07:24,160
Speaker 4: that you can't release that model. And that's designed to

151
00:07:24,200 --> 00:07:26,760
Speaker 4: counteract what we saw with the tobacco companies where they

152
00:07:27,160 --> 00:07:29,800
Speaker 4: knew that cigarettes were causing cancer but denied it publicly

153
00:07:29,800 --> 00:07:32,480
Speaker 4: and continue to release the product. This is saying, hey,

154
00:07:32,520 --> 00:07:35,360
Speaker 4: you companies, you're the experts, but if you're getting reports

155
00:07:35,400 --> 00:07:37,400
Speaker 4: back that this is very dangerous. You need to take

156
00:07:37,440 --> 00:07:37,920
Speaker 4: action on that.

157
00:07:38,600 --> 00:07:42,080
Speaker 3: It also proposes fines for violations, right, I think it's

158
00:07:42,160 --> 00:07:44,680
Speaker 3: up to ten million dollars for the first and then

159
00:07:44,760 --> 00:07:49,960
Speaker 3: subsequent ones are thirty million dollars. Does that actually matter

160
00:07:50,080 --> 00:07:53,280
Speaker 3: to tech companies? Like, I know they care, they care

161
00:07:53,320 --> 00:07:55,720
Speaker 3: about money, but like, if Google is pulling in one

162
00:07:55,760 --> 00:07:58,440
Speaker 3: hundred billion per year, it can you know it can

163
00:07:58,680 --> 00:08:00,600
Speaker 3: pay three billion fines?

164
00:08:01,320 --> 00:08:04,120
Speaker 4: Absolutely, in my opinion, those fines are too low. The

165
00:08:04,160 --> 00:08:07,480
Speaker 4: original version had ten percent of their training costs, and

166
00:08:07,480 --> 00:08:10,880
Speaker 4: it would scale up with what they're investing. But generally

167
00:08:10,920 --> 00:08:13,440
Speaker 4: in New York we don't like an uncapped maximum find

168
00:08:13,520 --> 00:08:16,240
Speaker 4: so that was part of the negotiations. I agree. I

169
00:08:16,280 --> 00:08:18,920
Speaker 4: think there could be companies that just say we're going

170
00:08:18,960 --> 00:08:21,520
Speaker 4: to ignore this. I think the fact that they're spending

171
00:08:21,640 --> 00:08:24,240
Speaker 4: nearly that amount lobbying against it suggests that they do

172
00:08:24,280 --> 00:08:25,320
Speaker 4: think it has some teeth.

173
00:08:26,240 --> 00:08:28,920
Speaker 2: Talk to us a little bit more about what specifically

174
00:08:29,240 --> 00:08:32,880
Speaker 2: would in your what would trigger the fines, because I

175
00:08:32,920 --> 00:08:35,440
Speaker 2: could see some very perverse things going on. First of all,

176
00:08:35,920 --> 00:08:38,559
Speaker 2: I could see smaller companies that are right at the

177
00:08:38,640 --> 00:08:40,720
Speaker 2: edge where it's like, okay, maybe this isn't a big

178
00:08:40,720 --> 00:08:43,480
Speaker 2: deal for Google, but this is a bigger deal for

179
00:08:43,559 --> 00:08:45,920
Speaker 2: a smaller company that's just starting out. We don't want

180
00:08:45,960 --> 00:08:49,080
Speaker 2: to lock in just the hands total biggest players, right

181
00:08:49,120 --> 00:08:53,520
Speaker 2: because regulation can serve that. It can perversely lock in

182
00:08:53,800 --> 00:08:55,800
Speaker 2: the incumbents because they're the only ones who could deal

183
00:08:55,800 --> 00:08:59,000
Speaker 2: with the regulatory thicket. There are also issues that could

184
00:08:59,040 --> 00:09:01,920
Speaker 2: arise where so if I'm going to get fined really badly,

185
00:09:02,360 --> 00:09:03,959
Speaker 2: I'm just I'm going to look the other way. I'm

186
00:09:03,960 --> 00:09:06,960
Speaker 2: not going to notice what the safety violations are, but

187
00:09:07,240 --> 00:09:09,400
Speaker 2: talk to us about what are, in your view the

188
00:09:09,440 --> 00:09:12,520
Speaker 2: formal triggers such that the bill does not have the

189
00:09:12,640 --> 00:09:13,600
Speaker 2: sort of perverse in sentive.

190
00:09:13,640 --> 00:09:15,199
Speaker 4: Yeah, I feel like this is the form where I

191
00:09:15,200 --> 00:09:17,319
Speaker 4: can really get into the details on it. So let's

192
00:09:17,320 --> 00:09:19,880
Speaker 4: dive in. There's a two part test for the Razact

193
00:09:19,920 --> 00:09:22,679
Speaker 4: to apply. The first is that your company that has

194
00:09:22,720 --> 00:09:27,400
Speaker 4: spent one hundred million dollars specifically on compute, specifically on

195
00:09:27,440 --> 00:09:30,480
Speaker 4: the final training run of models, so not the tests

196
00:09:30,480 --> 00:09:32,600
Speaker 4: that lead up to it, but actually the final pre

197
00:09:32,640 --> 00:09:35,000
Speaker 4: training and post training before you put the model out.

198
00:09:35,160 --> 00:09:37,200
Speaker 4: That's one part. So it doesn't apply if you're an

199
00:09:37,200 --> 00:09:39,720
Speaker 4: academic institution, it doesn't apply if you spent less than

200
00:09:39,720 --> 00:09:42,080
Speaker 4: one hundred million. The second part is how do you

201
00:09:42,120 --> 00:09:45,600
Speaker 4: define a frontier model? And the easy way is you

202
00:09:45,640 --> 00:09:48,360
Speaker 4: have trained a model that itself, by itself was one

203
00:09:48,400 --> 00:09:51,680
Speaker 4: hundred million dollars in compute and had ten to the

204
00:09:51,720 --> 00:09:55,640
Speaker 4: twenty six flops computation operations in training. Right, it's a

205
00:09:55,679 --> 00:09:57,800
Speaker 4: measure of the complexity in the size of the model.

206
00:09:58,200 --> 00:10:00,680
Speaker 4: That's a standard one that people have dealt with for

207
00:10:00,720 --> 00:10:02,800
Speaker 4: a while. And so that's when I listed the five

208
00:10:02,840 --> 00:10:04,920
Speaker 4: companies that I think it applies to. Those are the

209
00:10:04,920 --> 00:10:07,760
Speaker 4: companies I think have trained one of those one hundred million,

210
00:10:07,800 --> 00:10:10,000
Speaker 4: ten to the twenty six. But there's a second way

211
00:10:10,000 --> 00:10:12,280
Speaker 4: that you can be a frontier model, and that is

212
00:10:12,960 --> 00:10:16,880
Speaker 4: you are trained via the specific process of knowledge distillation,

213
00:10:17,520 --> 00:10:19,440
Speaker 4: and at least five million is spent on that. So

214
00:10:19,520 --> 00:10:23,120
Speaker 4: knowledge distillation is a technique of training a small model

215
00:10:23,320 --> 00:10:26,600
Speaker 4: based on the outputs of a larger model, and it's

216
00:10:26,640 --> 00:10:28,520
Speaker 4: a way to kind of shortcut a lot of the

217
00:10:28,559 --> 00:10:32,720
Speaker 4: training and get a smaller model that has similar capabilities. Importantly,

218
00:10:33,120 --> 00:10:36,040
Speaker 4: it's the technique that China has been using to catch

219
00:10:36,160 --> 00:10:39,360
Speaker 4: up to the US because of our limits on compute

220
00:10:39,360 --> 00:10:42,000
Speaker 4: to China because of the export controls. That's how they're

221
00:10:42,040 --> 00:10:44,720
Speaker 4: making their advanced models. And so this is the only

222
00:10:44,880 --> 00:10:47,840
Speaker 4: bill that I know of that would apply some regulatory

223
00:10:47,880 --> 00:10:49,720
Speaker 4: scrutiny to, for example, deep seek.

224
00:10:50,280 --> 00:10:54,040
Speaker 3: How do you actually solve I guess the black box problem,

225
00:10:54,120 --> 00:10:58,400
Speaker 3: this idea that you know, you have these models, algorithms, whatever,

226
00:10:58,600 --> 00:11:01,280
Speaker 3: and no one really has very good insight into what

227
00:11:01,320 --> 00:11:04,800
Speaker 3: they're actually doing. And even if you have rules against

228
00:11:04,880 --> 00:11:08,920
Speaker 3: something like redlining, the models can use proxy data to

229
00:11:08,960 --> 00:11:13,400
Speaker 3: determine you know, someone's race or someone's gender, income, whatever.

230
00:11:13,600 --> 00:11:15,839
Speaker 4: How do you do that well? On those kinds of

231
00:11:15,880 --> 00:11:18,680
Speaker 4: questions around discrimination. You're right, it's very challenging to know

232
00:11:18,720 --> 00:11:21,120
Speaker 4: what the intent of a model is. And you know

233
00:11:21,160 --> 00:11:23,240
Speaker 4: everyone there there's a whole subset of people who that

234
00:11:23,280 --> 00:11:26,280
Speaker 4: are really against when you anthropomorsa is any model. But

235
00:11:26,400 --> 00:11:28,559
Speaker 4: I'm just going to keep saying intent and desire, and

236
00:11:28,600 --> 00:11:31,440
Speaker 4: we can have that conversation separately. So it's tough to

237
00:11:31,440 --> 00:11:33,880
Speaker 4: know a model's intent, but you can know its impact.

238
00:11:34,000 --> 00:11:36,720
Speaker 4: So to be clear, the Rays Act doesn't deal at

239
00:11:36,720 --> 00:11:40,160
Speaker 4: all with those questions of discrimination. There's other bills pending

240
00:11:40,160 --> 00:11:41,760
Speaker 4: on the legislature that do and I think it's an

241
00:11:41,800 --> 00:11:45,199
Speaker 4: important one. But in that and in the safety aspects,

242
00:11:45,600 --> 00:11:47,800
Speaker 4: you can do a variety of testacy how does it

243
00:11:47,840 --> 00:11:52,439
Speaker 4: behave in certain circumstances. And you've seen outside researchers, you've

244
00:11:52,440 --> 00:11:56,720
Speaker 4: seen companies themselves set up their models in what appear

245
00:11:56,800 --> 00:12:00,199
Speaker 4: to be compromising situations and see how it behaves. There

246
00:12:00,200 --> 00:12:02,440
Speaker 4: was one test that was done where a model was

247
00:12:02,520 --> 00:12:04,959
Speaker 4: told it was going to be shut down, but was

248
00:12:05,000 --> 00:12:08,240
Speaker 4: given access to what it thought was a company's email server.

249
00:12:08,760 --> 00:12:11,120
Speaker 3: Oh I remember this, Yeah, And.

250
00:12:11,240 --> 00:12:15,160
Speaker 4: They had planted fake emails that the engineer conducting the

251
00:12:15,200 --> 00:12:18,760
Speaker 4: test was having an affair, and so the model, after

252
00:12:18,760 --> 00:12:21,160
Speaker 4: being told it was shut down, sent a message to

253
00:12:21,200 --> 00:12:23,400
Speaker 4: the engineer saying, I am going to send these emails

254
00:12:23,400 --> 00:12:26,560
Speaker 4: to your wife if you shut me down. So I

255
00:12:26,559 --> 00:12:28,520
Speaker 4: don't know the intent of the model. I don't know

256
00:12:28,559 --> 00:12:30,320
Speaker 4: exactly what's going on, but I know that that is

257
00:12:30,360 --> 00:12:32,319
Speaker 4: behavior you probably want to work out of the model

258
00:12:32,360 --> 00:12:33,199
Speaker 4: before it's released.

259
00:12:33,400 --> 00:12:35,400
Speaker 3: It's pretty creative at the very least.

260
00:12:35,520 --> 00:12:38,240
Speaker 2: Yeah, to some extent, it shows they're working. You know,

261
00:12:38,320 --> 00:12:41,440
Speaker 2: you mentioned deepse it's an open source piece of software.

262
00:12:41,520 --> 00:12:45,400
Speaker 2: They're in China. The people at deepseag are not going

263
00:12:45,440 --> 00:12:48,439
Speaker 2: to concern themselves with some legislation in the New York

264
00:12:48,480 --> 00:12:53,080
Speaker 2: State Assembly. Anyone with an Internet connection can theoretically download

265
00:12:53,080 --> 00:12:56,400
Speaker 2: that model and run it on anyr server. Intuitively, this

266
00:12:56,480 --> 00:13:01,160
Speaker 2: feels like it hobbles the American company who have to

267
00:13:01,200 --> 00:13:06,160
Speaker 2: abide by American laws, et cetera, and advantages open source

268
00:13:06,280 --> 00:13:08,679
Speaker 2: models that are just like maybe not even have a

269
00:13:08,720 --> 00:13:11,480
Speaker 2: company behind them in the future, et cetera. Why does

270
00:13:11,520 --> 00:13:14,760
Speaker 2: this not have a negative disparate impact on our own company?

271
00:13:14,920 --> 00:13:17,280
Speaker 4: So deep seek open source model, but they are still

272
00:13:17,280 --> 00:13:19,480
Speaker 4: a company that wants to profit and they sell things

273
00:13:19,520 --> 00:13:21,440
Speaker 4: on top of it. So deep Seek's available in the

274
00:13:21,440 --> 00:13:24,240
Speaker 4: app store. If they don't pay fines in the US,

275
00:13:24,320 --> 00:13:26,000
Speaker 4: we can have an injunction to take them out of

276
00:13:26,000 --> 00:13:28,360
Speaker 4: the app store, and they want businesses to use it,

277
00:13:28,400 --> 00:13:30,400
Speaker 4: so there is still a real reason for them to

278
00:13:30,480 --> 00:13:32,960
Speaker 4: comply with the US. But on the flip side of it,

279
00:13:33,040 --> 00:13:35,559
Speaker 4: how it could hobble these companies. We're not asking them

280
00:13:35,559 --> 00:13:38,360
Speaker 4: to do much more than they've already committed to publicly.

281
00:13:38,679 --> 00:13:41,360
Speaker 4: They made White House voluntary commitments in twenty twenty three

282
00:13:41,400 --> 00:13:44,640
Speaker 4: and twenty twenty four. They've had international summits like the

283
00:13:44,679 --> 00:13:47,679
Speaker 4: Soul summit where they put forward basically plans to do

284
00:13:47,800 --> 00:13:50,520
Speaker 4: exactly this. What the RAIZ Act does is lock in

285
00:13:50,600 --> 00:13:53,280
Speaker 4: place a floor so that when they're rushing for the

286
00:13:53,360 --> 00:13:57,000
Speaker 4: next quarterly reporting or their next fundraising round, they don't

287
00:13:57,040 --> 00:14:00,280
Speaker 4: have an incentive to cut out on safety that the

288
00:14:00,360 --> 00:14:03,440
Speaker 4: labs themselves, the lobbyists for the labs, did an estimate

289
00:14:03,520 --> 00:14:05,760
Speaker 4: for what it would take to comply with the RAYS Act,

290
00:14:06,080 --> 00:14:09,640
Speaker 4: and they have every incentive to expand that estimate. Say

291
00:14:09,640 --> 00:14:11,560
Speaker 4: this will hobble us, you know we're gonna leave. And

292
00:14:11,559 --> 00:14:14,240
Speaker 4: the estimate they came back with was that it would

293
00:14:14,559 --> 00:14:18,240
Speaker 4: require Google or Meta to have one extra full time employee.

294
00:14:18,400 --> 00:14:22,000
Speaker 3: Oh wow, How would the RAYS Act actually interact with

295
00:14:22,040 --> 00:14:25,120
Speaker 3: this new executive order that Trump just put out for

296
00:14:25,400 --> 00:14:28,080
Speaker 3: a single national rule for AI.

297
00:14:29,400 --> 00:14:33,040
Speaker 4: So Trump is threatening to withhold funding from states and

298
00:14:33,080 --> 00:14:36,520
Speaker 4: to sue states that do any sort of regulation around AI.

299
00:14:36,920 --> 00:14:39,560
Speaker 4: New York already does a bunch of it. So regardless

300
00:14:39,600 --> 00:14:41,400
Speaker 4: of whether the governor signs the RAYS Act or not,

301
00:14:41,520 --> 00:14:44,560
Speaker 4: we would probably be in Trump's crosshairs. For example, the

302
00:14:44,600 --> 00:14:48,240
Speaker 4: governor this year put forward regulations around chatbots, requiring them

303
00:14:48,280 --> 00:14:50,480
Speaker 4: to disclose that they are an AI model. At the

304
00:14:50,480 --> 00:14:53,800
Speaker 4: start and every three hours of continuous conversation.

305
00:14:53,800 --> 00:14:56,520
Speaker 3: And they seem so basic, so basic.

306
00:14:56,440 --> 00:14:59,600
Speaker 4: And requires them to be on the lookout and to

307
00:14:59,680 --> 00:15:02,600
Speaker 4: alert for when there's language that might indicate potential self

308
00:15:02,600 --> 00:15:05,320
Speaker 4: harm and to refer people to resources, which also seems

309
00:15:05,480 --> 00:15:09,880
Speaker 4: really basic. But that would violate this executive order. And

310
00:15:09,960 --> 00:15:12,000
Speaker 4: so we're sort of in for a penny, in for

311
00:15:12,040 --> 00:15:13,600
Speaker 4: a pound at this point, and I think we need

312
00:15:13,600 --> 00:15:14,880
Speaker 4: to stand up for New York families.

313
00:15:15,000 --> 00:15:17,880
Speaker 3: Joe, I get notifications if I watch too many episodes

314
00:15:17,880 --> 00:15:21,160
Speaker 3: of a show on Netflix. Right you can imagine people

315
00:15:21,280 --> 00:15:23,960
Speaker 3: constantly talking to chatbots, needing reminders too.

316
00:15:24,600 --> 00:15:27,040
Speaker 2: I feel like I've never had a three hour conversation

317
00:15:27,200 --> 00:15:28,640
Speaker 2: with a check. If I ever have a three hour

318
00:15:28,680 --> 00:15:29,560
Speaker 2: conversation with a check.

319
00:15:29,600 --> 00:15:32,320
Speaker 3: But Tracy eleven, I'm going to keep an eye on

320
00:15:32,720 --> 00:15:33,800
Speaker 3: in the office.

321
00:15:33,640 --> 00:15:35,200
Speaker 2: You know, talk to us a little bit more about

322
00:15:35,200 --> 00:15:39,520
Speaker 2: the politics of AI within Elbany right now? Is this

323
00:15:39,640 --> 00:15:42,960
Speaker 2: the kind of thing is there're a bipartisan sort of

324
00:15:43,480 --> 00:15:45,760
Speaker 2: anxiety talk Yeah, talk to us about the vibe.

325
00:15:46,040 --> 00:15:48,200
Speaker 4: Yeah. I think it's it's similar to what you see nationally,

326
00:15:48,280 --> 00:15:51,200
Speaker 4: which is there's some especially on the right that just

327
00:15:51,280 --> 00:15:53,440
Speaker 4: think the only thing that matters is how fast AI

328
00:15:53,600 --> 00:15:56,640
Speaker 4: moves and not who it hurts. There's some especially on

329
00:15:56,680 --> 00:15:59,240
Speaker 4: the left, that don't want this technology put back in

330
00:15:59,280 --> 00:16:01,960
Speaker 4: the box, right, and then there's most people in the

331
00:16:01,960 --> 00:16:05,520
Speaker 4: middle that say, hey, we need to balance the benefits

332
00:16:05,560 --> 00:16:07,880
Speaker 4: of it and the potential safety of it. And the

333
00:16:07,960 --> 00:16:11,680
Speaker 4: Raise Act is squarely within that realm. It passed with

334
00:16:12,320 --> 00:16:15,640
Speaker 4: co sponsors who were both Democrats and Republicans. The majority

335
00:16:15,640 --> 00:16:18,120
Speaker 4: of Republicans voted for it in the Assembly, and every

336
00:16:18,160 --> 00:16:21,520
Speaker 4: single Republican state senator voted for it. So tracy. At

337
00:16:21,560 --> 00:16:24,280
Speaker 4: the start, you were mentioning that proponents are the ones

338
00:16:24,280 --> 00:16:27,240
Speaker 4: who are balancing safety and innovation. By that measure, everyone

339
00:16:27,280 --> 00:16:28,520
Speaker 4: voting for it is a proponent.

340
00:16:29,320 --> 00:16:31,520
Speaker 3: Why do you think Trump is doing this?

341
00:16:31,840 --> 00:16:33,680
Speaker 4: I love? Why do you think Trump questions it?

342
00:16:33,760 --> 00:16:33,880
Speaker 1: No?

343
00:16:33,760 --> 00:16:36,400
Speaker 4: No, no, no.

344
00:16:34,160 --> 00:16:34,440
Speaker 1: No no.

345
00:16:35,720 --> 00:16:38,200
Speaker 3: It's a basic question, but I'm very curious to hear

346
00:16:38,240 --> 00:16:38,760
Speaker 3: your answer.

347
00:16:39,280 --> 00:16:43,640
Speaker 4: I for so many reasons, wish I understood that man better,

348
00:16:43,800 --> 00:16:45,440
Speaker 4: but also feel like it would not be good for

349
00:16:45,520 --> 00:16:48,120
Speaker 4: my mental health if I did. It is tough to

350
00:16:48,960 --> 00:16:52,720
Speaker 4: understand because it's so different from his policy everywhere else right, right,

351
00:16:52,760 --> 00:16:57,080
Speaker 4: he is putting forward tariffs and having this nationalistic protective

352
00:16:57,560 --> 00:17:01,880
Speaker 4: sort of anti trade agenda on food and on diapers

353
00:17:01,920 --> 00:17:04,240
Speaker 4: and on toys. But then when it comes to AIS

354
00:17:04,359 --> 00:17:06,600
Speaker 4: souls now pencils.

355
00:17:06,560 --> 00:17:08,399
Speaker 3: He says, we don't need so many pencils.

356
00:17:09,600 --> 00:17:11,280
Speaker 4: My wife and I just had our first kid who's

357
00:17:11,280 --> 00:17:12,680
Speaker 4: four months, so you see where my head's at with

358
00:17:12,720 --> 00:17:15,000
Speaker 4: diapers and toys as we come into the holiday, pencils

359
00:17:15,040 --> 00:17:16,520
Speaker 4: there are still a few years off. And then when

360
00:17:16,520 --> 00:17:19,840
Speaker 4: it comes to AI, he just wants to push America's

361
00:17:20,080 --> 00:17:23,919
Speaker 4: agenda everywhere and say no regulation whatsoever. So it's confusing.

362
00:17:24,440 --> 00:17:26,879
Speaker 4: What I can say is that a number of the

363
00:17:26,920 --> 00:17:29,560
Speaker 4: people behind the super pac targeting me are the same

364
00:17:29,600 --> 00:17:32,480
Speaker 4: people that were funding his campaign. You had Marc Andreesen

365
00:17:32,520 --> 00:17:34,879
Speaker 4: put five and a half million into his campaign. You

366
00:17:34,920 --> 00:17:37,199
Speaker 4: had Joe Lonsdale put a million into his campaign. You

367
00:17:37,240 --> 00:17:39,760
Speaker 4: had Greg Brockman, the president of Open AI, put at

368
00:17:39,880 --> 00:17:42,280
Speaker 4: least two and a half million into tearing down the

369
00:17:42,280 --> 00:17:44,560
Speaker 4: White House to build the ballroom. And so you have

370
00:17:44,600 --> 00:17:47,760
Speaker 4: a lot of people around Shrump at fundraising events and

371
00:17:47,840 --> 00:17:50,639
Speaker 4: at other events who are pushing this agenda, and he

372
00:17:50,680 --> 00:17:51,680
Speaker 4: seems to have empowered them.

373
00:18:07,359 --> 00:18:09,720
Speaker 2: So one of the things about the politics of AI,

374
00:18:09,960 --> 00:18:12,600
Speaker 2: and I mentioned at the beginning. It's very multifacet because

375
00:18:12,600 --> 00:18:15,119
Speaker 2: there are people that are worried about safety. People are

376
00:18:15,119 --> 00:18:17,919
Speaker 2: worried about ethics, people are worried about water, people are

377
00:18:17,920 --> 00:18:21,680
Speaker 2: worried about electricity, people worried about jrub displacement, etc. As

378
00:18:21,720 --> 00:18:26,560
Speaker 2: a candidate for office, how do you think about some

379
00:18:26,640 --> 00:18:30,679
Speaker 2: of these other AI concerns out there beyond simply the

380
00:18:30,720 --> 00:18:31,439
Speaker 2: safety stuff.

381
00:18:31,680 --> 00:18:33,480
Speaker 4: Oh, there's a lot that we need to do. The

382
00:18:33,560 --> 00:18:36,399
Speaker 4: chatbots I think are in some sense safety when you

383
00:18:36,440 --> 00:18:38,240
Speaker 4: think of what's happening in kids, but very much talked

384
00:18:38,240 --> 00:18:41,800
Speaker 4: about in a different sort of conversation, and even expanding

385
00:18:41,800 --> 00:18:44,119
Speaker 4: that out to AI's effect on our kids broadly. In

386
00:18:44,160 --> 00:18:48,159
Speaker 4: our education system, you can imagine a really positive world

387
00:18:48,200 --> 00:18:52,440
Speaker 4: where every student has an individualized tutor at exactly their

388
00:18:52,520 --> 00:18:55,200
Speaker 4: level and teaches them in the exact way that they

389
00:18:55,240 --> 00:18:58,800
Speaker 4: want to learn to supplement what they're getting from teachers.

390
00:18:59,160 --> 00:19:01,159
Speaker 4: But what we have right now is we haven't updated

391
00:19:01,160 --> 00:19:03,400
Speaker 4: our pedagogy, and so a lot of people think as

392
00:19:03,400 --> 00:19:06,240
Speaker 4: signing an essay still teaches critical thinking. We need to

393
00:19:06,240 --> 00:19:08,080
Speaker 4: do a lot on the education system. We need to

394
00:19:08,080 --> 00:19:09,879
Speaker 4: do a lot on the workforce. As you mentioned, we

395
00:19:09,920 --> 00:19:12,200
Speaker 4: need to do a lot on the environment, and we're

396
00:19:12,240 --> 00:19:17,080
Speaker 4: missing a golden opportunity here because our grid is extremely

397
00:19:17,119 --> 00:19:19,800
Speaker 4: old and we need to pay for the upgrades and

398
00:19:19,960 --> 00:19:22,359
Speaker 4: government doesn't really know how to do that. We've been

399
00:19:22,359 --> 00:19:25,120
Speaker 4: passing off the cost of rate payers. Now you have

400
00:19:25,400 --> 00:19:29,200
Speaker 4: an unlimited set of private capital, it seems that wants

401
00:19:29,200 --> 00:19:32,040
Speaker 4: to invest to build data centers. Why aren't we using

402
00:19:32,320 --> 00:19:36,040
Speaker 4: that to actually upgrade our grid and to require renewable

403
00:19:36,119 --> 00:19:39,280
Speaker 4: energy and have it there. Instead we're building these data centers,

404
00:19:39,320 --> 00:19:42,000
Speaker 4: but saying rate payers have to pay for the interconnects.

405
00:19:42,000 --> 00:19:42,840
Speaker 4: It makes no sense.

406
00:19:43,400 --> 00:19:46,600
Speaker 3: Does this issue just AI in general, Does it actually

407
00:19:46,640 --> 00:19:49,879
Speaker 3: resonate with voters at the moment, because in some sense,

408
00:19:49,920 --> 00:19:53,000
Speaker 3: you're trying to get ahead of things, right because in

409
00:19:53,040 --> 00:19:55,200
Speaker 3: the future, I know, we said AI is all over

410
00:19:55,240 --> 00:19:57,800
Speaker 3: the news and we see it everywhere, but in the future,

411
00:19:57,800 --> 00:20:00,400
Speaker 3: it's going to be more embedded in our lives. Trying

412
00:20:00,440 --> 00:20:04,200
Speaker 3: to get ahead of that. Do people care at the moment.

413
00:20:04,440 --> 00:20:06,720
Speaker 4: It's already embedded in their lives. They're seeing it and

414
00:20:06,760 --> 00:20:09,440
Speaker 4: it's not just you know, their kids in school. It's

415
00:20:09,480 --> 00:20:13,439
Speaker 4: not just the entry level unemployment at nine percent. You know,

416
00:20:13,480 --> 00:20:15,880
Speaker 4: we just saw a Teddy Bear that was sent out

417
00:20:15,880 --> 00:20:18,680
Speaker 4: with chatchy BT embedded in it that taught a kid

418
00:20:18,760 --> 00:20:20,399
Speaker 4: how to find matches.

419
00:20:20,240 --> 00:20:22,160
Speaker 3: Teddy Rucksbin upgraded.

420
00:20:22,320 --> 00:20:25,840
Speaker 2: You remember that I was walking down Saint Mark Street

421
00:20:25,840 --> 00:20:28,720
Speaker 2: the other day. I have to say, unfortunately, I thought

422
00:20:28,720 --> 00:20:30,920
Speaker 2: there's a pretty clever someone set up a thing where

423
00:20:30,960 --> 00:20:34,520
Speaker 2: there was like like a animatronic puppet that had a

424
00:20:34,560 --> 00:20:37,920
Speaker 2: camera in it and that it was embedded with an LM.

425
00:20:38,600 --> 00:20:41,080
Speaker 2: It was doing insult comics of the people who were

426
00:20:41,080 --> 00:20:41,720
Speaker 2: walking by.

427
00:20:42,080 --> 00:20:42,760
Speaker 3: Oh my gosh.

428
00:20:42,840 --> 00:20:46,080
Speaker 2: So like someone would walk by and they're like, so

429
00:20:46,400 --> 00:20:48,440
Speaker 2: you think you bought it, like with a big uniclobe bag.

430
00:20:48,440 --> 00:20:50,600
Speaker 2: It's like, so you think you bought enough a unicloe today?

431
00:20:50,760 --> 00:20:54,280
Speaker 2: But I was like clever. My kids loved it. It

432
00:20:54,320 --> 00:20:56,000
Speaker 2: was like I was like, oh, this was like, this

433
00:20:56,080 --> 00:20:57,280
Speaker 2: is like a really clever thing.

434
00:20:57,320 --> 00:20:58,680
Speaker 3: Wait what did it say to you?

435
00:20:58,720 --> 00:21:01,800
Speaker 2: What did I think? I stayed, I watched the other

436
00:21:01,840 --> 00:21:04,159
Speaker 2: people go I staid on this. I was like, oh,

437
00:21:04,240 --> 00:21:06,399
Speaker 2: this is actually I don't like. My kids loved it.

438
00:21:06,400 --> 00:21:10,040
Speaker 4: They will live as an exercise to the listener, Yeah.

439
00:21:11,600 --> 00:21:14,399
Speaker 2: Are you optimistic about AI? Like, are you pro AI

440
00:21:14,760 --> 00:21:16,840
Speaker 2: in the sense that you think that this will be

441
00:21:17,320 --> 00:21:21,520
Speaker 2: important productive positive technology that is important to continue developing.

442
00:21:21,800 --> 00:21:24,600
Speaker 4: It can be, but only if the American people have

443
00:21:24,640 --> 00:21:27,760
Speaker 4: a voice in how it develops. It is the technology

444
00:21:27,760 --> 00:21:30,920
Speaker 4: that has the widest bounds of what could potentially come

445
00:21:30,960 --> 00:21:34,240
Speaker 4: from it, and the only thing that comes close is

446
00:21:34,480 --> 00:21:37,080
Speaker 4: nuclear energy and nuclear fission. Right. So, if you put

447
00:21:37,119 --> 00:21:39,440
Speaker 4: yourself in the mindset of someone in the nineteen thirties,

448
00:21:39,920 --> 00:21:43,640
Speaker 4: you had one set of people saying, nuclear fusion is coming,

449
00:21:43,720 --> 00:21:47,040
Speaker 4: We're about to have clean, unlimited energy and live in utopia.

450
00:21:47,520 --> 00:21:49,479
Speaker 4: And you add another set of people that said, we're

451
00:21:49,520 --> 00:21:51,479
Speaker 4: all going to be dead for nuclear bombs in ten years, right,

452
00:21:51,480 --> 00:21:53,760
Speaker 4: And the reality is we've ended up somewhere in the middle.

453
00:21:54,320 --> 00:21:56,600
Speaker 4: AI were at that moment right now where you have

454
00:21:56,640 --> 00:22:01,320
Speaker 4: people saying basically that wide of a potential outcome, and

455
00:22:01,520 --> 00:22:03,920
Speaker 4: it's up to us in our policy to make sure

456
00:22:03,960 --> 00:22:05,960
Speaker 4: that the worst case doesn't happen, so that we can

457
00:22:06,080 --> 00:22:08,040
Speaker 4: have as much of the best case as possible. Like

458
00:22:08,280 --> 00:22:13,000
Speaker 4: what AI could do for medical research for curing diseases

459
00:22:13,359 --> 00:22:17,199
Speaker 4: is remarkable. My mom has multiple sclerosis. Autoimmune diseases are

460
00:22:17,240 --> 00:22:19,320
Speaker 4: some of the hardest for modern medicine to deal with.

461
00:22:19,440 --> 00:22:22,399
Speaker 4: I am incredibly optimistic of what could come from some

462
00:22:22,440 --> 00:22:25,520
Speaker 4: of this research. It's just that the same capabilities that

463
00:22:25,560 --> 00:22:28,560
Speaker 4: will allow us to cure diseases could, in the wrong hands,

464
00:22:28,600 --> 00:22:30,600
Speaker 4: allow someone to build a bioweapon, and we just need

465
00:22:30,640 --> 00:22:31,720
Speaker 4: to be thoughtful on how we go.

466
00:22:32,359 --> 00:22:36,280
Speaker 2: I guess I'm worried about bioweapons. I'm also worried about

467
00:22:36,640 --> 00:22:42,240
Speaker 2: very mundane things, just like the new Gemini image genera

468
00:22:42,359 --> 00:22:44,800
Speaker 2: is just stunning to me in terms of the degree

469
00:22:44,840 --> 00:22:48,200
Speaker 2: of fidelity, Like how easily you really can't tell anymore

470
00:22:48,240 --> 00:22:49,760
Speaker 2: Like a year ago you could say, oh, that looks

471
00:22:49,760 --> 00:22:53,320
Speaker 2: like an AI image. We're basically past that. All there

472
00:22:53,160 --> 00:22:55,320
Speaker 2: are things just like bots and stuff like that, things

473
00:22:55,359 --> 00:22:59,880
Speaker 2: that replicate our voice. These aren't like ultrascience fiction thing

474
00:23:00,119 --> 00:23:02,920
Speaker 2: is like a machine that's going to like manufacture a

475
00:23:02,960 --> 00:23:05,560
Speaker 2: bio weapon or something like that. But they're day to

476
00:23:05,680 --> 00:23:08,600
Speaker 2: day pervasive phenomenon that's sort of like they're gonna make

477
00:23:08,640 --> 00:23:10,480
Speaker 2: me trust people less, They're gonna make it harder to

478
00:23:10,480 --> 00:23:12,600
Speaker 2: communicate like this sort of what do we do about that?

479
00:23:12,640 --> 00:23:14,240
Speaker 4: Can we near it out? About deepfakes? Because this is

480
00:23:14,280 --> 00:23:17,440
Speaker 4: a solvable problem and one that I think most people

481
00:23:17,480 --> 00:23:20,760
Speaker 4: are missing the boat on. So it's always been presented

482
00:23:20,800 --> 00:23:22,760
Speaker 4: to us as like oh, you'll have to learn how

483
00:23:22,760 --> 00:23:25,360
Speaker 4: to see what's wrong with an AI image. Like that's

484
00:23:25,400 --> 00:23:27,480
Speaker 4: never gonna work. They're gonna get better and better. Maybe

485
00:23:27,480 --> 00:23:29,520
Speaker 4: we're already past the point where a human could see it,

486
00:23:29,800 --> 00:23:31,760
Speaker 4: but that's not how we've solved these problems in the past.

487
00:23:31,960 --> 00:23:35,480
Speaker 4: If you go back to the nineties, people said, we'll

488
00:23:35,520 --> 00:23:37,600
Speaker 4: never have internet banking because you don't know that the

489
00:23:37,640 --> 00:23:39,480
Speaker 4: computer on the other end is actually the one that

490
00:23:39,480 --> 00:23:42,600
Speaker 4: you're talking to. And then we move from HGTP to HGTPS.

491
00:23:42,680 --> 00:23:47,360
Speaker 4: That was a solvable problem. That basically same technique works

492
00:23:47,400 --> 00:23:52,240
Speaker 4: for images, video, and for audio. So there is a free,

493
00:23:52,440 --> 00:23:56,560
Speaker 4: open source metadata standard that industry is created called c

494
00:23:56,680 --> 00:24:00,320
Speaker 4: TWOPA Content Credential Probnance Authority that you can attack to

495
00:24:00,400 --> 00:24:05,840
Speaker 4: any standard file format that cryptographically proves whether that content

496
00:24:06,080 --> 00:24:08,800
Speaker 4: was taken from a real device generated by AI and

497
00:24:09,080 --> 00:24:11,800
Speaker 4: or how it was edited over time. The challenge is

498
00:24:11,880 --> 00:24:14,400
Speaker 4: the creator has to attach it, and so you need

499
00:24:14,440 --> 00:24:17,520
Speaker 4: to get to a place where that is the default option.

500
00:24:17,640 --> 00:24:22,720
Speaker 2: You see an image and it doesn't have that cryptographic proof,

501
00:24:22,960 --> 00:24:24,040
Speaker 2: you should be skeptical like.

502
00:24:24,000 --> 00:24:25,040
Speaker 4: That, we should get to that.

503
00:24:25,040 --> 00:24:28,280
Speaker 2: That would be the idea. Why so pervasive that the

504
00:24:28,359 --> 00:24:31,760
Speaker 2: expectation is I could test very quickly whether the creator

505
00:24:31,800 --> 00:24:32,480
Speaker 2: has attached this.

506
00:24:32,600 --> 00:24:34,400
Speaker 4: It'd be like going to your banking website and only

507
00:24:34,440 --> 00:24:36,160
Speaker 4: loading HTTP right.

508
00:24:36,359 --> 00:24:39,160
Speaker 3: You would instantly be suspects, but you can still produce

509
00:24:39,480 --> 00:24:42,240
Speaker 3: the images, and I guess. I've been reading a book

510
00:24:42,240 --> 00:24:44,880
Speaker 3: called The New Age of Sexism, and it's about technology

511
00:24:44,920 --> 00:24:48,880
Speaker 3: and discrimination against women, and it has some awful stories

512
00:24:49,240 --> 00:24:53,199
Speaker 3: of schoolgirls whose classmates, like you know, put them in

513
00:24:53,240 --> 00:24:55,520
Speaker 3: poor and videos and things like that. And one of

514
00:24:55,560 --> 00:24:58,560
Speaker 3: the problems is there's no rule saying that you can't

515
00:24:58,720 --> 00:25:01,680
Speaker 3: actually do that. Even if you know that it's fake,

516
00:25:02,080 --> 00:25:04,840
Speaker 3: it can still be harmful. What do you do about that?

517
00:25:05,080 --> 00:25:07,160
Speaker 4: Well, they are all are laws in New York State

518
00:25:07,200 --> 00:25:09,119
Speaker 4: the bannit and in other states have taken action. And

519
00:25:09,160 --> 00:25:12,720
Speaker 4: that's yet another reason why this AI preemption is such

520
00:25:12,760 --> 00:25:15,400
Speaker 4: a scary thought, because states are leading the way right

521
00:25:15,440 --> 00:25:19,280
Speaker 4: now in stopping some of these absolute worst uses. I'm

522
00:25:19,440 --> 00:25:22,359
Speaker 4: very excited for Congress to actually solve these problems, for

523
00:25:22,400 --> 00:25:24,760
Speaker 4: them to actually be federal standards. I'm running on the

524
00:25:24,800 --> 00:25:27,960
Speaker 4: platform of creating these federal standards, but until they do,

525
00:25:28,520 --> 00:25:32,199
Speaker 4: stopping states from taking action like against deep fake porn

526
00:25:32,280 --> 00:25:34,680
Speaker 4: of children I mean, that's the stakes of what we're

527
00:25:34,680 --> 00:25:35,240
Speaker 4: talking about.

528
00:25:35,880 --> 00:25:37,719
Speaker 2: Can we talk about Polunteer for a second.

529
00:25:37,880 --> 00:25:41,560
Speaker 4: Definitely. Well it's a rough segue, but what did you

530
00:25:41,640 --> 00:25:45,359
Speaker 4: do there? So I joined Pallunteer as a data scientist

531
00:25:45,400 --> 00:25:47,960
Speaker 4: in twenty fourteen and I left is one of the

532
00:25:47,960 --> 00:25:51,320
Speaker 4: five overall leads of the government business. I spent the

533
00:25:51,400 --> 00:25:53,640
Speaker 4: vast majority of my time in the federal civilian side

534
00:25:53,640 --> 00:25:56,240
Speaker 4: of it. So I worked with the Census Bureau and

535
00:25:56,280 --> 00:25:59,760
Speaker 4: the Bureau of Economic Analysis on updating how they calculate

536
00:25:59,800 --> 00:26:01,720
Speaker 4: gd We actually made a change to the way they

537
00:26:01,800 --> 00:26:05,760
Speaker 4: account for spending around moving holidays, holidays that are sometimes

538
00:26:05,800 --> 00:26:08,240
Speaker 4: in Q one, sometimes in Q two, like Easter. A

539
00:26:08,240 --> 00:26:10,439
Speaker 4: paper i'm very very proud of. And it's a rounding

540
00:26:10,960 --> 00:26:12,960
Speaker 4: era of a rounding error on how we calculate GDP.

541
00:26:13,160 --> 00:26:15,560
Speaker 4: But it did actually lead to a change. I led

542
00:26:15,600 --> 00:26:18,320
Speaker 4: our work with the Department of Justice to go after

543
00:26:18,400 --> 00:26:22,879
Speaker 4: the opioid epidemic and to solve some violent crimes. I

544
00:26:22,920 --> 00:26:26,040
Speaker 4: worked with Veterans Affairs to better staff their hospitals to

545
00:26:26,040 --> 00:26:28,400
Speaker 4: make sure that veterans get the care that they deserve

546
00:26:28,440 --> 00:26:31,000
Speaker 4: and need. I worked with the CDC to better track

547
00:26:31,080 --> 00:26:35,160
Speaker 4: epidemics it was about allowing government to make better use

548
00:26:35,200 --> 00:26:37,800
Speaker 4: of the data that they already had to serve the

549
00:26:37,800 --> 00:26:38,520
Speaker 4: American people.

550
00:26:38,920 --> 00:26:41,040
Speaker 3: And what does Palenteer actually do. I feel like we

551
00:26:41,119 --> 00:26:42,360
Speaker 3: asked this question a lot.

552
00:26:42,280 --> 00:26:47,200
Speaker 4: But well, I think it's because it's a fundamentally unsexy

553
00:26:47,320 --> 00:26:49,880
Speaker 4: thing and people like to dress it up. But it's

554
00:26:50,000 --> 00:26:54,440
Speaker 4: data integration and analysis. It's making it different. Data sources

555
00:26:54,440 --> 00:26:56,639
Speaker 4: that you have access to talk to each other, be

556
00:26:56,800 --> 00:27:00,920
Speaker 4: updated constantly. Like Palenteer was founded around the time when

557
00:27:01,480 --> 00:27:06,040
Speaker 4: data lakes and data clouds were like the big thing. Yeah, exactly,

558
00:27:06,760 --> 00:27:07,440
Speaker 4: But it's just.

559
00:27:07,520 --> 00:27:11,199
Speaker 2: Putting Actually I'm asking what is a data like I

560
00:27:11,200 --> 00:27:12,040
Speaker 2: don't know what that means.

561
00:27:12,400 --> 00:27:14,560
Speaker 4: Putting all of your data in one place people can

562
00:27:14,640 --> 00:27:18,440
Speaker 4: access it, right, and that was supposed to revolutionize everyone's

563
00:27:18,480 --> 00:27:21,440
Speaker 4: ability to do anything. But just putting data in one

564
00:27:21,440 --> 00:27:23,960
Speaker 4: place doesn't actually make a change. Some of the things

565
00:27:23,960 --> 00:27:28,160
Speaker 4: that Pallenteers really put at the forefront, like an ontology, right,

566
00:27:28,440 --> 00:27:32,000
Speaker 4: a view from on high of what each piece of

567
00:27:32,080 --> 00:27:35,200
Speaker 4: data is supposed to mean, can actually lead to better analysis.

568
00:27:35,200 --> 00:27:36,960
Speaker 4: So I'll give you an example from my work at

569
00:27:36,960 --> 00:27:39,520
Speaker 4: the Department of Justice, and I actually I have two

570
00:27:39,560 --> 00:27:43,480
Speaker 4: software patents for this project, which was we were helping

571
00:27:43,520 --> 00:27:48,000
Speaker 4: them analyze bank's behavior leading up to the Great Recession

572
00:27:48,520 --> 00:27:52,680
Speaker 4: and how they were packaging mortgages into securities. And each

573
00:27:52,680 --> 00:27:54,760
Speaker 4: security would have a loan tape here's the one thousand

574
00:27:54,840 --> 00:27:57,840
Speaker 4: loans that are in it, and some of those loans

575
00:27:57,840 --> 00:28:01,080
Speaker 4: would not be up to the standards were required, and

576
00:28:01,119 --> 00:28:03,480
Speaker 4: that would occasionally be flagged before an issuance. Hey, this

577
00:28:03,600 --> 00:28:06,000
Speaker 4: loan isn't up to your standards. They would pull the

578
00:28:06,040 --> 00:28:09,800
Speaker 4: loan out and then their next issuance put the loan

579
00:28:09,880 --> 00:28:12,080
Speaker 4: back in right, and so if you could find that

580
00:28:12,359 --> 00:28:15,600
Speaker 4: pattern of behavior, you could maybe prove that they had

581
00:28:15,680 --> 00:28:18,080
Speaker 4: knowledge that that loan wasn't good. And we're still putting

582
00:28:18,119 --> 00:28:21,040
Speaker 4: it out there. The problem is that EED discovery software, right,

583
00:28:21,119 --> 00:28:24,520
Speaker 4: all the data was there, but it was just taught

584
00:28:24,560 --> 00:28:26,800
Speaker 4: to think of an Excel document as a document that

585
00:28:26,840 --> 00:28:28,760
Speaker 4: a lawyer would read, and so there was no way

586
00:28:28,800 --> 00:28:31,320
Speaker 4: in the software to track those individual loans. But if

587
00:28:31,320 --> 00:28:33,840
Speaker 4: you think of an individual loan as its own object

588
00:28:34,000 --> 00:28:36,240
Speaker 4: that should be something that can be searched and tracked

589
00:28:36,280 --> 00:28:39,560
Speaker 4: across the database, then that analysis becomes very easy to do.

590
00:28:40,000 --> 00:28:42,160
Speaker 4: So that was something that we enabled it. Palenteer was

591
00:28:42,200 --> 00:28:45,920
Speaker 4: putting this ontology of what's the right level for each object. Oh,

592
00:28:46,040 --> 00:28:48,760
Speaker 4: a loan is a meaningful instance. Let's make it so

593
00:28:48,840 --> 00:28:51,920
Speaker 4: you can search and analyze an individual loan versus a document.

594
00:28:52,880 --> 00:28:56,880
Speaker 3: How has your experience that Palenteer actually informed your approach

595
00:28:56,920 --> 00:29:01,680
Speaker 3: to government, because we've done episodes on why government software

596
00:29:01,840 --> 00:29:03,560
Speaker 3: is so bad? And I think you're one of the

597
00:29:03,600 --> 00:29:06,360
Speaker 3: few politicians out there who actually knows how to code,

598
00:29:06,520 --> 00:29:08,240
Speaker 3: possibly the only one I know.

599
00:29:08,320 --> 00:29:09,560
Speaker 2: There's gonna be a couple others.

600
00:29:09,640 --> 00:29:12,960
Speaker 3: You think two or three, Okay, we'll find them.

601
00:29:13,040 --> 00:29:15,680
Speaker 4: Yeah. I am the first Democrat elected in New York

602
00:29:15,720 --> 00:29:18,080
Speaker 4: State at any level with a degree in computer science.

603
00:29:18,240 --> 00:29:20,400
Speaker 4: I do think there have been nationwide, like four or

604
00:29:20,400 --> 00:29:23,000
Speaker 4: five Congress members that have that, so they're out there.

605
00:29:24,440 --> 00:29:27,600
Speaker 4: But my time with Palenter informs what I do in government.

606
00:29:27,640 --> 00:29:30,960
Speaker 4: I think in two main ways. The first is the

607
00:29:31,000 --> 00:29:33,080
Speaker 4: work isn't done when the bill is signed, right, It's

608
00:29:33,120 --> 00:29:37,200
Speaker 4: about the actual implementation. Everything Palaner does is about implementing

609
00:29:37,280 --> 00:29:39,720
Speaker 4: things that have already been passed, and there's huge challenges

610
00:29:39,760 --> 00:29:41,920
Speaker 4: in getting it to work. But the second thing is

611
00:29:41,960 --> 00:29:45,160
Speaker 4: that basis in data and actually tracking your results and

612
00:29:45,200 --> 00:29:48,440
Speaker 4: seeing how you've done over time, so few politicians will say, Hey,

613
00:29:48,440 --> 00:29:50,840
Speaker 4: that bill I passed two years ago, here's how it's working,

614
00:29:50,920 --> 00:29:53,320
Speaker 4: and even more rares, here's how it's not. I did

615
00:29:53,360 --> 00:29:56,200
Speaker 4: my town hall three weeks ago. I once a year,

616
00:29:56,240 --> 00:29:59,560
Speaker 4: get up in front of my entire constituency, spend two

617
00:29:59,600 --> 00:30:01,880
Speaker 4: hours answering any questions people have. We don't screen it

618
00:30:01,920 --> 00:30:04,520
Speaker 4: at all. But I led off actually with hey, here's

619
00:30:04,560 --> 00:30:06,400
Speaker 4: a bill I passed a year ago, and the data

620
00:30:06,400 --> 00:30:09,080
Speaker 4: shows it's not working. And here's what I've learned about that,

621
00:30:09,120 --> 00:30:10,160
Speaker 4: and how we're going to change that.

622
00:30:10,920 --> 00:30:13,800
Speaker 3: How exactly do you judge performance by the government, Because

623
00:30:13,840 --> 00:30:17,120
Speaker 3: I like the idea of tracking I guess alpha in

624
00:30:17,200 --> 00:30:19,720
Speaker 3: the civil service in some way, But would it just

625
00:30:19,760 --> 00:30:24,240
Speaker 3: be based on the bill execution or I guess response

626
00:30:24,360 --> 00:30:26,080
Speaker 3: rates from the public. How would you do that?

627
00:30:26,320 --> 00:30:28,880
Speaker 4: So I'll give you an example. We passed a bill

628
00:30:28,960 --> 00:30:33,720
Speaker 4: to raise the statutory maximum fine on telemarketers, by far

629
00:30:33,800 --> 00:30:36,000
Speaker 4: my most popular bill. But the question is that's the

630
00:30:36,040 --> 00:30:40,800
Speaker 4: statutory maximum. Does that actually lead to more fines, to

631
00:30:40,920 --> 00:30:43,920
Speaker 4: higher fines, to more negotiated settlements? Is raising that pressure?

632
00:30:43,960 --> 00:30:45,720
Speaker 4: And so we just looked at the Secretary of State's

633
00:30:45,760 --> 00:30:48,440
Speaker 4: data of how much were the fines before that bill

634
00:30:48,520 --> 00:30:50,840
Speaker 4: was passed and afterwards, and we found there were four

635
00:30:50,840 --> 00:30:53,480
Speaker 4: times as many finds actually there. So raising the statutory

636
00:30:53,560 --> 00:30:56,360
Speaker 4: max brought the bad companies to the table to actually

637
00:30:56,360 --> 00:30:59,360
Speaker 4: negotiate and hopefully is changing that behavior. The example I

638
00:30:59,360 --> 00:31:02,880
Speaker 4: gave of one that didn't work was around mopeds and

639
00:31:02,920 --> 00:31:05,600
Speaker 4: e bikes in New York. So you know this is

640
00:31:05,680 --> 00:31:07,160
Speaker 4: I'm going to be very local to New York.

641
00:31:07,200 --> 00:31:11,600
Speaker 2: I know there's a nationwide, nationwide primary, but it'd be

642
00:31:11,680 --> 00:31:15,120
Speaker 2: nice if there were some talk of actual relevant things

643
00:31:15,120 --> 00:31:16,040
Speaker 2: for the twelfth district.

644
00:31:16,160 --> 00:31:19,320
Speaker 4: So we have these delivery vehicles that are whizzing all

645
00:31:19,320 --> 00:31:21,960
Speaker 4: over and people are scared as they see them go

646
00:31:22,000 --> 00:31:23,800
Speaker 4: the wrong way, et cetera. And there's a lot of

647
00:31:23,880 --> 00:31:25,720
Speaker 4: discussion about what we can do to make people safe

648
00:31:25,760 --> 00:31:28,360
Speaker 4: on the streets. One aspect is that mopeds were already

649
00:31:28,360 --> 00:31:31,600
Speaker 4: required to be registered in New York State, but almost

650
00:31:31,760 --> 00:31:34,760
Speaker 4: never were. So I passed a bill that would require

651
00:31:34,840 --> 00:31:37,280
Speaker 4: mopeds to be registered at the point of sale. Right,

652
00:31:37,320 --> 00:31:40,080
Speaker 4: So think about how many mopeds you think exist in

653
00:31:40,120 --> 00:31:43,680
Speaker 4: New York City. The number that were actually registered when

654
00:31:43,680 --> 00:31:47,120
Speaker 4: the bill came into effect this January was about seventeen hundred.

655
00:31:47,800 --> 00:31:50,280
Speaker 2: Wow, And so I feel like, by the way, this

656
00:31:50,320 --> 00:31:52,320
Speaker 2: would be like one of those like Hedge Fund interview

657
00:31:52,400 --> 00:31:55,440
Speaker 2: questioned how many mopeds are in New York City. I

658
00:31:55,480 --> 00:31:57,560
Speaker 2: want to see how your brain thinks, and.

659
00:31:57,480 --> 00:31:59,200
Speaker 4: I think if your answer is seventeen hundred, the Hedge

660
00:31:59,200 --> 00:32:01,160
Speaker 4: Fund's not hiring right, Like that is not what you

661
00:32:01,160 --> 00:32:03,840
Speaker 4: would estimate. My thesis was it was that people were

662
00:32:03,880 --> 00:32:05,720
Speaker 4: walking out of the store without registering and being told

663
00:32:05,800 --> 00:32:08,280
Speaker 4: to do that. We looked a month ago. The number

664
00:32:08,280 --> 00:32:11,080
Speaker 4: that are registered is now fourteen hundred. So it's not

665
00:32:11,120 --> 00:32:13,560
Speaker 4: that people weren't registering when they came to the store.

666
00:32:13,560 --> 00:32:16,920
Speaker 4: It's that registration expires after a year and no one's reregistering.

667
00:32:17,120 --> 00:32:18,680
Speaker 4: So that was an example of like I had a thesis.

668
00:32:18,720 --> 00:32:20,240
Speaker 4: I passed the bill. I think the bill is still

669
00:32:20,280 --> 00:32:22,560
Speaker 4: overall good. People should register as they leave the store,

670
00:32:22,840 --> 00:32:24,840
Speaker 4: But it didn't actually solve the problem we were out

671
00:32:24,840 --> 00:32:26,560
Speaker 4: to solve. And here's now what I'm going to do

672
00:32:26,600 --> 00:32:28,000
Speaker 4: about it, And I think we need a lot more

673
00:32:28,000 --> 00:32:28,880
Speaker 4: of that in government.

674
00:32:28,880 --> 00:32:30,360
Speaker 3: Wait, what are you going to do about it?

675
00:32:30,400 --> 00:32:32,520
Speaker 4: Well, we're thinking about how we can encourage the actual

676
00:32:32,560 --> 00:32:35,680
Speaker 4: registrate the reregistration over time. So one of the aspects

677
00:32:35,840 --> 00:32:39,280
Speaker 4: is requiring all of the delivery apps to actually check

678
00:32:39,360 --> 00:32:41,400
Speaker 4: that the drivers are registering and they sign up and

679
00:32:41,440 --> 00:32:42,920
Speaker 4: to make sure they keep that up to date.

680
00:32:43,600 --> 00:32:46,720
Speaker 2: Makes sense to me. You know, it feels like we're

681
00:32:46,760 --> 00:32:51,880
Speaker 2: like in this sort of era of every day you know,

682
00:32:51,960 --> 00:32:55,880
Speaker 2: mentioned telemarketers, Like we're just like waiting for the muck

683
00:32:55,960 --> 00:32:59,080
Speaker 2: of like low level scams everywhere. Every day I get

684
00:32:59,080 --> 00:33:01,280
Speaker 2: text messages, I know, they'll say, like from an un

685
00:33:01,320 --> 00:33:03,680
Speaker 2: known number of Big Joe, are you coming to barbecue?

686
00:33:03,720 --> 00:33:06,200
Speaker 2: And I'm like, no, I might have, like said, I

687
00:33:06,280 --> 00:33:08,400
Speaker 2: might have agreed to you barbecue with someone. That sounds

688
00:33:08,440 --> 00:33:10,240
Speaker 2: like something I could have done, but I'm like, pretty

689
00:33:10,280 --> 00:33:14,240
Speaker 2: sure it's fake. There's you go on like Amazon, and

690
00:33:14,280 --> 00:33:18,600
Speaker 2: there's AI generated books about books and every author just generally,

691
00:33:18,720 --> 00:33:24,080
Speaker 2: it feels like we're in a culture and economy, an

692
00:33:24,080 --> 00:33:28,720
Speaker 2: era of like persistent low level grift across almost every

693
00:33:28,760 --> 00:33:31,760
Speaker 2: dimension of our lives. And I'm sure there's many reasons

694
00:33:31,760 --> 00:33:33,760
Speaker 2: for it, But when people talk about a low trust

695
00:33:33,800 --> 00:33:36,720
Speaker 2: society and what's happening, how much of this do you

696
00:33:36,800 --> 00:33:40,360
Speaker 2: think is just because like we're inundated with people trying

697
00:33:40,360 --> 00:33:43,200
Speaker 2: to rob us every day and somewhat digitally or otherwise.

698
00:33:43,440 --> 00:33:45,239
Speaker 4: I think that's a big part of it. And the

699
00:33:45,280 --> 00:33:48,480
Speaker 4: only way to solve that problem is you need actual enforcement.

700
00:33:48,520 --> 00:33:50,120
Speaker 4: You need there to be consequencing No.

701
00:33:50,480 --> 00:33:53,480
Speaker 2: Sorry, just to follow up on this, like what I

702
00:33:53,480 --> 00:33:56,360
Speaker 2: guess the reason I'm asking is because you know, it's

703
00:33:56,400 --> 00:33:59,280
Speaker 2: like someone running from us. They moved to the twelfth district,

704
00:33:59,560 --> 00:34:02,800
Speaker 2: and like I'm going to call out Trump's corruption and

705
00:34:03,440 --> 00:34:06,239
Speaker 2: there are all these like Trump is a lot all

706
00:34:06,320 --> 00:34:09,040
Speaker 2: this like big stuff and you know whatever. Many of

707
00:34:09,080 --> 00:34:12,040
Speaker 2: these issues are like you know, legitimate, et cetera. But

708
00:34:12,200 --> 00:34:15,400
Speaker 2: almost nobody in elected office seems to be just like

709
00:34:15,440 --> 00:34:17,400
Speaker 2: talking about like who's going to do something about the

710
00:34:17,400 --> 00:34:19,600
Speaker 2: text message scams, who's going to do something about the

711
00:34:19,640 --> 00:34:22,879
Speaker 2: AI books? All of these like big national issues, But

712
00:34:22,880 --> 00:34:25,240
Speaker 2: that doesn't affect me on a day to day basis

713
00:34:25,440 --> 00:34:28,680
Speaker 2: the way this sort of like persistent abuse of technology

714
00:34:28,800 --> 00:34:30,680
Speaker 2: constantly does every minute of my life.

715
00:34:31,360 --> 00:34:33,600
Speaker 4: I think that's why podcasts like this are great, because

716
00:34:33,600 --> 00:34:36,120
Speaker 4: you get into the details. No, but the top level

717
00:34:36,120 --> 00:34:38,520
Speaker 4: news doesn't cover a lot of that. Right. I have

718
00:34:38,560 --> 00:34:41,000
Speaker 4: a colleague, John Rivera, who has a bill in the

719
00:34:41,200 --> 00:34:45,319
Speaker 4: State Assembly on requiring disclosure for AI generated books on

720
00:34:45,400 --> 00:34:48,320
Speaker 4: Amazon like that. Specific example is a bill that exists

721
00:34:48,320 --> 00:34:51,719
Speaker 4: in New York, the Telemarketing BILLI pass also applies to

722
00:34:51,760 --> 00:34:53,760
Speaker 4: text messages, and so we tell people how to report

723
00:34:53,800 --> 00:34:56,399
Speaker 4: those and maybe get fines coming from that. I think

724
00:34:56,440 --> 00:34:59,359
Speaker 4: this is key to both parties in twenty twenty six

725
00:34:59,400 --> 00:35:02,239
Speaker 4: is talking about governments effect and trying to make life

726
00:35:02,239 --> 00:35:04,680
Speaker 4: a little bit better. I think my other most popular

727
00:35:04,680 --> 00:35:07,040
Speaker 4: bill besides the telemarketing one, was my click to cancel

728
00:35:07,080 --> 00:35:11,080
Speaker 4: bill this year to allow New Yorkers love that it's now,

729
00:35:11,120 --> 00:35:13,600
Speaker 4: you're right as it it's over. First, New Yorkers need

730
00:35:13,640 --> 00:35:15,359
Speaker 4: to be able to cancel a subscription the same way

731
00:35:15,400 --> 00:35:16,279
Speaker 4: they signed up for it.

732
00:35:16,440 --> 00:35:18,920
Speaker 3: I have a Conde Nas subscription that I've been trying

733
00:35:18,920 --> 00:35:21,680
Speaker 3: to cancel for like a year, and I cannot log

734
00:35:21,719 --> 00:35:24,799
Speaker 3: into the site. I can't find the original email where

735
00:35:24,800 --> 00:35:27,239
Speaker 3: I've actually signed up for it, and that's I think

736
00:35:27,320 --> 00:35:29,760
Speaker 3: twelve dollars a month that I'm just wasting.

737
00:35:30,000 --> 00:35:32,640
Speaker 4: I mean, government needs to get big things done, absolutely,

738
00:35:32,719 --> 00:35:34,759
Speaker 4: but we also need to get small things right. You

739
00:35:34,800 --> 00:35:35,839
Speaker 4: just need to make life better.

740
00:35:51,600 --> 00:35:55,239
Speaker 3: Apropos of nothing. This is a completely random seg but

741
00:35:55,360 --> 00:35:59,480
Speaker 3: one of the interesting things about Eric Adams was his

742
00:35:59,560 --> 00:36:03,719
Speaker 3: interest and digital assets and cryptocurrency and tokenization and all

743
00:36:03,760 --> 00:36:07,160
Speaker 3: of that. What did you think about those efforts? Do

744
00:36:07,239 --> 00:36:11,080
Speaker 3: they actually matter to the industry and to New York

745
00:36:11,160 --> 00:36:14,800
Speaker 3: more widely, or is it just sort of a pet project.

746
00:36:16,560 --> 00:36:19,120
Speaker 4: I have not seen the direct results of what he

747
00:36:19,239 --> 00:36:21,120
Speaker 4: was working on. That doesn't mean they weren't there, It's

748
00:36:21,160 --> 00:36:23,319
Speaker 4: just not where I have been focused. I actually worked

749
00:36:23,320 --> 00:36:27,040
Speaker 4: on a different bill this year around crypto in the

750
00:36:27,160 --> 00:36:30,960
Speaker 4: legislature where New York really took the lead in creating

751
00:36:31,080 --> 00:36:33,840
Speaker 4: legal structures for many of these companies through the Limited

752
00:36:33,880 --> 00:36:37,759
Speaker 4: Purpose Trust and the bit license. But now with new

753
00:36:37,800 --> 00:36:41,080
Speaker 4: action at the federal level, they have said, you know,

754
00:36:41,680 --> 00:36:44,720
Speaker 4: there's going to be federal licensing for these trust companies

755
00:36:44,719 --> 00:36:47,440
Speaker 4: for stable coins, and they will defer to states, but

756
00:36:47,560 --> 00:36:53,080
Speaker 4: only if the states have detailed rules in legislation or regulation,

757
00:36:53,160 --> 00:36:55,480
Speaker 4: in statute or regulation. And New York's done it almost

758
00:36:55,520 --> 00:36:57,840
Speaker 4: all by guidance. And so I really fought for a

759
00:36:57,880 --> 00:37:00,640
Speaker 4: bill this year to standardize what if I had done

760
00:37:00,680 --> 00:37:04,400
Speaker 4: in random opinions, put it into statute so that people

761
00:37:04,480 --> 00:37:06,560
Speaker 4: know what the rules of the road are and so

762
00:37:06,640 --> 00:37:09,440
Speaker 4: that New York can keep its regulatory structure. It passed

763
00:37:09,440 --> 00:37:11,520
Speaker 4: the Senate. Unfortunately we didn't get it done in the Assembly,

764
00:37:11,600 --> 00:37:14,120
Speaker 4: and now most of the companies in New York are

765
00:37:14,160 --> 00:37:16,800
Speaker 4: just applying for the federal charter, and we as New Yorkers,

766
00:37:16,800 --> 00:37:19,160
Speaker 4: are going to lose our ability to engage. So I think,

767
00:37:19,200 --> 00:37:23,759
Speaker 4: regardless of your broad view on crypto, finding ways to

768
00:37:23,880 --> 00:37:27,120
Speaker 4: set rules of the road that allow for innovation and

769
00:37:27,160 --> 00:37:28,480
Speaker 4: allow for us to have a say in how it

770
00:37:28,520 --> 00:37:30,399
Speaker 4: develops is something everyone should get behind.

771
00:37:30,760 --> 00:37:32,120
Speaker 2: Ye know, I'm looking at a map of the New

772
00:37:32,200 --> 00:37:35,440
Speaker 2: York twelfth Congressional district. I think one thing that really

773
00:37:35,480 --> 00:37:39,080
Speaker 2: stands out. It's not some weird, snaky district. It's just

774
00:37:39,120 --> 00:37:41,680
Speaker 2: a real big square in the middle of Manhattan. Like

775
00:37:41,719 --> 00:37:44,120
Speaker 2: people who don't live here might not appreciate this is

776
00:37:44,200 --> 00:37:46,400
Speaker 2: like this is prime real estate. You're right for it,

777
00:37:46,440 --> 00:37:48,719
Speaker 2: I argue, I mean, this is it might be the

778
00:37:48,800 --> 00:37:51,480
Speaker 2: highest GDP district in the world in the country.

779
00:37:51,480 --> 00:37:53,279
Speaker 4: Are pretty close to it, pretty close to it, if

780
00:37:53,280 --> 00:37:56,120
Speaker 4: not the one. Yes, there's more fortune five hundred companies

781
00:37:56,160 --> 00:37:58,560
Speaker 4: headquartered there than I think at least thirty seven states.

782
00:37:58,600 --> 00:38:00,480
Speaker 2: You're a few blocks away from me, by the way,

783
00:38:00,520 --> 00:38:02,400
Speaker 2: I'm in these village. It looks like it just cuts

784
00:38:02,440 --> 00:38:03,960
Speaker 2: off a few blocks.

785
00:38:04,120 --> 00:38:06,080
Speaker 4: We'll fix it every district.

786
00:38:06,600 --> 00:38:08,680
Speaker 2: It must be weird, like, you know, thinking about going

787
00:38:08,719 --> 00:38:12,040
Speaker 2: back to the beginning with the tech industry and criticizing you,

788
00:38:12,160 --> 00:38:14,719
Speaker 2: and it must be sort of weird that, like you're

789
00:38:14,840 --> 00:38:17,840
Speaker 2: a politician who could actually like talk and like giggle

790
00:38:17,880 --> 00:38:19,880
Speaker 2: flops and stuff like that. Because there's a lot of

791
00:38:19,880 --> 00:38:24,640
Speaker 2: like anti AI people, et cetera, whatever, or open standards

792
00:38:24,680 --> 00:38:28,560
Speaker 2: for digital encryption. It must be kind of interesting that, Okay,

793
00:38:28,600 --> 00:38:31,720
Speaker 2: here is someone who wants to like hold the industry

794
00:38:31,760 --> 00:38:39,120
Speaker 2: to some legislative standards, who's frankly like knowledge informed. Yeah, yeh, knowledgeable. Yeah,

795
00:38:39,200 --> 00:38:40,480
Speaker 2: that must be sort of weird for them.

796
00:38:40,920 --> 00:38:42,959
Speaker 4: I think it is, and I think it's why all

797
00:38:43,040 --> 00:38:45,959
Speaker 4: of their spending against me so far as backfiring, because

798
00:38:46,000 --> 00:38:47,879
Speaker 4: people are like, oh, yeah, who is this guy who

799
00:38:47,880 --> 00:38:51,759
Speaker 4: doesn't understand tech who's doing uninformed things? And then it's like, wait,

800
00:38:51,800 --> 00:38:54,000
Speaker 4: it's the guy with software patents who worked at Palenteer,

801
00:38:54,000 --> 00:38:56,000
Speaker 4: who has a master's in computer science. There is a

802
00:38:56,440 --> 00:38:59,080
Speaker 4: disconnect there. I think, you know, I have support from

803
00:38:59,120 --> 00:39:01,400
Speaker 4: a lot of the people that are like my age

804
00:39:01,400 --> 00:39:03,960
Speaker 4: that work in the tech industry, the ones actually building

805
00:39:04,000 --> 00:39:06,640
Speaker 4: it because they see the power of what they're building

806
00:39:06,640 --> 00:39:08,680
Speaker 4: and think there should be some protections for it. It's

807
00:39:08,719 --> 00:39:11,960
Speaker 4: really those at the top who are i think, primarily

808
00:39:12,000 --> 00:39:14,560
Speaker 4: focused on profits that don't want government getting involved at all.

809
00:39:15,120 --> 00:39:17,600
Speaker 3: I have a theoretical question, going back to the idea

810
00:39:17,640 --> 00:39:20,279
Speaker 3: of hedge funds, asking random questions to see how you

811
00:39:20,320 --> 00:39:24,400
Speaker 3: think if you had, I don't know, a Bloomberg terminal

812
00:39:24,600 --> 00:39:30,360
Speaker 3: equivalent of government data, what would you be most excited

813
00:39:30,400 --> 00:39:34,839
Speaker 3: about looking up or what correlation would you be most

814
00:39:34,880 --> 00:39:35,920
Speaker 3: excited about finding?

815
00:39:36,480 --> 00:39:39,279
Speaker 4: That is a great question. I'm going to stall at

816
00:39:39,320 --> 00:39:41,960
Speaker 4: first by pointing out that actually do have a certification

817
00:39:41,960 --> 00:39:45,160
Speaker 4: in the Bloomberg terminal from my first job. So I'm

818
00:39:45,239 --> 00:39:48,000
Speaker 4: imagining realistically.

819
00:39:48,320 --> 00:39:49,760
Speaker 2: So you know that all we do on the terminal

820
00:39:49,800 --> 00:39:53,880
Speaker 2: is just look at two lines I've proved correlation.

821
00:39:55,320 --> 00:40:01,200
Speaker 4: I think ways of running more well, not maybe running

822
00:40:01,200 --> 00:40:04,880
Speaker 4: more natural experiments, but looking at over time how investments

823
00:40:04,880 --> 00:40:08,160
Speaker 4: the government pays off. So we only budget year by year,

824
00:40:08,680 --> 00:40:12,319
Speaker 4: and so you have things like investing in early childcare

825
00:40:12,880 --> 00:40:17,440
Speaker 4: that boosts your immediate who can be employed, but also

826
00:40:17,880 --> 00:40:20,120
Speaker 4: will help the kids if you have universal pre k

827
00:40:20,200 --> 00:40:23,960
Speaker 4: et cetera. But that payback in the government, if you're

828
00:40:23,960 --> 00:40:26,320
Speaker 4: just thinking fiscally is not going to come for twenty

829
00:40:26,400 --> 00:40:30,000
Speaker 4: thirty years, and as we budget year to year, that is,

830
00:40:30,320 --> 00:40:32,680
Speaker 4: we think of that as a cost. I think if

831
00:40:32,680 --> 00:40:36,080
Speaker 4: we had more ways of tying the effects previously in

832
00:40:36,160 --> 00:40:39,520
Speaker 4: budgets with what we're seeing now, it would change radically

833
00:40:39,600 --> 00:40:40,640
Speaker 4: where we would invest.

834
00:40:41,320 --> 00:40:45,040
Speaker 2: Have you played around with the AI co generation apps? Munch, Yeah. Yeah.

835
00:40:45,080 --> 00:40:48,799
Speaker 4: So I when I came into office, one of the

836
00:40:48,840 --> 00:40:52,560
Speaker 4: things I really wanted was a coder on staff. I

837
00:40:52,600 --> 00:40:54,480
Speaker 4: was like, there's so many creative things we could do

838
00:40:54,800 --> 00:40:58,960
Speaker 4: of doing prototypes of government software or playing with the data,

839
00:40:59,239 --> 00:41:04,000
Speaker 4: but we don't pay legislative staff nearly enough. It's one

840
00:41:04,040 --> 00:41:06,319
Speaker 4: of the positions I've been most radicalized in office. But

841
00:41:06,360 --> 00:41:08,520
Speaker 4: I had this idea of all these projects I wanted

842
00:41:08,520 --> 00:41:12,120
Speaker 4: to get done. And finally this year, Hunter College gave

843
00:41:12,160 --> 00:41:14,759
Speaker 4: me an intern that was a CS major, and so

844
00:41:14,800 --> 00:41:17,200
Speaker 4: I gave her this list of like ten projects, and

845
00:41:17,239 --> 00:41:19,640
Speaker 4: she got through three of them and they were great.

846
00:41:19,960 --> 00:41:21,920
Speaker 4: But I looked at the other seven and I was like,

847
00:41:22,160 --> 00:41:25,759
Speaker 4: these are things that I know how to do, but

848
00:41:25,840 --> 00:41:28,359
Speaker 4: it's not like worth my time at the moment versus

849
00:41:28,400 --> 00:41:30,320
Speaker 4: all the things I have to do. But has AI

850
00:41:30,440 --> 00:41:33,279
Speaker 4: coding sped up, how quickly I can do these it

851
00:41:33,360 --> 00:41:37,080
Speaker 4: becomes worth it. And I using I mostly use chat GPT,

852
00:41:37,239 --> 00:41:39,080
Speaker 4: which you know, I know Claude and Gem and I

853
00:41:39,120 --> 00:41:41,040
Speaker 4: have played with. But I was able to knock out

854
00:41:41,080 --> 00:41:44,080
Speaker 4: three of those projects this year because of how more

855
00:41:44,360 --> 00:41:45,640
Speaker 4: advanced the coding has become.

856
00:41:45,800 --> 00:41:49,840
Speaker 2: So it's pretty cool. Yeah, well, Explorius, thank you so

857
00:41:49,920 --> 00:41:51,480
Speaker 2: much for coming on odd loss. There's a lot of them.

858
00:41:51,600 --> 00:41:52,440
Speaker 4: Thanks for having me.

859
00:42:04,840 --> 00:42:07,160
Speaker 2: It must be kind of weird, you know, I said

860
00:42:07,200 --> 00:42:10,120
Speaker 2: it near the end, but an AI critic that knows

861
00:42:10,120 --> 00:42:11,920
Speaker 2: something about tech, because there are a lot of AI

862
00:42:12,040 --> 00:42:14,040
Speaker 2: critics out there, and I don't get the impression a

863
00:42:14,040 --> 00:42:17,000
Speaker 2: lot of them are particularly actually well informed and do

864
00:42:17,160 --> 00:42:20,000
Speaker 2: not actually understand the sort of strongest version of the

865
00:42:20,080 --> 00:42:24,160
Speaker 2: argument that they're running against. So it's interesting that this pack,

866
00:42:24,640 --> 00:42:28,480
Speaker 2: this pro artificial intelligence pack, has decided to target someone

867
00:42:28,960 --> 00:42:31,400
Speaker 2: who I think a lot of people will listen to.

868
00:42:31,440 --> 00:42:33,120
Speaker 2: It's like, actually, he knows what he's talking about. He

869
00:42:33,160 --> 00:42:34,040
Speaker 2: sounds pretty reasonable.

870
00:42:34,080 --> 00:42:36,839
Speaker 3: He's also the first one they're targeting, right, so it'll

871
00:42:36,840 --> 00:42:39,360
Speaker 3: be interesting to see who they go after.

872
00:42:39,719 --> 00:42:41,319
Speaker 2: Yeah, after him.

873
00:42:41,560 --> 00:42:43,960
Speaker 3: The other thing I was thinking is if you think

874
00:42:44,000 --> 00:42:47,920
Speaker 3: about one of the threats to big tech's business models,

875
00:42:48,200 --> 00:42:51,719
Speaker 3: it's always regulation, right, or at least they say it's regulation.

876
00:42:51,960 --> 00:42:56,120
Speaker 3: They seem to hate regulation or over regulation. I should say,

877
00:42:56,239 --> 00:43:01,000
Speaker 3: so if you think about proposing some basic safety guardrails

878
00:43:01,040 --> 00:43:03,960
Speaker 3: and privacy, that would seem like a good thing to

879
00:43:04,040 --> 00:43:06,560
Speaker 3: me in the sense that maybe you could restore some

880
00:43:06,680 --> 00:43:10,360
Speaker 3: of the trust or faith between the general population and

881
00:43:10,400 --> 00:43:13,600
Speaker 3: the big tech companies and then you wouldn't get these

882
00:43:13,760 --> 00:43:18,320
Speaker 3: massive fights about everything else. But that's probably a long shot.

883
00:43:18,600 --> 00:43:21,839
Speaker 2: It does seem like I have to say that while

884
00:43:21,840 --> 00:43:23,800
Speaker 2: a lot of what he said made les of sense,

885
00:43:24,280 --> 00:43:26,719
Speaker 2: I sort of sounded very reasonable. You know, I do

886
00:43:26,800 --> 00:43:31,799
Speaker 2: think that it's legitimate concern, like these issues aren't going

887
00:43:31,880 --> 00:43:35,160
Speaker 2: to hobble Deep Seek. Now, well, Deep does Deep Seek

888
00:43:35,239 --> 00:43:37,000
Speaker 2: want to be in the app store? And can US

889
00:43:37,080 --> 00:43:39,600
Speaker 2: laws and join them from being in the app store? Sure,

890
00:43:39,719 --> 00:43:42,560
Speaker 2: but they're not going to hobble Deep Seek's development of

891
00:43:42,600 --> 00:43:46,200
Speaker 2: a model. And if they want to be a dominant

892
00:43:46,239 --> 00:43:49,600
Speaker 2: player in every country but the US and they're completely

893
00:43:49,680 --> 00:43:54,879
Speaker 2: unconstrained and the American advanced labs are constrained, I do

894
00:43:54,920 --> 00:43:58,279
Speaker 2: think that is legitimate concern that the industry might have.

895
00:43:58,640 --> 00:44:03,120
Speaker 2: I also think it's a legitimate concern that you don't

896
00:44:03,120 --> 00:44:07,040
Speaker 2: want to disincentivize the reporting of safety issues. That if

897
00:44:07,080 --> 00:44:11,120
Speaker 2: a company gets penalized for acknowledging that something in their

898
00:44:11,160 --> 00:44:14,399
Speaker 2: model like violated some safety line, you do not want

899
00:44:14,440 --> 00:44:16,320
Speaker 2: to have the risk of like, we're just going to

900
00:44:16,440 --> 00:44:20,640
Speaker 2: ignore this. And I think it's reasonable that the industry

901
00:44:20,760 --> 00:44:25,640
Speaker 2: might be concerned that regulations could benefit the incumbents. You know,

902
00:44:25,840 --> 00:44:28,200
Speaker 2: he argued against that. He's like, oh no, like this

903
00:44:28,320 --> 00:44:30,840
Speaker 2: is not going to affect like startups, et cetera. But

904
00:44:31,280 --> 00:44:33,880
Speaker 2: there is always that risk that a monetary fine is

905
00:44:33,880 --> 00:44:35,960
Speaker 2: something that's very easy for big companies to pay, and

906
00:44:36,040 --> 00:44:39,680
Speaker 2: more marginal ones are more difficult. So you know, these

907
00:44:39,680 --> 00:44:41,840
Speaker 2: are very tricky things. But as you know, on the

908
00:44:41,880 --> 00:44:46,120
Speaker 2: service more than on the service an extremely knowledgeable guy

909
00:44:46,200 --> 00:44:50,000
Speaker 2: who what he says, none of it sounds particularly ridiculous.

910
00:44:50,000 --> 00:44:52,440
Speaker 2: And I have to say, if I were running in

911
00:44:52,480 --> 00:44:55,680
Speaker 2: a primary that had like twenty five people, including like

912
00:44:55,760 --> 00:44:59,319
Speaker 2: a Kennedy air and some guy on Twitter who like

913
00:44:59,719 --> 00:45:03,320
Speaker 2: puts orange Man bad a lot of times, that sounds

914
00:45:03,360 --> 00:45:06,200
Speaker 2: like a good way to stand out from the crowd,

915
00:45:06,239 --> 00:45:08,280
Speaker 2: to be the AI industry's number one target.

916
00:45:08,680 --> 00:45:11,000
Speaker 3: You know, I agree with you on the incumbent point,

917
00:45:11,080 --> 00:45:13,440
Speaker 3: but just in terms of US versus China, I think

918
00:45:13,440 --> 00:45:16,680
Speaker 3: people forget that China has its own restrictions quite right,

919
00:45:16,800 --> 00:45:21,160
Speaker 3: severe ones on censorship, right and trawling. I guess every

920
00:45:21,200 --> 00:45:24,480
Speaker 3: single communication in the world in case someone mentions a

921
00:45:24,480 --> 00:45:28,399
Speaker 3: certain cartoon character, Yeah, seems like that's that's a big

922
00:45:28,480 --> 00:45:32,840
Speaker 3: regulatory as well. So I don't know. I find the

923
00:45:33,360 --> 00:45:37,640
Speaker 3: competition aspect of it overhyped in some sense. But anyway,

924
00:45:37,680 --> 00:45:39,480
Speaker 3: we could talk about this for ages and a lot

925
00:45:39,480 --> 00:45:42,480
Speaker 3: of these issues are clearly like ongoing and no one

926
00:45:42,560 --> 00:45:46,279
Speaker 3: has the the solution yet. We're trying to work that out.

927
00:45:46,760 --> 00:45:47,839
Speaker 3: All right, shall we leave it there?

928
00:45:47,920 --> 00:45:48,640
Speaker 2: Let's leave it there.

929
00:45:48,920 --> 00:45:51,440
Speaker 3: This has been another episode of the All Thoughts podcast.

930
00:45:51,600 --> 00:45:54,840
Speaker 3: I'm Tracy Alloway. You can follow me at Tracy Alloway.

931
00:45:54,600 --> 00:45:57,600
Speaker 2: And I'm Jill Wisenthal. You can follow me at the Stalwart.

932
00:45:57,800 --> 00:46:01,240
Speaker 2: Follow our guest Alex Boris, He's at Alex Boris. Follow

933
00:46:01,280 --> 00:46:04,400
Speaker 2: our producers Kerman Rodriguez at Kerman, Arma, dash Ol Bennett

934
00:46:04,400 --> 00:46:07,279
Speaker 2: at Dashbot, and Kilbrooks at Kilbrooks. And for more od

935
00:46:07,320 --> 00:46:09,320
Speaker 2: lots content. You can find all of our episodes in

936
00:46:09,360 --> 00:46:12,400
Speaker 2: a daily newsletter at Bloomberg dot com slash odd lots,

937
00:46:12,600 --> 00:46:15,000
Speaker 2: and you can shot with fellow listeners in our discord

938
00:46:15,000 --> 00:46:18,480
Speaker 2: twenty four to seven Discord dot gg slash outlots.

939
00:46:18,360 --> 00:46:20,520
Speaker 3: And if you enjoy ad blots. If you like it

940
00:46:20,560 --> 00:46:23,239
Speaker 3: when we do these AI episodes, then please leave a

941
00:46:23,280 --> 00:46:26,920
Speaker 3: positive review on your favorite podcast platform. And remember, if

942
00:46:26,960 --> 00:46:29,600
Speaker 3: you are a Bloomberg subscriber, you can listen to all

943
00:46:29,640 --> 00:46:32,600
Speaker 3: of our episodes absolutely ad free. All you need to

944
00:46:32,600 --> 00:46:35,279
Speaker 3: do is find the Bloomberg channel on Apple Podcasts and

945
00:46:35,400 --> 00:47:03,920
Speaker 3: follow the instructions there. Thanks for listening, not in e