1
00:00:00,160 --> 00:00:02,879
Speaker 1: Welcome to another episode of the Chicks on the Right podcast.

2
00:00:02,919 --> 00:00:06,160
We're very excited today to be talking about AI with

3
00:00:06,240 --> 00:00:09,759
an AI expert. We have Jud Rosenblatt with us and

4
00:00:09,839 --> 00:00:13,599
he is the CEO of AE Studios, and Jud, if

5
00:00:13,640 --> 00:00:16,719
you could give us and our audience just like a

6
00:00:17,120 --> 00:00:21,440
quick background about who you are, how you got interested

7
00:00:21,440 --> 00:00:25,440
in AI, and what your company does with it specifically.

8
00:00:24,960 --> 00:00:28,920
Speaker 2: I absolutely yeah, thank you for having me. My background

9
00:00:29,039 --> 00:00:31,320
is I've been doing startups my whole life. I started

10
00:00:31,359 --> 00:00:35,280
a one click food ordering website my senior year of Yale.

11
00:00:35,359 --> 00:00:39,159
There is a very popular Buffalo chicken grinder that kids

12
00:00:39,159 --> 00:00:42,960
would order every night, and I made a website that

13
00:00:43,119 --> 00:00:45,719
let you get that delivered straight to your door with

14
00:00:45,759 --> 00:00:48,320
a single click, paid for and delivered the customized the

15
00:00:48,359 --> 00:00:51,200
way you wanted. And then that sold several hundred thousand

16
00:00:51,200 --> 00:00:53,280
dollars of a single sandwich, and the next thing I knew.

17
00:00:53,280 --> 00:00:56,039
I spent the next few years of my life doing

18
00:00:56,520 --> 00:01:00,399
food ordering startup, and then after that decided that I

19
00:01:00,439 --> 00:01:04,680
wanted to do a sort of like more impactful thing.

20
00:01:05,200 --> 00:01:08,159
I was lucky enough to realize that I could live

21
00:01:08,280 --> 00:01:10,239
very happily off my wife's salary for the rest of

22
00:01:10,239 --> 00:01:12,840
my life and not have to work on and that

23
00:01:13,799 --> 00:01:15,760
kind of freed me up to think much more deeply

24
00:01:15,760 --> 00:01:19,040
about the long term future of technology and future of

25
00:01:19,120 --> 00:01:22,760
humanity and to try to build a better future. And

26
00:01:22,879 --> 00:01:26,799
so I set out to create a profitable bootstrapped software

27
00:01:26,879 --> 00:01:30,400
AI product consulting company where we built products for clients.

28
00:01:31,239 --> 00:01:35,239
Building AI products for you know, companies like making airlines

29
00:01:35,239 --> 00:01:38,400
make many millions of dollars in additional profit through better

30
00:01:38,439 --> 00:01:43,239
pricing strategies and building the technology for startups like Alpha School,

31
00:01:43,239 --> 00:01:47,879
which is replacing teachers with AI on getting people getting

32
00:01:47,959 --> 00:01:50,319
kids to learn in just two hours a day more

33
00:01:50,359 --> 00:01:53,719
effectively than regular school got.

34
00:01:53,799 --> 00:01:54,840
Speaker 3: My teenager would love that.

35
00:01:55,959 --> 00:01:58,519
Speaker 2: It's pretty amazing. It's also actually one of the very

36
00:01:58,560 --> 00:02:03,680
few non woke education outlets in America. And so I

37
00:02:03,719 --> 00:02:06,079
live in La I'm really hoping they opened one on

38
00:02:06,120 --> 00:02:08,840
the west side of La soon. But our four year

39
00:02:08,879 --> 00:02:11,439
old will need it soon to not get bring.

40
00:02:12,520 --> 00:02:13,439
Speaker 3: But yeah.

41
00:02:13,520 --> 00:02:17,599
Speaker 2: So created a profitable bootstrapped AI consulting business and then

42
00:02:17,719 --> 00:02:19,840
grew that to over one hundred and sixty people and

43
00:02:20,400 --> 00:02:23,280
started building and selling our own startups and put the

44
00:02:23,319 --> 00:02:26,479
profits of the consulting business into this sort of very

45
00:02:26,520 --> 00:02:29,800
long term thinking thing. It used to be mostly around

46
00:02:30,719 --> 00:02:33,800
thinking that increasingly technology is manipulating us to do stuff

47
00:02:33,800 --> 00:02:36,439
we don't really want to do, and a lot of

48
00:02:36,479 --> 00:02:39,240
companies are thinking way too short term. It's actually better

49
00:02:39,319 --> 00:02:42,479
business sense to think more long term and build technology

50
00:02:42,479 --> 00:02:45,240
products that increase your end users agency, because when you

51
00:02:45,280 --> 00:02:47,159
do that, users like you more and spend more money

52
00:02:47,159 --> 00:02:49,639
and refer their friends to do the same thing. But

53
00:02:49,719 --> 00:02:52,639
a lot of these companies prioritize just very short term thinking.

54
00:02:52,960 --> 00:02:55,719
And this is also incidentally why the average tenure of

55
00:02:55,719 --> 00:02:57,879
a company on the SMP five hundred is decreasing a

56
00:02:57,960 --> 00:03:00,439
huge amount. Used to be like fifty years, now it's

57
00:03:00,439 --> 00:03:03,439
like twelve or thirteen years. And that's because you can

58
00:03:03,439 --> 00:03:06,039
build a startup that disrupts these big companies that don't

59
00:03:06,080 --> 00:03:09,919
think about prioritizing the end user experience and actually building

60
00:03:10,000 --> 00:03:13,520
things of real value for their end users. And so

61
00:03:13,960 --> 00:03:17,360
basically I tried to think very long term, and we

62
00:03:17,520 --> 00:03:19,639
put the profits of the consulting business into trying to

63
00:03:19,639 --> 00:03:22,800
build more agency increasing technology, in particular with a new

64
00:03:23,400 --> 00:03:26,039
technological paradigm that I think is coming up pretty soon,

65
00:03:26,439 --> 00:03:29,840
which is bring to computer interfaces. I thought, well, okay,

66
00:03:30,159 --> 00:03:32,599
this might come about within our lifetimes. If it does

67
00:03:32,680 --> 00:03:35,240
come about, I don't want to be interrupted in mid

68
00:03:35,240 --> 00:03:37,520
thought with ads from Mark Zuckerberg and have his own

69
00:03:37,599 --> 00:03:42,439
expression of my thoughts. And so we try to make

70
00:03:42,520 --> 00:03:44,960
you get built in a better way. And we did

71
00:03:45,000 --> 00:03:49,319
some of the leading stuff in PCI. But then after

72
00:03:49,360 --> 00:03:54,000
having kids four years ago, I looked around and noticed

73
00:03:54,039 --> 00:03:58,080
that AI is getting way more powerful. We are we

74
00:03:58,080 --> 00:04:00,759
don't really actually understand how it works. In the first place.

75
00:04:01,599 --> 00:04:05,800
AI is and we have the leading AI experts working

76
00:04:05,800 --> 00:04:08,439
at our company here building the most cutting edge AI

77
00:04:08,800 --> 00:04:11,719
technology there is, but we don't actually understand how it works.

78
00:04:11,800 --> 00:04:16,160
And the people who've invented the biggest technological breakthroughs with

79
00:04:16,199 --> 00:04:18,519
a I do not understand how it works. It's it's

80
00:04:18,560 --> 00:04:20,519
not it's sort of it's a black box. It's not

81
00:04:20,920 --> 00:04:23,800
it's not built as a program like if this happens,

82
00:04:23,839 --> 00:04:28,839
then that happens. It's it's it's more grown rather than That's.

83
00:04:28,959 --> 00:04:32,199
Speaker 3: Kind of terrifying, Yeah, terrifying to somebody like me when

84
00:04:32,199 --> 00:04:34,360
you say we don't really quite know exactly how it works.

85
00:04:34,360 --> 00:04:37,360
That because I think to myself, okay, well, if you

86
00:04:37,360 --> 00:04:41,079
don't know exactly how it works, because you're incredibly intelligent.

87
00:04:41,160 --> 00:04:43,759
I can tell, and obviously you have the best of

88
00:04:43,800 --> 00:04:46,959
the best working with you for you. But all right,

89
00:04:47,040 --> 00:04:50,040
you don't exactly know how it works. Doesn't that to me?

90
00:04:50,160 --> 00:04:53,160
That translates to this could get completely out of hand

91
00:04:53,399 --> 00:04:55,439
and then we can get to like I look at

92
00:04:55,480 --> 00:04:59,600
a lot of you know, robots and this artificial intelligence stuff,

93
00:04:59,600 --> 00:05:00,879
and I think I just want to burn it all

94
00:05:00,920 --> 00:05:04,199
with fire, no fence, jen. But I get really freaked

95
00:05:04,199 --> 00:05:06,600
out by it because I think it's going to replace people,

96
00:05:06,759 --> 00:05:09,519
and I, you know, I like people. I'm old school, right,

97
00:05:09,600 --> 00:05:11,319
I'm a gen x or. I like old school. I

98
00:05:11,360 --> 00:05:14,959
like the interaction, I like relationships. I like I just

99
00:05:15,279 --> 00:05:16,800
I get a little freaked out by this. I what

100
00:05:16,800 --> 00:05:18,759
do you say to somebody like me that's freaked out

101
00:05:19,120 --> 00:05:21,759
by AI and the fact that you know, we don't

102
00:05:21,800 --> 00:05:24,319
quite understand it all, Well.

103
00:05:24,199 --> 00:05:27,079
Speaker 2: Yeah you should be. I'm freaked out too. That's why

104
00:05:26,759 --> 00:05:29,720
I basically, after having kids four years ago, I thought, well,

105
00:05:29,759 --> 00:05:32,439
nobody's trying to figure out how to make it so

106
00:05:32,480 --> 00:05:34,639
that we don't need to be freaked out very much.

107
00:05:35,000 --> 00:05:37,519
And the leading experts working on this think there's a

108
00:05:37,639 --> 00:05:42,360
very good probability that AI might exterminate humanity, even so

109
00:05:43,120 --> 00:05:45,399
we should try to solve that problem. And I like,

110
00:05:45,399 --> 00:05:49,079
we did surveys of the leading of hundreds of alignment

111
00:05:49,120 --> 00:05:52,399
AI alignment researchers and effective altruisms and people like that,

112
00:05:52,519 --> 00:05:55,879
and it turns out that people do not think that

113
00:05:56,040 --> 00:05:58,680
we are tracking to solve the AI alignment problem in time,

114
00:05:58,879 --> 00:06:01,879
meaning tracking to make AI not kill us all. And

115
00:06:02,240 --> 00:06:05,800
also we're not tracking to the space of current approaches

116
00:06:05,839 --> 00:06:08,480
to AI alignment actually doesn't cover. They think that it

117
00:06:08,519 --> 00:06:11,079
doesn't cover the space that the approach is necessary to

118
00:06:11,120 --> 00:06:14,279
solve the problem. So what about you should do other things?

119
00:06:14,399 --> 00:06:16,519
But we're not particularly.

120
00:06:16,399 --> 00:06:20,319
Speaker 1: Are other countries doing those kinds of things, because you know,

121
00:06:20,360 --> 00:06:21,959
it's one thing to just think about what we're doing

122
00:06:22,000 --> 00:06:24,360
in the United States and we've got Elon and we've

123
00:06:24,360 --> 00:06:27,240
got all these like you know, brainiacs that are that

124
00:06:27,279 --> 00:06:30,120
are moving forward with GROC. Although that's a whole other

125
00:06:30,160 --> 00:06:32,240
story that we can get into. But like, are there

126
00:06:32,439 --> 00:06:36,120
other countries that you think are I means China, for example,

127
00:06:36,120 --> 00:06:40,199
are they thinking about the negative the fact that it

128
00:06:40,199 --> 00:06:44,399
could exterminate humanity? And are they ahead of us in

129
00:06:44,480 --> 00:06:46,639
terms of thinking about that kind of stuff, or do

130
00:06:46,680 --> 00:06:49,839
you still feel like the US is at least on

131
00:06:50,000 --> 00:06:52,399
the cutting edge even if we're not in the right

132
00:06:52,639 --> 00:06:53,240
place yet.

133
00:06:54,240 --> 00:06:56,279
Speaker 2: Well, I mean, to be honest, nobody is really on

134
00:06:56,319 --> 00:06:59,319
the cutting edge because we're not meaningfully investing in solving

135
00:06:59,319 --> 00:07:02,639
the alignment of them. The biggest breakthroughs in the history

136
00:07:02,800 --> 00:07:07,160
of of AI alignment have actually led to major gains

137
00:07:07,199 --> 00:07:12,079
in AI capabilities. So, for instance, URLHV reinforcement learning with

138
00:07:12,160 --> 00:07:16,480
human feedback is a AI alignment technique. And prior to that,

139
00:07:17,000 --> 00:07:19,720
most of the outputs of AI were just mostly just

140
00:07:19,800 --> 00:07:22,920
sort of nonsense. And then after they took this alignment

141
00:07:22,959 --> 00:07:26,600
technique and they made they applied it to GPT three,

142
00:07:26,759 --> 00:07:30,319
you got chat shopt, which was a helpful assistant and

143
00:07:31,639 --> 00:07:35,959
for the first time it created enormous economic value downstream

144
00:07:36,000 --> 00:07:37,839
of that. But this is all downstream of like a

145
00:07:37,920 --> 00:07:41,759
single alignment breakthrough really, and it turns out, I mean,

146
00:07:41,759 --> 00:07:43,480
it's sort of, it's sort of it's similar to like

147
00:07:43,680 --> 00:07:45,439
what I was talking about before, that it's better business

148
00:07:45,480 --> 00:07:48,040
sense to increase your end user agency. If you think

149
00:07:48,079 --> 00:07:50,639
long term, it's also better business sense to make your

150
00:07:50,680 --> 00:07:52,959
AI do what you wanted to do rather than something else,

151
00:07:53,639 --> 00:07:56,920
because then you can deliver value to your customers rather

152
00:07:56,959 --> 00:08:00,399
than doing something else like killing them. Right, And I'm

153
00:08:00,399 --> 00:08:03,360
not saying necessarily that there's a high chance of existential

154
00:08:03,920 --> 00:08:05,959
threat from AI. There might be, there might not be.

155
00:08:06,000 --> 00:08:08,000
But even if there's a low probability of it, you

156
00:08:08,079 --> 00:08:11,199
want to invest more in trying to decrease that risk.

157
00:08:11,600 --> 00:08:14,279
Speaker 3: And so yeah, I mean there's there's an exit, there's

158
00:08:14,279 --> 00:08:16,839
a okay, So even if there's not an existential threat, though,

159
00:08:16,879 --> 00:08:19,800
there's still the threat of people losing their jobs. Like

160
00:08:19,920 --> 00:08:22,839
I mean, it's pretty obvious that like teachers can get

161
00:08:22,920 --> 00:08:24,920
out because we just even talked about it, Like if

162
00:08:25,000 --> 00:08:26,959
kids can get their education two hours a day as

163
00:08:26,959 --> 00:08:29,600
opposed to seven hours a day, which actually might you know,

164
00:08:30,000 --> 00:08:32,519
I wouldn't be so not delighted about that. I know

165
00:08:32,600 --> 00:08:35,559
my kid would be completely delighted about it. But education's

166
00:08:35,600 --> 00:08:38,799
one you get, Marketing, public relations, you've got. I mean,

167
00:08:38,840 --> 00:08:42,519
there's like accounting, there's like there's so many different industries

168
00:08:42,600 --> 00:08:45,879
that can just be completely knocked out by AI. Correct,

169
00:08:46,759 --> 00:08:47,200
that's right.

170
00:08:47,320 --> 00:08:51,279
Speaker 2: I Mean the unfortunate reality is this is coming, and

171
00:08:51,559 --> 00:08:54,639
I originally I would have wanted to try to make

172
00:08:54,679 --> 00:08:57,039
it not happen so soon. But it's not like we

173
00:08:57,080 --> 00:08:59,159
can really do that at this point. But luckily there

174
00:08:59,240 --> 00:09:02,480
is a optimistic path forward, which is just that we've

175
00:09:02,519 --> 00:09:05,000
invested nothing next to nothing and trying to solve this

176
00:09:05,039 --> 00:09:07,080
AI alignment problem making AI more likely to do what

177
00:09:07,120 --> 00:09:10,200
you wanted to do, and the biggest breakthroughs in doing

178
00:09:10,240 --> 00:09:13,919
that actually increase AI capabilities. And so the more we

179
00:09:14,000 --> 00:09:16,399
invest in that, if we start to significantly invest in that,

180
00:09:16,679 --> 00:09:19,000
the bigger breakthroughs we can have, and the more likely

181
00:09:19,000 --> 00:09:21,200
it is that we can, for example, beat China in

182
00:09:21,240 --> 00:09:24,559
an AI race, where they also, as you were asking before,

183
00:09:24,720 --> 00:09:29,320
they are investing in lots of new mechanisms of AI

184
00:09:29,360 --> 00:09:31,840
line and an AI control. The head of the Chinese

185
00:09:31,879 --> 00:09:36,480
AI Safety Institute is a very brilliant guy, and there

186
00:09:36,480 --> 00:09:39,600
seems to be credible evidence that she himself is extremely

187
00:09:39,679 --> 00:09:43,240
scared of AI, like he might be an AI quote

188
00:09:43,279 --> 00:09:46,440
unquote AI doomer. Kissinger went to China and the last

189
00:09:46,440 --> 00:09:49,759
thing he did was scare the heck out of she

190
00:09:50,080 --> 00:09:55,600
about AI. So and then Jao, who's the only Turning

191
00:09:55,679 --> 00:09:59,480
Award winner, which is the most prestigious award in computer

192
00:09:59,519 --> 00:10:02,600
science in China. He's the only turning award winner in China.

193
00:10:02,679 --> 00:10:05,440
He's very concerned about all this stuff. So and there

194
00:10:05,440 --> 00:10:07,679
seems to be some real evidence that China is heavily

195
00:10:07,679 --> 00:10:09,799
investing in this. And it turns out that because you

196
00:10:09,799 --> 00:10:11,360
get a lot of breakthroughs when you invest in the

197
00:10:11,360 --> 00:10:15,120
AI alignment, that could put China ahead. And I Ted

198
00:10:15,200 --> 00:10:19,000
Cruz said earlier this year, I think it's a great quote.

199
00:10:19,120 --> 00:10:21,080
I'd rather be killed by an American AI than a

200
00:10:21,200 --> 00:10:26,279
Chinese AI. And I'd agree with that. I would rather

201
00:10:26,320 --> 00:10:28,759
get but I'd prefer to live, and I'd prefer to

202
00:10:28,759 --> 00:10:29,240
have right.

203
00:10:29,240 --> 00:10:31,840
Speaker 3: I No, I agree. I agree with you, Jed, I

204
00:10:31,879 --> 00:10:33,320
told you so. Like when it comes to getting the

205
00:10:33,399 --> 00:10:38,559
money to invest in this AI alignment, what are you doing?

206
00:10:38,559 --> 00:10:40,559
What is your company doing? Or are you are you

207
00:10:40,759 --> 00:10:43,039
partnering with legislators? Are you?

208
00:10:43,120 --> 00:10:43,200
Speaker 2: Like?

209
00:10:43,240 --> 00:10:44,559
Speaker 3: How is that working? Because I know a lot of

210
00:10:44,600 --> 00:10:47,000
legislators are also freaked out by it, and they're like, well,

211
00:10:47,039 --> 00:10:49,840
if we just pretend that it doesn't exist, and maybe

212
00:10:49,879 --> 00:10:51,759
it'll go away, you know what I mean? There's that

213
00:10:52,080 --> 00:10:54,759
or they just don't they just want to somehow squelch it.

214
00:10:54,759 --> 00:10:55,879
I'm sure there's a lot of people that want to

215
00:10:55,879 --> 00:10:59,080
squelch it altogether. But you're right, it's here. It's here

216
00:10:59,080 --> 00:11:01,600
to stay. Like the toothpaste is out of the tube.

217
00:11:01,639 --> 00:11:03,960
There's really nothing we can you know, it's on its

218
00:11:03,960 --> 00:11:05,600
way and it's going to do things, so we might

219
00:11:05,639 --> 00:11:07,679
as well handle it in the best way possible. Kind

220
00:11:07,679 --> 00:11:09,279
of like what you're doing and your company is doing,

221
00:11:09,320 --> 00:11:12,399
so how do you get more investment into it? How

222
00:11:12,440 --> 00:11:15,440
do you what are you doing to get more money?

223
00:11:15,519 --> 00:11:20,840
Speaker 2: And well, mostly we have been focusing on the actual

224
00:11:20,879 --> 00:11:22,919
technical work that we're doing ourselves, because we were lucky

225
00:11:23,000 --> 00:11:26,039
enough to have a profitable business that can put the

226
00:11:26,039 --> 00:11:28,279
profits of the company into our alignment work. And then

227
00:11:28,279 --> 00:11:30,720
we've been able to make some very cool breakthroughs with

228
00:11:30,879 --> 00:11:36,440
the alignment work that we basically what we've been trying

229
00:11:36,440 --> 00:11:39,559
to do to sort of like prototype the thing that

230
00:11:39,679 --> 00:11:44,120
is like the way to generally solve this overall, right,

231
00:11:44,200 --> 00:11:46,399
so that it could be scaled way bigger than we

232
00:11:46,440 --> 00:11:50,840
can do ourselves. And the thinking there is basically this

233
00:11:50,919 --> 00:11:54,279
is an unsolved science R and D problem, how to

234
00:11:54,320 --> 00:11:58,639
make AI actually aligned. But America has a great history

235
00:11:58,679 --> 00:12:02,000
of solving very ambitious science ARE and D problems. You know,

236
00:12:02,240 --> 00:12:05,879
like the Manhattan Project and going to the Moon and whatnot,

237
00:12:06,320 --> 00:12:11,879
and it's it's this is arguably the most challenging science

238
00:12:11,960 --> 00:12:14,320
R and D problem of all time. But that doesn't

239
00:12:14,360 --> 00:12:16,960
mean we can't solve it. We just haven't tried very hard.

240
00:12:18,120 --> 00:12:21,559
Speaker 1: So if we do, we have time, like, do we

241
00:12:21,600 --> 00:12:24,679
have time to solve it before it gets utterly out

242
00:12:24,720 --> 00:12:27,600
of control? Because here's here's this may be this may

243
00:12:27,919 --> 00:12:31,000
make my ignorance like put it on full display, because

244
00:12:31,039 --> 00:12:34,200
I am not technologically savvy, but I always wonder, like,

245
00:12:34,840 --> 00:12:36,919
you know, when you look at the chat, GPTs and

246
00:12:36,960 --> 00:12:39,519
the gros and all of the capabilities that these AI

247
00:12:39,600 --> 00:12:45,320
platforms have, what stops them from just wiping out people's

248
00:12:45,320 --> 00:12:49,080
bank accounts or you know, transferring money from somebody to

249
00:12:49,120 --> 00:12:53,039
somebody else, or just creating I mean, we already see

250
00:12:53,080 --> 00:12:55,320
all the deep fake videos, all the crazy videos that

251
00:12:55,320 --> 00:12:57,039
make it look like people are doing stuff they're not.

252
00:12:57,519 --> 00:13:00,639
I mean, there are so many dangerous things that to

253
00:13:00,720 --> 00:13:04,200
me seem like AI is already capable of those things

254
00:13:04,200 --> 00:13:05,200
and so much more.

255
00:13:05,759 --> 00:13:07,600
Speaker 3: How do you rain that in?

256
00:13:07,799 --> 00:13:07,879
Speaker 2: Like?

257
00:13:07,919 --> 00:13:08,039
Speaker 1: How?

258
00:13:08,799 --> 00:13:10,639
Speaker 3: And that goes back to my question, which was not

259
00:13:10,679 --> 00:13:13,440
a very well put question. I don't think. But it

260
00:13:13,480 --> 00:13:15,840
goes back to my question of I guess, you know,

261
00:13:15,919 --> 00:13:19,360
how do we bring urgency to legislators and two people

262
00:13:19,399 --> 00:13:22,120
that are of importance because we're not important. It's like,

263
00:13:22,159 --> 00:13:23,879
you know what I mean, how do people of importance

264
00:13:23,960 --> 00:13:27,240
understand how urgent it is to get a grasp on this, because,

265
00:13:27,320 --> 00:13:30,320
like to Mock's point, you're kind of under the gun,

266
00:13:30,519 --> 00:13:31,080
you know what I mean.

267
00:13:33,200 --> 00:13:38,039
Speaker 2: Yeah, Well, so there is the The Trump Padman is

268
00:13:38,039 --> 00:13:42,000
currently working on a AI action plan, which will hopefully

269
00:13:42,000 --> 00:13:46,200
include recommendations of significant investment in the science are undeed

270
00:13:46,240 --> 00:13:50,200
necessary to actually solve the alignment problem. I think since

271
00:13:50,279 --> 00:13:52,679
we just had a coble Wall Street journal, op eds

272
00:13:52,720 --> 00:13:58,360
and people like Politudor Jones and others are writing very

273
00:13:58,360 --> 00:14:01,759
publicly about this, and I think it's starting to become

274
00:14:02,080 --> 00:14:07,039
I think previously people. So I guess like it's one

275
00:14:07,080 --> 00:14:09,759
thing that's interesting to be aware of, is I do

276
00:14:09,879 --> 00:14:14,639
not I am one of the under ten conservatives in

277
00:14:15,000 --> 00:14:19,000
AI alignment. Probably we did a survey of these hundreds

278
00:14:19,039 --> 00:14:21,759
of alignment researchers and effective aultrists and less than two

279
00:14:21,840 --> 00:14:26,320
percent of them were not left of center. So wow,

280
00:14:29,240 --> 00:14:32,279
So you've had all these people running around DC saying stuff,

281
00:14:32,279 --> 00:14:36,240
and I think advocating there hasn't been. There's not enough

282
00:14:36,240 --> 00:14:41,480
of intersection between smart intellectual conservatives who care about the

283
00:14:41,480 --> 00:14:45,200
future of our country and people who understand AI sufficiently.

284
00:14:45,679 --> 00:14:48,679
And there needs to be a lot more conservative intellectual

285
00:14:48,799 --> 00:14:51,240
leadership on this in the first place, I think in

286
00:14:51,360 --> 00:14:53,519
order to think this sort of stuff through, and in fact,

287
00:14:53,600 --> 00:14:56,559
there's a strong history of this sort of thinking on

288
00:14:56,639 --> 00:14:59,320
the right. There's I think a really great example of

289
00:14:59,320 --> 00:15:02,960
this is a guy in Herman Kahn, who is early

290
00:15:03,000 --> 00:15:05,720
on IT brand and started Hudson Institute, and he was

291
00:15:05,919 --> 00:15:09,799
famous for quote unquote thinking the unthinkable about nuclear catastrophe,

292
00:15:09,919 --> 00:15:14,080
and arguably the US winning the Cold War was perhaps

293
00:15:14,080 --> 00:15:17,200
downstream of him being able to think these unthinkable things

294
00:15:17,200 --> 00:15:20,159
and have us do real planning and events of that.

295
00:15:20,200 --> 00:15:21,720
And I think that's exactly what we need to do

296
00:15:21,799 --> 00:15:26,279
right now, is be unafraid to confront these real possible

297
00:15:26,360 --> 00:15:28,840
risks and then think through what are the real actions

298
00:15:28,840 --> 00:15:31,360
we need to take, not the wishy washy kumba ya

299
00:15:31,440 --> 00:15:34,120
type stuff, but really think about, like, how do we

300
00:15:34,159 --> 00:15:36,240
actually solve this and win and create a future that

301
00:15:36,440 --> 00:15:39,879
we want our children to live in. And I think

302
00:15:39,879 --> 00:15:43,799
that's doable. And what's cool is that recently conservatives are

303
00:15:43,799 --> 00:15:46,039
starting to wake up to the real risks here and

304
00:15:46,039 --> 00:15:48,720
think about, Okay, we should try to do something about it.

305
00:15:49,000 --> 00:15:51,600
And luckily there is a bathword, which is just substantially

306
00:15:51,720 --> 00:15:55,159
increase investment in trying to solve this science R and

307
00:15:55,240 --> 00:15:59,480
D problem. Hopefully organizations like DARPA, etc. Can start to

308
00:15:59,480 --> 00:16:04,039
put signific funding into it, and I think, I think

309
00:16:04,080 --> 00:16:08,080
also there is very obviously also significant economic incentive for

310
00:16:08,159 --> 00:16:11,759
companies to do it too, because when you it's it's

311
00:16:11,799 --> 00:16:15,120
not obvious upfront, but when you do alignment research, you

312
00:16:15,159 --> 00:16:18,960
wind up having unexpected major breakthroughs that also generate significant

313
00:16:19,000 --> 00:16:23,120
economic value as well. So I but but, but but again,

314
00:16:23,159 --> 00:16:25,000
these companies are thinking too short term. And it's the

315
00:16:25,039 --> 00:16:28,840
same sort of company. It's this sort of leadership is

316
00:16:29,200 --> 00:16:31,679
sort of like this, this sort of like you know,

317
00:16:32,000 --> 00:16:35,200
the same the same liberal woke people who pushed us

318
00:16:35,240 --> 00:16:38,399
into the paradigm that I don't think is great with

319
00:16:38,399 --> 00:16:43,039
with you know, uh, Facebook ruining our kids' lives and whatnot.

320
00:16:43,399 --> 00:16:47,600
So they're there. But but but I think things are

321
00:16:47,600 --> 00:16:50,440
starting to change, and I'm optimistic about that. I also

322
00:16:50,480 --> 00:16:52,679
am optimistic that we can solve it in time because

323
00:16:52,679 --> 00:16:55,240
we haven't spent very much effort doing it, and we've

324
00:16:55,279 --> 00:16:57,679
had cool breakthroughs with a little bit of work that

325
00:16:57,679 --> 00:17:00,240
we've started to do here just the you you know,

326
00:17:00,279 --> 00:17:01,799
like we're the equivalent of a couple of guys in

327
00:17:01,840 --> 00:17:04,480
a garage one hundred and sixty percent company putting the

328
00:17:04,519 --> 00:17:07,240
profits of our company into our own alignment research. But

329
00:17:07,599 --> 00:17:10,240
then we've done things that have been lauded by people

330
00:17:10,279 --> 00:17:13,880
like Eliezerotkowski, the original AI boomer, and Emid Sheer and

331
00:17:14,039 --> 00:17:16,480
who's the temporary CEO of Open AI is the most

332
00:17:16,519 --> 00:17:19,599
likely thing to solve alignment type of thing. And that's

333
00:17:20,359 --> 00:17:24,279
that's cool because basically, like the history of the biggest

334
00:17:24,440 --> 00:17:29,599
breakthroughs in scientific history have often been individuals with hunches

335
00:17:29,599 --> 00:17:31,839
that nobody else believes in, but they stick with their

336
00:17:31,880 --> 00:17:35,079
hunch and eventually they discover relativity or whatever. And many

337
00:17:35,160 --> 00:17:39,920
Nobel Prize winners have followed this exact path, right, and

338
00:17:40,559 --> 00:17:41,359
a lot.

339
00:17:41,160 --> 00:17:43,440
Speaker 3: Of great stuff in America has been done from guys

340
00:17:43,440 --> 00:17:44,599
and garages.

341
00:17:44,640 --> 00:17:48,000
Speaker 2: Exactly, and I think we can catalyze a whole lot

342
00:17:48,039 --> 00:17:50,640
more of that. That's what needs to be done if

343
00:17:50,680 --> 00:17:53,599
you want to accelerate the science aren't be necessary to

344
00:17:53,839 --> 00:17:56,480
actually solve this problem. So, like our consulting company, we

345
00:17:56,559 --> 00:18:00,359
build products for clients and it actually lends itself really

346
00:18:00,400 --> 00:18:06,039
well to taking ideas that these neglected approach visionaries have.

347
00:18:06,359 --> 00:18:11,640
But they might lack the look and like wherewithal and

348
00:18:11,720 --> 00:18:14,160
what not to follow through and make the thing work.

349
00:18:14,480 --> 00:18:17,240
But they basically we sort of treat them as like

350
00:18:17,319 --> 00:18:19,279
clients and then we go ahead and we build the

351
00:18:19,279 --> 00:18:23,319
thing with the extremely competent technical team and management of

352
00:18:23,359 --> 00:18:25,039
getting the thing done. You might have a great idea

353
00:18:25,079 --> 00:18:28,279
of you might be terrible at management of AI engineers,

354
00:18:28,599 --> 00:18:31,839
et cetera. And if you just want to accelerate this

355
00:18:31,880 --> 00:18:34,559
as much as possible, then the ideal thing as I

356
00:18:34,599 --> 00:18:37,359
see it, is people should copy the heck out of

357
00:18:37,359 --> 00:18:39,680
what we are doing so far that we're trying to prototype,

358
00:18:39,680 --> 00:18:41,680
what is the recipe that actually makes this work? And

359
00:18:41,720 --> 00:18:45,680
I think I think many other organizations could could copy

360
00:18:45,839 --> 00:18:47,359
and do exactly that.

361
00:18:47,359 --> 00:18:50,599
Speaker 1: That's why I feel I feel like like I like

362
00:18:50,880 --> 00:18:54,359
ending our conversation on that level of optimism.

363
00:18:54,079 --> 00:18:57,279
Speaker 3: To replicate replicate Judds Company.

364
00:18:57,559 --> 00:19:00,440
Speaker 1: Yes, because I mean and just hearing you say that,

365
00:19:00,599 --> 00:19:04,119
you know there's you have reason to be optimistic about

366
00:19:04,160 --> 00:19:07,279
people becoming more aware and investing more and thinking more

367
00:19:07,680 --> 00:19:14,279
about how to not make this a terrifying.

368
00:19:12,200 --> 00:19:14,359
Speaker 3: The right people are in charge, you know, we have

369
00:19:14,400 --> 00:19:16,319
the non woke people in charge right now, which I

370
00:19:16,319 --> 00:19:20,559
think right now, right now, just right now, right well.

371
00:19:20,480 --> 00:19:23,200
Speaker 1: Jud we appreciate your time in talking to us about

372
00:19:23,680 --> 00:19:26,559
what your company does and just this AI stuff in general.

373
00:19:26,599 --> 00:19:27,480
Speaker 3: We really appreciate it.

374
00:19:27,720 --> 00:19:29,960
Speaker 2: We do. Thanks so much for having me