1
00:00:15,604 --> 00:00:27,724
Speaker 1: Pushkin. Welcome back to Risky Business, a show about making

2
00:00:27,764 --> 00:00:31,724
Speaker 1: better decisions. I'm Maria Kanikova and I'm Nate Silver. So

3
00:00:31,884 --> 00:00:34,324
Speaker 1: today on the show, note, we've got we've got a

4
00:00:34,404 --> 00:00:39,284
Speaker 1: pretty interesting Risky Business e news week. We've got GPT

5
00:00:39,444 --> 00:00:43,444
Speaker 1: five being released, which is maybe less of a deal

6
00:00:43,484 --> 00:00:46,244
Speaker 1: than the launch hype announcements were, but still still a

7
00:00:46,284 --> 00:00:49,044
Speaker 1: big deal. And then we have some interesting trade stuff

8
00:00:49,084 --> 00:00:52,364
Speaker 1: going on with AI chips right with Nvidio.

9
00:00:52,804 --> 00:00:54,244
Speaker 2: Interesting is an interesting reader?

10
00:00:55,244 --> 00:00:57,884
Speaker 1: Yes, well, well we've got gp.

11
00:00:57,884 --> 00:01:00,244
Speaker 2: Now there have been five GPT since the last GTA

12
00:01:00,484 --> 00:01:02,364
Speaker 2: Grand Theft Donald. Sorry to throw that in there.

13
00:01:03,124 --> 00:01:18,164
Speaker 1: Fascinating to well, before we before we get into it, Nate,

14
00:01:18,244 --> 00:01:21,084
Speaker 1: I just wanted to say, you know, we're taping this

15
00:01:21,244 --> 00:01:24,124
Speaker 1: on the twelfth of August Tuesday, so a few days

16
00:01:24,164 --> 00:01:27,804
Speaker 1: before listeners will hear it. But congratulations. Today is the

17
00:01:27,924 --> 00:01:32,004
Speaker 1: launch of the paperback of On the Edge.

18
00:01:32,004 --> 00:01:34,164
Speaker 2: On the Edge the Guard We're seeing Everything, the bestseller

19
00:01:34,284 --> 00:01:38,684
Speaker 2: by groundbreaking author Nate Silver. Yeah, so the paperback is out.

20
00:01:38,724 --> 00:01:40,924
Speaker 2: There's a new forward, or I think it's called a

21
00:01:41,004 --> 00:01:44,804
Speaker 2: preface technically. I was just at Barnes and Noble signing

22
00:01:45,204 --> 00:01:48,084
Speaker 2: some copies. A little sweaty. I walked here from there.

23
00:01:48,164 --> 00:01:49,764
Speaker 2: It's hot in the middle of August in New York,

24
00:01:49,804 --> 00:01:53,604
Speaker 2: breaking news. But yeah, no, I think the book holds

25
00:01:53,644 --> 00:01:55,764
Speaker 2: up really well. It covers a lot of really risky

26
00:01:55,804 --> 00:01:59,444
Speaker 2: business esque topics. And you know, it's a big book,

27
00:01:59,684 --> 00:02:02,124
Speaker 2: cheaper now with paback. It fits better on a shelf,

28
00:02:02,804 --> 00:02:05,044
Speaker 2: not quite as thick, and there's new content. So I

29
00:02:05,044 --> 00:02:07,604
Speaker 2: would strongly recommend, of course. I mean, you know, a

30
00:02:07,604 --> 00:02:09,964
Speaker 2: little biased here, but like, thank you.

31
00:02:09,924 --> 00:02:14,164
Speaker 1: Of course. Well, I'm excited to read the new preface,

32
00:02:14,764 --> 00:02:17,804
Speaker 1: and yeah, I definitely recommend the book to everyone. I

33
00:02:17,884 --> 00:02:20,804
Speaker 1: will try to repost the review I did of it

34
00:02:20,924 --> 00:02:24,884
Speaker 1: on my substack so that people can get reintroduced to

35
00:02:24,924 --> 00:02:29,124
Speaker 1: it one more time. Anyway, it's a fantastic book. Congrats Nate.

36
00:02:29,724 --> 00:02:35,004
Speaker 1: And let's get into some rivarian topics, like the release

37
00:02:35,204 --> 00:02:40,404
Speaker 1: of GPT five from open Ai. Nate, have you had

38
00:02:40,444 --> 00:02:43,724
Speaker 1: a chance to use GPT five yet we don't.

39
00:02:43,604 --> 00:02:45,964
Speaker 2: Have much choice, you know, they kind of steer you

40
00:02:46,124 --> 00:02:49,044
Speaker 2: into GPT five if you And at first I'm like, oh, okay,

41
00:02:49,084 --> 00:02:51,324
Speaker 2: I guess I pay for the pro plan and so

42
00:02:51,324 --> 00:02:52,924
Speaker 2: I'm like, oh wow, I'm one of the privileged few

43
00:02:52,964 --> 00:02:56,804
Speaker 2: and now I don't think there's an easy way to

44
00:02:56,804 --> 00:02:58,724
Speaker 2: get back into all the all the old family of

45
00:02:58,764 --> 00:03:01,764
Speaker 2: GPT products that we knew in love before, Like people

46
00:03:01,764 --> 00:03:04,884
Speaker 2: did get it attached to the different models for different reasons,

47
00:03:04,924 --> 00:03:08,444
Speaker 2: and so now it kind of is implicitly trying to

48
00:03:08,524 --> 00:03:11,364
Speaker 2: figure out what you want, which is kind of part

49
00:03:11,364 --> 00:03:13,204
Speaker 2: of I mean, it's interesting. Right. So on the one hand,

50
00:03:13,324 --> 00:03:14,604
Speaker 2: you know, I kind of made the joke before about

51
00:03:14,644 --> 00:03:17,324
Speaker 2: kind of comparing it to like a video game release.

52
00:03:17,364 --> 00:03:20,004
Speaker 2: But anything new, if we release a new presidential model

53
00:03:20,204 --> 00:03:25,044
Speaker 2: or something, right, anything new might have some kinks and bugs.

54
00:03:25,164 --> 00:03:27,204
Speaker 2: I mean, as much as you might say, okay, we

55
00:03:27,244 --> 00:03:30,284
Speaker 2: have to have it perfect before it's ships, I don't

56
00:03:30,324 --> 00:03:32,564
Speaker 2: think it's practical because like most people are gonna only

57
00:03:32,684 --> 00:03:34,444
Speaker 2: learn thing. I mean, you know, I think they did

58
00:03:34,444 --> 00:03:36,524
Speaker 2: discover that's not doing what like groc did and calling

59
00:03:36,564 --> 00:03:37,804
Speaker 2: itself Hitler for example.

60
00:03:37,964 --> 00:03:40,844
Speaker 1: Right, Yeah, I mean I don't know if we could

61
00:03:40,844 --> 00:03:42,684
Speaker 1: really call that a win. Like that seems to be

62
00:03:42,764 --> 00:03:44,244
Speaker 1: like a baseline.

63
00:03:43,804 --> 00:03:47,524
Speaker 2: Or like Google Gemini drawing multicultural Nazis, you know, so

64
00:03:47,604 --> 00:03:49,604
Speaker 2: that the bar is pretty low. But no, I think

65
00:03:49,604 --> 00:03:51,684
Speaker 2: it has a little bit of new car, new model

66
00:03:52,044 --> 00:03:54,004
Speaker 2: kind of smell a little bit. The first thing I

67
00:03:54,004 --> 00:03:55,684
Speaker 2: asked it to do was, my partner are planning a

68
00:03:55,724 --> 00:04:01,844
Speaker 2: trip to Bestly, Scandinavia or the Nordics, technically, right, I

69
00:04:01,844 --> 00:04:04,164
Speaker 2: want to visit these cities. Give us in our itinerary, right,

70
00:04:04,244 --> 00:04:07,284
Speaker 2: and like thinks and things and has something that seemed

71
00:04:07,284 --> 00:04:09,284
Speaker 2: plausible to me, very detailed, right, And then I'm like,

72
00:04:10,004 --> 00:04:12,444
Speaker 2: emails use a PBF and it like freaks out right

73
00:04:12,604 --> 00:04:16,124
Speaker 2: like shows some complicated like you know code, Python code,

74
00:04:16,164 --> 00:04:17,524
Speaker 2: and it's like, I don't know how I know this.

75
00:04:17,564 --> 00:04:20,124
Speaker 2: I don't I know this, So you know it's it's

76
00:04:20,164 --> 00:04:21,684
Speaker 2: up and down. Well, hey, at.

77
00:04:21,604 --> 00:04:23,524
Speaker 1: Least it gave you a plausible itinerary.

78
00:04:23,724 --> 00:04:24,284
Speaker 2: I so.

79
00:04:26,444 --> 00:04:29,684
Speaker 1: We've been warned. We were given an explanation by Sam

80
00:04:29,724 --> 00:04:31,644
Speaker 1: Altman because I tried to test it, you know, when

81
00:04:31,644 --> 00:04:35,364
Speaker 1: it was just released, and apparently the auto switching tool

82
00:04:35,484 --> 00:04:38,244
Speaker 1: was down, so it seemed like it was a lot

83
00:04:38,324 --> 00:04:40,444
Speaker 1: dumber than it was that it was supposed to be.

84
00:04:40,684 --> 00:04:42,204
Speaker 1: So one of the one of the things that they're

85
00:04:42,284 --> 00:04:44,324
Speaker 1: kind of really touting on this model is that it

86
00:04:44,364 --> 00:04:47,244
Speaker 1: automatically knows what you want right, whether it should think

87
00:04:47,284 --> 00:04:49,804
Speaker 1: deeply or not, which does not actually seem to be

88
00:04:49,804 --> 00:04:53,164
Speaker 1: the case. Even now when the swings back, it's like.

89
00:04:53,124 --> 00:04:54,564
Speaker 2: Hey, who got a cheek here? I don't think we

90
00:04:54,604 --> 00:04:57,684
Speaker 2: need too much deep thinking. She probably wants a quick answer,

91
00:04:57,684 --> 00:04:58,564
Speaker 2: get back to the cooking.

92
00:04:59,404 --> 00:05:04,164
Speaker 1: That's that's exactly honestly, that's that's been my experience. And

93
00:05:04,244 --> 00:05:06,524
Speaker 1: one of the first things I actually tested it on

94
00:05:06,764 --> 00:05:08,764
Speaker 1: because it says that, you know, one of the things

95
00:05:08,764 --> 00:05:12,324
Speaker 1: that it's much better on is on hallucination. So I

96
00:05:12,364 --> 00:05:15,644
Speaker 1: actually gave it a psyche question and asked for some

97
00:05:15,804 --> 00:05:19,964
Speaker 1: sources and like papers, and it unfortunately still hallucinates when

98
00:05:20,404 --> 00:05:22,764
Speaker 1: you go down that route. I was because so I'm

99
00:05:22,764 --> 00:05:26,404
Speaker 1: writing a piece this week about kindness contagion. You know,

100
00:05:26,444 --> 00:05:29,364
Speaker 1: when someone is nice, you know, how does that spread?

101
00:05:29,724 --> 00:05:32,284
Speaker 1: And so I asked it some questions about the research

102
00:05:32,284 --> 00:05:34,964
Speaker 1: in that field. And I didn't really need its help

103
00:05:34,964 --> 00:05:36,524
Speaker 1: in the sense that I know the field pretty well,

104
00:05:36,524 --> 00:05:38,364
Speaker 1: but like, I was just curious to see what it

105
00:05:38,404 --> 00:05:40,724
Speaker 1: would come up with. And it gave me some good stuff,

106
00:05:40,764 --> 00:05:43,604
Speaker 1: but it also gave me some stuff that just simply

107
00:05:43,644 --> 00:05:46,684
Speaker 1: does not exist. And I know, you know, as you say,

108
00:05:46,804 --> 00:05:48,764
Speaker 1: and as a lot of people say, like you need

109
00:05:48,764 --> 00:05:51,404
Speaker 1: to know what to ask it but the reason I

110
00:05:51,444 --> 00:05:54,244
Speaker 1: tested this specifically was because one of their big claims

111
00:05:54,404 --> 00:05:58,124
Speaker 1: was no more hallucination basically, which is not true.

112
00:05:58,404 --> 00:06:01,284
Speaker 2: Yeah, I've used it for you know, I almost always

113
00:06:01,284 --> 00:06:04,404
Speaker 2: put the articles I'm writing for the newsletter through a

114
00:06:04,444 --> 00:06:07,004
Speaker 2: copy of it in fact check, in the GPT models

115
00:06:07,044 --> 00:06:08,924
Speaker 2: or cloud. Sometimes I think this is one of the

116
00:06:08,964 --> 00:06:13,524
Speaker 2: most useful things that AIS are good with. It took

117
00:06:13,604 --> 00:06:15,804
Speaker 2: seven and a half minutes for what, from Nate is

118
00:06:15,804 --> 00:06:18,684
Speaker 2: a relatively short article seventy birds or something, Right, I

119
00:06:18,724 --> 00:06:21,004
Speaker 2: am using, like the thinking version, right.

120
00:06:21,124 --> 00:06:23,364
Speaker 1: Did you tell it to use the thinking version or

121
00:06:23,364 --> 00:06:23,764
Speaker 1: did it?

122
00:06:24,364 --> 00:06:27,684
Speaker 2: I think I was on thinking by default, right, And

123
00:06:28,044 --> 00:06:30,444
Speaker 2: usually I say have a high threshold, and I didn't

124
00:06:30,444 --> 00:06:32,884
Speaker 2: say that. But it's really nitpicky. It's like, you know,

125
00:06:32,924 --> 00:06:36,604
Speaker 2: I made some line about like Elon Musk is like

126
00:06:36,604 --> 00:06:40,684
Speaker 2: tweeting out like anime smut was a term I used, right,

127
00:06:40,844 --> 00:06:43,084
Speaker 2: I already tone that down from porn. And it's like

128
00:06:43,564 --> 00:06:47,644
Speaker 2: you should be more careful, you should say NSFW images

129
00:06:48,564 --> 00:06:51,564
Speaker 2: smut implies an editorial stance, you know. So like it's

130
00:06:51,604 --> 00:06:53,524
Speaker 2: just like it was kind of nitty as a poker term, right,

131
00:06:53,564 --> 00:06:58,564
Speaker 2: and like I I had it actually a friend sent

132
00:06:58,644 --> 00:07:00,484
Speaker 2: me a poker hand because i'd we wrote before or

133
00:07:00,484 --> 00:07:02,964
Speaker 2: I wrote before about how Chick's bad at poker, and

134
00:07:03,004 --> 00:07:04,724
Speaker 2: it played the first hand really well, and then I'm like, okay,

135
00:07:04,724 --> 00:07:06,404
Speaker 2: simulate more hands, and then it was still kind of

136
00:07:06,404 --> 00:07:10,004
Speaker 2: not not great. Maybe a little better, right, Yeah, No,

137
00:07:10,124 --> 00:07:13,084
Speaker 2: it's a little weird because I don't look to me,

138
00:07:13,324 --> 00:07:15,684
Speaker 2: you kind of had four point five and you had

139
00:07:15,724 --> 00:07:18,964
Speaker 2: like oh three and oh one. Right, Like I noticed

140
00:07:19,804 --> 00:07:21,804
Speaker 2: in the spring, I thought some of the reasoning quote

141
00:07:21,884 --> 00:07:23,964
Speaker 2: unquote models or the type of stuff I'm doing, which

142
00:07:24,004 --> 00:07:26,724
Speaker 2: is not using GPT as a chat bot, right, it's

143
00:07:26,724 --> 00:07:28,524
Speaker 2: like a workaid, you know. I thought there was an

144
00:07:28,524 --> 00:07:30,404
Speaker 2: improvement then, and I kind of feel like there's a

145
00:07:30,444 --> 00:07:34,724
Speaker 2: bit less this time. You know. Also, the open air

146
00:07:34,804 --> 00:07:36,604
Speaker 2: models or some of them are kind of slow, right,

147
00:07:36,644 --> 00:07:40,244
Speaker 2: I've found sometimes like Clode or Gemini or even groc

148
00:07:40,324 --> 00:07:42,684
Speaker 2: will like spit out things faster, and so you.

149
00:07:42,564 --> 00:07:46,044
Speaker 1: Know, and sometimes that matters, by the way, like sometimes

150
00:07:46,564 --> 00:07:48,924
Speaker 1: you wanted to take time, but sometimes like you're like, okay,

151
00:07:48,924 --> 00:07:52,004
Speaker 1: come on, like let's let's get let's get moving, and

152
00:07:52,244 --> 00:07:53,884
Speaker 1: uh yeah that if.

153
00:07:53,804 --> 00:07:55,484
Speaker 2: You can get me a simple you know, and often

154
00:07:55,524 --> 00:07:56,804
Speaker 2: like in a fact check, I mean today I had

155
00:07:56,804 --> 00:07:58,564
Speaker 2: to go to the pricking Barnes and Noble, right, and

156
00:07:58,604 --> 00:08:02,004
Speaker 2: so like the time pressure is often relevant, especially, you know,

157
00:08:02,044 --> 00:08:04,324
Speaker 2: and oftentimes I also use the models a lot for

158
00:08:04,444 --> 00:08:08,724
Speaker 2: like little programming task right, Like I forget the programming

159
00:08:08,764 --> 00:08:10,724
Speaker 2: using language US is called STATA. Some people would say

160
00:08:10,724 --> 00:08:11,884
Speaker 2: it's kind of a fake, but you know it's a

161
00:08:12,364 --> 00:08:15,524
Speaker 2: it's a real language, right, and I'm like, I forget

162
00:08:15,524 --> 00:08:18,204
Speaker 2: how I do this instat or full disclosure, even Excel.

163
00:08:18,204 --> 00:08:20,964
Speaker 2: I'll be like, oh guy, what's the complicated formula for this? Right?

164
00:08:21,004 --> 00:08:25,404
Speaker 2: And like it seems to me to be seventy to

165
00:08:25,444 --> 00:08:28,284
Speaker 2: eighty percent reliable the AI models in general, for like

166
00:08:29,004 --> 00:08:30,924
Speaker 2: you know, where I want a snippet of code that's like,

167
00:08:31,004 --> 00:08:34,724
Speaker 2: you know, three to ten lines long, right, you can

168
00:08:34,804 --> 00:08:36,204
Speaker 2: kind of vibe code, right, You're just like I want

169
00:08:36,244 --> 00:08:38,764
Speaker 2: to do this right, and it's like, you know, it's

170
00:08:38,804 --> 00:08:40,364
Speaker 2: pretty smart about it. I mean, every now and then

171
00:08:40,364 --> 00:08:42,324
Speaker 2: they won't work. And you know, I always tell people

172
00:08:42,364 --> 00:08:45,484
Speaker 2: like these things screw up when they're trying to chain

173
00:08:45,564 --> 00:08:48,244
Speaker 2: together different steps and they don't really quite I mean

174
00:08:48,244 --> 00:08:50,004
Speaker 2: they're trying to train themselves, right, but they don't quite

175
00:08:50,044 --> 00:08:52,924
Speaker 2: know how to stop right, and so it's like, okay,

176
00:08:52,964 --> 00:08:56,124
Speaker 2: when I build a model, we're going to NFL model.

177
00:08:57,764 --> 00:09:00,644
Speaker 2: You know, I stop at every point and ask, okay,

178
00:09:00,684 --> 00:09:03,884
Speaker 2: does this output make sense? Do you form a bunch

179
00:09:03,884 --> 00:09:07,804
Speaker 2: of complicated discoperations sort from the bottom to the top. Right,

180
00:09:07,964 --> 00:09:10,244
Speaker 2: you know, hopefully Tom brady ease list is one of

181
00:09:10,284 --> 00:09:13,764
Speaker 2: the best quarterbacks, right and Ryan Leaf is one of

182
00:09:13,804 --> 00:09:16,124
Speaker 2: the worst. But like, yeah, I'm keeping myself in the loop.

183
00:09:16,124 --> 00:09:17,284
Speaker 2: And for that kind of thing, it's like, okay, I'll

184
00:09:17,284 --> 00:09:19,164
Speaker 2: just say fifteen minutes trying to figure how to program

185
00:09:19,204 --> 00:09:22,124
Speaker 2: this or you can debug, right, You're like, why isn't

186
00:09:22,124 --> 00:09:24,684
Speaker 2: this working? As somebody's clawed and it was like, because

187
00:09:24,724 --> 00:09:26,964
Speaker 2: you made a typo. You misspell the word nate. That's

188
00:09:26,964 --> 00:09:29,164
Speaker 2: why you've been tearing your hair out for forty minutes.

189
00:09:29,244 --> 00:09:33,244
Speaker 2: Is like you misspell the word yeah. So yeah, no, look,

190
00:09:33,644 --> 00:09:37,244
Speaker 2: I think it's kind of halfway a branding it did

191
00:09:37,324 --> 00:09:39,644
Speaker 2: it did your size or some people seem to vouch

192
00:09:39,644 --> 00:09:40,844
Speaker 2: for this a lot. I don't know.

193
00:09:41,164 --> 00:09:43,964
Speaker 1: Well, you know, obviously, like I don't code right, and

194
00:09:44,044 --> 00:09:46,044
Speaker 1: people have said that it seems to be a lot

195
00:09:46,044 --> 00:09:49,444
Speaker 1: better at coding certain things, which great, you know, if

196
00:09:49,444 --> 00:09:51,964
Speaker 1: it is great, but when you're what you're saying actually

197
00:09:52,004 --> 00:09:55,164
Speaker 1: like seventy to eighty percent to someone like me, that

198
00:09:55,244 --> 00:09:57,804
Speaker 1: makes me actually much less likely to use it because

199
00:09:57,964 --> 00:10:00,204
Speaker 1: I'm not you. I don't have that background, so I

200
00:10:00,284 --> 00:10:03,804
Speaker 1: can't do I can't always do like a check to

201
00:10:03,844 --> 00:10:07,284
Speaker 1: figure out, you know, does this make sense? Right, And

202
00:10:07,364 --> 00:10:09,364
Speaker 1: in the sense that I don't have that technical base,

203
00:10:09,644 --> 00:10:12,444
Speaker 1: I can't review it in any real way, and so

204
00:10:12,604 --> 00:10:15,124
Speaker 1: I need it to be accurate, right, because I don't

205
00:10:15,124 --> 00:10:18,484
Speaker 1: trust myself to spot any potentially would.

206
00:10:18,324 --> 00:10:21,404
Speaker 2: Say the old way is to look at a manual

207
00:10:21,564 --> 00:10:26,084
Speaker 2: or like stock overflow or whatever, and there you're starting

208
00:10:26,084 --> 00:10:28,164
Speaker 2: through a lot of crap too. Right, it might not

209
00:10:28,204 --> 00:10:30,924
Speaker 2: be pertinent to your particular KSE, or it might be

210
00:10:30,964 --> 00:10:33,764
Speaker 2: an old version of software or instead of there are

211
00:10:33,764 --> 00:10:37,004
Speaker 2: lots of little fussy language things with local variables and

212
00:10:37,044 --> 00:10:38,844
Speaker 2: scalers and one of those things mean and what the

213
00:10:38,844 --> 00:10:41,364
Speaker 2: different rules are and stuff like that. Right, it slightly

214
00:10:41,404 --> 00:10:45,124
Speaker 2: fussy language, and it's good for a handle that kind

215
00:10:45,124 --> 00:10:48,804
Speaker 2: of thing. And again, to me, I'm working in ways

216
00:10:48,804 --> 00:10:51,564
Speaker 2: where it's fairly failure proof. Right, You're doing one thing,

217
00:10:52,404 --> 00:10:54,124
Speaker 2: you have an expectation for what that will do to

218
00:10:54,124 --> 00:10:56,484
Speaker 2: transform the data set, right, and if it doesn't happen,

219
00:10:56,524 --> 00:10:59,084
Speaker 2: then it won't work anyway. Right, But like the notion

220
00:10:59,164 --> 00:11:01,124
Speaker 2: of like I'm just gonna sit back here and trust

221
00:11:01,124 --> 00:11:02,644
Speaker 2: it to do all these things, I mean, I think

222
00:11:02,644 --> 00:11:07,564
Speaker 2: it's probably, you know, to code an entire NFL model,

223
00:11:07,644 --> 00:11:11,124
Speaker 2: which involves a lot of original research collection and involves

224
00:11:11,164 --> 00:11:14,484
Speaker 2: a lot of like knowledge about the sport, knowledge how

225
00:11:14,524 --> 00:11:16,484
Speaker 2: to build models right, a lot of trial and error,

226
00:11:16,564 --> 00:11:19,484
Speaker 2: Like you know, I don't think the AIS are particularly

227
00:11:19,564 --> 00:11:20,804
Speaker 2: close to doing that kind of work.

228
00:11:21,644 --> 00:11:25,404
Speaker 1: So I think that you just made a really important point,

229
00:11:25,404 --> 00:11:29,044
Speaker 1: which is to just not expect the world from it

230
00:11:29,564 --> 00:11:32,684
Speaker 1: and to like to know what it can and can't do, which,

231
00:11:32,724 --> 00:11:35,204
Speaker 1: by the way, already takes a certain user intelligence and

232
00:11:35,324 --> 00:11:38,564
Speaker 1: like knowledge to say, Okay, you know what, I don't

233
00:11:38,564 --> 00:11:40,284
Speaker 1: trust it to do this, but I do trust it

234
00:11:40,324 --> 00:11:42,524
Speaker 1: to do that. So there is you know, even though

235
00:11:43,044 --> 00:11:46,244
Speaker 1: GPT five was kind of hype, does you know you

236
00:11:46,284 --> 00:11:48,924
Speaker 1: don't have to think anymore. You still actually kind of

237
00:11:49,004 --> 00:11:51,484
Speaker 1: do right in order to get the outputs that you want,

238
00:11:51,884 --> 00:11:54,004
Speaker 1: and to realize, Okay, I can trust this task, but

239
00:11:54,044 --> 00:11:56,364
Speaker 1: not this task. I can I can get it to

240
00:11:56,364 --> 00:11:58,204
Speaker 1: do this, but not that and I think that that,

241
00:11:58,484 --> 00:12:01,644
Speaker 1: you know, that human in the loop is still very

242
00:12:01,684 --> 00:12:04,084
Speaker 1: much a thing and still very much needs to be

243
00:12:04,124 --> 00:12:04,444
Speaker 1: a thing.

244
00:12:04,884 --> 00:12:06,444
Speaker 2: Yeah. There, I just kind of keep like a little

245
00:12:06,524 --> 00:12:11,524
Speaker 2: running mental tracker of like here my expectations for AI models,

246
00:12:11,684 --> 00:12:14,124
Speaker 2: LLM SLARGE language models, and are they exceeding those or

247
00:12:14,364 --> 00:12:17,484
Speaker 2: were falling short? Right? I mean? And then do weird thing,

248
00:12:17,524 --> 00:12:19,764
Speaker 2: you know, I like I had a situation where, like

249
00:12:20,524 --> 00:12:24,084
Speaker 2: I had a bunch of latitudes and longitudes of NFL

250
00:12:24,124 --> 00:12:26,484
Speaker 2: stadiums that we'd code up quickly, right, and we're like,

251
00:12:26,524 --> 00:12:28,284
Speaker 2: I want to reverse look these up and tell me

252
00:12:28,324 --> 00:12:30,244
Speaker 2: what city they're near, right as a double check, and

253
00:12:30,284 --> 00:12:33,684
Speaker 2: like put Atlanta as like Chattanooga and just a little

254
00:12:33,684 --> 00:12:36,084
Speaker 2: it's you know, and so like for things like that,

255
00:12:36,404 --> 00:12:39,444
Speaker 2: because for data, I want all my data to be perfect, right,

256
00:12:39,484 --> 00:12:41,924
Speaker 2: I don't want to misattribute one city that doesn't really

257
00:12:41,924 --> 00:12:44,524
Speaker 2: matter if they have the you know, Falcons playing in Chattanooga,

258
00:12:44,564 --> 00:12:46,644
Speaker 2: it doesn't matter that much, right, and like and so

259
00:12:46,684 --> 00:12:51,524
Speaker 2: for that kind of thing, you know, I I still

260
00:12:51,564 --> 00:12:54,644
Speaker 2: would rather have my, yeah, my research assistant do it

261
00:12:54,724 --> 00:12:57,844
Speaker 2: or me do it myself. Right, things that require like

262
00:12:57,844 --> 00:12:59,324
Speaker 2: a lot of person But you know, it's very frustating,

263
00:12:59,404 --> 00:13:01,484
Speaker 2: use them enough where I have like particular rules where

264
00:13:01,484 --> 00:13:03,484
Speaker 2: I think they're likely to be helpful and not and

265
00:13:03,524 --> 00:13:06,364
Speaker 2: how how safely can you fail and things like that,

266
00:13:06,404 --> 00:13:09,324
Speaker 2: But like, yeah, I mean my general view is that

267
00:13:10,124 --> 00:13:12,124
Speaker 2: getting kind of savant like and that they're not very

268
00:13:12,164 --> 00:13:15,444
Speaker 2: bright about some things and they're freaking genius is about

269
00:13:15,484 --> 00:13:20,684
Speaker 2: others as opposed to this notion of like general intelligence,

270
00:13:20,684 --> 00:13:22,844
Speaker 2: where I mean, but you know, look, even the things

271
00:13:22,844 --> 00:13:25,324
Speaker 2: it's worse at, it's like as good as like a

272
00:13:26,444 --> 00:13:28,084
Speaker 2: you know, high school sophomore or so, you know, I mean,

273
00:13:28,084 --> 00:13:29,844
Speaker 2: it's not terrible and there aren't too many things where

274
00:13:29,844 --> 00:13:32,684
Speaker 2: it's terrible, right, But like, yeah.

275
00:13:32,204 --> 00:13:35,204
Speaker 1: Yeah, well, I think I think it really really depends

276
00:13:35,244 --> 00:13:36,044
Speaker 1: on what you're asking it.

277
00:13:36,324 --> 00:13:36,484
Speaker 2: You know.

278
00:13:36,564 --> 00:13:39,444
Speaker 1: One of the main things that I've read about people

279
00:13:39,604 --> 00:13:43,324
Speaker 1: kind of responding to which highlights an issue that you know,

280
00:13:43,364 --> 00:13:45,084
Speaker 1: Sam All was like, Oh, we didn't realize how big

281
00:13:45,084 --> 00:13:48,044
Speaker 1: of an issue this was, which I think is very

282
00:13:48,084 --> 00:13:50,524
Speaker 1: interesting because people have been trying to say it's an issue.

283
00:13:50,724 --> 00:13:53,924
Speaker 1: Is the change in voice, right, the fact that past

284
00:13:53,964 --> 00:13:57,084
Speaker 1: models were very psychophantic. You and I were talking before

285
00:13:57,084 --> 00:13:58,564
Speaker 1: taping today and I was like, we should really be

286
00:13:58,604 --> 00:14:02,524
Speaker 1: pronouncing the psychophantic because there's been some there's been some

287
00:14:02,604 --> 00:14:07,644
Speaker 1: real psycho behavior here, and when they introduced GPT five,

288
00:14:08,724 --> 00:14:13,604
Speaker 1: the default tone of voice was very different. Right, They

289
00:14:13,724 --> 00:14:16,284
Speaker 1: did try to address this, and then they got within

290
00:14:16,724 --> 00:14:18,764
Speaker 1: not even twenty it didn't even take twenty four hours.

291
00:14:18,804 --> 00:14:22,524
Speaker 1: Just like immediately they got all of this pushback with

292
00:14:22,564 --> 00:14:25,084
Speaker 1: people saying, no, you know, I've lost my boyfriend, I've

293
00:14:25,084 --> 00:14:28,084
Speaker 1: lost my best friend, I've lost the person who told

294
00:14:28,124 --> 00:14:31,724
Speaker 1: me I was a genius. And it makes you realize

295
00:14:32,044 --> 00:14:34,724
Speaker 1: how many people were using this really not for what

296
00:14:34,764 --> 00:14:38,004
Speaker 1: it's intended, and something that can be incredibly bad for

297
00:14:38,164 --> 00:14:42,644
Speaker 1: a lot of things, right, mental health, just social connections,

298
00:14:42,684 --> 00:14:47,084
Speaker 1: all of these things that you know people people were like, whoa, whoa, whoa?

299
00:14:47,164 --> 00:14:52,724
Speaker 1: What happened to my significant other? And they brought it

300
00:14:52,804 --> 00:14:56,444
Speaker 1: back right, so now you can actually select that voice again.

301
00:14:56,724 --> 00:14:58,844
Speaker 1: We have the default voice, but we also have you know,

302
00:14:58,924 --> 00:15:03,364
Speaker 1: the listener of voice, and there are a few other voices.

303
00:15:03,564 --> 00:15:06,524
Speaker 1: None of their descriptions actually map onto what that actually is.

304
00:15:06,844 --> 00:15:09,484
Speaker 1: I was reading there was like a cynic and I

305
00:15:09,484 --> 00:15:11,884
Speaker 1: don't even remember what they said the cynic voice was,

306
00:15:12,084 --> 00:15:14,004
Speaker 1: but I was like, that's not what a cynic is

307
00:15:14,324 --> 00:15:17,084
Speaker 1: like it was, it was they're really they're really weird.

308
00:15:17,244 --> 00:15:21,284
Speaker 1: They're really weird descriptors. But basically you can get you

309
00:15:21,284 --> 00:15:24,764
Speaker 1: can get GPT to interact with you in different voices,

310
00:15:25,084 --> 00:15:30,004
Speaker 1: and I you know, my reaction to that is like

311
00:15:30,604 --> 00:15:33,804
Speaker 1: you shouldn't always give the people what they want in

312
00:15:33,844 --> 00:15:36,484
Speaker 1: a lot of ways, like this was bad and like

313
00:15:36,564 --> 00:15:39,444
Speaker 1: you've fixed it, like don't go don't go unfixing it,

314
00:15:39,484 --> 00:15:42,084
Speaker 1: because even though you fixed it, they didn't fully fix it, right,

315
00:15:42,124 --> 00:15:44,844
Speaker 1: they just made it less overt, which you know, subtle

316
00:15:44,924 --> 00:15:48,484
Speaker 1: psychopancy can also be bad. But they've tried, they at

317
00:15:48,564 --> 00:15:51,644
Speaker 1: least initially tried, and now they've really gone back on

318
00:15:51,724 --> 00:15:55,124
Speaker 1: that immediately caving to pressure. And if you always give

319
00:15:55,164 --> 00:15:57,524
Speaker 1: the public what it wants, like we've talked about p

320
00:15:57,684 --> 00:15:59,684
Speaker 1: doom and like a lot of people.

321
00:15:59,404 --> 00:16:02,084
Speaker 2: Want things you really should not be getting these, it's

322
00:16:02,084 --> 00:16:04,964
Speaker 2: pretty hard kind of in an equilibrium, like not to

323
00:16:05,004 --> 00:16:09,364
Speaker 2: optimize for what drives engagement in the show. I mean,

324
00:16:09,524 --> 00:16:11,084
Speaker 2: you know, on the one, thecept something that they don't

325
00:16:11,084 --> 00:16:13,804
Speaker 2: need revenue like right away, but like, yeah, no, I

326
00:16:14,004 --> 00:16:17,524
Speaker 2: think sometimes these models like you give them an inch

327
00:16:17,564 --> 00:16:19,604
Speaker 2: and they take a mile with it, right, like if

328
00:16:19,644 --> 00:16:22,004
Speaker 2: you look at what happened with ROC when it had

329
00:16:22,084 --> 00:16:28,164
Speaker 2: its moment, the prompts that system prompts that elon or

330
00:16:28,324 --> 00:16:34,564
Speaker 2: ate an unnamed engineer XAI was using. We're like not

331
00:16:34,724 --> 00:16:38,404
Speaker 2: that wild, right, but like if you go down a

332
00:16:38,444 --> 00:16:41,004
Speaker 2: rabbit hole, we keep kind of getting like reinforcement feedback.

333
00:16:41,044 --> 00:16:42,724
Speaker 2: This is good, this is good. And you know, you

334
00:16:42,764 --> 00:16:45,804
Speaker 2: know open aies models used to have this thing where

335
00:16:46,364 --> 00:16:48,204
Speaker 2: did you like answer A or answer beast. They're now

336
00:16:48,204 --> 00:16:52,564
Speaker 2: outsourcing it to some of their users, right, and like yeah,

337
00:16:52,604 --> 00:16:57,884
Speaker 2: I mean, look, you know what are you kind of

338
00:16:58,524 --> 00:17:02,764
Speaker 2: optimizing for objectively? Right? It's kind of easier when you

339
00:17:02,804 --> 00:17:04,804
Speaker 2: are trying to train it on like a math problem

340
00:17:05,164 --> 00:17:10,204
Speaker 2: where there's an objectively correct answer, right. You know, the

341
00:17:10,324 --> 00:17:11,844
Speaker 2: NFL model turning to build the end of the day,

342
00:17:12,124 --> 00:17:14,404
Speaker 2: how accurate is it to predict NFL games? Right? That's

343
00:17:14,444 --> 00:17:15,324
Speaker 2: that's the bottom line.

344
00:17:15,764 --> 00:17:17,484
Speaker 1: You actually have a metric, right.

345
00:17:17,684 --> 00:17:20,604
Speaker 2: And for you know, for feedback or goodness of an

346
00:17:20,604 --> 00:17:24,524
Speaker 2: answer that's more subjective, then it's a lot trickier, right.

347
00:17:25,804 --> 00:17:27,444
Speaker 2: You know, I think we've seen with some of the

348
00:17:27,884 --> 00:17:31,404
Speaker 2: you know, some of the reason that like Grock hasn't

349
00:17:31,404 --> 00:17:34,644
Speaker 2: had as much reinforcement learning training right or or or

350
00:17:35,004 --> 00:17:37,844
Speaker 2: you see that right? You know, the works are pretty rough.

351
00:17:37,924 --> 00:17:41,124
Speaker 2: I'm mixing metaphors here and yeah, I mean there it's

352
00:17:41,124 --> 00:17:44,244
Speaker 2: a weird technology, and I think people understand more about

353
00:17:44,244 --> 00:17:46,244
Speaker 2: how these models work than before. And like, by the way,

354
00:17:46,324 --> 00:17:49,764
Speaker 2: like one reason to be optimistic unless you're a doomer.

355
00:17:49,804 --> 00:17:52,244
Speaker 2: I guess it's just like the amount of like human

356
00:17:52,364 --> 00:17:57,164
Speaker 2: another capital being poured into AI research, right is like,

357
00:17:57,884 --> 00:18:00,124
Speaker 2: you know quite something, right, It wouldn't surprise me at

358
00:18:00,164 --> 00:18:01,404
Speaker 2: these companies start saying, oh, we have a lot of

359
00:18:01,404 --> 00:18:04,964
Speaker 2: smart people's to kind of spin off technologies and energy

360
00:18:04,764 --> 00:18:07,884
Speaker 2: or quantum computing or whatever else.

361
00:18:08,004 --> 00:18:11,364
Speaker 1: Right, yet, know, I think that there's so much promise,

362
00:18:11,484 --> 00:18:14,044
Speaker 1: but I think that with this particular thing, the incentives

363
00:18:14,084 --> 00:18:16,484
Speaker 1: are misaligned at least for now, which you were kind

364
00:18:16,524 --> 00:18:19,564
Speaker 1: of hinting at, right in the sense that sure, like

365
00:18:19,644 --> 00:18:22,524
Speaker 1: they don't necessarily need it. But if people if this

366
00:18:22,644 --> 00:18:25,444
Speaker 1: is one of the things that fuels revenue growth, right,

367
00:18:25,724 --> 00:18:28,604
Speaker 1: that people want to feel not lonely. They want you know,

368
00:18:28,684 --> 00:18:32,924
Speaker 1: someone who's kind of reinforcing their ideas that they interact more,

369
00:18:33,284 --> 00:18:36,524
Speaker 1: which is good, right if you're actually paying more in

370
00:18:36,644 --> 00:18:38,964
Speaker 1: order to be able to kind of spend more time

371
00:18:39,004 --> 00:18:42,124
Speaker 1: with the model, and they're interacting more when they feel like,

372
00:18:42,164 --> 00:18:44,084
Speaker 1: this is my girlfriend, this is my boyfriend, this is

373
00:18:44,404 --> 00:18:48,124
Speaker 1: you know, my best friend, This is my counselor my psychiatrist,

374
00:18:48,164 --> 00:18:51,124
Speaker 1: you know whatever, it is my teacher. There was a

375
00:18:51,164 --> 00:18:54,604
Speaker 1: guy who was who Cash Hill just wrote about in

376
00:18:54,644 --> 00:18:55,884
Speaker 1: the New York Times.

377
00:18:56,004 --> 00:18:56,484
Speaker 2: Who.

378
00:18:58,004 --> 00:19:01,604
Speaker 1: Believed that he'd created a new mathematical theory right that

379
00:19:02,524 --> 00:19:06,284
Speaker 1: solved everything, and like and tried to actually he tried

380
00:19:06,324 --> 00:19:08,644
Speaker 1: to fact check the delusion, which was crazy. He's like,

381
00:19:08,964 --> 00:19:11,524
Speaker 1: I feel like I'm sounding crazy and the a I

382
00:19:11,604 --> 00:19:14,284
Speaker 1: was like, no, You're absolutely not crazy. You're the most

383
00:19:14,444 --> 00:19:17,804
Speaker 1: everyone else is crazy, right like you're saying so. So

384
00:19:18,044 --> 00:19:21,284
Speaker 1: was one of these very strange kind of things where

385
00:19:21,284 --> 00:19:23,284
Speaker 1: he went in. By the way, his first question was

386
00:19:23,524 --> 00:19:26,884
Speaker 1: he wanted it to explain how you got the value

387
00:19:26,884 --> 00:19:29,284
Speaker 1: of pie, because this was someone who never finished high

388
00:19:29,284 --> 00:19:32,364
Speaker 1: school and his son needed to help with homework, and

389
00:19:32,404 --> 00:19:35,244
Speaker 1: so he just asked chat japt about pie. That was

390
00:19:35,324 --> 00:19:38,764
Speaker 1: the start of this insane rabbit hole. And as and

391
00:19:38,804 --> 00:19:41,044
Speaker 1: I'm not I mean, we're talking about chat JEPT because

392
00:19:41,084 --> 00:19:45,124
Speaker 1: of GPT five release. But this doesn't just apply here.

393
00:19:45,164 --> 00:19:47,844
Speaker 1: We're just talking about. You know, that's Lum's in general

394
00:19:47,924 --> 00:19:51,204
Speaker 1: probably will be susceptible to this. I'm not sure, right

395
00:19:51,404 --> 00:19:54,684
Speaker 1: because all of these examples are from CHATJAPT But you know,

396
00:19:54,764 --> 00:19:58,924
Speaker 1: you have this innocuous question that then leads someone to

397
00:19:59,044 --> 00:20:02,964
Speaker 1: a very detrimental spiral. And it's not a standalone case, right,

398
00:20:03,284 --> 00:20:05,964
Speaker 1: And now we know how many cases that were below

399
00:20:06,084 --> 00:20:09,724
Speaker 1: the radar because nothing bad, nothing quote unquote bad happened,

400
00:20:09,764 --> 00:20:14,004
Speaker 1: yet there were given the outcry when the model shifted.

401
00:20:14,364 --> 00:20:18,524
Speaker 1: And I don't think honestly, like, do we really trust

402
00:20:18,844 --> 00:20:21,804
Speaker 1: a company that's clearly profit driven? We know, I mean

403
00:20:22,244 --> 00:20:26,004
Speaker 1: Sam Milton's companies profit driven. I know, I know it's

404
00:20:26,124 --> 00:20:27,564
Speaker 1: it's crazy.

405
00:20:26,964 --> 00:20:30,084
Speaker 2: Gambling a casino, But do we.

406
00:20:30,084 --> 00:20:33,164
Speaker 1: Really trust a company that's profit driven, right, that's bottom

407
00:20:33,204 --> 00:20:37,324
Speaker 1: line driven to sure, they will fix all these other

408
00:20:37,364 --> 00:20:39,604
Speaker 1: problems if it's good for them, But do we trust

409
00:20:39,604 --> 00:20:43,804
Speaker 1: them to fix these things that are really undermining mental health?

410
00:20:44,044 --> 00:20:46,284
Speaker 1: We've seen that meta, you know, Facebook, all these none

411
00:20:46,324 --> 00:20:49,364
Speaker 1: of them have right for years, they've known that there

412
00:20:49,404 --> 00:20:51,644
Speaker 1: are issues that have existed and they haven't addressed that.

413
00:20:51,764 --> 00:20:55,764
Speaker 2: I'm not as convinced about this particular problem. I mean

414
00:20:55,764 --> 00:20:57,444
Speaker 2: because first of all, what it's substituting for. Is it

415
00:20:57,444 --> 00:21:01,444
Speaker 2: subsitting for Twitter or Reddit forums or some other dark

416
00:21:01,484 --> 00:21:02,324
Speaker 2: corner of the Internet.

417
00:21:02,364 --> 00:21:06,964
Speaker 1: Potentially No, because from a psychological standpoint, there's immediate reinforcement

418
00:21:07,004 --> 00:21:09,084
Speaker 1: and a conversation that goes back and forth, which is

419
00:21:09,204 --> 00:21:12,124
Speaker 1: very very different for the brain than like saying something

420
00:21:12,164 --> 00:21:12,604
Speaker 1: and then.

421
00:21:12,844 --> 00:21:14,324
Speaker 2: On Twitter your media feva.

422
00:21:15,244 --> 00:21:17,684
Speaker 1: But it's not quite the same thing. Just from a

423
00:21:17,684 --> 00:21:20,964
Speaker 1: psychological standpoint, it can be much more pernicious when you

424
00:21:21,004 --> 00:21:24,364
Speaker 1: feel like you're talking to an actual person and personalities

425
00:21:24,404 --> 00:21:29,484
Speaker 1: do start developing. I mean, what do you like? Obviously

426
00:21:29,524 --> 00:21:31,244
Speaker 1: this is this is not ideal. What do you think

427
00:21:31,324 --> 00:21:31,844
Speaker 1: is ideal?

428
00:21:31,964 --> 00:21:32,124
Speaker 2: Right?

429
00:21:32,164 --> 00:21:35,684
Speaker 1: If you're interacting with chat GPT? What personality do you

430
00:21:35,684 --> 00:21:38,844
Speaker 1: want me personally? I want no personality whatsoever. I just

431
00:21:38,884 --> 00:21:40,364
Speaker 1: wanted to give me the damn facts.

432
00:21:40,884 --> 00:21:43,404
Speaker 2: You can give it custom instructions, right, which is like appendittus.

433
00:21:43,404 --> 00:21:48,644
Speaker 2: So I tell it like be honest and straightforward site sources.

434
00:21:48,764 --> 00:21:51,084
Speaker 2: It's fine to speculate, but if you are speculating, label

435
00:21:51,124 --> 00:21:53,724
Speaker 2: it speculation that kind of thing, right, And.

436
00:21:53,684 --> 00:21:56,044
Speaker 1: It sometimes lies about that too, Like when I say

437
00:21:56,044 --> 00:21:59,044
Speaker 1: give me sources and it lies about the sources and

438
00:21:59,164 --> 00:22:01,284
Speaker 1: you say hey, the source doesn't exist, and it says

439
00:22:01,284 --> 00:22:04,644
Speaker 1: something like you caught me. And I'm like, okay, well

440
00:22:05,564 --> 00:22:09,044
Speaker 1: you're you're ostensibly following instructions, but you're not really And

441
00:22:09,044 --> 00:22:11,404
Speaker 1: I actually, I don't know, I didn't try that you

442
00:22:11,484 --> 00:22:14,164
Speaker 1: caught me thing with GPT five. I don't know if

443
00:22:14,164 --> 00:22:17,684
Speaker 1: it's still going to do kind of a version of that.

444
00:22:17,844 --> 00:22:20,484
Speaker 1: But Nate, I mean, if we're looking at the reliability

445
00:22:20,484 --> 00:22:24,644
Speaker 1: of these models, right, you had mentioned the Chattanooga example

446
00:22:24,764 --> 00:22:28,284
Speaker 1: right where it thought that Chattanooga and Atlanta were interchanging.

447
00:22:28,524 --> 00:22:31,284
Speaker 2: Basically, well, Atlanta wasn't in my gazzeteer file. I forgot

448
00:22:31,324 --> 00:22:32,724
Speaker 2: about it. Man. It ha'd a litle arch just sitting

449
00:22:32,724 --> 00:22:35,124
Speaker 2: in Georgia. But I fixed it now, right.

450
00:22:35,324 --> 00:22:37,764
Speaker 1: But if it's getting things like Chattanooga wrong, like you

451
00:22:37,804 --> 00:22:41,364
Speaker 1: start questioning how how good its output was on other things.

452
00:22:41,444 --> 00:22:43,564
Speaker 2: Yeah, and if there are you know whatever, four hundred

453
00:22:43,604 --> 00:22:46,884
Speaker 2: rows of data and it screws up five, it's a lot.

454
00:22:47,084 --> 00:22:51,044
Speaker 2: And the analogy here is something like, Okay, if you

455
00:22:51,084 --> 00:22:54,124
Speaker 2: take the subway in New York, you don't have to

456
00:22:54,124 --> 00:22:57,604
Speaker 2: like look up the timetables because it's rarely more than

457
00:22:57,684 --> 00:22:59,604
Speaker 2: like a five or seven minute wait for a train, right,

458
00:22:59,564 --> 00:23:00,004
Speaker 2: you just go on.

459
00:23:00,044 --> 00:23:02,324
Speaker 1: Station unless it's the B, in which case you'll be

460
00:23:02,364 --> 00:23:03,004
Speaker 1: waiting forever.

461
00:23:03,804 --> 00:23:06,604
Speaker 2: Sorry, B fans, I'm on the L now. It's a

462
00:23:06,644 --> 00:23:09,364
Speaker 2: real it's a real adult train. Rent there.

463
00:23:11,164 --> 00:23:13,164
Speaker 1: The B stands for bullshit.

464
00:23:14,404 --> 00:23:19,604
Speaker 2: Bullshit Okay, And that's when it's part of your continuous workflow, right,

465
00:23:19,924 --> 00:23:21,844
Speaker 2: chechup is more like you go to like the amcrak

466
00:23:21,964 --> 00:23:25,604
Speaker 2: station and you have to plan around it a little bit. Right,

467
00:23:25,644 --> 00:23:27,484
Speaker 2: It's not it's you know, it's not kind of the

468
00:23:27,524 --> 00:23:32,524
Speaker 2: bonus productivity from from kind of turning your brain off

469
00:23:32,604 --> 00:23:35,404
Speaker 2: and and just having to handle the work. You know.

470
00:23:35,964 --> 00:23:39,044
Speaker 2: With that said, you know, I've come home uh knights

471
00:23:39,084 --> 00:23:40,964
Speaker 2: when I'm tired or busier and had a couple of

472
00:23:41,004 --> 00:23:42,684
Speaker 2: glasses of wine, right, and it sure is nights then

473
00:23:42,724 --> 00:23:44,564
Speaker 2: to be like, oh, I forget the stupid fucking command

474
00:23:44,644 --> 00:23:47,084
Speaker 2: and static, Can you just tell me what it is? Right?

475
00:23:47,604 --> 00:23:49,324
Speaker 2: But I'm you know, still using it in like a

476
00:23:49,324 --> 00:23:54,004
Speaker 2: piecemeal way.

477
00:23:53,324 --> 00:23:54,764
Speaker 1: And we'll be back right after this.

478
00:24:07,044 --> 00:24:11,444
Speaker 2: All this might be kind of not bad for AI safety, right,

479
00:24:11,484 --> 00:24:12,644
Speaker 2: Like I kind of think.

480
00:24:12,484 --> 00:24:15,924
Speaker 1: Like, yeah, well I think it depends. So I don't

481
00:24:15,924 --> 00:24:19,484
Speaker 1: know the nitty gritty of what the improvements were, but

482
00:24:19,524 --> 00:24:21,444
Speaker 1: some of the some of the reviews that I read,

483
00:24:22,284 --> 00:24:26,524
Speaker 1: some have said that the lack of transparency into kind

484
00:24:26,524 --> 00:24:28,404
Speaker 1: of what's being used and how it's being used is

485
00:24:28,444 --> 00:24:31,324
Speaker 1: actually potentially not great. That before you could, you could

486
00:24:31,364 --> 00:24:34,444
Speaker 1: query much better to actually figure out, you know, what

487
00:24:34,484 --> 00:24:36,924
Speaker 1: processes were being used, what models were being used, et cetera.

488
00:24:37,044 --> 00:24:40,724
Speaker 1: Now you can't, and that if some quoting gets to

489
00:24:40,804 --> 00:24:43,084
Speaker 1: a certain point that it can do like some self

490
00:24:43,124 --> 00:24:47,204
Speaker 1: replicating things. And this is kind of what we talked

491
00:24:47,204 --> 00:24:50,684
Speaker 1: about briefly when we were talking about the AI twenty

492
00:24:50,724 --> 00:24:55,484
Speaker 1: twenty seven report, that some of those things are potentially

493
00:24:56,524 --> 00:24:58,044
Speaker 1: becoming closer to being a.

494
00:24:58,004 --> 00:25:01,044
Speaker 2: Real Yeah, I mean this is almost another show, right,

495
00:25:01,084 --> 00:25:04,244
Speaker 2: but like you know, humans currently have an important role,

496
00:25:04,524 --> 00:25:06,844
Speaker 2: a couple of important roles in the process, right, one

497
00:25:06,844 --> 00:25:09,324
Speaker 2: of which is to provide the corpus all the time

498
00:25:09,564 --> 00:25:11,684
Speaker 2: that the models are trained on, the second of which

499
00:25:11,724 --> 00:25:16,684
Speaker 2: is with reinforcement learning. And like again, math problems are

500
00:25:16,724 --> 00:25:19,844
Speaker 2: a weird exception in that you kind of you kind

501
00:25:19,844 --> 00:25:22,724
Speaker 2: of know the answers, right, there's objectively correct answer, it's

502
00:25:22,724 --> 00:25:24,884
Speaker 2: just really hard to figure out, right, and a human

503
00:25:24,924 --> 00:25:28,724
Speaker 2: can say, oh, that's correct, right with other things. It's

504
00:25:28,724 --> 00:25:32,004
Speaker 2: it's trickier if you're if you're, you know, wanting to

505
00:25:32,084 --> 00:25:35,844
Speaker 2: patent some novel protein that could be used in drug discovery, right,

506
00:25:35,924 --> 00:25:37,484
Speaker 2: then you got to test that and make sure it

507
00:25:37,524 --> 00:25:39,844
Speaker 2: actually works. Right, And like so if you don't have

508
00:25:39,844 --> 00:25:41,884
Speaker 2: a human reinforcement learning, if you if you're trying to

509
00:25:41,924 --> 00:25:44,244
Speaker 2: train corpses that are beyond them, I mean there again,

510
00:25:44,284 --> 00:25:46,884
Speaker 2: there are a case where you can extrapolate and get

511
00:25:47,524 --> 00:25:50,484
Speaker 2: fifty percent than the best human or or two hundred

512
00:25:50,484 --> 00:25:53,684
Speaker 2: percent better. Right, the notion of like an explosion of superintelligence,

513
00:25:53,724 --> 00:25:56,604
Speaker 2: I think you know, I mean, these things still can't

514
00:25:56,604 --> 00:26:00,164
Speaker 2: book good fucking delta flight to Chicago, right, I'm probably

515
00:26:00,284 --> 00:26:01,084
Speaker 2: the year they can.

516
00:26:01,244 --> 00:26:03,724
Speaker 1: But by the way, by the way that talking about

517
00:26:03,764 --> 00:26:07,204
Speaker 1: like security risks, this isn't p doom, but like personal

518
00:26:07,244 --> 00:26:10,044
Speaker 1: security risk as it becomes better at doing that, the

519
00:26:10,404 --> 00:26:13,124
Speaker 1: chance of someone hacking and like actually being able to

520
00:26:13,324 --> 00:26:16,364
Speaker 1: insert some malicious code without your knowledge so that you

521
00:26:16,484 --> 00:26:18,724
Speaker 1: end up, you know, so that they can steal credit

522
00:26:18,724 --> 00:26:21,324
Speaker 1: card information, et cetera, et cetera. Don't underestimate. I'm not

523
00:26:21,364 --> 00:26:25,964
Speaker 1: talking about you, but I think we as a society underestimating.

524
00:26:27,684 --> 00:26:30,604
Speaker 2: Maybe even more than to Google, you know what I mean.

525
00:26:30,724 --> 00:26:33,564
Speaker 1: It's crazy and that is available somewhere, right They're storing

526
00:26:33,604 --> 00:26:35,924
Speaker 1: all of that data, and so that means that it

527
00:26:35,964 --> 00:26:38,044
Speaker 1: can be hacked, and if it can be hacked, it

528
00:26:38,124 --> 00:26:39,644
Speaker 1: will be hacked at some point. I think that that's

529
00:26:39,724 --> 00:26:41,764
Speaker 1: kind of the rule of the Internet right now. I've

530
00:26:41,764 --> 00:26:43,924
Speaker 1: spent enough time with you know, con artists and bad

531
00:26:43,964 --> 00:26:46,524
Speaker 1: actors that I know that, like, there's always someone, if

532
00:26:46,564 --> 00:26:49,884
Speaker 1: there's a new technology, there's someone one step ahead figuring out, Okay,

533
00:26:49,884 --> 00:26:54,764
Speaker 1: how do we exploit this? Does it call me Maria? Now?

534
00:26:54,924 --> 00:26:55,844
Speaker 2: Okaymate?

535
00:26:55,964 --> 00:27:00,764
Speaker 1: Sometimes no, it hasn't called me Maria. But I've tried not.

536
00:27:01,124 --> 00:27:02,964
Speaker 1: I mean, obviously you have to sign into it, but

537
00:27:03,044 --> 00:27:08,924
Speaker 1: I try to uh minimize uh minimize my sharing of

538
00:27:09,004 --> 00:27:13,204
Speaker 1: any personal information. But that only takes you so far.

539
00:27:13,444 --> 00:27:15,444
Speaker 1: But yeah, I think that that's kind of a personal

540
00:27:15,444 --> 00:27:20,484
Speaker 1: security risk that people are probably underestimating. I'm guessing there's

541
00:27:20,524 --> 00:27:23,244
Speaker 1: some people who are very well aware of it, but

542
00:27:23,604 --> 00:27:27,724
Speaker 1: I would hesitate before giving it access to you know,

543
00:27:27,764 --> 00:27:33,604
Speaker 1: my travel plans, access to any itineraries, credit cards, et cetera,

544
00:27:33,604 --> 00:27:37,604
Speaker 1: et cetera, because those are things that seem like there

545
00:27:37,644 --> 00:27:41,324
Speaker 1: could be vulnerabilities and we see you know that as

546
00:27:41,364 --> 00:27:43,964
Speaker 1: you said, you know, this is a this launch has

547
00:27:44,004 --> 00:27:46,524
Speaker 1: the new car smell, so like at the beginning, like

548
00:27:46,564 --> 00:27:47,924
Speaker 1: there are going to be bugs. There are going to

549
00:27:48,004 --> 00:27:50,604
Speaker 1: be issues, and sure eventually like some of them are

550
00:27:50,604 --> 00:27:52,684
Speaker 1: going to get sorted out, but there's always going to

551
00:27:52,684 --> 00:27:56,444
Speaker 1: be like the initial data breach that prompts it to

552
00:27:56,484 --> 00:27:58,964
Speaker 1: get sorted out. And I don't want to be part

553
00:27:58,964 --> 00:28:00,364
Speaker 1: of that initial data breach.

554
00:28:00,604 --> 00:28:02,764
Speaker 2: Yeah. I don't know that these giant tech companies had

555
00:28:02,764 --> 00:28:06,124
Speaker 2: behaved in particularly trustworthy they have, right.

556
00:28:06,244 --> 00:28:08,484
Speaker 1: Right, Yeah, I don't think they have. So I don't

557
00:28:08,524 --> 00:28:10,764
Speaker 1: know how much we want to we want to give

558
00:28:10,804 --> 00:28:14,044
Speaker 1: them and what you know, Nate, there might be some

559
00:28:14,284 --> 00:28:16,284
Speaker 1: there might be someone who's like, uh oh, does Nate

560
00:28:16,644 --> 00:28:19,324
Speaker 1: upload all of the specifics of his models to these

561
00:28:19,364 --> 00:28:22,204
Speaker 1: Maybe we should hack Nate's uh chat GPTs that we

562
00:28:22,204 --> 00:28:25,084
Speaker 1: can steal his model and improve on it. I mean

563
00:28:25,204 --> 00:28:28,244
Speaker 1: that seems silly, but like it actually makes corporate espionage

564
00:28:28,284 --> 00:28:30,924
Speaker 1: those types of things much easier too. It's what I've

565
00:28:30,964 --> 00:28:33,924
Speaker 1: always said with you know, con artists, that it's become

566
00:28:34,004 --> 00:28:37,204
Speaker 1: so much easier, and the barrier of entry to conning

567
00:28:37,204 --> 00:28:39,684
Speaker 1: people has become so much lower simply because we share

568
00:28:39,724 --> 00:28:45,484
Speaker 1: so much information online unthinkingly, and so it becomes the

569
00:28:45,484 --> 00:28:47,604
Speaker 1: case where before it would take someone a lot of

570
00:28:47,884 --> 00:28:50,044
Speaker 1: kind of research to try to figure out, you know, oh,

571
00:28:50,084 --> 00:28:51,764
Speaker 1: what are the things you know? Where does Nate like

572
00:28:51,804 --> 00:28:53,484
Speaker 1: to go? What does he? You know? What are the

573
00:28:53,524 --> 00:28:56,324
Speaker 1: pressure points that I can How can I approach him?

574
00:28:56,364 --> 00:28:57,124
Speaker 1: Et cetera, et cetera.

575
00:28:57,124 --> 00:28:59,444
Speaker 2: And maybe you can trade out on poker tells, right, yeh,

576
00:28:59,564 --> 00:29:03,964
Speaker 2: watch five hours of Who's a Poker Player? But yeah,

577
00:29:04,084 --> 00:29:06,804
Speaker 2: Adam Hendrix and figure out like what are his tells? Right?

578
00:29:06,884 --> 00:29:09,724
Speaker 1: But but now I'm you know, con artists can use

579
00:29:09,724 --> 00:29:11,884
Speaker 1: all that information like very quickly because you've shared it.

580
00:29:12,084 --> 00:29:15,124
Speaker 1: And with chat GPT, people are sharing so much right

581
00:29:15,164 --> 00:29:17,204
Speaker 1: on such a personal level, not thinking that this will

582
00:29:17,244 --> 00:29:19,644
Speaker 1: become public, and I don't know why they think that

583
00:29:19,724 --> 00:29:22,444
Speaker 1: it won't. So it's a very you know, it's a

584
00:29:22,564 --> 00:29:25,804
Speaker 1: very interesting conundrum. And I think there are so many

585
00:29:25,884 --> 00:29:27,644
Speaker 1: amazing things that are kind of come out of this,

586
00:29:27,724 --> 00:29:30,044
Speaker 1: and then some very dystopian things and some things that

587
00:29:30,324 --> 00:29:32,124
Speaker 1: will potentially really hurt individual Yeah.

588
00:29:32,124 --> 00:29:34,044
Speaker 2: I mean, look, I'm not even talking about the audio

589
00:29:34,124 --> 00:29:36,684
Speaker 2: and video caare I mean, look, it remains the case

590
00:29:36,764 --> 00:29:39,964
Speaker 2: that if you beamed to twenty twenty five from twenty

591
00:29:40,164 --> 00:29:44,524
Speaker 2: twenty right, you would be amazed by what these models

592
00:29:44,684 --> 00:29:46,924
Speaker 2: can do. Absolut would be considered a freak if you

593
00:29:47,004 --> 00:29:50,404
Speaker 2: had predicted five years ago that you have this machine

594
00:29:50,444 --> 00:29:53,084
Speaker 2: that for many things can like pass the touring tests.

595
00:29:53,084 --> 00:29:55,364
Speaker 2: Some areaster don't use Turing tests, but like it's you know,

596
00:29:55,684 --> 00:29:58,364
Speaker 2: it's basically giving you plausible human level performance across a

597
00:29:58,404 --> 00:30:02,284
Speaker 2: variety of cognitive tasks, deficient in some, excellent in others. Right,

598
00:30:02,324 --> 00:30:04,484
Speaker 2: Like that still is quite amazing, And like part of

599
00:30:04,524 --> 00:30:07,524
Speaker 2: what I'm reacting to is like, you know, where is

600
00:30:07,644 --> 00:30:11,244
Speaker 2: the hype relative to the reality. And it felt like

601
00:30:11,444 --> 00:30:14,684
Speaker 2: a year ago it was like, Okay, people outside of

602
00:30:14,684 --> 00:30:17,764
Speaker 2: Filekan Valley are just not seeing at all the power

603
00:30:17,844 --> 00:30:19,564
Speaker 2: of this, and now they kind of do. And I

604
00:30:19,564 --> 00:30:22,244
Speaker 2: still think that kind of like the political types are

605
00:30:22,284 --> 00:30:26,524
Speaker 2: like significantly behind the curve and calling them chatbots or whatever.

606
00:30:26,564 --> 00:30:30,644
Speaker 2: But like also like you know, you read these really

607
00:30:30,684 --> 00:30:32,924
Speaker 2: smart researchers saying that, oh, we think there's gonna be

608
00:30:33,764 --> 00:30:36,804
Speaker 2: a singularity in two years, right, and I'm like, you know, look,

609
00:30:36,844 --> 00:30:38,764
Speaker 2: you can drive. There's a lot of trucks you can

610
00:30:38,844 --> 00:30:41,644
Speaker 2: drive in between. Oh, it's just a chatbot and Singularity

611
00:30:41,684 --> 00:30:44,084
Speaker 2: by twenty twenty seven, right, right, Yeah, it feels like

612
00:30:44,204 --> 00:30:45,684
Speaker 2: pretty safe bounds.

613
00:30:46,644 --> 00:30:48,204
Speaker 1: I totally agree with that.

614
00:30:48,644 --> 00:30:50,964
Speaker 2: You know, just to come back to the question of

615
00:30:51,004 --> 00:30:54,724
Speaker 2: your ideal AI chat bot personality, this sounds like a

616
00:30:54,804 --> 00:30:57,244
Speaker 2: weirdly like the Howard's Stern someone or something. But Maria,

617
00:30:57,284 --> 00:30:59,684
Speaker 2: what's your what's your what's floats your boat?

618
00:31:00,124 --> 00:31:04,644
Speaker 1: Well, Howard? You know, earlier I had said that I

619
00:31:04,724 --> 00:31:06,924
Speaker 1: want an AI that just gives me the facts, right, Like,

620
00:31:06,964 --> 00:31:09,564
Speaker 1: I do not want the damn thing to have a personality.

621
00:31:10,164 --> 00:31:12,964
Speaker 1: This is an AI, it's a computer. Like, this is

622
00:31:13,004 --> 00:31:17,044
Speaker 1: not my friend, and I don't want its opinions. I

623
00:31:17,124 --> 00:31:19,884
Speaker 1: just want it to kind of give me factual answers.

624
00:31:20,044 --> 00:31:22,764
Speaker 1: Now I know that that's not actually possible because, as

625
00:31:22,804 --> 00:31:26,244
Speaker 1: we've talked about many many times, like I'm probably not

626
00:31:26,364 --> 00:31:29,924
Speaker 1: asking it about math problems because I don't have any

627
00:31:30,044 --> 00:31:32,804
Speaker 1: use for that. I'm probably asking it for things that

628
00:31:33,924 --> 00:31:38,444
Speaker 1: will inevitably be opinion tinged because you know, the inputs

629
00:31:38,444 --> 00:31:41,364
Speaker 1: were made by humans. But yeah, I want it to

630
00:31:41,644 --> 00:31:44,684
Speaker 1: kind of be as as neutral as possible, And like.

631
00:31:45,204 --> 00:31:47,284
Speaker 2: Do you not get were you familiar with FIVEY Fox?

632
00:31:47,284 --> 00:31:48,244
Speaker 2: Does that mean anything to you?

633
00:31:48,284 --> 00:31:48,324
Speaker 1: No?

634
00:31:49,044 --> 00:31:51,564
Speaker 2: Five E Fox was like the mascot of five thirty

635
00:31:51,604 --> 00:31:54,764
Speaker 2: eight models, right, it was a cartoon fox.

636
00:31:54,884 --> 00:31:57,084
Speaker 1: Oh, I've seen the picture of the cartoon fox.

637
00:31:57,604 --> 00:32:00,844
Speaker 2: I'm presenting my modelty like, maybe i'd be a good personality, right, Yeah, little,

638
00:32:00,924 --> 00:32:02,124
Speaker 2: you know, a little furry animal.

639
00:32:02,564 --> 00:32:06,724
Speaker 1: I am so so Nate. Are you familiar with Microsoft

640
00:32:06,764 --> 00:32:12,844
Speaker 1: Office's paper Cliff Gliffe? Oh my god, I think we

641
00:32:12,884 --> 00:32:14,044
Speaker 1: all have stories about clipping.

642
00:32:14,164 --> 00:32:15,804
Speaker 2: I mean, there are things about paper clips and AI.

643
00:32:16,084 --> 00:32:17,284
Speaker 2: You do you probably don't want it to.

644
00:32:17,444 --> 00:32:19,444
Speaker 1: No, we do not want paper clips anywhere near our

645
00:32:19,484 --> 00:32:23,644
Speaker 1: AI models. That the paper clip problem has given us

646
00:32:23,764 --> 00:32:26,204
Speaker 1: enough headaches. By the way, for those of you who

647
00:32:26,284 --> 00:32:29,244
Speaker 1: aren't familiar with the paper clip problem, you know, you

648
00:32:29,484 --> 00:32:32,044
Speaker 1: might know Microsoft clipping, but not the paper clip problem.

649
00:32:32,244 --> 00:32:36,924
Speaker 1: It's an AI philosophy problem first proposed by Nick Bostrom

650
00:32:37,004 --> 00:32:39,164
Speaker 1: about basically how paper clips can cause the end of

651
00:32:39,164 --> 00:32:42,164
Speaker 1: the world. Well, we'll talk more about it in today's

652
00:32:42,284 --> 00:32:46,364
Speaker 1: Pushkin plus Day. What about you, what's your ideal personality?

653
00:32:46,804 --> 00:32:48,804
Speaker 2: Yeah? Like, I mean, you know, my custom instructions are

654
00:32:48,844 --> 00:32:53,244
Speaker 2: to be straight forward, to provide a lot of detail.

655
00:32:53,444 --> 00:32:55,644
Speaker 2: I'm not looking for the AI for like, you.

656
00:32:55,604 --> 00:33:00,484
Speaker 1: Don't want emotional, you don't want fiery foxy or fighty foxy.

657
00:33:00,644 --> 00:33:04,684
Speaker 2: I take you fivey fox. No, I want five fox.

658
00:33:04,764 --> 00:33:07,244
Speaker 1: Yeah, all right, five e fox. So you want five

659
00:33:07,244 --> 00:33:11,084
Speaker 1: e fox. I want to be like a cartoon, except

660
00:33:11,124 --> 00:33:12,084
Speaker 1: not a paper club.

661
00:33:12,204 --> 00:33:12,964
Speaker 2: Not a paper club.

662
00:33:15,444 --> 00:33:18,484
Speaker 1: Let's take a little break, Nate, and then talk about

663
00:33:18,564 --> 00:33:21,964
Speaker 1: n video and another element of AI, the chips that

664
00:33:22,044 --> 00:33:35,444
Speaker 1: make it happen, Nate, this has been such an AI

665
00:33:35,964 --> 00:33:38,604
Speaker 1: E AI E. That's a weird word, but you know

666
00:33:38,644 --> 00:33:39,084
Speaker 1: what I mean.

667
00:33:39,284 --> 00:33:39,524
Speaker 2: Week.

668
00:33:40,004 --> 00:33:44,324
Speaker 1: The other kind of big news has been in video

669
00:33:44,604 --> 00:33:48,964
Speaker 1: and the fact that you know, we've gone through quite

670
00:33:49,084 --> 00:33:52,484
Speaker 1: the cycle on in video where at first there was

671
00:33:52,564 --> 00:33:55,404
Speaker 1: a ban on in Nvidia selling its chips to China.

672
00:33:56,084 --> 00:33:59,524
Speaker 1: Then within the last month the ban was softened and

673
00:34:00,444 --> 00:34:02,644
Speaker 1: Trump announced that they had kind of reached a deal

674
00:34:03,044 --> 00:34:06,444
Speaker 1: where in video was going to be able to sell

675
00:34:06,524 --> 00:34:09,044
Speaker 1: some of its H two O chips to China. And

676
00:34:09,084 --> 00:34:12,764
Speaker 1: then all of a sudden, there was an announcement that

677
00:34:13,244 --> 00:34:16,044
Speaker 1: now Nvidia can sell these chips as long as the

678
00:34:16,284 --> 00:34:20,884
Speaker 1: US government gets fifteen percent of the profits. So this

679
00:34:21,044 --> 00:34:23,724
Speaker 1: is starting to seem a lot like we're now in

680
00:34:23,764 --> 00:34:28,244
Speaker 1: the world of the Godfather or the Sopranos and less

681
00:34:28,284 --> 00:34:30,444
Speaker 1: like we're in the world of the US government. Give

682
00:34:30,444 --> 00:34:32,124
Speaker 1: me a taste, Give me a taste, Nate, and then

683
00:34:32,164 --> 00:34:35,324
Speaker 1: you can do whatever you want. But Papa wants his

684
00:34:35,364 --> 00:34:37,124
Speaker 1: taste of the action. Yeah.

685
00:34:37,204 --> 00:34:39,884
Speaker 2: Look, I mean Trump had his, or the White House had.

686
00:34:39,884 --> 00:34:41,604
Speaker 2: It's like ai action Plan, which you talked about a

687
00:34:41,604 --> 00:34:44,084
Speaker 2: couple weeks ago, and like, you know, people I trust

688
00:34:44,084 --> 00:34:46,204
Speaker 2: thought it wasn't that bad. But like these are people who,

689
00:34:46,244 --> 00:34:48,004
Speaker 2: you know, one thing you might think is good for

690
00:34:48,124 --> 00:34:51,364
Speaker 2: Trump are good for I don't know. Right, what I

691
00:34:51,404 --> 00:34:53,764
Speaker 2: call the river in the book having more influence over

692
00:34:53,804 --> 00:34:55,724
Speaker 2: the White House is like they're all really competitive. They

693
00:34:55,724 --> 00:34:57,204
Speaker 2: want to be China. They want to be China, right,

694
00:34:57,244 --> 00:35:00,164
Speaker 2: And here the US now has like an incentive for

695
00:35:00,244 --> 00:35:03,244
Speaker 2: the best chips in the world to be sold to China. Right,

696
00:35:03,484 --> 00:35:05,364
Speaker 2: I haven't tried. I assume it's going to like the

697
00:35:05,404 --> 00:35:07,564
Speaker 2: Treasury and it is not like Trump's personal stash. But

698
00:35:07,604 --> 00:35:12,404
Speaker 2: like that seems that seems a bit weird. And like granted, okay,

699
00:35:12,804 --> 00:35:15,884
Speaker 2: you manufacture a chip and it's kind of hard for

700
00:35:16,084 --> 00:35:18,324
Speaker 2: China not to get it. Eventually, I'm sure, there are

701
00:35:18,324 --> 00:35:21,364
Speaker 2: black markets and gray markets, although you know, people have

702
00:35:21,444 --> 00:35:23,884
Speaker 2: used the parallel like nuclear facile material. We track that

703
00:35:24,324 --> 00:35:27,044
Speaker 2: pretty carefully potentially, But yeah, no, I imagine that. I mean,

704
00:35:27,124 --> 00:35:29,604
Speaker 2: let's go with that analogy. Right, It's like, oh, okay,

705
00:35:29,644 --> 00:35:33,044
Speaker 2: you can sell radioactive material to Iran.

706
00:35:33,084 --> 00:35:35,084
Speaker 1: Right as long as as long as hes goverment get.

707
00:35:36,524 --> 00:35:40,244
Speaker 2: Think China's Iran or exactly right, but they are the

708
00:35:41,604 --> 00:35:46,364
Speaker 2: right now only country that's competitive with us on AI, right, yeah,

709
00:35:46,324 --> 00:35:48,044
Speaker 2: the third place, right, maybe maybe the least you know,

710
00:35:48,084 --> 00:35:49,164
Speaker 2: let's give them some two.

711
00:35:49,164 --> 00:35:51,924
Speaker 1: Yeah, no, it's kind of crazy. And and Trump just

712
00:35:52,324 --> 00:35:54,964
Speaker 1: says that the more advanced black Well chips, He's like,

713
00:35:54,964 --> 00:35:57,164
Speaker 1: oh yeah, I'm open to us selling those as well

714
00:35:57,524 --> 00:35:59,684
Speaker 1: if we can also get a percentage. So, all of

715
00:35:59,724 --> 00:36:01,964
Speaker 1: a sudden, the national security concerns and you and I

716
00:36:02,004 --> 00:36:04,924
Speaker 1: did a whole segment about the AI Action Plan, and

717
00:36:05,124 --> 00:36:07,124
Speaker 1: the one thing that was kind of a concern in

718
00:36:07,164 --> 00:36:09,524
Speaker 1: it was China, right, And all of a sudden that

719
00:36:09,524 --> 00:36:11,684
Speaker 1: seems to have gone out the window. If we can

720
00:36:11,764 --> 00:36:14,564
Speaker 1: you know, grease the palm a little bit and get

721
00:36:14,644 --> 00:36:17,524
Speaker 1: get the fifteen percent kick back, and so from you know,

722
00:36:17,604 --> 00:36:19,644
Speaker 1: all of a sudden, you realize, well, it was just

723
00:36:19,684 --> 00:36:22,084
Speaker 1: a talking point, right, like, it really didn't matter. It

724
00:36:22,124 --> 00:36:25,204
Speaker 1: wasn't national security doesn't matter. If we can if we

725
00:36:25,244 --> 00:36:29,404
Speaker 1: can get a percentage of this. By the way, after

726
00:36:29,884 --> 00:36:35,364
Speaker 1: these announcements, China has itself said, hey, companies, as in

727
00:36:35,604 --> 00:36:38,804
Speaker 1: like Ali Baba, you know, Chinese companies, we don't want

728
00:36:38,844 --> 00:36:42,044
Speaker 1: you buying these US chips because we're worried that they're

729
00:36:42,044 --> 00:36:45,204
Speaker 1: going to kind of insert location tracking and back doors

730
00:36:45,244 --> 00:36:47,284
Speaker 1: and all these things into them. We want you to

731
00:36:47,324 --> 00:36:51,804
Speaker 1: be buying local That would be smart. And Video said, no, no, absolutely,

732
00:36:51,844 --> 00:36:54,364
Speaker 1: we would never do that. But so they say, you know,

733
00:36:54,604 --> 00:36:58,244
Speaker 1: we want you to buy local Huawei chips and you.

734
00:36:58,084 --> 00:37:02,204
Speaker 2: Want your local pitches and exactly.

735
00:37:03,124 --> 00:37:06,844
Speaker 1: So China is actually like a little bit skeptical of this,

736
00:37:07,324 --> 00:37:10,004
Speaker 1: but it's not a law, and you know, it's not

737
00:37:10,124 --> 00:37:13,284
Speaker 1: actually clear yet how it's going to be enforced because

738
00:37:13,324 --> 00:37:17,164
Speaker 1: the agency that issued this directive doesn't actually have enforcement power.

739
00:37:17,844 --> 00:37:19,884
Speaker 1: And by the way, there's also already been a pre

740
00:37:20,004 --> 00:37:22,164
Speaker 1: order of I think it was like seven hundred thousand

741
00:37:22,284 --> 00:37:24,844
Speaker 1: something like that. There's been a pre order of a

742
00:37:24,884 --> 00:37:28,924
Speaker 1: shit ton in other words, of chips that are presumably

743
00:37:28,924 --> 00:37:32,164
Speaker 1: going to start getting shipped now and so the now

744
00:37:32,284 --> 00:37:34,564
Speaker 1: you know, we talk a lot about incentives, but they're

745
00:37:34,604 --> 00:37:36,884
Speaker 1: just for the US. Like now, it's all out of whack.

746
00:37:37,284 --> 00:37:39,684
Speaker 1: The other part of this, by the way, is that

747
00:37:40,124 --> 00:37:42,484
Speaker 1: there is a ban on the in the Constitution on

748
00:37:42,524 --> 00:37:46,324
Speaker 1: export taxes, and so I can see there being a

749
00:37:46,404 --> 00:37:49,004
Speaker 1: legal challenge here. I mean, this kind of is an

750
00:37:49,044 --> 00:37:53,884
Speaker 1: export tax Actually, think about it A fifteen Like they

751
00:37:53,884 --> 00:37:54,724
Speaker 1: can probably.

752
00:37:54,644 --> 00:37:57,284
Speaker 2: Ban in the Constitution of export taxes. Yeah, I didn't know.

753
00:37:57,364 --> 00:38:00,044
Speaker 1: Yeah, there's a ban in the Constitution on export taxes,

754
00:38:00,324 --> 00:38:05,524
Speaker 1: so we cannot lovey export taxes on companies. Yeah. So

755
00:38:05,684 --> 00:38:07,444
Speaker 1: if you can argue that this is.

756
00:38:07,764 --> 00:38:11,164
Speaker 2: We're going to change constitution, then it seems.

757
00:38:10,364 --> 00:38:13,004
Speaker 1: I mean, I think that we have seen that this

758
00:38:13,044 --> 00:38:17,644
Speaker 1: particular administration has no problems changing the constitution or trying

759
00:38:17,644 --> 00:38:20,684
Speaker 1: to change the facts when the facts don't agree.

760
00:38:20,764 --> 00:38:24,644
Speaker 2: If you were allowed one free constitutional amendment, what would

761
00:38:24,644 --> 00:38:24,884
Speaker 2: you do?

762
00:38:27,004 --> 00:38:28,644
Speaker 1: One free constitutional amendment?

763
00:38:28,724 --> 00:38:32,524
Speaker 2: One free I guess you're by FIAT allowed to enact

764
00:38:32,524 --> 00:38:33,204
Speaker 2: a constitutionalism.

765
00:38:33,244 --> 00:38:35,924
Speaker 1: Oh, I know, it's a really good question. I don't

766
00:38:35,924 --> 00:38:37,804
Speaker 1: have a ready made answer for it to you. Is

767
00:38:37,844 --> 00:38:38,844
Speaker 1: it something you've thought about?

768
00:38:39,164 --> 00:38:42,044
Speaker 2: It's harder to produce some practice than in theory, but

769
00:38:42,284 --> 00:38:44,804
Speaker 2: like to ban gerrymandering is.

770
00:38:44,724 --> 00:38:45,844
Speaker 1: One that would be amazing.

771
00:38:46,004 --> 00:38:48,244
Speaker 2: Yeah, yeah, pretty good one that would be a really

772
00:38:48,284 --> 00:38:50,124
Speaker 2: good one. I mean the Senate, I mean, don't mean

773
00:38:50,164 --> 00:38:54,604
Speaker 2: fucking you know, I think we should sell North Dakota

774
00:38:54,644 --> 00:38:57,684
Speaker 2: to Canada.

775
00:38:57,924 --> 00:39:00,324
Speaker 1: All right, all right, I need a little bit more

776
00:39:00,404 --> 00:39:01,284
Speaker 1: explanation here.

777
00:39:02,084 --> 00:39:05,644
Speaker 2: Too many fucking Dakota's. It's their nice states. It's such

778
00:39:05,684 --> 00:39:08,724
Speaker 2: a there's surprising beauty in South Dakota in particular, I'll

779
00:39:08,724 --> 00:39:10,564
Speaker 2: tell you that much. But I don't think that Dakotas

780
00:39:10,644 --> 00:39:12,244
Speaker 2: need four centers between them.

781
00:39:12,284 --> 00:39:14,684
Speaker 1: This is true. I mean, I do think that the

782
00:39:14,724 --> 00:39:19,644
Speaker 1: senatorial representative kind of model of government is broken.

783
00:39:20,524 --> 00:39:22,404
Speaker 2: You trade North Dakota for Greenland.

784
00:39:27,084 --> 00:39:29,164
Speaker 1: This podcast is devolving.

785
00:39:30,684 --> 00:39:32,844
Speaker 2: Like, oh, I've just came home from Denmark. Oh you

786
00:39:32,884 --> 00:39:38,764
Speaker 2: mean Fargo, Denmark. Never mind, we're typing in the studio.

787
00:39:38,804 --> 00:39:40,364
Speaker 2: If you're listening to the auto version, we have our

788
00:39:40,404 --> 00:39:43,124
Speaker 2: producer turning around and.

789
00:39:43,204 --> 00:39:46,564
Speaker 1: Take it back to the video, Yes, take it back

790
00:39:46,604 --> 00:39:50,924
Speaker 1: to n video. So, yeah, we we have this incredibly

791
00:39:51,004 --> 00:39:55,164
Speaker 1: perverse situation right where the incentives are just now completely fucked. Now,

792
00:39:55,244 --> 00:39:57,844
Speaker 1: let's assume the reason we got on our tangent is

793
00:39:57,884 --> 00:40:01,244
Speaker 1: because of the export taxes. Let's assume that legally this

794
00:40:01,364 --> 00:40:04,204
Speaker 1: is upheld, that the fifteen percent is allowed to stand.

795
00:40:04,444 --> 00:40:08,004
Speaker 1: It also sets an incredibly dangerous precedent for you know,

796
00:40:08,044 --> 00:40:12,964
Speaker 1: for everything, for like, it's a really it's a really

797
00:40:13,004 --> 00:40:16,284
Speaker 1: scary proposition to think that, Now, well, as long as

798
00:40:16,324 --> 00:40:18,644
Speaker 1: you kind of give a kickback to the US government,

799
00:40:18,764 --> 00:40:21,884
Speaker 1: you're fine. So we do see this kind of quid

800
00:40:21,884 --> 00:40:24,884
Speaker 1: pro quo mentality where like, you know, you help me,

801
00:40:24,964 --> 00:40:28,044
Speaker 1: I help you, And that is something that norm that

802
00:40:28,084 --> 00:40:30,204
Speaker 1: should not be happening in a healthy democracy.

803
00:40:31,404 --> 00:40:33,724
Speaker 2: Yeah, I mean I think I think that train flew

804
00:40:33,804 --> 00:40:36,644
Speaker 2: right past the station. Marie, I don't, yah, you know,

805
00:40:36,484 --> 00:40:39,444
Speaker 2: you look, it's it's part of It's the same thing

806
00:40:39,444 --> 00:40:42,884
Speaker 2: with the tariff policy. Right. Why are we doing these

807
00:40:42,964 --> 00:40:46,604
Speaker 2: tariffs that try to encourage the Eastern American made products.

808
00:40:46,644 --> 00:40:49,284
Speaker 2: Is it's trying to do industrial policy, is trying to

809
00:40:49,284 --> 00:40:50,764
Speaker 2: do foreign policy right, or just trying to make a

810
00:40:50,764 --> 00:40:54,364
Speaker 2: bunch of money for the government, right. And sometimes Republicans

811
00:40:54,364 --> 00:40:56,924
Speaker 2: are kind of caught or tariff defenders are kind of

812
00:40:56,964 --> 00:41:00,524
Speaker 2: caught in between saying, you know, actually the ideal amount

813
00:41:00,524 --> 00:41:02,324
Speaker 2: of money is terifts make is zero, because then we

814
00:41:02,404 --> 00:41:04,284
Speaker 2: on shore everything and people saying, hey, that's going to

815
00:41:04,324 --> 00:41:07,524
Speaker 2: make up for like a loss of tax revenue elsewhere. Right,

816
00:41:07,564 --> 00:41:09,884
Speaker 2: And it's kind of the same thing with this with

817
00:41:09,964 --> 00:41:12,884
Speaker 2: this China thing and again, you know and videos like okay, sure, yeah,

818
00:41:12,924 --> 00:41:17,124
Speaker 2: by the way, I own a video stoic, right, so yeah, yeah.

819
00:41:16,844 --> 00:41:19,604
Speaker 1: All right, yeah, well no, but you can actually envision

820
00:41:19,644 --> 00:41:22,604
Speaker 1: a future, right where for companies they're like when they're

821
00:41:22,684 --> 00:41:25,764
Speaker 1: calculating profit, they're like, and this is the cut, Like

822
00:41:25,884 --> 00:41:27,924
Speaker 1: just like before you had to like this is the

823
00:41:27,964 --> 00:41:29,804
Speaker 1: cut we give to the mob boss, like, this is

824
00:41:29,844 --> 00:41:32,164
Speaker 1: the cut that goes to Trump for it's the cost

825
00:41:32,204 --> 00:41:35,284
Speaker 1: of doing business. And so we're willing, we willing to

826
00:41:35,324 --> 00:41:35,564
Speaker 1: do that.

827
00:41:35,804 --> 00:41:40,164
Speaker 2: The Italians, you know, Italians, French, we're acting like the

828
00:41:40,284 --> 00:41:45,564
Speaker 2: fucking Europeans, you know some yeah, yeah, the southern less

829
00:41:45,564 --> 00:41:48,084
Speaker 2: efficient Southern Europeans. Yeah.

830
00:41:48,284 --> 00:41:52,764
Speaker 1: So so it's a obviously, I mean, it's an the

831
00:41:52,844 --> 00:41:54,804
Speaker 1: understatement of the day to say it's not a good look.

832
00:41:55,044 --> 00:41:58,164
Speaker 1: But it also doesn't just it doesn't bode well for

833
00:41:58,524 --> 00:42:00,924
Speaker 1: a lot of things. Yeah, so I think, you know,

834
00:42:01,764 --> 00:42:06,444
Speaker 1: bottom line, like this really is I think bad for

835
00:42:06,684 --> 00:42:11,884
Speaker 1: our economic prospects and for the way that other countries

836
00:42:11,924 --> 00:42:15,764
Speaker 1: see US as well. Right, like that matters, especially when

837
00:42:16,164 --> 00:42:20,484
Speaker 1: our currency matters, and kind of our reliability as a

838
00:42:20,524 --> 00:42:23,164
Speaker 1: trading partner and as a lender, and all these things matter.

839
00:42:23,604 --> 00:42:27,444
Speaker 1: So reputation, reputation matters, and foreign reputation matters, and in

840
00:42:27,524 --> 00:42:29,564
Speaker 1: other stupid things. By the way, we're we're going to

841
00:42:29,644 --> 00:42:34,724
Speaker 1: have be hosting Putin on US soil, even though he

842
00:42:35,164 --> 00:42:38,124
Speaker 1: has a warrant out for his arrest. Really nice to

843
00:42:38,164 --> 00:42:38,404
Speaker 1: see you.

844
00:42:39,084 --> 00:42:41,764
Speaker 2: He's a warrant who has an unrest warrant in the

845
00:42:41,844 --> 00:42:42,404
Speaker 2: US for him.

846
00:42:42,844 --> 00:42:45,884
Speaker 1: Well, the ic C has one, Okay, so he technically

847
00:42:45,884 --> 00:42:48,244
Speaker 1: cannot leave Russia because any other country would have to be.

848
00:42:48,324 --> 00:42:53,204
Speaker 2: Kind of fun a Vladimir, that would be funny.

849
00:42:53,244 --> 00:42:59,164
Speaker 1: Funny that is not happening. Yeah, for war crimes against

850
00:42:59,164 --> 00:43:03,084
Speaker 1: EU crime is it in Alaska or something? It's in Alaska. Yeah, yeah,

851
00:43:03,124 --> 00:43:05,364
Speaker 1: so we'll see what happens there. But anyway, all of

852
00:43:05,364 --> 00:43:10,164
Speaker 1: this not great news. So we have yeah, we have

853
00:43:10,204 --> 00:43:12,924
Speaker 1: a mixed bag for you today on a risky business

854
00:43:12,964 --> 00:43:18,004
Speaker 1: and export taxes. Let's see, let's see what happens on

855
00:43:18,164 --> 00:43:20,284
Speaker 1: the legal side of things, and if this is in

856
00:43:20,324 --> 00:43:23,764
Speaker 1: fact deemed an export tax every time will be challenged.

857
00:43:23,484 --> 00:43:25,724
Speaker 2: GPT five hallucinates they have to pay the US government

858
00:43:25,724 --> 00:43:26,284
Speaker 2: fifteen cents.

859
00:43:26,324 --> 00:43:29,844
Speaker 1: Some of that that would be That would be something, Nate,

860
00:43:29,884 --> 00:43:37,004
Speaker 1: that would be something. Let us know what you think

861
00:43:37,004 --> 00:43:39,524
Speaker 1: of the show. Reach out to us at Risky Business

862
00:43:39,644 --> 00:43:42,884
Speaker 1: at pushkin dot fm. And by the way, if you're

863
00:43:42,884 --> 00:43:45,724
Speaker 1: a Pushkin Plus subscriber, we have some bonus content for

864
00:43:45,804 --> 00:43:48,204
Speaker 1: you that's coming up right after the credits.

865
00:43:48,564 --> 00:43:50,804
Speaker 2: And if you're not subscribing any come on really, but

866
00:43:50,884 --> 00:43:54,004
Speaker 2: consider consider signing up for just six ninety nine a month.

867
00:43:54,004 --> 00:43:56,204
Speaker 2: You get access to all that premium content and ad

868
00:43:56,204 --> 00:43:58,644
Speaker 2: for listening across Pushkin's entire network of shows.

869
00:43:59,524 --> 00:44:02,924
Speaker 1: Risky Business is hosted by me Maria Tanakova.

870
00:44:02,244 --> 00:44:04,524
Speaker 2: And by me Nate Silver. The show is a co

871
00:44:04,564 --> 00:44:08,404
Speaker 2: production of Pushian Industries and iHeartMedia. This episode was produced

872
00:44:08,404 --> 00:44:12,484
Speaker 2: by Isabella Carter. Our associate producer is Sonya Gerwick. Sally

873
00:44:12,524 --> 00:44:15,724
Speaker 2: Hilm is our editor, and our executive producer is Jacob Goldstein.

874
00:44:16,324 --> 00:44:17,644
Speaker 2: Mixing by Sarah Briguer.

875
00:44:18,404 --> 00:44:20,764
Speaker 1: If you like the show, please rate and review us

876
00:44:20,804 --> 00:44:23,364
Speaker 1: so other people can find us too. But please only

877
00:44:23,444 --> 00:44:25,884
Speaker 1: rate and review if you like the show, because you

878
00:44:25,924 --> 00:44:28,284
Speaker 1: know we like good reviews. Thanks for tuning in.