1
00:00:00,349 --> 00:00:01,010
this book says

2
00:00:01,349 --> 00:00:02,370
you can call the direction right

3
00:00:02,669 --> 00:00:03,790
and still lose

4
00:00:04,189 --> 00:00:05,910
a 25 year old prodigy

5
00:00:05,948 --> 00:00:07,910
who entered Columbia University at 15

6
00:00:08,028 --> 00:00:09,750
graduated top of his class at 19

7
00:00:09,829 --> 00:00:11,910
spent time on the Superalignment team

8
00:00:11,948 --> 00:00:14,147
and wrote a 165 page report

9
00:00:14,148 --> 00:00:18,428
laying out the compute memory and power bottlenecks AI is about to hit

10
00:00:18,428 --> 00:00:19,870
in crystal clear terms

11
00:00:20,068 --> 00:00:22,829
that report was translated and shared all over the world

12
00:00:22,908 --> 00:00:26,789
and it nearly defined the public narrative of AI trading for the next two years

13
00:00:27,277 --> 00:00:29,078
then he started a fund

14
00:00:29,317 --> 00:00:30,315
in two years

15
00:00:30,317 --> 00:00:33,198
it grew from just over 200 million dollars to 45 billion dollars

16
00:00:33,357 --> 00:00:35,957
with a net return of 439 percent in the first half of the year

17
00:00:36,277 --> 00:00:36,777
then

18
00:00:37,076 --> 00:00:37,537
in one month

19
00:00:37,877 --> 00:00:39,357
it lost 67 percent

20
00:00:39,637 --> 00:00:43,238
and its entire public equity book was sold off to a hedge fund

21
00:00:44,018 --> 00:00:46,178
his name is Aschenbrenner

22
00:00:46,617 --> 00:00:48,738
his read on AI has not failed

23
00:00:49,338 --> 00:00:50,358
compute demand is still there

24
00:00:50,857 --> 00:00:52,619
memory profits are still growing

25
00:00:52,778 --> 00:00:54,659
power constraints have not gone away

26
00:00:55,098 --> 00:00:56,510
and yet he still lost

27
00:00:56,710 --> 00:00:57,010
hello

28
00:00:57,190 --> 00:00:57,650
everyone

29
00:00:57,830 --> 00:00:58,830
I am Leo Wang

30
00:00:59,229 --> 00:01:01,229
recently a friend sent me a book

31
00:01:01,229 --> 00:01:02,290
called The Cognitive Machine

32
00:01:02,909 --> 00:01:06,991
the author is an observer who has written from Palo Alto in Silicon Valley for years

33
00:01:07,390 --> 00:01:08,771
the book runs nearly 500 pages

34
00:01:09,069 --> 00:01:10,309
it does not explain how the tech works

35
00:01:10,310 --> 00:01:12,150
it does not teach you which AI tools to use

36
00:01:12,269 --> 00:01:14,191
from start to finish it does just one thing

37
00:01:14,349 --> 00:01:17,191
it teaches you how to make judgments in the age of AI

38
00:01:17,984 --> 00:01:18,805
after I finished it

39
00:01:19,185 --> 00:01:21,146
one feeling hit me hard

40
00:01:21,424 --> 00:01:23,024
the harshest thing about this book

41
00:01:23,025 --> 00:01:25,025
is not telling you how powerful AI is

42
00:01:25,224 --> 00:01:26,263
it is telling you

43
00:01:26,265 --> 00:01:27,566
that even if you understand all of it

44
00:01:27,745 --> 00:01:29,066
you can still lose

45
00:01:29,405 --> 00:01:29,885
why

46
00:01:30,344 --> 00:01:32,066
because you have mixed up two things

47
00:01:32,465 --> 00:01:33,525
one is called cognition

48
00:01:34,025 --> 00:01:35,346
the other is called position

49
00:01:35,625 --> 00:01:36,605
cognition answers

50
00:01:36,864 --> 00:01:38,185
what you see

51
00:01:38,424 --> 00:01:39,366
position answers

52
00:01:39,625 --> 00:01:40,986
where you stand

53
00:01:41,224 --> 00:01:43,225
these two questions sound alike

54
00:01:43,344 --> 00:01:45,246
but they are completely different

55
00:01:45,525 --> 00:01:48,606
Aschenbrenner is the most extreme example

56
00:01:48,725 --> 00:01:52,083
he wrote his case on AI infrastructure too well

57
00:01:52,085 --> 00:01:54,166
so well that the whole market could understand it

58
00:01:54,484 --> 00:01:57,365
after that 165 page report came out

59
00:01:57,525 --> 00:01:59,644
compute memory power

60
00:01:59,645 --> 00:02:01,845
were no longer just his own judgment

61
00:02:01,925 --> 00:02:03,585
they became investment memos

62
00:02:03,844 --> 00:02:06,645
the common language of social media and dinner tables

63
00:02:06,964 --> 00:02:10,206
when we talked about carbon based business we covered reflexivity

64
00:02:10,404 --> 00:02:12,644
once everyone knows a pattern

65
00:02:12,645 --> 00:02:14,806
the pattern itself stops working

66
00:02:15,771 --> 00:02:17,252
a street is getting a subway line

67
00:02:17,292 --> 00:02:19,653
of course that raises the value of nearby homes

68
00:02:19,931 --> 00:02:22,292
but when the zoning notice has been posted for six months

69
00:02:22,371 --> 00:02:24,333
and prices are already up 40 percent

70
00:02:24,371 --> 00:02:26,372
if you walk in with that news now

71
00:02:26,452 --> 00:02:28,050
what you buy is not a secret

72
00:02:28,052 --> 00:02:31,195
it is the price after everyone believed the secret

73
00:02:31,274 --> 00:02:34,635
one kind of return comes from seeing something earlier than the market

74
00:02:34,714 --> 00:02:38,554
the other comes from more and more people believing the same thing

75
00:02:38,633 --> 00:02:40,434
as money pushes prices up

76
00:02:40,913 --> 00:02:41,575
on the way up

77
00:02:41,953 --> 00:02:43,795
the two kinds of money are nearly impossible to tell apart

78
00:02:44,233 --> 00:02:45,553
only when the tide goes out

79
00:02:45,553 --> 00:02:47,795
do you learn which kind you actually earned

80
00:02:48,500 --> 00:02:50,022
there is a line in this book

81
00:02:50,101 --> 00:02:51,702
that I read over and over

82
00:02:52,141 --> 00:02:52,482
it says

83
00:02:53,020 --> 00:02:55,901
position shapes destiny more than value does

84
00:02:56,340 --> 00:02:56,841
honestly

85
00:02:57,141 --> 00:02:59,061
the first time I read that line

86
00:02:59,101 --> 00:03:00,341
it felt a bit absolute

87
00:03:00,740 --> 00:03:02,642
does personal value not matter at all

88
00:03:03,141 --> 00:03:06,022
your ability your insight your experience

89
00:03:06,060 --> 00:03:06,922
do none of them count

90
00:03:07,661 --> 00:03:09,047
later I figured it out

91
00:03:09,367 --> 00:03:10,988
the author is not saying value does not matter

92
00:03:11,286 --> 00:03:12,166
he is saying

93
00:03:12,166 --> 00:03:14,528
value answers what you can do

94
00:03:14,686 --> 00:03:16,148
position asks a different set of questions

95
00:03:16,807 --> 00:03:18,067
who has to go through you

96
00:03:18,527 --> 00:03:19,787
who can replace you

97
00:03:20,166 --> 00:03:21,428
whose permission you depend on

98
00:03:21,846 --> 00:03:22,347
when something goes wrong

99
00:03:22,807 --> 00:03:24,628
whose name is on the last line

100
00:03:25,286 --> 00:03:27,968
value can be seen in a single demo

101
00:03:28,047 --> 00:03:30,468
position often has to wait until a platform pulls its API

102
00:03:30,807 --> 00:03:32,028
clients sign new contracts

103
00:03:32,327 --> 00:03:34,326
or capital starts demanding cash

104
00:03:34,327 --> 00:03:35,368
before it shows

105
00:03:36,045 --> 00:03:38,046
the book uses a vivid comparison

106
00:03:38,325 --> 00:03:40,366
it says as models move forward

107
00:03:40,646 --> 00:03:43,287
many startups look like they have a moat

108
00:03:43,485 --> 00:03:45,086
but they are really standing on sand

109
00:03:45,445 --> 00:03:47,366
before the model gets that far

110
00:03:47,406 --> 00:03:49,007
you look very valuable

111
00:03:49,205 --> 00:03:50,365
once the model arrives

112
00:03:50,366 --> 00:03:52,485
you realize the ground you stand on

113
00:03:52,485 --> 00:03:54,326
is being swallowed by the sea

114
00:03:54,786 --> 00:03:57,188
he calls this the middle layer death list

115
00:03:57,626 --> 00:03:58,808
which companies will die

116
00:03:59,307 --> 00:04:00,868
the ones stuck between two systems

117
00:04:00,946 --> 00:04:04,588
that only translate relay assemble and package

118
00:04:04,946 --> 00:04:06,088
which companies will live

119
00:04:06,626 --> 00:04:07,927
those holding compliance licenses

120
00:04:08,106 --> 00:04:09,505
those holding physical endpoints

121
00:04:09,506 --> 00:04:10,727
those holding brand trust

122
00:04:10,867 --> 00:04:12,592
those holding the right to evaluate

123
00:04:13,111 --> 00:04:15,552
the book tells a story that really stings

124
00:04:15,750 --> 00:04:16,971
an indie developer

125
00:04:17,231 --> 00:04:18,951
through a packaging accident

126
00:04:18,991 --> 00:04:22,992
got the full source code of the orchestration system at a major AI model company

127
00:04:23,151 --> 00:04:24,831
more than 1900 files

128
00:04:25,191 --> 00:04:27,951
he spent three days reading it cover to cover

129
00:04:28,071 --> 00:04:31,232
then he set his own code repo to private

130
00:04:31,350 --> 00:04:32,711
and posted a message saying

131
00:04:32,951 --> 00:04:33,612
after reading the source

132
00:04:33,871 --> 00:04:35,951
I kind of want to quit AI

133
00:04:36,211 --> 00:04:36,691
why

134
00:04:37,110 --> 00:04:40,191
because he found that the product layer he thought he was building

135
00:04:40,230 --> 00:04:41,552
did not exist at all

136
00:04:41,790 --> 00:04:43,711
he thought he was building the application layer

137
00:04:43,790 --> 00:04:47,992
but their orchestration engine alone was 46 thousand lines of code

138
00:04:48,631 --> 00:04:52,350
a bare API call uses maybe 20 percent of what a model can do

139
00:04:52,350 --> 00:04:54,711
that system pushed it close to 80 percent

140
00:04:54,951 --> 00:04:56,591
he thought he was building a tower

141
00:04:56,711 --> 00:04:58,562
he had not even touched the foundation

142
00:04:58,762 --> 00:05:01,320
this framework does not just apply to companies

143
00:05:01,322 --> 00:05:02,403
it applies to people too

144
00:05:02,882 --> 00:05:04,521
there is a line in the book that sounds cold

145
00:05:04,521 --> 00:05:06,082
but on reflection is very accurate

146
00:05:06,281 --> 00:05:07,082
it says

147
00:05:07,082 --> 00:05:09,481
to judge whether AI will replace a job

148
00:05:09,481 --> 00:05:11,041
do not look at how high end it is

149
00:05:11,041 --> 00:05:14,043
look at whether its core rules can be written as code

150
00:05:14,242 --> 00:05:15,022
if it can be coded

151
00:05:15,242 --> 00:05:17,560
whether it is reviewing contracts or reading scans

152
00:05:17,562 --> 00:05:19,082
it is within range of AI

153
00:05:19,242 --> 00:05:20,103
what cannot be coded

154
00:05:20,242 --> 00:05:21,240
like trust

155
00:05:21,242 --> 00:05:22,723
like the sense of responsibility when things go wrong

156
00:05:22,762 --> 00:05:25,723
that makes you step up and sign your own name

157
00:05:25,762 --> 00:05:27,403
is the last safe zone

158
00:05:28,361 --> 00:05:31,283
this is the same point we made in the Sequoia episode

159
00:05:31,481 --> 00:05:33,882
once a skill gets solved by technology at scale

160
00:05:34,002 --> 00:05:36,200
what you pay for is no longer the price of knowledge

161
00:05:36,202 --> 00:05:37,864
but the price of trust

162
00:05:38,183 --> 00:05:39,284
the book has another concept

163
00:05:39,424 --> 00:05:40,703
that I find very precise

164
00:05:40,703 --> 00:05:41,705
it is called the base

165
00:05:42,224 --> 00:05:44,065
he says AI is a multiplier

166
00:05:44,503 --> 00:05:46,142
everyone has a base

167
00:05:46,143 --> 00:05:49,145
your existing insight resources and position

168
00:05:49,583 --> 00:05:50,743
after AI arrives

169
00:05:50,743 --> 00:05:53,184
it does not pull everyone to the same line

170
00:05:53,304 --> 00:05:56,784
it multiplies every base by a huge factor

171
00:05:57,143 --> 00:05:57,965
people with a big base

172
00:05:58,304 --> 00:05:59,585
get amplified more

173
00:06:00,024 --> 00:06:00,885
people with a small base

174
00:06:01,304 --> 00:06:02,625
stay small after the multiplication

175
00:06:03,574 --> 00:06:05,574
he uses a very simple bit of math

176
00:06:06,014 --> 00:06:06,754
two and three

177
00:06:07,134 --> 00:06:07,855
differ by one

178
00:06:08,334 --> 00:06:10,814
2 to the 10th power is 1024

179
00:06:11,053 --> 00:06:13,935
3 to the 10th power is 59049

180
00:06:14,454 --> 00:06:16,375
at first they look only slightly apart

181
00:06:16,574 --> 00:06:19,095
after amplification the gap is dozens of times

182
00:06:19,995 --> 00:06:22,036
another line in the book stuck with me

183
00:06:22,435 --> 00:06:24,015
he says AI is a mirror

184
00:06:24,234 --> 00:06:25,395
not a genie

185
00:06:25,794 --> 00:06:27,436
a genie can grant your wishes

186
00:06:27,554 --> 00:06:30,955
a mirror just shows more clearly what you already are

187
00:06:31,275 --> 00:06:32,135
for people with a high base

188
00:06:32,395 --> 00:06:34,195
it reflects amplified strengths

189
00:06:34,554 --> 00:06:35,376
for people with a low base

190
00:06:35,635 --> 00:06:38,158
it reflects a widening gap

191
00:06:38,477 --> 00:06:41,518
this also explains a fact many people do not want to face

192
00:06:41,758 --> 00:06:44,119
AI is not a tool for closing gaps

193
00:06:44,157 --> 00:06:46,078
it is a tool for widening them

194
00:06:46,318 --> 00:06:48,039
I have felt this myself

195
00:06:48,318 --> 00:06:52,037
I use AI for research first drafts and organizing material

196
00:06:52,037 --> 00:06:53,518
it really is many times faster

197
00:06:53,758 --> 00:06:54,679
but I know

198
00:06:54,758 --> 00:06:57,739
the speed depends on knowing what to look for and how to ask

199
00:06:58,157 --> 00:07:00,838
and how to judge what is usable once I get results

200
00:07:01,157 --> 00:07:03,636
take someone completely new to this field

201
00:07:03,638 --> 00:07:04,876
with the same tool

202
00:07:04,878 --> 00:07:08,559
they might just churn out a pile of things that look like answers faster

203
00:07:08,827 --> 00:07:10,428
there is a great passage in the book

204
00:07:10,627 --> 00:07:13,788
he says when a senior lawyer drafts with AI

205
00:07:13,867 --> 00:07:16,428
one glance tells him which argument will not hold

206
00:07:16,627 --> 00:07:19,226
someone with little experience reading the same text

207
00:07:19,226 --> 00:07:20,868
is more likely to believe it

208
00:07:21,106 --> 00:07:23,387
because every sentence sounds real

209
00:07:23,707 --> 00:07:26,747
the most dangerous thing about AI is not giving you a wrong answer

210
00:07:26,827 --> 00:07:29,587
it is giving you a plausible answer too fast

211
00:07:30,354 --> 00:07:31,153
on that note

212
00:07:31,155 --> 00:07:34,234
I want to talk about another idea in this book that moved me

213
00:07:34,234 --> 00:07:35,116
called distillation

214
00:07:35,515 --> 00:07:37,955
distillation is a technical term in the AI industry

215
00:07:38,035 --> 00:07:41,636
it means a small model learning from the outputs of a large model

216
00:07:41,755 --> 00:07:45,116
getting most of what the large model can do at a tiny cost

217
00:07:45,354 --> 00:07:47,794
the author applies this idea to companies and people

218
00:07:47,794 --> 00:07:49,155
and I think it fits perfectly

219
00:07:49,395 --> 00:07:50,975
he says distillation can copy answers

220
00:07:51,275 --> 00:07:52,356
but not questions

221
00:07:52,794 --> 00:07:53,415
what does that mean

222
00:07:53,874 --> 00:07:55,953
you can spend a thousandth of the cost

223
00:07:55,955 --> 00:07:58,316
to reach where someone else is right now

224
00:07:58,434 --> 00:08:00,034
but you cannot get their speed

225
00:08:00,035 --> 00:08:02,109
let alone the source of their acceleration

226
00:08:02,428 --> 00:08:03,589
he offers an analogy

227
00:08:03,748 --> 00:08:05,750
two ships on the same route

228
00:08:05,789 --> 00:08:07,589
one has a compass and charts

229
00:08:07,709 --> 00:08:09,990
the other just follows the taillight of the ship ahead

230
00:08:10,188 --> 00:08:12,510
in clear weather you cannot tell them apart

231
00:08:12,628 --> 00:08:15,269
when fog rolls in or the lead ship turns

232
00:08:15,308 --> 00:08:16,584
the difference shows

233
00:08:16,702 --> 00:08:19,144
if an organization only distills for years

234
00:08:19,262 --> 00:08:22,504
the muscle it builds is for catching up inside a frame set by others

235
00:08:22,983 --> 00:08:26,144
the first people promoted are the ones good at optimizing metrics

236
00:08:26,582 --> 00:08:28,983
when the company finally decides to create something original

237
00:08:29,103 --> 00:08:32,423
it finds that organizational muscle cannot be swapped out in one quarter

238
00:08:32,903 --> 00:08:35,384
this makes me think of a lot of copycat startups

239
00:08:35,743 --> 00:08:37,741
whether copying a business model

240
00:08:37,742 --> 00:08:39,583
or copying a content style

241
00:08:39,862 --> 00:08:41,504
it looks efficient in the short run

242
00:08:41,623 --> 00:08:44,182
but in the long run what you train is not your own judgment

243
00:08:44,183 --> 00:08:45,823
it is a follower reflex

244
00:08:46,882 --> 00:08:47,883
speaking of startups

245
00:08:47,961 --> 00:08:50,123
another passage in the book stuck with me

246
00:08:50,402 --> 00:08:51,422
he says it three times in a row

247
00:08:51,802 --> 00:08:52,801
do not start a company

248
00:08:52,802 --> 00:08:53,503
do not start a company

249
00:08:53,802 --> 00:08:54,802
do not start a company

250
00:08:55,042 --> 00:08:55,822
he is not trying to scare you off

251
00:08:56,162 --> 00:08:57,042
he is saying

252
00:08:57,201 --> 00:09:00,363
AI has pushed the cost of building a product close to zero

253
00:09:00,561 --> 00:09:03,123
but the cost of owning a position has actually gone up

254
00:09:03,282 --> 00:09:05,682
most AI founders are not really building startups

255
00:09:05,762 --> 00:09:08,643
they are doing free market research for the big model companies

256
00:09:08,922 --> 00:09:11,042
you work hard to validate a need

257
00:09:11,081 --> 00:09:12,843
prove users will pay

258
00:09:13,002 --> 00:09:16,843
and then one platform update makes what you built a native feature

259
00:09:17,042 --> 00:09:18,802
he has a really brutal line

260
00:09:19,441 --> 00:09:19,782
he says

261
00:09:20,122 --> 00:09:21,641
people who start companies chasing market cap

262
00:09:21,642 --> 00:09:24,763
are often chasing a photo of a meal someone else already ate

263
00:09:25,143 --> 00:09:26,621
there is another part of the book

264
00:09:26,622 --> 00:09:28,784
that made me stop and think again and again

265
00:09:28,982 --> 00:09:31,363
it is his iron rule for AI native companies

266
00:09:31,823 --> 00:09:34,424
the right to produce and the right to evaluate must be separate

267
00:09:34,742 --> 00:09:36,981
he says the biggest trap of the AI era

268
00:09:36,982 --> 00:09:38,424
is not that AI cannot do the work

269
00:09:38,622 --> 00:09:41,784
it is that the work AI does looks too convincing

270
00:09:42,023 --> 00:09:44,784
it gives you a wrong answer in the most confident tone

271
00:09:44,982 --> 00:09:47,024
if you have no independent reviewer

272
00:09:47,063 --> 00:09:49,463
you will march confidently all the way into a crash

273
00:09:49,903 --> 00:09:50,644
what he means is

274
00:09:51,023 --> 00:09:52,822
AI failures are not obvious

275
00:09:52,823 --> 00:09:54,424
they get structurally amplified

276
00:09:54,462 --> 00:09:56,044
and output is so cheap

277
00:09:56,382 --> 00:09:57,823
there is no natural brake

278
00:09:58,722 --> 00:10:00,841
if a person makes a mistake

279
00:10:00,842 --> 00:10:02,204
it affects one client

280
00:10:02,483 --> 00:10:04,802
if AI makes the same mistake

281
00:10:04,802 --> 00:10:07,004
it might hit ten thousand clients at once

282
00:10:07,163 --> 00:10:09,604
because it outputs in bulk from templates

283
00:10:09,842 --> 00:10:11,682
so for an AI native company

284
00:10:11,682 --> 00:10:13,882
the core is not how many AI tools it uses

285
00:10:13,883 --> 00:10:18,163
it is whether the org structure separates the doers from the reviewers

286
00:10:18,523 --> 00:10:19,682
AI executes

287
00:10:19,682 --> 00:10:22,643
humans set rules review and carry responsibility

288
00:10:23,245 --> 00:10:26,287
I think this applies beyond companies

289
00:10:26,725 --> 00:10:29,407
it applies to everyone who uses AI

290
00:10:30,006 --> 00:10:31,747
if you treat AI as the executor

291
00:10:32,166 --> 00:10:33,726
and yourself as the evaluator

292
00:10:33,806 --> 00:10:34,966
that is healthy

293
00:10:35,205 --> 00:10:37,086
if you treat AI as the judge

294
00:10:37,205 --> 00:10:39,407
and stop checking results yourself

295
00:10:39,526 --> 00:10:41,046
sooner or later things go wrong

296
00:10:42,102 --> 00:10:44,022
one line in the book really moved me

297
00:10:44,301 --> 00:10:45,102
he says

298
00:10:45,181 --> 00:10:48,220
in the end a person either designs the orchestration system

299
00:10:48,222 --> 00:10:49,963
or is managed by it

300
00:10:50,701 --> 00:10:52,142
it sounds a bit cold

301
00:10:52,541 --> 00:10:53,203
but think about it

302
00:10:53,502 --> 00:10:54,983
he is talking about position

303
00:10:55,342 --> 00:10:56,860
are you the one who makes the rules

304
00:10:56,862 --> 00:10:58,323
or the one who follows them

305
00:10:58,941 --> 00:11:00,781
are you the one who designs the system

306
00:11:00,781 --> 00:11:02,483
or the one scheduled by it

307
00:11:03,142 --> 00:11:04,220
in the AI era

308
00:11:04,222 --> 00:11:06,141
that dividing line will get clearer and clearer

309
00:11:06,142 --> 00:11:07,463
and more and more brutal

310
00:11:08,011 --> 00:11:09,932
knowing has never been a ticket in

311
00:11:10,091 --> 00:11:13,212
sometimes knowing is just the most refined way of watching from the sidelines

312
00:11:13,731 --> 00:11:14,852
have you ever wondered

313
00:11:14,971 --> 00:11:17,130
why so many people talk about AI

314
00:11:17,131 --> 00:11:19,232
but never actually step in

315
00:11:19,851 --> 00:11:20,791
the book says something

316
00:11:21,011 --> 00:11:22,011
that I think is spot on

317
00:11:22,331 --> 00:11:24,552
it says what holds them back is not intelligence

318
00:11:24,930 --> 00:11:25,852
it is pride

319
00:11:26,331 --> 00:11:28,411
because the moment you really step in

320
00:11:28,491 --> 00:11:31,251
you find the thing about this machine that hurts most

321
00:11:31,331 --> 00:11:33,011
was never that it is faster than you

322
00:11:33,371 --> 00:11:34,771
it is that it makes you see

323
00:11:34,851 --> 00:11:37,491
you are not worth as much as you imagined

324
00:11:37,971 --> 00:11:39,052
but flip it around

325
00:11:39,131 --> 00:11:41,531
that is exactly what makes it most valuable

326
00:11:42,211 --> 00:11:45,251
if you are willing to accept what that mirror shows

327
00:11:45,451 --> 00:11:49,169
you get a chance to see what your real base is

328
00:11:49,170 --> 00:11:51,092
and then find ways to grow it

329
00:11:51,639 --> 00:11:52,099
in the end

330
00:11:52,399 --> 00:11:54,319
this book made one thing clear to me

331
00:11:54,399 --> 00:11:55,660
those of us who make content

332
00:11:55,799 --> 00:11:56,757
who invest

333
00:11:56,758 --> 00:11:57,920
who build startups

334
00:11:57,958 --> 00:12:01,679
are busy every day chasing new tools new models new trends

335
00:12:01,878 --> 00:12:04,679
but the author spends nearly 500 pages telling you

336
00:12:04,718 --> 00:12:06,038
tools will keep changing

337
00:12:06,039 --> 00:12:07,260
models will keep upgrading

338
00:12:07,519 --> 00:12:09,120
trends will keep rotating

339
00:12:09,159 --> 00:12:10,757
what really decides your fate

340
00:12:10,758 --> 00:12:12,438
is not which wave you caught

341
00:12:12,439 --> 00:12:14,368
but which layer you stand on

342
00:12:14,567 --> 00:12:16,567
at the end of the book he asks three questions

343
00:12:16,767 --> 00:12:18,488
that nobody can answer yet

344
00:12:18,966 --> 00:12:19,788
as intelligence gets cheaper

345
00:12:20,407 --> 00:12:21,748
does power get more concentrated

346
00:12:22,446 --> 00:12:24,608
when AI starts improving AI

347
00:12:24,807 --> 00:12:26,547
can humans still set the pace

348
00:12:27,287 --> 00:12:28,947
who gets to wait for the future to pay off

349
00:12:29,606 --> 00:12:30,268
these three questions

350
00:12:30,767 --> 00:12:32,528
I cannot answer them either after finishing the book

351
00:12:32,726 --> 00:12:34,927
but I think he put one thing well

352
00:12:35,126 --> 00:12:37,327
he says judgment is not shutting the door

353
00:12:37,407 --> 00:12:39,327
it is first taking a position

354
00:12:39,527 --> 00:12:42,146
and then watching where the wind comes from

355
00:12:42,584 --> 00:12:45,466
back to the 25 year old prodigy from the start

356
00:12:45,704 --> 00:12:47,665
Aschenbrenner may well be right

357
00:12:47,944 --> 00:12:51,185
his read on AI will likely be proven in the end

358
00:12:51,584 --> 00:12:54,746
but the market will not simply because a man turns out right

359
00:12:54,785 --> 00:12:57,346
hand him back his old seat

360
00:12:57,865 --> 00:13:00,106
cognition answers what you see

361
00:13:00,265 --> 00:13:03,986
position decides whether you can hold on to what you saw until the end

362
00:13:04,693 --> 00:13:07,094
this is what woke me up most in this book

363
00:13:07,613 --> 00:13:09,294
it is not selling anxiety

364
00:13:09,373 --> 00:13:10,974
or predicting the future

365
00:13:11,294 --> 00:13:12,395
it is like a mirror

366
00:13:12,733 --> 00:13:14,334
that makes you stop and take a look

367
00:13:14,414 --> 00:13:15,194
and ask yourself

368
00:13:15,774 --> 00:13:17,495
the position I stand in now

369
00:13:17,534 --> 00:13:20,214
if a model makes this a native feature next month

370
00:13:20,294 --> 00:13:21,155
what do I have left

371
00:13:21,894 --> 00:13:23,495
if you can answer that

372
00:13:23,654 --> 00:13:25,775
you probably do not need to fear AI

373
00:13:26,134 --> 00:13:27,454
if you cannot

374
00:13:27,573 --> 00:13:29,375
then no matter how sharp your insight

375
00:13:29,534 --> 00:13:33,333
you may just be on a stretch of sand the sea is swallowing

376
00:13:33,333 --> 00:13:34,694
running faster and that is all

377
00:13:35,054 --> 00:13:35,714
I am Leo Wang

378
00:13:36,014 --> 00:13:37,092
see you next time