1
00:00:00,110 --> 00:00:02,240
S1: What do you call an endpoint security product that works

2
00:00:02,240 --> 00:00:05,600
S1: perfectly but makes users miserable a failure? The old approach

3
00:00:05,600 --> 00:00:08,420
S1: to endpoint security is to lock down employee devices and

4
00:00:08,420 --> 00:00:11,870
S1: roll out changes through forced restarts, but it just doesn't work.

5
00:00:11,900 --> 00:00:14,720
S1: It is miserable because they've got a mountain of support tickets.

6
00:00:14,720 --> 00:00:17,570
S1: Employees start using personal devices just to get their work done,

7
00:00:17,570 --> 00:00:19,880
S1: and executives opt out the first time it makes them

8
00:00:19,880 --> 00:00:22,190
S1: late for a meeting. You can have a successful security

9
00:00:22,190 --> 00:00:24,920
S1: implementation unless you work with end users, and that's where

10
00:00:24,920 --> 00:00:28,100
S1: Kaleid comes in. Their user first device trust solution notifies

11
00:00:28,100 --> 00:00:30,590
S1: users as soon as it detects an issue on their device,

12
00:00:30,590 --> 00:00:32,720
S1: and teaches them how to solve it without needing help

13
00:00:32,720 --> 00:00:35,930
S1: from it. That way, untrusted devices are blocked from authenticating,

14
00:00:35,930 --> 00:00:38,810
S1: but users don't stay blocked. Kaleidos is designed for companies

15
00:00:38,810 --> 00:00:42,080
S1: with Okta, and it works on macOS, windows, Linux, and

16
00:00:42,080 --> 00:00:44,240
S1: mobile devices. So if you have Okta and you're looking

17
00:00:44,240 --> 00:00:47,690
S1: for a device trust solution that respects your team, visit collider.com.

18
00:00:47,960 --> 00:00:50,600
S1: Unsupervised learning to watch a demo and see how it works.

19
00:00:50,600 --> 00:00:58,280
S1: That's collider.com unsupervised learning. Welcome to Unsupervised Learning, a security, AI,

20
00:00:58,280 --> 00:01:00,830
S1: and meaning focused podcast that looks at how best to

21
00:01:00,830 --> 00:01:04,459
S1: thrive as humans in a post AI world. It combines

22
00:01:04,459 --> 00:01:08,300
S1: original ideas, analysis, and mental models to bring not just

23
00:01:08,300 --> 00:01:11,209
S1: the news, but why it matters and how to respond.

24
00:01:15,540 --> 00:01:18,330
S1: All right. Welcome to unsupervised learning. This is Daniel Meisler.

25
00:01:18,360 --> 00:01:24,119
S1: All right. 436. In about a bunch of AI stuff here. So, uh,

26
00:01:24,569 --> 00:01:27,780
S1: big religious day for me on Monday. It was, uh,

27
00:01:27,780 --> 00:01:31,800
S1: WWDC by Apple. And what I said after it was Apple,

28
00:01:31,800 --> 00:01:35,190
S1: just one. I they caught up and passed everyone basically

29
00:01:35,190 --> 00:01:38,190
S1: in one shot. And I think this is because the

30
00:01:38,190 --> 00:01:40,229
S1: thing that they get that most people don't is it

31
00:01:40,230 --> 00:01:42,690
S1: it's all about the platform. It's all about the integrations.

32
00:01:42,690 --> 00:01:45,809
S1: It's all about the ecosystem. And they're essentially the only

33
00:01:45,810 --> 00:01:49,200
S1: one who understands that and is also good at tech.

34
00:01:49,200 --> 00:01:52,380
S1: And they also have this massive platform in the iPhone. Right.

35
00:01:52,380 --> 00:01:55,260
S1: So it's just they're just winning. They're so far ahead

36
00:01:55,260 --> 00:01:57,960
S1: of everyone in terms of long term thinking. Like it's

37
00:01:57,960 --> 00:02:00,660
S1: not even close now. This one has since been cleared up,

38
00:02:00,660 --> 00:02:02,820
S1: but at the time it wasn't clear what part of

39
00:02:02,820 --> 00:02:06,870
S1: what they were doing was actually Apple stuff. And what

40
00:02:06,870 --> 00:02:09,420
S1: part was OpenAI. I feel like they could have done

41
00:02:09,419 --> 00:02:11,820
S1: a better job explaining that because I was paying attention,

42
00:02:11,820 --> 00:02:14,940
S1: and I still kind of didn't get it. Looking back,

43
00:02:14,940 --> 00:02:18,030
S1: it was pretty clear somehow I missed it. And I,

44
00:02:18,030 --> 00:02:20,190
S1: I mean, I used to work there, I've worked on

45
00:02:20,190 --> 00:02:23,460
S1: AI quite a bit and I like, I love AI

46
00:02:23,460 --> 00:02:25,620
S1: and I love Apple, and I still didn't get it.

47
00:02:25,620 --> 00:02:27,510
S1: So I feel like they could have been more clear.

48
00:02:27,510 --> 00:02:30,120
S1: So the answer to this question is that most of

49
00:02:30,120 --> 00:02:33,060
S1: what they showed was actually all Apple stuff. It was

50
00:02:33,060 --> 00:02:37,440
S1: not pure OpenAI. And that that's a huge difference, right?

51
00:02:37,440 --> 00:02:40,980
S1: So so essentially what the situation is, is, I would

52
00:02:40,980 --> 00:02:43,980
S1: say probably 90% of what they showed. I don't know,

53
00:02:43,980 --> 00:02:46,170
S1: you can figure it out for yourself. Like if you

54
00:02:46,169 --> 00:02:48,510
S1: did the the math or whatever, it doesn't really matter.

55
00:02:48,510 --> 00:02:51,840
S1: The point is like 90% roughly of everything they talked

56
00:02:51,840 --> 00:02:56,610
S1: about on device, the cloud stuff, the whole new cloud infrastructure,

57
00:02:56,610 --> 00:03:00,870
S1: which is absolutely insane. They basically built a new AI

58
00:03:00,900 --> 00:03:05,250
S1: cloud infrastructure that's completely secure, where, you know, they don't

59
00:03:05,250 --> 00:03:08,160
S1: even have access to the content that's on there. They

60
00:03:08,160 --> 00:03:11,220
S1: don't give that access to any third parties. It's essentially

61
00:03:11,220 --> 00:03:15,270
S1: like a cloud version of the local, very secure enclave.

62
00:03:15,270 --> 00:03:17,580
S1: I'm not sure it's quite that level, but it's pretty

63
00:03:17,580 --> 00:03:20,670
S1: much that level. It is better than anything that exists

64
00:03:20,669 --> 00:03:23,280
S1: in the industry right now in terms of like secure

65
00:03:23,280 --> 00:03:27,180
S1: cloud compute, and it's just extraordinary. And they generated it

66
00:03:27,180 --> 00:03:30,239
S1: just so that they could do these AI features. So

67
00:03:30,240 --> 00:03:33,840
S1: that is essentially their AI story is they're going to

68
00:03:33,840 --> 00:03:37,140
S1: have deep context about you because they're Apple. And this

69
00:03:37,140 --> 00:03:40,050
S1: is essentially life OS is what we're talking about. And

70
00:03:40,050 --> 00:03:43,620
S1: so they have all that information about you and they

71
00:03:43,620 --> 00:03:46,290
S1: have the ability to bring that to AI. And what

72
00:03:46,290 --> 00:03:48,090
S1: they've done is they've made it so that if you

73
00:03:48,090 --> 00:03:50,610
S1: could do it locally, it does it locally, and if

74
00:03:50,610 --> 00:03:52,470
S1: you can't do it locally, it goes up to the

75
00:03:52,470 --> 00:03:56,820
S1: cloud using AI models by Apple. So these are Apple

76
00:03:56,820 --> 00:04:00,660
S1: models running on Apple silicon right. It's not even it's

77
00:04:00,660 --> 00:04:04,140
S1: not even Nvidia stuff. They're using their own chips also

78
00:04:04,140 --> 00:04:07,350
S1: in the cloud and obviously on device. So that's the

79
00:04:07,350 --> 00:04:12,480
S1: whole ecosystem. Now that is for all the context data okay.

80
00:04:12,480 --> 00:04:16,590
S1: That's for basically all of the personal data that you have,

81
00:04:16,589 --> 00:04:18,810
S1: which they don't want to share with other people and

82
00:04:18,810 --> 00:04:21,210
S1: they want to be very, very private. Now in the

83
00:04:21,210 --> 00:04:23,880
S1: case of like looking things up like, oh, get the

84
00:04:23,880 --> 00:04:27,270
S1: current weather and stuff like that. That's where they partnered

85
00:04:27,270 --> 00:04:31,919
S1: with the new Siri, with the new with OpenAI. Right.

86
00:04:31,920 --> 00:04:34,529
S1: So that's that's what they did. So essentially what they

87
00:04:34,529 --> 00:04:38,820
S1: did was they said, okay for live searches, OpenAI is

88
00:04:38,820 --> 00:04:42,120
S1: really good at that. You know, fact checking, stuff like that.

89
00:04:42,120 --> 00:04:44,700
S1: Just just calling the internet. And maybe they're not great

90
00:04:44,700 --> 00:04:46,620
S1: at fact checking, but you know what I mean? So

91
00:04:46,620 --> 00:04:51,420
S1: they're leveraging third party AI for a very specific thing,

92
00:04:51,420 --> 00:04:54,539
S1: but they're not like uploading all this stuff to OpenAI.

93
00:04:54,540 --> 00:04:57,599
S1: It's like the exact opposite of what they're doing. They

94
00:04:57,600 --> 00:05:00,570
S1: built this whole new infrastructure just to keep it all

95
00:05:00,570 --> 00:05:03,299
S1: very secure and very private. So it's very much two

96
00:05:03,300 --> 00:05:06,030
S1: different things. Apple AI, which is like 90% of what

97
00:05:06,029 --> 00:05:08,580
S1: they talked about, which has the on prem and the

98
00:05:08,580 --> 00:05:12,060
S1: cloud component. And then this other third party integration, which

99
00:05:12,060 --> 00:05:14,640
S1: is with ChatGPT. And they even talked about yeah, they're

100
00:05:14,640 --> 00:05:17,670
S1: going to do like they're looking at possibly doing Gemini.

101
00:05:17,670 --> 00:05:20,609
S1: But the point is other third party services for those

102
00:05:20,610 --> 00:05:24,539
S1: kind of integrations. But don't confuse it. That is not

103
00:05:24,570 --> 00:05:28,920
S1: what they're doing with our personal data. Like so Live Bovary.

104
00:05:29,070 --> 00:05:31,290
S1: A whole bunch of people were like, well, how do

105
00:05:31,290 --> 00:05:34,409
S1: I find an android? Like, I can't believe they're sending

106
00:05:34,410 --> 00:05:36,510
S1: all my data to OpenAI. And it's like, that is

107
00:05:36,510 --> 00:05:38,729
S1: not what they said. It's not what they said. It's

108
00:05:38,730 --> 00:05:42,270
S1: like the opposite of what they said. So, um, already

109
00:05:42,270 --> 00:05:46,950
S1: running the demos or the betas for iOS 18, uh,

110
00:05:46,950 --> 00:05:50,490
S1: remarkably stable, but largely because they didn't really change that

111
00:05:50,490 --> 00:05:54,060
S1: much yet, and they definitely haven't added like the AI

112
00:05:54,089 --> 00:05:57,600
S1: services yet. So like, Siri is still the same. That

113
00:05:57,600 --> 00:06:00,570
S1: stuff is coming out like later this year, they said.

114
00:06:00,570 --> 00:06:03,029
S1: So I feel like I didn't really get the AI

115
00:06:03,029 --> 00:06:06,300
S1: features that I was looking for immediately in the beta,

116
00:06:06,300 --> 00:06:09,480
S1: which I was hoping to get. But whatever it happens

117
00:06:09,480 --> 00:06:13,560
S1: when it happens went and trained kickboxing, I was. Bad

118
00:06:13,560 --> 00:06:16,800
S1: at it. It was quite sad. I was like five

119
00:06:16,800 --> 00:06:19,950
S1: times worse than a beginner, I think, because it's hard

120
00:06:19,950 --> 00:06:23,070
S1: for me to like, do things slowly and I'm like

121
00:06:23,070 --> 00:06:25,529
S1: imagining things in my brain, but my body won't do

122
00:06:25,529 --> 00:06:28,350
S1: that thing that's in my brain. I'm just very sort

123
00:06:28,350 --> 00:06:30,810
S1: of confused. So I think I'm going to do some more,

124
00:06:30,810 --> 00:06:34,380
S1: like learning the basics of the body coordination, like on

125
00:06:34,380 --> 00:06:37,890
S1: a bag, and then do it with live people after that.

126
00:06:37,890 --> 00:06:41,039
S1: That's my current plan. Plus I'm going to be doing

127
00:06:41,040 --> 00:06:43,320
S1: a Gits as well. And uh, one of the most

128
00:06:43,320 --> 00:06:47,010
S1: important conversations ever on AI safety. Absolutely. Got to watch

129
00:06:47,010 --> 00:06:51,960
S1: this thing. It's, uh, Leopold Aschenbrenner and Dwarkesh Patel. I've

130
00:06:51,960 --> 00:06:54,630
S1: talked about this a million times, but it is really,

131
00:06:54,630 --> 00:06:58,890
S1: really good. Wrote a new fabric pattern called Capture Thinkers Work.

132
00:06:58,890 --> 00:07:03,150
S1: So you basically send in, like Hayek or somebody into,

133
00:07:03,150 --> 00:07:07,170
S1: you know, fabric SP capture thinkers work. And I mean,

134
00:07:07,170 --> 00:07:10,800
S1: it's extraordinary. In fact, I can show you, okay, so

135
00:07:10,800 --> 00:07:15,180
S1: if we do a capture thinkers work here, let's do this.

136
00:07:15,180 --> 00:07:18,030
S1: Let's first let's blow it up some and then we'll

137
00:07:18,030 --> 00:07:22,740
S1: do who's a philosopher? Hume. We'll send that to capture

138
00:07:22,740 --> 00:07:26,310
S1: thinkers work. So comes up with a single line sentence

139
00:07:26,310 --> 00:07:30,900
S1: up here. Most impactful ideas primary advice or teachings. The

140
00:07:30,900 --> 00:07:33,960
S1: works their main works. So you can go build a

141
00:07:33,960 --> 00:07:37,830
S1: reading list based on this top quotes. Something is human

142
00:07:37,830 --> 00:07:41,460
S1: if it emphasizes the role of experienced custom and sentiment

143
00:07:41,460 --> 00:07:45,540
S1: over abstract reason and innate ideas, and like advice of

144
00:07:45,540 --> 00:07:50,760
S1: how to like implement human concepts into your life. So yeah,

145
00:07:50,760 --> 00:07:53,100
S1: love this pattern. It's a great way of like thinking

146
00:07:53,100 --> 00:07:56,430
S1: about someone's work. So for example, if you do like

147
00:07:56,460 --> 00:07:58,800
S1: S or Perl, which it doesn't matter if you spell

148
00:07:58,800 --> 00:08:01,290
S1: it wrong, for example, it looks like maybe I spelled

149
00:08:01,290 --> 00:08:03,930
S1: it wrong because it corrected it to one R and

150
00:08:03,930 --> 00:08:07,290
S1: one L, but yeah. Background where they were born, the

151
00:08:07,290 --> 00:08:11,580
S1: school psychotherapy just really, really powerful. Got a thread here

152
00:08:11,580 --> 00:08:14,280
S1: on how to fight a good mentor. I, I put

153
00:08:14,280 --> 00:08:16,230
S1: this out a very long time ago, but I think

154
00:08:16,230 --> 00:08:18,239
S1: it's worth sharing. I'm going to go ahead and open

155
00:08:18,240 --> 00:08:22,260
S1: it up. So it's like yeah, being mentored by someone

156
00:08:22,260 --> 00:08:25,560
S1: ahead of you can really change your life. Don't overuse flattery.

157
00:08:25,560 --> 00:08:30,600
S1: Ask something very specific. Behave like a future peer like don't.

158
00:08:30,690 --> 00:08:34,980
S1: Don't grovel. Basically, show your you've already done a bunch

159
00:08:34,980 --> 00:08:38,040
S1: of work and, you know, don't ask basic questions that

160
00:08:38,040 --> 00:08:41,370
S1: you could have got by reading their blog or their book.

161
00:08:41,370 --> 00:08:43,709
S1: I mean, to some extent you can, but you don't

162
00:08:43,710 --> 00:08:46,380
S1: want to overdo that and ask for an opinion on

163
00:08:46,380 --> 00:08:48,900
S1: something you've created. So it's like, look, I've done work,

164
00:08:48,900 --> 00:08:52,979
S1: I've created something. I'm curious about a very specific question

165
00:08:52,980 --> 00:08:55,320
S1: about what you think about this thing you don't want

166
00:08:55,320 --> 00:08:58,469
S1: to ask open ended things. Kind of a general rule

167
00:08:58,470 --> 00:09:01,980
S1: here is like, don't give them work. You're you're being

168
00:09:01,980 --> 00:09:05,760
S1: respectful of their time by asking a pretty small thing

169
00:09:05,760 --> 00:09:08,640
S1: and sort of like presenting yourself as a peer and

170
00:09:08,640 --> 00:09:10,770
S1: you're like, hey, look, I'm looking for feedback on this.

171
00:09:10,770 --> 00:09:12,630
S1: I know you've done it a lot, and that's kind

172
00:09:12,630 --> 00:09:15,030
S1: of the vibe. Or you can offer an improvement or

173
00:09:15,030 --> 00:09:19,800
S1: construction a constructive like comment or feedback on something they've

174
00:09:19,800 --> 00:09:23,400
S1: done and most importantly, like produce something really cool and

175
00:09:23,400 --> 00:09:26,189
S1: show it to them. And then they're likely to just like,

176
00:09:26,190 --> 00:09:29,040
S1: treat you like a future peer and interact with you.

177
00:09:29,040 --> 00:09:31,829
S1: And that's the summary here. So and it got a

178
00:09:31,830 --> 00:09:35,670
S1: blog post around it okay. Let's see. Yeah. Adding source

179
00:09:35,670 --> 00:09:38,460
S1: names to stories now which we'll see here in a second.

180
00:09:38,460 --> 00:09:41,190
S1: So basically you'll know what the source is before you

181
00:09:41,220 --> 00:09:44,820
S1: click it. New member piece on the future of AI okay.

182
00:09:44,820 --> 00:09:46,980
S1: Yeah I'm going to go into this one. This one's

183
00:09:46,980 --> 00:09:51,690
S1: pretty major okay. Let's dive into this. All right. So

184
00:09:51,690 --> 00:09:57,210
S1: first thing is like my definition of AGI and ASI.

185
00:09:57,240 --> 00:10:02,640
S1: So my definition of AGI is can functionally replace an

186
00:10:02,640 --> 00:10:05,970
S1: average white collar worker in the US making 80 K

187
00:10:05,970 --> 00:10:09,540
S1: in 2023. And I say 2023 because it's going to

188
00:10:09,540 --> 00:10:12,570
S1: get weird. This definition will get more weird if you

189
00:10:12,570 --> 00:10:15,719
S1: don't lock it into a previous time before. Like AGI.

190
00:10:15,750 --> 00:10:19,500
S1: We're basically saying like, look, a white collar worker making

191
00:10:19,500 --> 00:10:22,469
S1: an average salary, right? Like 80 K in the US

192
00:10:22,470 --> 00:10:28,470
S1: in 2023. And ASI, I'm defining as an AGI that's

193
00:10:28,470 --> 00:10:32,040
S1: smarter and more capable than any human that's ever lived.

194
00:10:32,040 --> 00:10:34,980
S1: And what what this is essentially saying is like, if

195
00:10:34,980 --> 00:10:37,560
S1: you can functionally replace an average white collar worker in

196
00:10:37,559 --> 00:10:42,780
S1: the US, what that is referring to is generality, because

197
00:10:42,780 --> 00:10:45,150
S1: what I'm talking about is the ability for the boss

198
00:10:45,179 --> 00:10:47,610
S1: to say, you know what, you're actually not on that

199
00:10:47,610 --> 00:10:51,690
S1: project anymore. Go get with Julie and Lisa, and you're

200
00:10:51,690 --> 00:10:55,110
S1: being completely retasked to this other project, which is kind

201
00:10:55,110 --> 00:10:57,840
S1: of unrelated to that. And there is some issue of

202
00:10:57,840 --> 00:11:01,860
S1: like specialization, but we're assuming it's still within the specialization

203
00:11:01,860 --> 00:11:05,189
S1: of that person, and it's general enough where you could

204
00:11:05,190 --> 00:11:08,250
S1: be like, look, it's a completely different task. It's a

205
00:11:08,250 --> 00:11:10,860
S1: completely different project. It's got a different time frame, it's

206
00:11:10,860 --> 00:11:14,250
S1: got different goals, it's got different metrics. And what a

207
00:11:14,250 --> 00:11:17,640
S1: human does is basically say, okay, cool, I will erase

208
00:11:17,640 --> 00:11:20,970
S1: what I was thinking before. I will take the new requirements,

209
00:11:20,970 --> 00:11:22,560
S1: I will come up with a new plan and I

210
00:11:22,559 --> 00:11:24,270
S1: will start working on it, which means you got to

211
00:11:24,270 --> 00:11:26,610
S1: break it into tasks. You got to have a timeline,

212
00:11:26,640 --> 00:11:28,110
S1: you got to have a schedule, you got to have

213
00:11:28,110 --> 00:11:30,720
S1: all that. So in other words, it's not just narrowly

214
00:11:30,720 --> 00:11:35,310
S1: executing small tasks, it's also the planning stage. It's also

215
00:11:35,309 --> 00:11:39,030
S1: constant readjustment, constant evolution of like how do I best

216
00:11:39,030 --> 00:11:41,880
S1: do this. And obviously there's different tiers of quality of

217
00:11:41,880 --> 00:11:44,760
S1: people who can do that as well. But that's why

218
00:11:44,760 --> 00:11:49,410
S1: I say average white collar worker making $80,000. So it's

219
00:11:49,410 --> 00:11:53,400
S1: not like exceptional. And it's not like the worst. So

220
00:11:53,400 --> 00:11:57,000
S1: so given that as a definition of AGI, this general

221
00:11:57,000 --> 00:12:01,020
S1: intelligence that can like be retasked as well as do

222
00:12:01,020 --> 00:12:05,130
S1: specific tasks underneath plus answer emails plus be, you know,

223
00:12:05,130 --> 00:12:09,059
S1: communicative and, you know, basically function as a good employee.

224
00:12:09,059 --> 00:12:13,860
S1: That's essentially AGI. So given that what who is able

225
00:12:13,860 --> 00:12:16,319
S1: to do that and basically be a thinker and be

226
00:12:16,320 --> 00:12:20,429
S1: a producer of new ideas, basically smarter and more capable

227
00:12:20,429 --> 00:12:22,800
S1: than any human that's ever lived. And a good way

228
00:12:22,800 --> 00:12:26,400
S1: of thinking about this is like and Aschenbrenner talked about

229
00:12:26,400 --> 00:12:29,850
S1: this as well in the Türkische caste. So while using

230
00:12:29,850 --> 00:12:33,150
S1: their first names, Leopold talked about that in the Dwarkesh

231
00:12:33,150 --> 00:12:37,170
S1: podcast conversation. It's essentially like, what if you could just

232
00:12:37,170 --> 00:12:42,570
S1: make like 10,000 Einsteins or John von Neumann, right? That

233
00:12:42,570 --> 00:12:45,810
S1: would be insane. So just imagine John von Neumann here,

234
00:12:45,809 --> 00:12:48,989
S1: or Einstein or whoever you want to put as like

235
00:12:48,990 --> 00:12:51,630
S1: the smartest person. And just imagine you can have a

236
00:12:51,630 --> 00:12:54,390
S1: whole bunch of them who could go and do these tasks,

237
00:12:54,390 --> 00:12:58,110
S1: like AGI tasks like, you know, most importantly, come up

238
00:12:58,110 --> 00:13:00,720
S1: with new science, come up with new innovations, come up

239
00:13:00,720 --> 00:13:04,050
S1: with new medicines, but also, yeah, go do all this work,

240
00:13:04,050 --> 00:13:06,870
S1: figure out how how to do it better figure out

241
00:13:06,870 --> 00:13:08,429
S1: better ways to do all the work we have to

242
00:13:08,429 --> 00:13:10,830
S1: do and then go do it. So that's ASI. And

243
00:13:10,830 --> 00:13:14,339
S1: I'm saying essentially there's a big divide here between conscious

244
00:13:14,340 --> 00:13:17,580
S1: and unconscious. And it actually doesn't matter all that much.

245
00:13:17,610 --> 00:13:20,580
S1: A lot of people think that consciousness is on this

246
00:13:20,580 --> 00:13:22,890
S1: path over here. So it's like, oh, we're going to

247
00:13:22,890 --> 00:13:26,310
S1: get artificial narrow intelligence and then we're going to get AGI.

248
00:13:26,309 --> 00:13:28,440
S1: And then after that it becomes conscious and then it

249
00:13:28,440 --> 00:13:31,950
S1: becomes super intelligent or some combination thereof. But I think

250
00:13:31,950 --> 00:13:34,200
S1: the better way to think about this is that we

251
00:13:34,200 --> 00:13:37,200
S1: could stay completely over here and not even get consciousness,

252
00:13:37,200 --> 00:13:41,730
S1: but still have artificial super intelligence, right? So one does

253
00:13:41,730 --> 00:13:45,300
S1: not require the other. And yeah, I've got prediction for

254
00:13:45,630 --> 00:13:50,100
S1: a date later on. We are now inside the post itself. Yeah.

255
00:13:50,100 --> 00:13:52,560
S1: So how I see the paths of development and the

256
00:13:52,559 --> 00:13:54,959
S1: thing that got me talking about this, or thinking about

257
00:13:54,960 --> 00:13:58,350
S1: this a lot more, is just this whole Leopold and

258
00:13:58,350 --> 00:14:02,220
S1: Dwarkesh thing. Yeah. So I talked about that path and

259
00:14:02,220 --> 00:14:06,600
S1: that's the conversation. Yeah. Okay. So this is basically my

260
00:14:06,600 --> 00:14:09,420
S1: argument of like how we're going to get to AGI

261
00:14:09,420 --> 00:14:14,429
S1: and ASI. So I think this is a hypothesis right.

262
00:14:14,520 --> 00:14:17,190
S1: It's it's an idea. It's a thought only so many

263
00:14:17,190 --> 00:14:20,640
S1: components to understand the world. Some of these are easily

264
00:14:20,640 --> 00:14:23,850
S1: attainable and some are like impossible. So an impossible one

265
00:14:23,850 --> 00:14:26,790
S1: is like you can't calculate all the combinations of things

266
00:14:26,790 --> 00:14:31,770
S1: like numbers get crazy very quickly with combinations. So unless

267
00:14:31,770 --> 00:14:35,280
S1: you can brute force everything like like God or something,

268
00:14:35,280 --> 00:14:39,060
S1: it's really hard. But I think there are really realistic

269
00:14:39,060 --> 00:14:44,489
S1: and approachable abstractions of that complexity. So my intuition is

270
00:14:44,490 --> 00:14:47,430
S1: that there are likely abstractions of those that we haven't

271
00:14:47,430 --> 00:14:51,090
S1: yet attained. I think that's pretty obvious. And then an

272
00:14:51,090 --> 00:14:54,750
S1: analogy I use here is like fluid dynamics representing all

273
00:14:54,750 --> 00:14:58,680
S1: the molecules in a liquid, or Newtonian physics being an

274
00:14:58,680 --> 00:15:01,680
S1: abstraction for like quantum or whatever the bottom layer is.

275
00:15:01,680 --> 00:15:05,100
S1: So my friend's argument is I was walking with him

276
00:15:05,100 --> 00:15:08,190
S1: recently and like, I think it's really smart. He basically

277
00:15:08,190 --> 00:15:11,310
S1: said there's limitations to what we can do, even with

278
00:15:11,310 --> 00:15:14,700
S1: super intelligence, because of the one way nature of calculation.

279
00:15:14,700 --> 00:15:18,060
S1: So basically P and P, you could try lots of things,

280
00:15:18,060 --> 00:15:20,130
S1: but it's a one way function. You can see if

281
00:15:20,130 --> 00:15:23,430
S1: it works, but you can't like immediately see all the options.

282
00:15:23,430 --> 00:15:26,640
S1: And that's kind of like P and P. So he

283
00:15:26,640 --> 00:15:29,190
S1: came up with a great definition of intelligence, which I've

284
00:15:29,190 --> 00:15:32,730
S1: since seen like other people talking about as well, which

285
00:15:32,730 --> 00:15:35,880
S1: is like the ability to quickly reduce a search space

286
00:15:35,880 --> 00:15:39,060
S1: for a problem, like like pruning trees. So there's like

287
00:15:39,060 --> 00:15:42,510
S1: 100 trillion options. What can we do to get that

288
00:15:42,510 --> 00:15:45,990
S1: down to like 10 or 100 or 1000 or whatever

289
00:15:45,990 --> 00:15:49,050
S1: as fast as possible. And the faster you could do that,

290
00:15:49,050 --> 00:15:51,150
S1: the more you could do that, the more intelligent you are.

291
00:15:51,150 --> 00:15:54,330
S1: And actually reminds me a lot of quantum computing for

292
00:15:54,330 --> 00:15:57,120
S1: the first time. I've never seen like an intersection there,

293
00:15:57,120 --> 00:16:00,990
S1: but imagine that you could test multiple options all at once,

294
00:16:00,990 --> 00:16:03,960
S1: which is kind of the whole thing with quantum anyway,

295
00:16:03,960 --> 00:16:07,530
S1: so here are the components I see leading to superintelligence.

296
00:16:07,530 --> 00:16:10,500
S1: One is a super complex model of the world, like

297
00:16:10,500 --> 00:16:14,480
S1: how the cell works, how medicines. Interact with cells, you know,

298
00:16:14,480 --> 00:16:19,400
S1: molecular interaction with different body parts. So it's essentially like

299
00:16:19,400 --> 00:16:23,870
S1: testing drugs, but doing it with models instead of like

300
00:16:23,870 --> 00:16:27,590
S1: having to actually try it in labs and people and mice.

301
00:16:27,590 --> 00:16:31,010
S1: So think of that at like all levels of physics.

302
00:16:31,010 --> 00:16:35,600
S1: So like quantum atomic molecules, cells and bigger and bigger,

303
00:16:35,600 --> 00:16:40,010
S1: eventually you're at like psychology, sociology, whatever. And in some

304
00:16:40,010 --> 00:16:43,700
S1: cases it's just feeding our current understanding as text. But

305
00:16:43,700 --> 00:16:47,060
S1: even better, if you like, have actual recordings of these

306
00:16:47,060 --> 00:16:51,710
S1: things happening, similar to how Tesla taught its new full Self-Driving.

307
00:16:51,710 --> 00:16:55,040
S1: It's just like, yeah, this is here. Here are the

308
00:16:55,040 --> 00:16:58,220
S1: actual interactions. Learn from that. So the use case I

309
00:16:58,220 --> 00:17:02,510
S1: kept using for this was solving human aging. So let's

310
00:17:02,510 --> 00:17:04,939
S1: say it's like the telomeres or whatever the length of

311
00:17:04,940 --> 00:17:07,430
S1: the telomeres. And there's a whole bunch of different possible

312
00:17:07,430 --> 00:17:10,490
S1: mechanisms of how aging could be happening, and therefore how

313
00:17:10,490 --> 00:17:12,320
S1: we might be able to turn it off or reverse

314
00:17:12,320 --> 00:17:15,320
S1: it or whatever. But the most important thing is like

315
00:17:15,320 --> 00:17:18,320
S1: have a deep understanding of how everything works, so you

316
00:17:18,320 --> 00:17:20,690
S1: know where you could actually start to go and look.

317
00:17:20,690 --> 00:17:23,330
S1: So the next big thing for this, in my opinion,

318
00:17:23,330 --> 00:17:27,740
S1: is the patterns that analogies and metaphors. So links between things.

319
00:17:27,740 --> 00:17:30,679
S1: And we're already seeing like models get really good at that.

320
00:17:30,680 --> 00:17:32,719
S1: And I think the next one is the size of

321
00:17:32,720 --> 00:17:36,680
S1: the working memory. Right. So not exactly sure what this means,

322
00:17:36,680 --> 00:17:39,800
S1: but it's like it's it's like your Ram or your

323
00:17:39,800 --> 00:17:43,040
S1: functional memory. It's like, how much of this map can

324
00:17:43,040 --> 00:17:45,439
S1: you fit in your brain at once without having to

325
00:17:45,440 --> 00:17:47,300
S1: slide it over? And parts of the map fall off

326
00:17:47,300 --> 00:17:49,520
S1: the table and you can't see them. So maybe that's

327
00:17:49,520 --> 00:17:52,220
S1: a combination of multiple tiers. So you have like you

328
00:17:52,220 --> 00:17:55,160
S1: have like Ram, then you have like L1, L2 inside

329
00:17:55,160 --> 00:17:57,980
S1: of the CPU, which is really the brain, not really

330
00:17:57,980 --> 00:18:00,709
S1: memory but kind of all the same. Right. And then

331
00:18:00,710 --> 00:18:03,770
S1: you've got disk which is much slower, which has to

332
00:18:03,770 --> 00:18:07,400
S1: go into memory to get loaded up. And when there's

333
00:18:07,400 --> 00:18:10,580
S1: too much on disk and not enough Ram, like I said,

334
00:18:10,580 --> 00:18:13,490
S1: you got to swap. Right. So these are the types

335
00:18:13,490 --> 00:18:16,370
S1: of things that I think the overall size of that

336
00:18:16,369 --> 00:18:21,290
S1: memory that can do the patterns based on the world

337
00:18:21,290 --> 00:18:24,800
S1: model is essentially what gets us there. That's my theory.

338
00:18:24,800 --> 00:18:28,250
S1: And essentially this theory is that with scale inside of

339
00:18:28,250 --> 00:18:32,300
S1: neural networks, those three things just keep getting better and better,

340
00:18:32,300 --> 00:18:35,840
S1: plus some post-processing which we'll talk about, and a whole

341
00:18:35,840 --> 00:18:38,000
S1: bunch of hacks and tricks that are basically just going

342
00:18:38,000 --> 00:18:40,369
S1: to come with the industry, because my friend pointed out

343
00:18:40,369 --> 00:18:44,239
S1: that just the size of the model enough alone is

344
00:18:44,240 --> 00:18:47,270
S1: not going to be enough, and that Post-training is super critical.

345
00:18:47,270 --> 00:18:50,420
S1: And once I got what he was saying, like, we

346
00:18:50,420 --> 00:18:52,790
S1: definitely agree on that now, it was just something that

347
00:18:52,790 --> 00:18:55,520
S1: I wasn't aware of and that got us kind of

348
00:18:55,520 --> 00:18:58,010
S1: in line. So this is my way of capturing this.

349
00:18:58,010 --> 00:19:01,100
S1: So my hypothesis is that there are very few fundamental

350
00:19:01,100 --> 00:19:04,520
S1: components that allow AI to scale up from narrow AI

351
00:19:04,640 --> 00:19:08,720
S1: to AGI to ASI. So it's a world model, sufficient

352
00:19:08,720 --> 00:19:12,680
S1: training examples to allow deep enough ability to find patterns

353
00:19:12,680 --> 00:19:17,600
S1: and similarities between phenomenon within that world model, sufficient working memory,

354
00:19:17,600 --> 00:19:21,530
S1: and then ultimately the ability to like, model the scientific

355
00:19:21,530 --> 00:19:26,450
S1: method and test things. Ideally not having to actually do

356
00:19:26,450 --> 00:19:31,430
S1: that test, but being able to test things in an abstraction. Right.

357
00:19:31,430 --> 00:19:35,000
S1: And I'm not sure how smart or how good that is,

358
00:19:35,000 --> 00:19:38,869
S1: because if you're not testing, you're not really testing. I

359
00:19:38,869 --> 00:19:41,690
S1: think this might be possible. And yeah, like I said,

360
00:19:41,690 --> 00:19:45,770
S1: I'm least confident that number four is required or that

361
00:19:45,770 --> 00:19:49,129
S1: it's possible. However, I would say that these three are

362
00:19:49,130 --> 00:19:53,030
S1: amazing and possibly enough. And this one, if it were possible,

363
00:19:53,030 --> 00:19:55,640
S1: to whatever degree, it would massively enhance the other three.

364
00:19:55,640 --> 00:19:58,490
S1: And I think these are the main components. I mean,

365
00:19:58,490 --> 00:20:02,150
S1: I've spent relatively little time thinking about this. I could

366
00:20:02,150 --> 00:20:04,430
S1: revise it in the future. The other thing to say

367
00:20:04,430 --> 00:20:08,210
S1: about this is that number four, which is the testing piece,

368
00:20:08,210 --> 00:20:13,550
S1: might just be implied somehow if number one is sufficiently advanced,

369
00:20:13,550 --> 00:20:17,090
S1: if the world model is good enough, maybe we start

370
00:20:17,090 --> 00:20:20,090
S1: getting to this, I don't know. And maybe it requires

371
00:20:20,090 --> 00:20:24,080
S1: the patterns and similarities. And obviously I think a lot

372
00:20:24,080 --> 00:20:26,690
S1: of this requires the memory. So I guess what I'm

373
00:20:26,690 --> 00:20:29,270
S1: saying is the whole thing might be two big things

374
00:20:29,270 --> 00:20:31,909
S1: the complexity of the world model and the size of

375
00:20:31,910 --> 00:20:34,669
S1: the active memory that it's able to use. And maybe

376
00:20:34,670 --> 00:20:37,910
S1: the patterns and the testing stuff might come along with

377
00:20:37,910 --> 00:20:42,050
S1: those two. And I guess the most controversial thing is that,

378
00:20:42,050 --> 00:20:45,920
S1: and this kind of undergirds everything I'm saying is the

379
00:20:45,920 --> 00:20:50,270
S1: hypothesis that the universe has like an actual complexity that

380
00:20:50,270 --> 00:20:54,560
S1: is approachable or finite, and that models only need to

381
00:20:54,560 --> 00:21:00,410
S1: cross a certain threshold of depth of understanding or quality

382
00:21:00,410 --> 00:21:05,330
S1: of abstractions or depth of abstractions or height of abstractions

383
00:21:05,330 --> 00:21:10,459
S1: to become very similar to like a Laplacian demon in

384
00:21:10,460 --> 00:21:13,810
S1: the sense of like full knowledge. So I don't think

385
00:21:13,810 --> 00:21:16,869
S1: they can have full knowledge. I think that's God level

386
00:21:16,869 --> 00:21:21,189
S1: stuff that's supernatural. At least according to our current understanding

387
00:21:21,190 --> 00:21:24,760
S1: and my belief. But that if you hit a certain

388
00:21:24,760 --> 00:21:29,830
S1: threshold or depth of abstraction quality, it's functionally very similar.

389
00:21:29,830 --> 00:21:34,900
S1: That is what I'm basing my intuition, my hypothesis of

390
00:21:34,900 --> 00:21:39,490
S1: AGI and ASI on. I got another concept called human 3.0.

391
00:21:39,490 --> 00:21:43,060
S1: So it's essentially about how many people believe that they

392
00:21:43,060 --> 00:21:45,370
S1: have ideas that are useful to the world. How many

393
00:21:45,369 --> 00:21:48,280
S1: people believe they could write a short book, percentage of

394
00:21:48,280 --> 00:21:51,609
S1: a person's creativity that they're actively using in their life,

395
00:21:51,609 --> 00:21:55,419
S1: percentage of a person's total capabilities that they actually broadcast

396
00:21:55,420 --> 00:21:59,200
S1: the world as some sort of value, and the degree

397
00:21:59,200 --> 00:22:02,710
S1: to which someone lives as their full spectrum self. So

398
00:22:02,710 --> 00:22:07,869
S1: in human 2.0, we're essentially living and working for capitalism.

399
00:22:07,869 --> 00:22:10,780
S1: Capitalism is like the most important thing. So we're kind

400
00:22:10,780 --> 00:22:14,950
S1: of like 10%. We're putting like 10% of our value

401
00:22:14,950 --> 00:22:17,980
S1: in what we can offer to the world, I would

402
00:22:17,980 --> 00:22:20,560
S1: argue in our CV because it's like, oh, I'm really

403
00:22:20,560 --> 00:22:23,830
S1: good with Microsoft Excel, and I can send emails or whatever.

404
00:22:23,830 --> 00:22:26,380
S1: And it's like, is that really you? I don't think so.

405
00:22:26,380 --> 00:22:29,950
S1: The other aspect is that human 2.0, I would say

406
00:22:29,950 --> 00:22:34,570
S1: current model is essentially most people believing that they're not special,

407
00:22:34,570 --> 00:22:39,219
S1: most people believing that only special people write books, write blogs,

408
00:22:39,220 --> 00:22:44,380
S1: give presentations, do things like that, and regular people don't.

409
00:22:44,380 --> 00:22:49,060
S1: So two main things bifurcated lives broken into personal and professional.

410
00:22:49,060 --> 00:22:52,960
S1: This is 2.0, and the percentage of people who think

411
00:22:52,960 --> 00:22:54,939
S1: that they have something to offer the world, which I

412
00:22:54,940 --> 00:22:59,109
S1: would argue is extremely low and human 3.0 is the

413
00:22:59,109 --> 00:23:02,710
S1: transition to a world where people understand that everyone has

414
00:23:02,710 --> 00:23:04,900
S1: something to offer to the world, and they live as

415
00:23:04,900 --> 00:23:09,280
S1: full spectrum people. So it's like living for other humans,

416
00:23:09,400 --> 00:23:14,530
S1: living as humans, for humans, as opposed to living as

417
00:23:14,530 --> 00:23:19,540
S1: humans who like it's like living to work instead of

418
00:23:19,540 --> 00:23:22,810
S1: working to live and wouldn't even be working to live.

419
00:23:22,810 --> 00:23:25,270
S1: Human 3.0 would even be more advanced than that. It

420
00:23:25,270 --> 00:23:27,609
S1: would be. So it means like your public profile is

421
00:23:27,609 --> 00:23:30,850
S1: everything about you, right? How caring you are, how smart

422
00:23:30,850 --> 00:23:34,300
S1: you are, how funny you are, your favorite things in life,

423
00:23:34,300 --> 00:23:37,420
S1: your best ideas, the projects you like to work on,

424
00:23:37,420 --> 00:23:40,180
S1: the problems you think are important and meaningful in the world,

425
00:23:40,180 --> 00:23:44,380
S1: plus your technical skills, of course. And they know that

426
00:23:44,380 --> 00:23:48,070
S1: they're valuable because they are all these things, not just

427
00:23:48,070 --> 00:23:51,250
S1: what they can do for a corporation. And so my

428
00:23:51,250 --> 00:23:54,070
S1: thoughts on this are like, because you have to start wondering,

429
00:23:54,070 --> 00:23:57,700
S1: when you listen to the Leopold and Aakash conversation, it's like,

430
00:23:57,700 --> 00:24:01,420
S1: is human 3.0 in danger? Are we in danger of

431
00:24:01,420 --> 00:24:05,890
S1: losing a world where humans exist and do what they

432
00:24:05,890 --> 00:24:08,860
S1: do for other humans? Because AC is going to take

433
00:24:08,859 --> 00:24:12,940
S1: over and whatever turn us into whatever. Maybe, hopefully? Well,

434
00:24:12,940 --> 00:24:16,810
S1: I'm pushing for a possible AGI or a that allows

435
00:24:16,810 --> 00:24:20,410
S1: us to do human 3.0 either. It would be kind

436
00:24:20,410 --> 00:24:22,780
S1: of scary if it just took over and implemented it.

437
00:24:22,780 --> 00:24:25,359
S1: That would be scary. Hopefully it would arrive at something

438
00:24:25,359 --> 00:24:29,200
S1: like human 3.0 and keep us alive and thriving. Or

439
00:24:29,200 --> 00:24:31,690
S1: it doesn't take over. It's still used as a tool,

440
00:24:31,690 --> 00:24:35,260
S1: but it's in the hands of someone somewhat benign like

441
00:24:35,260 --> 00:24:40,540
S1: the US, hopefully remains benign. And at that point we

442
00:24:40,570 --> 00:24:44,920
S1: we basically give it the criteria of help us maintain

443
00:24:44,920 --> 00:24:47,859
S1: a society in which we move towards and maintain something

444
00:24:47,859 --> 00:24:51,010
S1: like human 3.0. Another thing I'm really worried about is

445
00:24:51,010 --> 00:24:54,910
S1: like how immigration will affect UBI. So if the government

446
00:24:54,910 --> 00:24:58,570
S1: is basically the jobs are gone, no, very few people

447
00:24:58,570 --> 00:25:01,660
S1: can do something that I can't. And that's like, whatever 10%

448
00:25:01,660 --> 00:25:04,270
S1: of the world or 1% of the world, or 25%

449
00:25:04,270 --> 00:25:08,230
S1: of the world, whatever. Different times, different numbers, different ways

450
00:25:08,230 --> 00:25:11,710
S1: of measuring. But let's say it's 10% and the other 90%

451
00:25:11,710 --> 00:25:15,670
S1: just aren't employable because, like Harari said, you know, they're

452
00:25:15,670 --> 00:25:18,070
S1: a useless class. They have nothing that they can offer

453
00:25:18,070 --> 00:25:20,979
S1: the economy. And this is human 2.0. Still, this is

454
00:25:20,980 --> 00:25:25,030
S1: not crossed over into this hopeful land of human 3.0.

455
00:25:25,030 --> 00:25:28,180
S1: In the meantime, I mean, because some people, like Leopold

456
00:25:28,180 --> 00:25:31,570
S1: are worried about ASI happening in like a number of

457
00:25:31,570 --> 00:25:36,520
S1: years and just like completely redoing society and the government's

458
00:25:36,520 --> 00:25:38,320
S1: obviously going to have to be involved there. They're going

459
00:25:38,320 --> 00:25:42,040
S1: to have to pay people to basically not riot, not

460
00:25:42,040 --> 00:25:48,100
S1: kill themselves and, you know, provide education, provide gaming, provide entertainment,

461
00:25:48,100 --> 00:25:52,210
S1: provide something some sort of basis for meaning that's going

462
00:25:52,210 --> 00:25:55,270
S1: to have to be included. But what does that look like?

463
00:25:55,270 --> 00:25:58,960
S1: I mean, there's like a welfare state and there's like

464
00:25:58,960 --> 00:26:01,750
S1: the the government's giving out too much money, and then

465
00:26:01,750 --> 00:26:04,540
S1: there's a whole nother level when it's actually UBI, because

466
00:26:04,540 --> 00:26:07,630
S1: no one's even able to work. Right. It's not that

467
00:26:07,630 --> 00:26:11,170
S1: they can't find a job when when people are actually hiring,

468
00:26:11,170 --> 00:26:13,780
S1: it's like, no. Nobody's hiring. All the work's being done

469
00:26:13,780 --> 00:26:18,490
S1: by AI or whatever, 90%, and AI is producing trillions

470
00:26:18,490 --> 00:26:22,060
S1: of dollars in value. And that raises another question. Okay,

471
00:26:22,060 --> 00:26:24,730
S1: you're making all this new stuff. You have all these

472
00:26:24,730 --> 00:26:29,020
S1: new creators, you have all this new AI generated productivity

473
00:26:29,140 --> 00:26:34,020
S1: and products and services. For who? Who's buying the stuff?

474
00:26:34,050 --> 00:26:36,929
S1: Who's buying the stuff? If the top 10% are the

475
00:26:36,930 --> 00:26:40,320
S1: only ones who can make things and sell things. Who's

476
00:26:40,320 --> 00:26:44,010
S1: buying the stuff? And the natural answer looks like, which

477
00:26:44,010 --> 00:26:49,140
S1: is quite gross. It's like, I guess the government taxes

478
00:26:49,140 --> 00:26:52,439
S1: all the income that's being made. Again, where's the income

479
00:26:52,440 --> 00:26:55,470
S1: coming from? It's almost like there's an exchange between the

480
00:26:55,470 --> 00:26:59,220
S1: government and the producers, and the producers are making all

481
00:26:59,220 --> 00:27:01,679
S1: this awesome stuff. The government is using all the awesome

482
00:27:01,680 --> 00:27:04,770
S1: stuff to generate all these awesome things. But then the

483
00:27:04,770 --> 00:27:07,590
S1: government has to take this money, which is very strange

484
00:27:07,590 --> 00:27:09,990
S1: word for it now, because it's not being generated by

485
00:27:09,990 --> 00:27:13,320
S1: the people and being taxed by the government. In fact,

486
00:27:13,320 --> 00:27:16,860
S1: it's like it's just a money idea. The money idea

487
00:27:16,859 --> 00:27:21,389
S1: goes to the government. The government then gives the population money,

488
00:27:21,390 --> 00:27:24,179
S1: the ability to purchase these goods and services, which are

489
00:27:24,180 --> 00:27:27,210
S1: being generated by the small percentage of the population to

490
00:27:27,210 --> 00:27:30,870
S1: the tune of whatever trillions of trillions upon trillions of dollars.

491
00:27:30,869 --> 00:27:33,150
S1: And then they give that money out to the people

492
00:27:33,150 --> 00:27:35,700
S1: and the people that can then use that to buy

493
00:27:35,700 --> 00:27:38,399
S1: the services that are being made by this 10%. And

494
00:27:38,400 --> 00:27:41,970
S1: that feels very not quite Ponzi, but like shell game,

495
00:27:41,970 --> 00:27:44,940
S1: it's like, oh, it's fake money. We're moving it around.

496
00:27:44,940 --> 00:27:48,149
S1: It just feels very strange when when the, the human

497
00:27:48,150 --> 00:27:53,700
S1: population is not producing not participating in that creation. Now

498
00:27:53,700 --> 00:27:57,510
S1: in human 3.0, I think they will be I'm not

499
00:27:57,510 --> 00:27:59,970
S1: sure exactly what that economy will look like. I'm sure

500
00:27:59,970 --> 00:28:03,090
S1: much smarter people can help me think about that. It's

501
00:28:03,090 --> 00:28:07,080
S1: essentially more of a human to human exchange barter system.

502
00:28:07,080 --> 00:28:10,470
S1: Obviously a currency like a maybe like a doctor, a

503
00:28:10,590 --> 00:28:13,740
S1: coffee type of currency or something. But you can also

504
00:28:13,740 --> 00:28:16,710
S1: buy regular goods and services, but you could also pay

505
00:28:16,710 --> 00:28:20,190
S1: people for making you happy or whatever. Like a lot

506
00:28:20,190 --> 00:28:22,200
S1: of people will be creating art and doing lots of

507
00:28:22,200 --> 00:28:25,169
S1: stuff like that. So I'm hoping there's a large economy

508
00:28:25,170 --> 00:28:28,080
S1: for sharing on this full spectrum that we talked about

509
00:28:28,080 --> 00:28:31,650
S1: with human 3.0, and that will involve money and stuff

510
00:28:31,650 --> 00:28:34,740
S1: like that. But what I'm talking about here is really

511
00:28:34,740 --> 00:28:38,040
S1: human 2.0 stuff. This is like this year, next year,

512
00:28:38,040 --> 00:28:40,530
S1: the next five years, the next ten years, you know,

513
00:28:40,530 --> 00:28:42,360
S1: what do we do in the meantime with all these

514
00:28:42,360 --> 00:28:45,240
S1: people who are going to be needing money? I mean,

515
00:28:45,240 --> 00:28:49,200
S1: what happens with an economy with millions of immigrants who

516
00:28:49,200 --> 00:28:54,780
S1: are undocumented, who have families, they need to feed them,

517
00:28:54,780 --> 00:28:57,120
S1: they need to house them. And what if the government

518
00:28:57,120 --> 00:28:59,790
S1: is like, well, what if the jobs go away? Well,

519
00:28:59,790 --> 00:29:02,430
S1: maybe they leave. Maybe they don't leave, though, and they're

520
00:29:02,430 --> 00:29:05,340
S1: very angry because they can no longer get work, and

521
00:29:05,340 --> 00:29:09,180
S1: they're also not being paid like the local citizens. I

522
00:29:09,180 --> 00:29:12,270
S1: don't know what the solution here is, but it's I'm

523
00:29:12,270 --> 00:29:14,130
S1: saying we need to think about it. So my thoughts

524
00:29:14,130 --> 00:29:18,810
S1: on conscious AI are, I think, fairly simple actually. I

525
00:29:18,810 --> 00:29:21,960
S1: think the way humans got consciousness is that it was

526
00:29:21,960 --> 00:29:25,979
S1: adaptive for helping us accelerate evolution. So basically winning and

527
00:29:25,980 --> 00:29:29,790
S1: losing is what powers, evolution, blame and praise are smaller

528
00:29:29,790 --> 00:29:33,330
S1: versions of winning and losing plants and insects win and

529
00:29:33,330 --> 00:29:37,770
S1: lose as well, evolutionarily, but it seems like only humans

530
00:29:37,770 --> 00:29:40,380
S1: and a few other species. And maybe it's a lot,

531
00:29:40,380 --> 00:29:44,820
S1: but it's a subset. Very small subset, have subjective experiences,

532
00:29:44,820 --> 00:29:47,760
S1: and it's not clear that anyone else other than humans

533
00:29:47,760 --> 00:29:50,970
S1: has like blame and praise. Well, maybe dogs do. I

534
00:29:50,970 --> 00:29:55,020
S1: feel like they understand, blame and praise. Good boy, bad girl, right?

535
00:29:55,020 --> 00:29:57,870
S1: I feel like that works based on their facial expressions.

536
00:29:57,870 --> 00:30:00,960
S1: Who knows if that's right. But anyway, bottom line is

537
00:30:00,960 --> 00:30:05,940
S1: blame and praise might empower evolution if we experience winning

538
00:30:05,940 --> 00:30:10,020
S1: and losing, being blamed and praised and feeling that at

539
00:30:10,020 --> 00:30:14,970
S1: a visceral level versus having no inter like immediate repercussions

540
00:30:14,970 --> 00:30:17,790
S1: of not doing well. So maybe evolution still gets you

541
00:30:17,790 --> 00:30:22,380
S1: there if you don't have blame and praise and subjective experience.

542
00:30:22,380 --> 00:30:26,340
S1: But I think this whole kind of container is how

543
00:30:26,340 --> 00:30:29,430
S1: we ended up being conscious, because it enables this, which

544
00:30:29,430 --> 00:30:33,300
S1: makes evolution better. That that's my guess. If that's right,

545
00:30:33,300 --> 00:30:37,440
S1: then consciousness might have simply emerged from evolution as our

546
00:30:37,440 --> 00:30:41,040
S1: brains got bigger and we became better at getting better,

547
00:30:41,040 --> 00:30:47,670
S1: which means it could emerge naturally through like self similar loops,

548
00:30:47,670 --> 00:30:51,030
S1: definitely some RL related stuff. Or we could kind of

549
00:30:51,030 --> 00:30:53,340
S1: just like hack the system and like add it on

550
00:30:53,340 --> 00:30:56,610
S1: and be like, hey, develop this so that you can

551
00:30:56,610 --> 00:30:59,280
S1: have this advantage or whatever. We can try to shortcut

552
00:30:59,280 --> 00:31:02,250
S1: what evolution did over millions of years. Yeah. And then

553
00:31:02,250 --> 00:31:05,490
S1: one question is like, well, if we hacked it that way,

554
00:31:05,490 --> 00:31:08,250
S1: would it be real consciousness? And me as a materialist,

555
00:31:08,250 --> 00:31:11,670
S1: I'm like, well, what is real consciousness like? I think

556
00:31:11,670 --> 00:31:15,330
S1: at some level it's all an illusion anyway. But practically,

557
00:31:15,330 --> 00:31:19,980
S1: if you feel yourself feeling and it hurts, that's real, right?

558
00:31:19,980 --> 00:31:24,420
S1: It doesn't matter how mechanistic the substrate below actually is,

559
00:31:24,420 --> 00:31:27,570
S1: because at some point you can pick along the way

560
00:31:27,570 --> 00:31:30,780
S1: and just be like, this is mechanical, this is mechanical. This.

561
00:31:30,920 --> 00:31:35,360
S1: A mechanical. Therefore it's all mechanical. Doesn't matter. Friendship still

562
00:31:35,360 --> 00:31:38,630
S1: is awesome. Love is awesome. Ice cream is awesome. Doesn't

563
00:31:38,630 --> 00:31:41,360
S1: make it any less awesome than it's made out of atoms, right?

564
00:31:41,360 --> 00:31:45,800
S1: So my AC prediction I'm ferm on my AGI prediction

565
00:31:45,800 --> 00:31:51,890
S1: of 2025 and 2028. Somewhere in there, 2026 2027 is

566
00:31:51,890 --> 00:31:54,590
S1: like ideal, the sweet spot for me. I'm going to

567
00:31:54,590 --> 00:32:01,100
S1: give soft prediction, soft prediction of AC 2027 to 31.

568
00:32:01,100 --> 00:32:04,760
S1: So what is that. That's three four plus two. So

569
00:32:04,760 --> 00:32:08,210
S1: that's roughly 2029. Don't feel great about it. That's like

570
00:32:08,210 --> 00:32:12,440
S1: a 30 to 40% prediction like confidence level. This one

571
00:32:12,440 --> 00:32:17,810
S1: I'm saying 90% by 2028. End of 2028 I'm saying

572
00:32:17,810 --> 00:32:21,080
S1: like 60% by end of 2025. So I feel a

573
00:32:21,080 --> 00:32:24,200
S1: lot more confident about that one. But who knows. Predictions

574
00:32:24,200 --> 00:32:27,080
S1: are hard, especially when they're about the future. Niels Bohr

575
00:32:27,080 --> 00:32:30,560
S1: said that. All right, unifying thoughts. Leopold, you want to

576
00:32:30,560 --> 00:32:35,390
S1: watch him on AI safety? Completely. Absolutely impressive guy. And

577
00:32:35,390 --> 00:32:37,190
S1: you want to go read his essays that are released

578
00:32:37,190 --> 00:32:40,520
S1: on this as well. Talked to Aakash Patel is becoming

579
00:32:40,520 --> 00:32:43,910
S1: one of my favorite podcasters ever right now. He is

580
00:32:43,910 --> 00:32:49,430
S1: basically an autodidact. He's a crazy reader. When Dwarkesh talks

581
00:32:49,430 --> 00:32:53,420
S1: to Cowan or Cowan, that's how he pronounces it Tyler Cowan.

582
00:32:53,420 --> 00:32:57,410
S1: When they talk, they jump around, they're like, oh yeah,

583
00:32:57,410 --> 00:32:59,360
S1: you know, I was I was knitting this thing the

584
00:32:59,360 --> 00:33:01,820
S1: other day and I was working. I was, you know,

585
00:33:01,820 --> 00:33:06,230
S1: making wood figurines on my lathe in the back backyard.

586
00:33:06,230 --> 00:33:09,230
S1: And I was thinking about, you know, whatever, Roman history.

587
00:33:09,230 --> 00:33:12,260
S1: And then I was thinking about, you know, Adam's economics.

588
00:33:12,260 --> 00:33:15,470
S1: And that reminds me of this one poem by Rilke

589
00:33:15,470 --> 00:33:18,020
S1: or whatever. And they just jump from here to there

590
00:33:18,020 --> 00:33:20,090
S1: and they're like, oh, that reminds me of this. Hey,

591
00:33:20,090 --> 00:33:22,490
S1: what about World War Two? And this thing happened and

592
00:33:22,490 --> 00:33:24,860
S1: they're just jumping from place to place and like, I

593
00:33:24,860 --> 00:33:26,810
S1: feel like I'm pretty good at that. I feel like

594
00:33:26,810 --> 00:33:31,070
S1: that's one of my superpowers. But Cowan, seeing Cowan do this,

595
00:33:31,070 --> 00:33:35,030
S1: it is insane. And Dwarkesh is like a mini Cowan.

596
00:33:35,030 --> 00:33:38,210
S1: He's like coming up so fast. It's the reason that

597
00:33:38,210 --> 00:33:42,740
S1: his interviews are so good. He's, like, fiercely curious and

598
00:33:42,740 --> 00:33:46,880
S1: an autodidact, reading, like, whatever, 12 books a second, whatever

599
00:33:46,880 --> 00:33:51,980
S1: his current metric is, it's insane. Anyway. Follow them. It's really,

600
00:33:51,980 --> 00:33:56,720
S1: really good stuff. And human 3.0. My best optimism here

601
00:33:56,720 --> 00:33:59,840
S1: is we get AGI, but we don't get AC for

602
00:33:59,840 --> 00:34:02,180
S1: a while, which will give us a chance to hopefully

603
00:34:02,180 --> 00:34:05,810
S1: move towards 3.0 or we get AC, but it's like

604
00:34:05,810 --> 00:34:09,320
S1: controlled by the US and or benign US actors and

605
00:34:09,320 --> 00:34:14,720
S1: we move towards human 3.0 or AC comes online, it's

606
00:34:14,719 --> 00:34:17,960
S1: benign and it kind of takes over and it builds

607
00:34:17,960 --> 00:34:21,260
S1: human 3.0 because that's the best future for humanity or

608
00:34:21,260 --> 00:34:23,900
S1: something that looks similar. Right. And this is what I

609
00:34:23,900 --> 00:34:27,020
S1: was saying before. I don't actually care how unlikely these

610
00:34:27,020 --> 00:34:31,610
S1: scenarios are, because it's kind of the only good path

611
00:34:31,610 --> 00:34:34,220
S1: that I could see happening from all this AI stuff

612
00:34:34,219 --> 00:34:39,330
S1: is we eventually get to human 3.0. A human world

613
00:34:39,330 --> 00:34:42,930
S1: for humans, and I would say future versions of humans,

614
00:34:42,930 --> 00:34:46,350
S1: which might be hybrids with AI or whatever. But you

615
00:34:46,350 --> 00:34:49,109
S1: got to maintain the humanity. We got to be doing

616
00:34:49,110 --> 00:34:52,440
S1: what we're doing for humans, not for the sake of tech.

617
00:34:52,440 --> 00:34:55,770
S1: So I'm optimistic about that. I am trying to help

618
00:34:55,770 --> 00:34:57,989
S1: build that. That's my whole purpose in, like doing all

619
00:34:57,989 --> 00:35:01,680
S1: of this so that that's what I'm shooting for. And

620
00:35:01,680 --> 00:35:04,050
S1: I don't care if it's a 20% chance of us

621
00:35:04,050 --> 00:35:06,120
S1: getting there. First of all, nobody can know what this

622
00:35:06,120 --> 00:35:09,239
S1: number is, so it doesn't really matter. But second of all,

623
00:35:09,239 --> 00:35:12,060
S1: if it's a 10% chance, guess what? I'm shooting for

624
00:35:12,060 --> 00:35:14,280
S1: the 10%. I'm trying to make it 11. Try to

625
00:35:14,280 --> 00:35:17,040
S1: make it a 50, whatever. Try to make it a 90.

626
00:35:17,070 --> 00:35:19,740
S1: If it's a 60% chance and it's a 40% chance,

627
00:35:19,739 --> 00:35:21,690
S1: we're all going to die. Whatever. I'm not going to

628
00:35:21,690 --> 00:35:25,049
S1: sit around thinking about the 40%. I mean, that's my vibe.

629
00:35:25,050 --> 00:35:29,160
S1: I am shooting for human 3.0. And if someone's like, yeah, well,

630
00:35:29,160 --> 00:35:31,200
S1: you don't know, China can take it over and blah

631
00:35:31,200 --> 00:35:33,570
S1: blah blah, AC and it can kill us all and

632
00:35:33,570 --> 00:35:38,190
S1: whatever paper clips and I just don't see the value

633
00:35:38,190 --> 00:35:42,630
S1: of going hard down the path of like, the doomers

634
00:35:42,630 --> 00:35:46,590
S1: because let's say they're right. Their answer is turn it off,

635
00:35:46,590 --> 00:35:50,760
S1: don't build it. That's not going to happen. Next answer.

636
00:35:50,760 --> 00:35:52,890
S1: They're like, I don't have a next answer. We should

637
00:35:52,890 --> 00:35:55,050
S1: stop talking about this. We should stop building it. You

638
00:35:55,050 --> 00:35:57,510
S1: have exited yourself from the adult table. You are no

639
00:35:57,510 --> 00:36:01,799
S1: longer in the conversation because of Moloch. We are building this.

640
00:36:01,800 --> 00:36:05,640
S1: It is happening. We are running full speed with scissors

641
00:36:05,640 --> 00:36:08,520
S1: in both hands. And there's nothing we can do to

642
00:36:08,520 --> 00:36:11,550
S1: stop that. My argument is, the only thing we could

643
00:36:11,550 --> 00:36:14,280
S1: do is try to aim it in a good direction.

644
00:36:14,280 --> 00:36:17,400
S1: Try to point towards human 3.0 and that's what I'm

645
00:36:17,400 --> 00:36:22,319
S1: trying to do. All right, huge diversion, but worth talking about.

646
00:36:22,320 --> 00:36:25,530
S1: That was the AI piece collection of thoughts and predictions

647
00:36:25,530 --> 00:36:29,669
S1: about AI. June of 2024. All right. Security got a

648
00:36:29,670 --> 00:36:35,070
S1: whole bunch of misinformation sneaking into like, little man stuffed news.

649
00:36:35,070 --> 00:36:38,460
S1: So this is like manosphere, like podcasts or whatever, or

650
00:36:38,460 --> 00:36:44,790
S1: news sites or whatever. They're actually republishing Russian propaganda network content.

651
00:36:44,790 --> 00:36:47,490
S1: So it's like, yeah, it can seep in and go

652
00:36:47,489 --> 00:36:49,710
S1: lots of different places. This is why you actually need

653
00:36:49,739 --> 00:36:53,580
S1: AI to be watching everything. And basically when a new

654
00:36:53,580 --> 00:36:59,069
S1: information or disinformation or misinformation propaganda campaign comes out, when

655
00:36:59,070 --> 00:37:02,940
S1: it comes out, we should categorize, okay, what the core

656
00:37:02,940 --> 00:37:06,630
S1: concept is, the core idea. It's trying to spread with AI,

657
00:37:06,630 --> 00:37:09,900
S1: and then you monitor all the different stories coming out.

658
00:37:09,900 --> 00:37:14,580
S1: It might be like Saskatchewan news from South Parish, you know,

659
00:37:14,580 --> 00:37:19,080
S1: Wisconsin or whatever. And they're spreading the same story, because

660
00:37:19,080 --> 00:37:21,480
S1: what the attackers will do is they'll figure out, okay,

661
00:37:21,480 --> 00:37:24,000
S1: where can we get it in? Where can we push

662
00:37:24,000 --> 00:37:26,550
S1: this narrative? And maybe, maybe the front door is closed,

663
00:37:26,550 --> 00:37:29,730
S1: but maybe there's a thousand different backdoors. That was that one.

664
00:37:29,730 --> 00:37:32,610
S1: TikTok had an issue with DMs. There was a zero

665
00:37:32,640 --> 00:37:34,980
S1: day in DMs where you click it and open it.

666
00:37:34,980 --> 00:37:39,090
S1: It's pretty bad. The current tactic for what? You basically

667
00:37:39,090 --> 00:37:43,170
S1: get compromised for compromise of your TikTok. If you did that,

668
00:37:43,170 --> 00:37:47,100
S1: and a new tactic TikTok is using to avoid the

669
00:37:47,100 --> 00:37:50,219
S1: lawsuit is basically to say they're going to build out

670
00:37:50,219 --> 00:37:53,310
S1: a separate algorithm and separate data just for us. So

671
00:37:53,310 --> 00:37:55,830
S1: we'll see if that works. New paper claims that GPT

672
00:37:55,830 --> 00:38:00,480
S1: four can autonomously hack zero day vulnerabilities. 53% success rate

673
00:38:00,510 --> 00:38:01,920
S1: going to be a lot of these papers. A lot

674
00:38:01,920 --> 00:38:04,380
S1: of these papers are not very good. A lot of

675
00:38:04,380 --> 00:38:07,440
S1: them are decent. I will keep reading and posting these

676
00:38:07,440 --> 00:38:10,770
S1: when the claims are interesting and they're from decent outfits,

677
00:38:10,770 --> 00:38:13,680
S1: but you want to watch the quality of the paper

678
00:38:13,680 --> 00:38:17,580
S1: very closely. Look, especially for how much of their testing

679
00:38:17,580 --> 00:38:20,250
S1: methodology that they publish. Did they show exactly what they

680
00:38:20,250 --> 00:38:22,380
S1: tried against these different things? Did they show all the

681
00:38:22,380 --> 00:38:25,890
S1: different examples? Did they, you know, give you all the

682
00:38:25,890 --> 00:38:29,220
S1: stuff to do the replication? If not, then they're kind

683
00:38:29,219 --> 00:38:31,410
S1: of like hiding the ball and you can't really trust

684
00:38:31,410 --> 00:38:35,550
S1: their score. And we got to analyze paper fabric pattern

685
00:38:35,550 --> 00:38:37,200
S1: that you can use for this. And when I ran

686
00:38:37,200 --> 00:38:39,840
S1: it against this one it got a six on rigor

687
00:38:39,840 --> 00:38:44,969
S1: benchmark and Ablations look good but some details missing and

688
00:38:44,969 --> 00:38:48,750
S1: stats lacking. That's why that's why we have fabric. Okay

689
00:38:48,750 --> 00:38:53,339
S1: Chinese drone photographer got snagged by the US Espionage Act

690
00:38:53,340 --> 00:38:58,410
S1: taking pics of military shipyards. So there's another thing that's

691
00:38:58,410 --> 00:39:00,690
S1: being talked about right now with China buying up a

692
00:39:00,690 --> 00:39:04,410
S1: whole bunch of land right next to military bases, they're

693
00:39:04,410 --> 00:39:07,680
S1: buying like massive amounts of farmland, and they start setting

694
00:39:07,680 --> 00:39:09,629
S1: up all this gear. It's like, oh, don't pay any

695
00:39:09,630 --> 00:39:12,600
S1: attention to us. We're just building some stuff over here.

696
00:39:12,600 --> 00:39:16,350
S1: So I hope Biden or Trump or whoever just like,

697
00:39:16,350 --> 00:39:20,009
S1: finds a list of these. And just like search and seizure,

698
00:39:20,040 --> 00:39:23,279
S1: good bye. Do not pass go. And maybe some of

699
00:39:23,280 --> 00:39:27,270
S1: those are benign, right. But I have to basically be like, well,

700
00:39:27,270 --> 00:39:33,210
S1: based on the, you know, 9566, you know, fucking billion

701
00:39:33,210 --> 00:39:37,210
S1: times that you've actually tried to steal from us. And

702
00:39:37,210 --> 00:39:40,779
S1: actually hacked us, then you know you're going to lose

703
00:39:40,780 --> 00:39:43,180
S1: some for the home team when you're actually doing something

704
00:39:43,180 --> 00:39:46,239
S1: benign and actually just setting up a farm next to

705
00:39:46,239 --> 00:39:49,239
S1: a military base on accident. And they didn't actually mean

706
00:39:49,239 --> 00:39:52,480
S1: anything by that. Too bad you ruined your reputation. Therefore

707
00:39:52,480 --> 00:39:55,480
S1: get out. Okay. OpenAI revealed that its AI tech was

708
00:39:55,480 --> 00:39:59,290
S1: used by Russia and China for sneaky influence ops. Yeah,

709
00:39:59,290 --> 00:40:02,109
S1: they've said this a few times and I like it.

710
00:40:02,110 --> 00:40:05,890
S1: I like the transparency. They're basically like, look, this keeps happening.

711
00:40:05,890 --> 00:40:07,960
S1: We keep catching it and we're going to try to

712
00:40:07,960 --> 00:40:13,210
S1: do better. All right. WWDC keynote got an analysis here.

713
00:40:13,210 --> 00:40:16,660
S1: This is a fabric analysis from a TechCrunch article. As

714
00:40:16,660 --> 00:40:19,450
S1: I talked about before, Apple basically killed the AI thing.

715
00:40:19,450 --> 00:40:21,460
S1: They also did a bunch of really cool stuff. One

716
00:40:21,460 --> 00:40:23,170
S1: of the things is you bring your phone next to

717
00:40:23,170 --> 00:40:26,140
S1: someone else, you can transfer them cash. I've been waiting

718
00:40:26,140 --> 00:40:29,020
S1: that for that forever. Problem is, a lot of people

719
00:40:29,020 --> 00:40:32,470
S1: won't have that feature turned on, or they'll have an

720
00:40:32,500 --> 00:40:35,110
S1: Android and you won't be able to do it like so.

721
00:40:35,110 --> 00:40:38,080
S1: A lot of people are switching to like Zelle or Venmo.

722
00:40:38,110 --> 00:40:42,400
S1: I don't like Venmo, and Zelle is slow. Whatever. I

723
00:40:42,400 --> 00:40:45,430
S1: wish more people had the ability to do these swipey

724
00:40:45,430 --> 00:40:49,330
S1: things and Apple things, but kind of rude to require

725
00:40:49,330 --> 00:40:52,149
S1: that everyone has an iPhone, right? I mean, that's just

726
00:40:52,150 --> 00:40:54,700
S1: not realistic and also not cool. All right. All the

727
00:40:54,700 --> 00:41:00,190
S1: different startups that Apple killed today always fun. After WWDC

728
00:41:00,190 --> 00:41:05,620
S1: or an OpenAI event, US needs 225,000 more cybersecurity workers.

729
00:41:05,620 --> 00:41:09,489
S1: I never trust numbers like this because it's hard to know. Like,

730
00:41:09,489 --> 00:41:12,610
S1: what are they asking? Are they asking it right? What

731
00:41:12,610 --> 00:41:15,190
S1: are the questions that they accept? Like, I just feel

732
00:41:15,190 --> 00:41:17,470
S1: like the numbers are bad. I feel like that number

733
00:41:17,469 --> 00:41:19,720
S1: is going to go down because of AI, and I

734
00:41:19,719 --> 00:41:21,790
S1: feel like it's going to go up because of AI.

735
00:41:21,820 --> 00:41:23,860
S1: So it's going to go down because of AI, because

736
00:41:23,860 --> 00:41:26,320
S1: the jobs are going to get easier to do without people,

737
00:41:26,320 --> 00:41:29,859
S1: but so much more stuff is going to get made

738
00:41:29,860 --> 00:41:33,340
S1: that we're just going to be injecting insecurity into the world,

739
00:41:33,340 --> 00:41:36,010
S1: especially when it's like getting made with AI. So it's

740
00:41:36,010 --> 00:41:38,439
S1: like AI going in, building a bunch of stuff and

741
00:41:38,440 --> 00:41:40,750
S1: have that stuff's going to be really nasty in terms

742
00:41:40,750 --> 00:41:43,810
S1: of security vulnerabilities. So it's like still going to need

743
00:41:43,810 --> 00:41:48,489
S1: people for a while. Like assuming all this AI replacement

744
00:41:48,489 --> 00:41:51,759
S1: stuff goes really, really fast. We're still talking about years

745
00:41:51,760 --> 00:41:55,719
S1: and years. I mean, super fast for like an 80%

746
00:41:55,719 --> 00:41:59,770
S1: replacement of all jobs, super fast. And these are crazy

747
00:41:59,770 --> 00:42:03,370
S1: numbers I'm just throwing out. But but like Nuclear fast

748
00:42:03,370 --> 00:42:06,969
S1: is like ten years now as far as starting and

749
00:42:06,969 --> 00:42:10,089
S1: actually having major impact, I think it's already starting and

750
00:42:10,090 --> 00:42:12,820
S1: it's already like a year or two from now we're

751
00:42:12,820 --> 00:42:14,800
S1: going to start feeling it. So I think that's going

752
00:42:14,800 --> 00:42:18,100
S1: to be really fast. There's going to be a regulatory response.

753
00:42:18,100 --> 00:42:21,790
S1: It's slow to make any change in any organization, and

754
00:42:21,790 --> 00:42:24,550
S1: the bigger and more bureaucratic it is, the slower it

755
00:42:24,550 --> 00:42:27,940
S1: will be. So just because something could happen in a

756
00:42:27,940 --> 00:42:31,000
S1: year or six months doesn't mean it will. And it's

757
00:42:31,000 --> 00:42:33,730
S1: more likely to happen in 2 or 3 or 4

758
00:42:33,730 --> 00:42:37,030
S1: or 10 years. Got to keep that in mind. It's

759
00:42:37,030 --> 00:42:39,400
S1: going to go very fast. It's going to go faster

760
00:42:39,400 --> 00:42:41,830
S1: than you think and slower than you think both at

761
00:42:41,830 --> 00:42:45,160
S1: the same time. Apple just hit the three. $3 trillion

762
00:42:45,160 --> 00:42:48,399
S1: mark again. But Nvidia jumped ahead. And now Nvidia is

763
00:42:48,400 --> 00:42:52,750
S1: even more valuable than Apple. And Microsoft is still number

764
00:42:52,750 --> 00:42:57,520
S1: one insane. Such as killing it over there. Cartwheel tool

765
00:42:57,550 --> 00:43:02,320
S1: olama CLE wire JS for us for UI design shade

766
00:43:02,320 --> 00:43:06,040
S1: map project that turns every mountain building and tree shadow

767
00:43:06,040 --> 00:43:08,800
S1: for any date and time. How to think like a

768
00:43:08,800 --> 00:43:13,510
S1: computer scientist. Pretty cool. Uh, interactive edition here. Starship just

769
00:43:13,510 --> 00:43:16,810
S1: landed its first one. Like all the other ones either

770
00:43:16,810 --> 00:43:19,300
S1: blew up on the way out or, uh, blew up

771
00:43:19,300 --> 00:43:23,080
S1: on the way in, but he finally landed one very impressive,

772
00:43:23,080 --> 00:43:29,830
S1: I think 5000 tons. So that's what, £20,000? Five times two, £2,000.

773
00:43:29,830 --> 00:43:33,730
S1: 2000 times 5000. Holy crap, that's a lot of pounds.

774
00:43:33,730 --> 00:43:38,290
S1: World's largest solar farm just went online. Three point gigawatt

775
00:43:38,290 --> 00:43:42,219
S1: Marty just went live in China. Okay, here's my question.

776
00:43:42,219 --> 00:43:45,969
S1: Why can't we build like ten of these in the US? Okay,

777
00:43:45,969 --> 00:43:48,880
S1: we know we need all this power for AI. So

778
00:43:48,880 --> 00:43:52,510
S1: I say we go Biden. Trump does this. You basically

779
00:43:52,510 --> 00:43:54,430
S1: go to each state and be like, look, I need

780
00:43:54,430 --> 00:43:59,860
S1: you to build 3.5GW in your state within five years.

781
00:43:59,860 --> 00:44:02,080
S1: And they're like, yeah, that's impossible. We'd have to hire

782
00:44:02,080 --> 00:44:04,930
S1: all these people. And the government's like, whatever. I mean,

783
00:44:04,930 --> 00:44:07,839
S1: we have money printers, so we're just going to print

784
00:44:07,840 --> 00:44:10,900
S1: all this money, we're going to inject it into the economy.

785
00:44:10,900 --> 00:44:15,100
S1: Most importantly, giving it to these training programs will, which

786
00:44:15,100 --> 00:44:17,140
S1: will train the people how to build the wind farms

787
00:44:17,140 --> 00:44:20,739
S1: and the solar farms, whatever the best renewable energy is

788
00:44:20,739 --> 00:44:24,520
S1: for that particular state, like Nevada should build like 100

789
00:44:24,520 --> 00:44:27,430
S1: of these things for for sun. I mean, I fly

790
00:44:27,430 --> 00:44:30,129
S1: over it all the time. There's nothing there. In fact,

791
00:44:30,130 --> 00:44:33,070
S1: you fly over the entire United States. There's nothing there.

792
00:44:33,070 --> 00:44:36,010
S1: It's all empty. And meanwhile the sun just hits it

793
00:44:36,010 --> 00:44:39,920
S1: every day. Like nine times out of ten. For every

794
00:44:39,920 --> 00:44:42,319
S1: day of the year the sun comes up. In fact,

795
00:44:42,320 --> 00:44:44,480
S1: I think it might be higher than 90% when the

796
00:44:44,480 --> 00:44:47,120
S1: sun comes up. In fact, it might be 100%. The

797
00:44:47,120 --> 00:44:49,910
S1: sun always comes up and it is hitting us with

798
00:44:49,910 --> 00:44:53,630
S1: like so much energy. And we just it soaks in,

799
00:44:53,630 --> 00:44:57,890
S1: you know, goes off into the atmosphere. Capture it. Okay, look,

800
00:44:57,890 --> 00:45:00,140
S1: we've got people who don't have jobs and don't have

801
00:45:00,140 --> 00:45:02,510
S1: meaning in their lives. We know that AI is going

802
00:45:02,510 --> 00:45:05,359
S1: to require all this power. So millions of people have

803
00:45:05,360 --> 00:45:07,310
S1: no meaning in their lives because they don't have work.

804
00:45:07,310 --> 00:45:10,790
S1: We talk about the loss of manufacturing jobs. Go build

805
00:45:10,790 --> 00:45:13,940
S1: all these solar panels, go build all these solar plants,

806
00:45:13,940 --> 00:45:16,700
S1: have the people working there. Eventually they'll be replaced by

807
00:45:16,700 --> 00:45:19,460
S1: robots in AI. That's fine. That'll take time like we

808
00:45:19,460 --> 00:45:23,210
S1: talked about before. In the meantime, let's go build 100

809
00:45:23,210 --> 00:45:29,270
S1: of these, whatever, 3.5GW facilities or ten gigawatt facilities, whatever.

810
00:45:29,270 --> 00:45:34,040
S1: Giant solar farms, giant wind farms. And we put millions

811
00:45:34,040 --> 00:45:37,069
S1: of people to work, which gives them meaning, and they're

812
00:45:37,070 --> 00:45:39,859
S1: building towards a renewable future. So, like, even if I

813
00:45:39,860 --> 00:45:42,740
S1: didn't happen whatever, we got renewable energy. Now we can

814
00:45:42,739 --> 00:45:44,960
S1: export that energy, we can sell it other places. We

815
00:45:44,960 --> 00:45:47,810
S1: just have it for whatever purpose. And you have to

816
00:45:47,810 --> 00:45:49,670
S1: build the pipelines as well, right? You got to get

817
00:45:49,670 --> 00:45:51,890
S1: that energy to where it needs to go. So that's

818
00:45:51,890 --> 00:45:56,360
S1: like pipes, cables, like, you know, wires, all that stuff,

819
00:45:56,360 --> 00:46:00,020
S1: all the infrastructure for moving energy. Where's the downside here?

820
00:46:00,020 --> 00:46:04,190
S1: It's win win jobs and and tons of energy and

821
00:46:04,190 --> 00:46:07,460
S1: no reliance on external energy. And we get to move

822
00:46:07,460 --> 00:46:10,969
S1: on AI faster. So really bothers me when I see

823
00:46:10,969 --> 00:46:15,080
S1: China understanding this and building so quickly. And we're like, man,

824
00:46:15,110 --> 00:46:18,350
S1: let's fight with each other instead. Oh, in nuclear, nuclear

825
00:46:18,350 --> 00:46:21,590
S1: is the other one. Okay, so USA's although I prefer solar,

826
00:46:21,590 --> 00:46:24,770
S1: but someone knows better than me. USA's solar panel manufacturing

827
00:46:24,770 --> 00:46:28,610
S1: capacity jumped by 71%. I like it, I like it.

828
00:46:28,610 --> 00:46:31,820
S1: Florida and Texas are leading. Let's go. What about Nevada?

829
00:46:31,820 --> 00:46:34,580
S1: Where are you at? Florida. Forget about it. It rains

830
00:46:34,580 --> 00:46:36,320
S1: all the time and it's sinking. It's not going to

831
00:46:36,320 --> 00:46:39,530
S1: work out. I guess Florida's energy generation would have to

832
00:46:39,530 --> 00:46:42,680
S1: be based on, like being flooded or winds. No, you

833
00:46:42,680 --> 00:46:45,469
S1: can't do can't do wind farms and hurricanes. That would

834
00:46:45,469 --> 00:46:48,500
S1: be a bit rough. All right. Humans. David Brooks argues

835
00:46:48,500 --> 00:46:51,110
S1: that progressive energy has shifted from the working class to

836
00:46:51,110 --> 00:46:55,340
S1: elite universities, making them more left leaning than ever. Points

837
00:46:55,340 --> 00:46:59,239
S1: to protests and progressive opinions. Yeah, it's all bad. Highly

838
00:46:59,239 --> 00:47:02,779
S1: recommend you read David Brooks stuff The Road to Character.

839
00:47:02,780 --> 00:47:08,000
S1: My favorite one is the Second Mountain. Absolutely unbelievable book. Unbelievable.

840
00:47:08,000 --> 00:47:10,129
S1: And I love his stuff. The New York Times, he's

841
00:47:10,130 --> 00:47:14,540
S1: like a centrist kind of conservative, but like, yeah, logical.

842
00:47:14,540 --> 00:47:16,700
S1: I like him. Okay. This one turned out to be

843
00:47:16,700 --> 00:47:19,790
S1: not great. A couple people. I looked at it again,

844
00:47:19,790 --> 00:47:22,700
S1: people don't like me linking to Substack. I'm going to

845
00:47:22,700 --> 00:47:25,759
S1: link to Substack. The question isn't is it Substack or

846
00:47:25,760 --> 00:47:29,360
S1: is it whatever, New York Times, because all these different

847
00:47:29,360 --> 00:47:32,210
S1: sources can be bad. What I care about is like

848
00:47:32,210 --> 00:47:36,050
S1: factually being wrong or like making claims and not backing

849
00:47:36,050 --> 00:47:38,810
S1: them up in some sort of in any sort of way,

850
00:47:38,810 --> 00:47:42,980
S1: or just being ideologue and just like being hateful or

851
00:47:42,980 --> 00:47:46,100
S1: rambling or whatever, like, well, I'm rambling, so not quite

852
00:47:46,100 --> 00:47:49,130
S1: the same. But the point is, the best people in

853
00:47:49,130 --> 00:47:53,060
S1: the world at thinking and being rigorous and having good

854
00:47:53,060 --> 00:47:57,950
S1: ideas are going to Substack, okay, and beehive and all

855
00:47:57,950 --> 00:48:02,150
S1: these other platforms and YouTube and TikTok. Even so, it's

856
00:48:02,150 --> 00:48:04,640
S1: not the source that I'm linking to. That's the problem.

857
00:48:04,640 --> 00:48:07,820
S1: The quality of the content is the problem because before

858
00:48:07,820 --> 00:48:11,300
S1: too long, CNBC and like Fox, first of all, Fox News,

859
00:48:11,300 --> 00:48:13,520
S1: I don't link to that usually. I'm sure they have

860
00:48:13,520 --> 00:48:16,610
S1: good news sometimes, but you get the point. It's not

861
00:48:16,610 --> 00:48:20,719
S1: about the type of source, okay? Blogs are just as

862
00:48:20,719 --> 00:48:25,549
S1: authoritative as YouTube as a lot of the media networks

863
00:48:25,550 --> 00:48:28,040
S1: at this point. It's more about the individual and the

864
00:48:28,040 --> 00:48:31,910
S1: actual article. I am making these links available so that

865
00:48:31,910 --> 00:48:34,910
S1: someone can my link numbers or my click numbers are

866
00:48:34,910 --> 00:48:38,750
S1: probably going to go down because they will be like, well,

867
00:48:38,750 --> 00:48:40,850
S1: I don't want to read from this guy, or I

868
00:48:40,850 --> 00:48:43,250
S1: don't want to read from Hacker News or Financial Times.

869
00:48:43,250 --> 00:48:45,950
S1: I hate that place or whatever. Hopefully you don't need

870
00:48:45,950 --> 00:48:48,620
S1: it anyway. Hopefully this one sentence is good enough. That's

871
00:48:48,620 --> 00:48:50,540
S1: the whole point of doing the summary. All right. New

872
00:48:50,540 --> 00:48:54,080
S1: York Times deep dive into devastation of Ukraine. Oh, this

873
00:48:54,080 --> 00:48:58,009
S1: story was absolutely nuts. And yeah, New York Times good stuff.

874
00:48:58,010 --> 00:49:00,740
S1: They're staying around at a company more than two years

875
00:49:00,739 --> 00:49:05,300
S1: could massively hurt your earning potential. US job openings hit

876
00:49:05,330 --> 00:49:10,070
S1: a three year low. Layoffs are surprisingly low, it says CNN.

877
00:49:10,070 --> 00:49:14,540
S1: See above for commentary on sources. Viagra improves blood flow

878
00:49:14,540 --> 00:49:19,520
S1: and could help to prevent dementia. University of Oxford 49%

879
00:49:19,520 --> 00:49:23,180
S1: of independents think Trump should drop out because of his

880
00:49:23,180 --> 00:49:26,480
S1: guilty verdict. I'm making a prediction. I think this effect

881
00:49:26,480 --> 00:49:29,180
S1: wears off and Trump will be pulling even with Biden

882
00:49:29,180 --> 00:49:31,040
S1: within a month. You know what I'm going to do?

883
00:49:31,040 --> 00:49:36,910
S1: Let's go look at 538 538 polls. Wow. Look at

884
00:49:36,910 --> 00:49:41,529
S1: the impact. Look at the impact of the guilty verdict. Yeah,

885
00:49:41,530 --> 00:49:45,640
S1: it happened right in here somewhere. Oh, look. No impact.

886
00:49:45,640 --> 00:49:48,549
S1: And maybe I'm reading this wrong. Maybe this doesn't account

887
00:49:48,550 --> 00:49:52,460
S1: for independence. I mean, this is this is a June

888
00:49:52,460 --> 00:49:57,489
S1: 12th 10th through 12th result. Echelon insights. I don't know

889
00:49:57,489 --> 00:50:01,839
S1: how big this thing is. Okay, YouGov I know that's respectable.

890
00:50:01,840 --> 00:50:06,670
S1: Biden 40, Trump 42. Okay. We got two different echelon

891
00:50:06,670 --> 00:50:10,930
S1: ones daily cause what do we got? 4545 tide. So

892
00:50:10,930 --> 00:50:15,250
S1: so I read this article and it's like, yeah, 49%

893
00:50:15,250 --> 00:50:18,310
S1: of independents think Trump should drop out after his guilty verdict.

894
00:50:18,310 --> 00:50:22,480
S1: Are these polls okay? That's one argument. Independents don't respond

895
00:50:22,480 --> 00:50:25,569
S1: to polls. And supposedly these independents are going to switch

896
00:50:25,570 --> 00:50:28,060
S1: the thing and they're going to be mad at Trump,

897
00:50:28,060 --> 00:50:30,730
S1: and it's going to go for Biden. I don't buy it.

898
00:50:30,940 --> 00:50:33,700
S1: If anything, in my opinion, this is just my opinion.

899
00:50:33,700 --> 00:50:37,600
S1: This is politics. Like, who knows anything, right? In my opinion,

900
00:50:37,600 --> 00:50:41,500
S1: these numbers are soft for Trump. I think Trump is

901
00:50:41,500 --> 00:50:44,380
S1: probably winning by another 5 or 10 points right now.

902
00:50:44,380 --> 00:50:48,730
S1: That's my guess. And I don't think him being convicted

903
00:50:48,730 --> 00:50:53,650
S1: is going to do anything for his numbers if if not,

904
00:50:53,650 --> 00:50:56,920
S1: make it better for him. Just ridiculous. This is why

905
00:50:56,920 --> 00:50:59,170
S1: I think about I. I literally do not look at

906
00:50:59,170 --> 00:51:03,520
S1: news anymore other other than like I brings me political

907
00:51:03,520 --> 00:51:06,009
S1: news and analyses for me and then I have to

908
00:51:06,010 --> 00:51:08,200
S1: go like read the story, I have to do my

909
00:51:08,200 --> 00:51:10,990
S1: analysis or whatever. But like I do not pay attention

910
00:51:10,989 --> 00:51:13,629
S1: to this stuff. I'm trying to build human 3.0 I

911
00:51:13,630 --> 00:51:17,080
S1: don't like. I'm just so disillusioned with everything. I'm just

912
00:51:17,080 --> 00:51:20,380
S1: like checked out. All right, X is doing adult content.

913
00:51:20,380 --> 00:51:24,009
S1: Good for them. Ernest Hemingway wrote a deeply moving letter

914
00:51:24,010 --> 00:51:27,460
S1: to friends who lost their son, saying, those who truly

915
00:51:27,460 --> 00:51:34,750
S1: live never really die. I love Maria Popova. The marginalien. Yeah,

916
00:51:34,750 --> 00:51:36,640
S1: I liked the old name that she had for that

917
00:51:36,640 --> 00:51:40,390
S1: site before. I can't remember what it was. The yellow site. Anyway,

918
00:51:40,390 --> 00:51:42,880
S1: one of the coolest life wisdom reads I've come across

919
00:51:42,880 --> 00:51:44,680
S1: in a while. This is a really cool one. I'll

920
00:51:44,680 --> 00:51:47,319
S1: put this one up. You can never expect honesty from

921
00:51:47,320 --> 00:51:50,230
S1: people who even lie to themselves. You'll be ten times

922
00:51:50,230 --> 00:51:52,600
S1: happier if you forgive your parents and stop blaming them

923
00:51:52,600 --> 00:51:56,230
S1: for your misery for yourself. From society's advice, most of

924
00:51:56,230 --> 00:51:58,510
S1: them have no idea of what they're doing. If you

925
00:51:58,510 --> 00:52:01,660
S1: continue to wait for the right time, you'll waste your

926
00:52:01,660 --> 00:52:04,540
S1: entire life and nothing will ever happen. It's a massive

927
00:52:04,540 --> 00:52:07,870
S1: list of these. They're quite, quite good. You'll lose 99%

928
00:52:07,870 --> 00:52:10,779
S1: of your close friends when you start to improve your life.

929
00:52:10,780 --> 00:52:14,020
S1: That's bleak. I wouldn't go with 99%, but I get

930
00:52:14,020 --> 00:52:17,920
S1: the point. You don't need 100 self-help books. All you

931
00:52:17,920 --> 00:52:21,280
S1: need is action and self-discipline. It's a great list. It's

932
00:52:21,280 --> 00:52:23,710
S1: a great list. You should check it out. Ideas and

933
00:52:23,710 --> 00:52:27,940
S1: analysis I think I figured out the simplest possible answer

934
00:52:27,940 --> 00:52:31,660
S1: for why Trump is still doing so well. Oh my god,

935
00:52:31,660 --> 00:52:35,230
S1: I didn't include the link. Oh man, I didn't include

936
00:52:35,230 --> 00:52:38,169
S1: the link in here. It's like, oh really? You did great.

937
00:52:38,170 --> 00:52:40,870
S1: Thanks for not sharing it. And recommendation for the week

938
00:52:40,870 --> 00:52:44,650
S1: is Tyler Cowen and Aakash Patel. As I talked about

939
00:52:44,650 --> 00:52:47,109
S1: in the aphorism of the week, live as if you

940
00:52:47,110 --> 00:52:49,810
S1: were to die tomorrow. Learn as if you were to

941
00:52:49,810 --> 00:52:52,960
S1: live forever. Live as if you were to die tomorrow.

942
00:52:52,960 --> 00:52:56,589
S1: Learn as if you were to live forever. Mahatma Gandhi.

943
00:52:57,719 --> 00:53:00,839
S1: Unsupervised Learning is produced and edited by Daniel Meisler on

944
00:53:00,840 --> 00:53:05,460
S1: a Neumann U87 AI microphone using Hindenburg. Intro and outro

945
00:53:05,460 --> 00:53:08,790
S1: music is by zombie with the why and to get

946
00:53:08,790 --> 00:53:10,859
S1: the text and links from this episode, sign up for

947
00:53:10,860 --> 00:53:16,509
S1: the newsletter version of the show at Daniel meisler.com/newsletter. We'll

948
00:53:16,510 --> 00:53:17,350
S1: see you next time.