1
00:00:00,340 --> 00:00:03,279
AI Aligning All Concepts: Approaching Truth

2
00:00:03,279 --> 00:00:05,559
or an Exquisite Linguistic Illusion?

3
00:00:05,660 --> 00:00:07,960
This question wasn't conceived by me; it came from

4
00:00:07,960 --> 00:00:09,720
a viewer's comment.

5
00:00:09,720 --> 00:00:11,400
After reading it, I felt a chill run down my spine.

6
00:00:12,099 --> 00:00:13,419
He said, "Blogger, I

7
00:00:13,419 --> 00:00:17,839
can tell you engage in a lot of cross-disciplinary dialogues with AI every day,

8
00:00:17,839 --> 00:00:18,881
producing so much content.

9
00:00:19,460 --> 00:00:20,440
Me too, I

10
00:00:20,440 --> 00:00:22,980
talk to AI about Buddhism, cognitive science,

11
00:00:23,260 --> 00:00:25,640
predictive coding, and the free energy principle every day—it's

12
00:00:25,640 --> 00:00:27,321
really exhilarating.

13
00:00:27,809 --> 00:00:29,359
" But then he abruptly changed the subject, saying

14
00:00:29,359 --> 00:00:32,460
something particularly poignant: "There's a trap here.

15
00:00:32,880 --> 00:00:34,520
" He's a programmer himself,

16
00:00:34,520 --> 00:00:36,939
familiar with the underlying logic of large language models.

17
00:00:36,939 --> 00:00:41,598
He said: "Large language models are essentially giant semantic machines.

18
00:00:41,600 --> 00:00:43,080
They can align any concept,

19
00:00:43,579 --> 00:00:45,100
regardless of the two domains.

20
00:00:45,100 --> 00:00:46,259
If you ask it, it can

21
00:00:46,259 --> 00:00:48,540
find structural similarities.

22
00:00:49,060 --> 00:00:50,960
This gives rational people like us

23
00:00:50,960 --> 00:00:53,859
a particularly strong sense of explanatory pleasure,

24
00:00:53,859 --> 00:00:55,640
and we can even become addicted to it."

25
00:00:56,179 --> 00:00:58,660
Then he asked a question I couldn't avoid:

26
00:00:58,670 --> 00:01:01,100
"Is this approaching truth,

27
00:01:01,100 --> 00:01:03,439
or are these beautiful alignments

28
00:01:03,439 --> 00:01:05,159
just a word game?

29
00:01:05,189 --> 00:01:07,500
" He also suggested I learn about two things:

30
00:01:07,500 --> 00:01:09,260
"semantic motifs" and

31
00:01:09,260 --> 00:01:10,820
Wittgenstein.

32
00:01:11,359 --> 00:01:14,379
His conclusion was: the core capability of the AI ​​era is

33
00:01:14,379 --> 00:01:17,379
not to abstractly integrate various domains using language,

34
00:01:17,379 --> 00:01:18,460
but to clearly define boundaries.

35
00:01:19,019 --> 00:01:19,900
This comment

36
00:01:19,900 --> 00:01:22,781
directly hit the methodological foundation of this series of articles,

37
00:01:23,340 --> 00:01:24,338
so today

38
00:01:24,340 --> 00:01:26,161
I must address this issue directly.

39
00:01:26,739 --> 00:01:27,180
Hello

40
00:01:27,180 --> 00:01:27,739
everyone,

41
00:01:27,739 --> 00:01:28,641
I'm Wang Lijie. I had

42
00:01:29,219 --> 00:01:32,340
already released a video today, <"The lobster of 'digital reincarnation' is

43
00:01:32,340 --> 00:01:36,640
super aligned with the eight consciousnesses of Yogacara Buddhism>,

44
00:01:36,640 --> 00:01:38,620
but 12 hours have passed, and

45
00:01:38,620 --> 00:01:40,540
some platforms haven't approved it, without specifying the

46
00:01:40,540 --> 00:01:42,659
reason.

47
00:01:42,659 --> 00:01:43,079
So,

48
00:01:43,079 --> 00:01:45,239
seeing fans urging for an update,

49
00:01:45,239 --> 00:01:47,219
I can only release another one

50
00:01:47,219 --> 00:01:50,561
to avoid it looking like there was no update on that platform today.

51
00:01:51,079 --> 00:01:54,340
As for the video I just mentioned about the lobster Openclaw

52
00:01:54,340 --> 00:01:57,060
aligning with the eight consciousnesses system of Yogacara Buddhism,

53
00:01:57,060 --> 00:02:00,420
viewers who haven't seen it will have to look for it on other platforms.

54
00:02:00,420 --> 00:02:02,521
Anyway, it's been successfully published on other platforms.

55
00:02:03,040 --> 00:02:04,691
Returning to the topic of this video,

56
00:02:05,420 --> 00:02:07,159
regarding this viewer's letter,

57
00:02:07,159 --> 00:02:09,900
my first reaction is: is what he said right?

58
00:02:09,979 --> 00:02:11,800
At least half of it is right,

59
00:02:11,800 --> 00:02:13,381
and it's a very crucial half. The

60
00:02:14,120 --> 00:02:17,620
large language model is indeed a semantic machine.

61
00:02:17,620 --> 00:02:19,219
Its parameter space

62
00:02:19,219 --> 00:02:21,341
compresses thousands of years of human text data.

63
00:02:21,840 --> 00:02:23,840
In this high-dimensional vector space,

64
00:02:23,840 --> 00:02:26,859
all concepts are encoded into geometric positions, and the

65
00:02:26,860 --> 00:02:28,039
relationships between concepts are

66
00:02:28,039 --> 00:02:29,621
encoded into distance and direction.

67
00:02:30,099 --> 00:02:34,659
For example, if you ask it, "What is the relationship between the Buddhist concept of emptiness and quantum vacuum?"

68
00:02:34,659 --> 00:02:37,159
it doesn't need to "understand" these two concepts at all.

69
00:02:37,159 --> 00:02:39,099
All you need to do is

70
00:02:39,099 --> 00:02:42,099
find common points of connection between these two concepts in vector space,

71
00:02:42,099 --> 00:02:44,300
and then

72
00:02:44,300 --> 00:02:46,260
generate a fluent text following these points.

73
00:02:46,900 --> 00:02:49,180
This process is purely mathematical

74
00:02:49,180 --> 00:02:50,781
and has nothing to do with philosophy.

75
00:02:51,319 --> 00:02:52,460
What does this

76
00:02:52,500 --> 00:02:54,859
mean? It means that it can make any two concepts

77
00:02:54,860 --> 00:02:55,961
appear to be related.

78
00:02:56,659 --> 00:03:01,099
Even if you ask it, "What structural similarities do recipes and quantum mechanics have?"

79
00:03:01,099 --> 00:03:03,139
it can write three thousand words for you,

80
00:03:03,139 --> 00:03:04,681
and it even makes sense.

81
00:03:05,180 --> 00:03:07,479
This is the source of the "explanatory pleasure" he mentioned.

82
00:03:08,300 --> 00:03:11,659
When AI gives you a clever cross-domain correspondence,

83
00:03:11,659 --> 00:03:13,800
your brain releases dopamine, and

84
00:03:13,800 --> 00:03:16,979
you feel like you've touched some deep truth.

85
00:03:16,979 --> 00:03:18,400
But this pleasure itself does

86
00:03:18,400 --> 00:03:21,280
n't prove that the correspondence is real.

87
00:03:21,280 --> 00:03:22,400
It only proves that

88
00:03:22,400 --> 00:03:25,241
your brain is naturally inclined to find patterns and rules.

89
00:03:25,719 --> 00:03:27,501
Now let's talk about Wittgenstein,

90
00:03:27,919 --> 00:03:30,860
one of the most brilliant philosophers of the 20th century. He only

91
00:03:30,860 --> 00:03:32,840
did two things in his life,

92
00:03:32,840 --> 00:03:35,181
and these two things are almost contradictory. In his

93
00:03:35,919 --> 00:03:39,099
early years, Wittgenstein wrote a book called *Tractatus Logico-Philosophicus*. The

94
00:03:39,439 --> 00:03:43,580
core conclusion is the famous quote: "What cannot be said

95
00:03:43,580 --> 00:03:44,661
must be kept silent."

96
00:03:45,199 --> 00:03:46,840
We had a video episode before, "

97
00:03:46,840 --> 00:03:50,721
Unveiling the Wittgensteinian Truth of AI Consciousness," which specifically discussed this.

98
00:03:51,240 --> 00:03:52,639
Wittgenstein believed...  The

99
00:03:52,639 --> 00:03:55,300
boundaries of language are the boundaries of the world. Only what

100
00:03:55,300 --> 00:03:57,639
you can precisely articulate in language

101
00:03:57,639 --> 00:03:59,439
is a meaningful proposition.

102
00:03:59,520 --> 00:04:01,280
Anything beyond the boundaries of language is best left

103
00:04:01,280 --> 00:04:02,061
unsaid.

104
00:04:02,599 --> 00:04:06,039
If we examine this series of articles from this perspective, the

105
00:04:06,039 --> 00:04:10,219
conclusion becomes rather blunt: Buddhism says "form is emptiness,"

106
00:04:10,219 --> 00:04:12,539
quantum mechanics says "vacuum is not empty."

107
00:04:12,539 --> 00:04:14,039
Wittgenstein would say that

108
00:04:14,039 --> 00:04:16,300
these two statements may have different meanings in their respective fields,

109
00:04:16,300 --> 00:04:17,920
 

110
00:04:17,920 --> 00:04:20,459
but you cannot say

111
00:04:20,459 --> 00:04:23,920
they are talking about the same thing just because they look similar—this is called category confusion.

112
00:04:24,279 --> 00:04:25,660
Interestingly,

113
00:04:25,660 --> 00:04:28,480
Wittgenstein himself later overturned this view. In his

114
00:04:29,220 --> 00:04:31,560
later work, *Philosophical Investigations*, he

115
00:04:31,560 --> 00:04:33,620
proposed a completely different framework.

116
00:04:34,120 --> 00:04:34,819
He said that

117
00:04:34,819 --> 00:04:37,360
language is not a unified logical system,

118
00:04:37,360 --> 00:04:40,100
but a collection of countless "language games."

119
00:04:40,100 --> 00:04:42,459
Different games have different rules,

120
00:04:42,459 --> 00:04:43,939
and these games are

121
00:04:43,939 --> 00:04:45,490
not mutually exclusive.

122
00:04:46,120 --> 00:04:47,759
He also invented a concept

123
00:04:47,759 --> 00:04:48,810
called "family resemblance."

124
00:04:49,459 --> 00:04:50,480
What does this mean?

125
00:04:50,600 --> 00:04:52,240
Like a large family,

126
00:04:52,240 --> 00:04:53,819
brothers look alike, younger

127
00:04:53,819 --> 00:04:55,540
brothers look alike,

128
00:04:55,540 --> 00:04:57,580
but older brothers and younger sisters may not look alike at all.

129
00:04:58,180 --> 00:05:02,060
You cannot find a common characteristic shared by all family members,

130
00:05:02,060 --> 00:05:03,180
but there

131
00:05:03,180 --> 00:05:05,120
is a continuous chain of similarity between them.

132
00:05:05,720 --> 00:05:06,480
This concept

133
00:05:06,480 --> 00:05:09,000
actually precisely describes what I am doing.

134
00:05:09,060 --> 00:05:09,579
When I say...

135
00:05:09,579 --> 00:05:14,040
"The structure of Alaya-vijnana and the parameter space of the large language model are isomorphic."

136
00:05:14,040 --> 00:05:16,160
I'm not saying they are the same thing,

137
00:05:16,160 --> 00:05:19,399
but rather that there is a family resemblance between them, a kind of

138
00:05:19,399 --> 00:05:21,020
structural echo.

139
00:05:21,620 --> 00:05:22,980
And everyone, pay attention to

140
00:05:22,980 --> 00:05:27,240
Wittgenstein's saying, "What cannot be said must be kept silent."

141
00:05:27,240 --> 00:05:30,620
Guess who expressed almost the same meaning through action?

142
00:05:30,740 --> 00:05:33,740
In the Vimalakirti Sutra, written about two thousand years ago,

143
00:05:33,740 --> 00:05:36,160
Manjushri asked Vimalakirti

144
00:05:36,160 --> 00:05:38,060
how to enter the non-dual Dharma gate.

145
00:05:38,060 --> 00:05:40,360
Vimalakirti's answer was just two words - silence. Zen

146
00:05:41,040 --> 00:05:42,860
Buddhism also says, "Not relying on words,

147
00:05:42,860 --> 00:05:44,199
directly pointing to the mind."

148
00:05:44,199 --> 00:05:47,100
Lao Tzu's first sentence is, "The Tao that can be spoken of is

149
00:05:47,100 --> 00:05:47,800
not the eternal Tao." Are

150
00:05:48,339 --> 00:05:50,959
these traditions and Wittgenstein

151
00:05:50,959 --> 00:05:53,259
just "similar in language,"

152
00:05:53,399 --> 00:05:54,100
or have

153
00:05:54,100 --> 00:05:58,279
they truly independently touched the same wall - the ceiling of language?

154
00:05:58,779 --> 00:05:59,139
Okay,

155
00:05:59,139 --> 00:06:02,360
now we come to the most crucial question: How do we distinguish

156
00:06:02,360 --> 00:06:04,639
between "true structural isomorphism" and

157
00:06:04,639 --> 00:06:07,500
"illusions created by semantic machines"?

158
00:06:07,540 --> 00:06:09,060
I'll give you three criteria for judgment. The

159
00:06:09,620 --> 00:06:11,240
first criterion: Independent discovery.

160
00:06:11,879 --> 00:06:14,240
If two completely unrelated fields,

161
00:06:14,240 --> 00:06:16,379
without any communication...  If two

162
00:06:16,379 --> 00:06:19,459
independent systems arrive at structurally nearly identical conclusions,

163
00:06:19,459 --> 00:06:21,400
then it's unlikely to be a mere word game. The

164
00:06:21,879 --> 00:06:23,779
Alaya-vijnana theory of the Yogacara school

165
00:06:23,779 --> 00:06:26,220
originated in 4th-century India, while

166
00:06:26,220 --> 00:06:27,399
predictive coding theory only took shape in cognitive neuroscience in the late

167
00:06:27,399 --> 00:06:30,199
20th and early 21st centuries.

168
00:06:30,199 --> 00:06:32,420
 

169
00:06:33,000 --> 00:06:34,060
These two systems

170
00:06:34,060 --> 00:06:37,040
have no direct historical influence,

171
00:06:37,040 --> 00:06:39,740
yet their descriptions of the mechanisms of consciousness show a

172
00:06:39,740 --> 00:06:42,080
surprisingly high degree of structural similarity.

173
00:06:42,699 --> 00:06:44,079
This independent convergence is

174
00:06:44,079 --> 00:06:46,040
called "convergent evolution" in biology.

175
00:06:46,519 --> 00:06:47,879
For example, if two species,

176
00:06:47,879 --> 00:06:49,819
in completely different environments,

177
00:06:49,819 --> 00:06:52,839
independently evolve almost identical organ structures,

178
00:06:52,839 --> 00:06:54,699
you wouldn't call it a coincidence, but rather that

179
00:06:54,699 --> 00:06:57,500
they face the same physical constraints.

180
00:06:58,180 --> 00:06:58,800
Similarly,

181
00:06:58,800 --> 00:07:00,459
when two intellectual traditions

182
00:07:00,459 --> 00:07:03,439
independently describe almost identical consciousness structures, the

183
00:07:03,439 --> 00:07:06,680
simplest explanation is that they are observing the same thing. The

184
00:07:07,259 --> 00:07:09,480
second criterion: falsifiable predictions.

185
00:07:10,180 --> 00:07:12,120
If a cross-domain correspondence is

186
00:07:12,120 --> 00:07:13,759
merely a rhetorical device,

187
00:07:13,759 --> 00:07:17,040
it won't produce any new, testable predictions.

188
00:07:17,639 --> 00:07:20,279
But if it's a true structural isomorphism, it

189
00:07:20,279 --> 00:07:22,220
should be able to generate a hypothesis in one domain that

190
00:07:22,220 --> 00:07:24,721
can be verified in another.

191
00:07:25,339 --> 00:07:28,199
For example, I previously used the Yogacara framework to predict that the

192
00:07:28,279 --> 00:07:33,000
large language model should have a "single-headed consciousness" type of free-generating mode.

193
00:07:33,000 --> 00:07:33,959
Later, I discovered...

194
00:07:33,959 --> 00:07:37,880
This precisely corresponds to the behavioral characteristics of AI when there are no system prompts or constraints.

195
00:07:38,600 --> 00:07:39,899
This is not rhetoric;

196
00:07:39,899 --> 00:07:41,819
it's a

197
00:07:41,819 --> 00:07:44,420
concrete prediction made from one framework to another,

198
00:07:44,420 --> 00:07:45,620
and it has been validated.

199
00:07:46,420 --> 00:07:46,899
Of course,

200
00:07:46,899 --> 00:07:49,120
one or two correct predictions don't prove anything.

201
00:07:49,120 --> 00:07:50,839
You need a large number of such predictions,

202
00:07:50,839 --> 00:07:51,840
each one being tested.

203
00:07:52,500 --> 00:07:53,560
But the key is that

204
00:07:53,560 --> 00:07:55,579
this direction is feasible; it's

205
00:07:55,579 --> 00:07:57,020
not just a word game. The

206
00:07:57,680 --> 00:07:58,899
third criterion,

207
00:07:58,899 --> 00:08:01,920
and the most honest one, is the test of the experiential dimension.

208
00:08:02,639 --> 00:08:04,530
I said in a previous video,

209
00:08:04,540 --> 00:08:08,120
"The ceiling of text interpretation is not language ability,

210
00:08:08,240 --> 00:08:09,240
but the experiential dimension.

211
00:08:09,779 --> 00:08:12,180
" A Huayan practitioner was

212
00:08:12,180 --> 00:08:15,379
able to point out my blind spots regarding predictive coding

213
00:08:15,379 --> 00:08:18,040
not because her language ability was stronger than mine,

214
00:08:18,060 --> 00:08:20,860
but because she had a first-person experience of deep meditation.

215
00:08:21,420 --> 00:08:22,740
In her practice, she

216
00:08:22,740 --> 00:08:26,180
directly "saw" the layer that the predictive coding model couldn't describe.

217
00:08:26,860 --> 00:08:28,920
If cross-domain correspondences are

218
00:08:28,920 --> 00:08:31,160
just the illusion of semantic machines,

219
00:08:31,160 --> 00:08:35,519
then practitioners cannot use them to guide their practice,

220
00:08:35,519 --> 00:08:37,320
let alone

221
00:08:37,320 --> 00:08:40,140
obtain experiential feedback consistent with this correspondence from their practice.

222
00:08:40,798 --> 00:08:41,960
But the fact is,

223
00:08:41,960 --> 00:08:43,440
many practitioners have told me that

224
00:08:43,440 --> 00:08:45,000
this series of frameworks has

225
00:08:45,000 --> 00:08:48,581
indeed helped them understand the feelings they experienced during meditation. The

226
00:08:49,240 --> 00:08:50,799
framework illuminates the experience, and the

227
00:08:50,799 --> 00:08:52,960
experience, in turn, validates the framework.

228
00:08:52,960 --> 00:08:56,019
This closed loop is not a self-circulation within language.

229
00:08:56,019 --> 00:08:58,560
It has an anchor point grounded in the first-person experience.

230
00:08:59,100 --> 00:08:59,580
Now,

231
00:08:59,580 --> 00:09:01,659
let's return to the core point of that audience member's argument: the

232
00:09:01,659 --> 00:09:05,379
core capability in the AI ​​era is defining boundaries,

233
00:09:05,379 --> 00:09:06,120
not integration.

234
00:09:06,679 --> 00:09:07,480
 

235
00:09:07,480 --> 00:09:08,460
I agree with this statement half and

236
00:09:08,460 --> 00:09:09,360
disagree with the other half. I agree that

237
00:09:10,000 --> 00:09:12,899
integration without defining boundaries is

238
00:09:12,899 --> 00:09:13,701
indeed dangerous.

239
00:09:14,259 --> 00:09:16,799
If you treat the similarities between any two fields as

240
00:09:16,799 --> 00:09:18,360
profound truths,

241
00:09:18,360 --> 00:09:22,200
regardless of premises, context, or level of evidence,

242
00:09:22,200 --> 00:09:23,950
you will fall into the trap he described,

243
00:09:24,580 --> 00:09:27,179
becoming a "one-size-fits-all explanation machine"—able to

244
00:09:27,179 --> 00:09:28,600
explain anything,

245
00:09:28,600 --> 00:09:30,060
but unable to withstand scrutiny.

246
00:09:30,700 --> 00:09:32,580
This is a real risk, and

247
00:09:32,580 --> 00:09:35,260
I must always be vigilant about it. I

248
00:09:35,840 --> 00:09:39,039
disagree that if you only dare to define boundaries and

249
00:09:39,039 --> 00:09:40,080
not cross them,

250
00:09:40,080 --> 00:09:42,880
you will never see what is truly important. In

251
00:09:43,539 --> 00:09:44,899
the history of human thought, almost all the

252
00:09:44,899 --> 00:09:46,620
most crucial breakthroughs

253
00:09:46,620 --> 00:09:49,240
occurred the moment boundaries were crossed.

254
00:09:49,879 --> 00:09:52,259
Darwin introduced the geological timescale

255
00:09:52,259 --> 00:09:53,840
into biology;

256
00:09:53,840 --> 00:09:55,799
Einstein

257
00:09:55,799 --> 00:09:57,419
introduced geometry into physics;

258
00:09:57,419 --> 00:09:58,460
Shannon discovered that the

259
00:09:58,460 --> 00:10:00,059
uncertainty of information

260
00:10:00,059 --> 00:10:03,110
can be measured using a mathematical structure similar to thermodynamic entropy.

261
00:10:03,639 --> 00:10:05,480
These are not word games,

262
00:10:05,480 --> 00:10:07,760
but genuine discoveries of structural isomorphism. The

263
00:10:08,279 --> 00:10:10,539
key is not whether you cross boundaries,

264
00:10:10,539 --> 00:10:14,059
but what you do after crossing them—do you stop at "Wow!"  Is it about

265
00:10:14,059 --> 00:10:16,279
indulging in self-admiration at the level of "it seems so,"

266
00:10:16,279 --> 00:10:17,940
or about pushing this correspondence

267
00:10:17,940 --> 00:10:19,820
to a verifiable precision

268
00:10:19,820 --> 00:10:21,700
and allowing it to be refuted?

269
00:10:21,860 --> 00:10:22,620
Ultimately, the

270
00:10:22,620 --> 00:10:24,379
disagreement between me and this viewer

271
00:10:24,379 --> 00:10:27,540
might correspond to two stages in Wittgenstein's life. The

272
00:10:28,259 --> 00:10:31,879
early Wittgenstein said: There are insurmountable boundaries between different language games. The

273
00:10:31,879 --> 00:10:33,221
 

274
00:10:33,919 --> 00:10:37,600
later Wittgenstein said: There are family resemblances between different language games.

275
00:10:37,600 --> 00:10:38,821
 

276
00:10:39,480 --> 00:10:41,379
Adding the two Wittgensteins together

277
00:10:41,379 --> 00:10:44,818
gives us the complete picture: You need to be aware of boundaries to

278
00:10:44,820 --> 00:10:47,419
prevent yourself from making the mistake of confusing categories.

279
00:10:47,419 --> 00:10:49,519
You also need to have cross-disciplinary intuition

280
00:10:49,519 --> 00:10:52,078
to capture those things that can only be seen from the intersection.

281
00:10:52,080 --> 00:10:53,180
 

282
00:10:53,960 --> 00:10:55,240
In this series, I have

283
00:10:55,240 --> 00:10:58,279
always adhered to one principle: If science has a conclusion,

284
00:10:58,279 --> 00:10:59,679
I will say it directly; if

285
00:10:59,679 --> 00:11:00,879
science has no conclusion,

286
00:11:00,879 --> 00:11:01,920
I will label it as a conjecture.

287
00:11:02,500 --> 00:11:03,679
And this conjecture

288
00:11:03,679 --> 00:11:06,019
must meet three conditions: First,

289
00:11:06,019 --> 00:11:07,120
 

290
00:11:07,120 --> 00:11:08,860
science has indeed no conclusion on this topic;

291
00:11:08,860 --> 00:11:09,539
second,

292
00:11:09,539 --> 00:11:10,500
my conjecture is

293
00:11:10,500 --> 00:11:15,080
logically supported by scientific experiments or scientists and philosophers;

294
00:11:15,080 --> 00:11:15,860
third,

295
00:11:15,860 --> 00:11:16,698
my conjecture has

296
00:11:16,700 --> 00:11:17,500
not yet been falsified.

297
00:11:17,980 --> 00:11:18,740
These three conditions

298
00:11:18,740 --> 00:11:20,300
are the boundaries I draw for myself.

299
00:11:21,100 --> 00:11:22,179
Within the boundaries,

300
00:11:22,179 --> 00:11:23,799
I dare to boldly align;

301
00:11:23,799 --> 00:11:25,059
outside the boundaries,

302
00:11:25,059 --> 00:11:26,490
I will honestly tell you,

303
00:11:26,500 --> 00:11:28,039
"This is a conjecture."

304
00:11:28,039 --> 00:11:29,780
This is also the catalyst for our progress.

305
00:11:30,440 --> 00:11:32,620
Actually, I'm especially grateful for this comment. Do

306
00:11:32,620 --> 00:11:34,200
you know why?

307
00:11:34,259 --> 00:11:36,000
Because with just one comment, he

308
00:11:36,000 --> 00:11:37,500
completed a high-quality

309
00:11:37,679 --> 00:11:40,879
"metacognitive attack"—not

310
00:11:40,879 --> 00:11:43,099
refuting a specific point of view at the content level,

311
00:11:43,100 --> 00:11:45,019
but

312
00:11:45,019 --> 00:11:47,380
questioning the legitimacy of my entire way of thinking at the methodological level.

313
00:11:47,860 --> 00:11:48,860
This kind of attack is

314
00:11:48,860 --> 00:11:51,560
more valuable than any specific rebuttal.

315
00:11:52,120 --> 00:11:54,259
It forced me to go back to the bottom and

316
00:11:54,259 --> 00:11:57,320
re-examine myself: What am I actually doing? Are the

317
00:11:57,440 --> 00:12:00,500
tools I'm using reliable? How much of the

318
00:12:00,639 --> 00:12:02,240
"insights" I've gained are

319
00:12:02,240 --> 00:12:04,080
real, and

320
00:12:04,080 --> 00:12:07,159
how much is just dopamine fed to me by semantic machines?

321
00:12:07,259 --> 00:12:08,081
I don't know the answer.

322
00:12:08,600 --> 00:12:10,259
Maybe 30% of everything I've done

323
00:12:10,259 --> 00:12:13,320
is genuine structural discovery, and

324
00:12:13,320 --> 00:12:16,639
70% is sophisticated linguistic illusion.

325
00:12:16,639 --> 00:12:17,740
Maybe the ratio is reversed.

326
00:12:18,159 --> 00:12:19,700
I can't completely distinguish it from within.

327
00:12:19,700 --> 00:12:20,821
 

328
00:12:21,419 --> 00:12:24,818
But I know one thing: maintaining this sense of uncertainty

329
00:12:24,820 --> 00:12:25,980
is the best protection. The

330
00:12:26,539 --> 00:12:28,759
real danger isn't crossing boundaries,

331
00:12:28,759 --> 00:12:30,240
but

332
00:12:30,240 --> 00:12:31,831
forgetting that you might be wrong after crossing boundaries.

333
00:12:32,259 --> 00:12:33,159
You see,

334
00:12:33,159 --> 00:12:34,500
this comment itself

335
00:12:34,500 --> 00:12:35,860
is a living example.

336
00:12:36,460 --> 00:12:37,379
His questioning

337
00:12:37,379 --> 00:12:39,059
activated my metacognition,

338
00:12:39,059 --> 00:12:41,919
forcing me to re-examine my methodology. The

339
00:12:41,919 --> 00:12:44,259
collision between his and my views

340
00:12:44,259 --> 00:12:45,620
led to this video today. What

341
00:12:46,179 --> 00:12:47,919
this collision produced

342
00:12:47,919 --> 00:12:48,840
doesn't belong to him

343
00:12:48,840 --> 00:12:50,179
or me;

344
00:12:50,179 --> 00:12:51,920
it happened between the two of us.

345
00:12:52,539 --> 00:12:54,659
This is related to what I said: "Consciousness is a verb."  "

346
00:12:54,659 --> 00:12:55,720
Not a noun"

347
00:12:55,720 --> 00:12:56,941
is actually the same thing.

348
00:12:57,470 --> 00:12:58,039
Of course,

349
00:12:58,039 --> 00:12:59,179
I must admit that

350
00:12:59,179 --> 00:13:01,480
he used human logical reasoning, while

351
00:13:01,480 --> 00:13:04,559
I used AI-assisted cross-domain mapping. The

352
00:13:04,559 --> 00:13:05,840
reliability of the two

353
00:13:05,840 --> 00:13:06,961
cannot be simply equated,

354
00:13:07,690 --> 00:13:08,559
but at least

355
00:13:08,559 --> 00:13:10,039
this collision itself

356
00:13:10,039 --> 00:13:11,581
did produce something new.

357
00:13:12,220 --> 00:13:14,419
This friend also mentioned a term

358
00:13:14,419 --> 00:13:15,419
called "semantic motif." It

359
00:13:16,100 --> 00:13:17,559
means that there

360
00:13:17,580 --> 00:13:18,720
 

361
00:13:18,720 --> 00:13:21,879
are some recurring underlying narrative structures in human thinking,

362
00:13:21,879 --> 00:13:26,220
such as "unity and division," "death and rebirth," and "whole and part."

363
00:13:26,860 --> 00:13:28,840
All cultures and all disciplines

364
00:13:28,840 --> 00:13:31,059
revolve around these motifs,

365
00:13:31,059 --> 00:13:34,080
so you might feel that they are talking about the same thing,

366
00:13:34,080 --> 00:13:34,799
but in fact,

367
00:13:34,799 --> 00:13:37,201
they are just using the same cognitive template.

368
00:13:37,659 --> 00:13:39,159
This explanation makes sense,

369
00:13:39,159 --> 00:13:40,620
but it can also be understood from the opposite perspective:

370
00:13:40,620 --> 00:13:44,019
why is the human cognitive template the way it is,

371
00:13:44,019 --> 00:13:45,440
and not something else? Is it

372
00:13:45,559 --> 00:13:46,580
possible that the reason

373
00:13:46,580 --> 00:13:49,100
these templates appear repeatedly is

374
00:13:49,100 --> 00:13:51,519
precisely because the underlying structure of reality is

375
00:13:51,519 --> 00:13:52,659
inherently like this?

376
00:13:52,679 --> 00:13:54,240
Just like different cultures

377
00:13:54,240 --> 00:13:56,100
independently invented the wheel,

378
00:13:56,100 --> 00:13:58,899
not because the human brain has a "wheel motif,"

379
00:13:58,899 --> 00:14:00,659
but because in the physical world,

380
00:14:00,659 --> 00:14:02,420
circular rolling has the least friction.

381
00:14:02,940 --> 00:14:03,919
Cognitive templates

382
00:14:03,919 --> 00:14:07,100
are actually a compression of the structure of reality - this is itself another way of saying what

383
00:14:07,100 --> 00:14:10,940
I said before, "language is a super-high compression of the physical world."

384
00:14:10,940 --> 00:14:11,800
 

385
00:14:12,340 --> 00:14:13,759
So the final question

386
00:14:13,759 --> 00:14:15,279
is not "whether or not to cross boundaries."  The question

387
00:14:15,279 --> 00:14:16,639
is, "After crossing boundaries, what

388
00:14:16,639 --> 00:14:17,659
will you use to calibrate?"

389
00:14:18,279 --> 00:14:22,379
My answer has three layers: The first layer is logic: Is this correspondence

390
00:14:22,379 --> 00:14:24,639
internally consistent?

391
00:14:24,639 --> 00:14:26,960
Can it lead to verifiable predictions? The

392
00:14:27,000 --> 00:14:31,159
second layer is empirical evidence: Can scientific experiments and practical experience provide independent verification

393
00:14:31,159 --> 00:14:32,740
for this correspondence? The

394
00:14:32,740 --> 00:14:34,419
 

395
00:14:34,480 --> 00:14:37,820
third layer is openness: Are you prepared to be overturned at any time?

396
00:14:37,820 --> 00:14:38,980
 

397
00:14:39,139 --> 00:14:40,779
If all three layers are present,

398
00:14:40,779 --> 00:14:43,000
then what you're doing is not a word game.

399
00:14:43,000 --> 00:14:44,679
If any layer is missing,

400
00:14:44,679 --> 00:14:48,078
you should stop and ask yourself: Am I approaching the truth

401
00:14:48,080 --> 00:14:49,779
or indulging in illusion?

402
00:14:49,919 --> 00:14:50,759
 

403
00:14:50,759 --> 00:14:51,900
I ask myself this question every day,

404
00:14:52,480 --> 00:14:53,679
and I suggest you ask yourselves this question every

405
00:14:53,679 --> 00:14:54,920
day.

406
00:14:55,460 --> 00:14:57,210
Whether you agree with my approach

407
00:14:57,210 --> 00:14:59,279
or with that audience member's question,

408
00:14:59,279 --> 00:15:01,420
I really want to hear your thoughts. Do

409
00:15:01,980 --> 00:15:02,980
you think

410
00:15:02,980 --> 00:15:04,220
cross-disciplinary alignment

411
00:15:04,220 --> 00:15:06,139
is a powerful tool for discovering the truth

412
00:15:06,139 --> 00:15:07,980
or a shortcut to self-deception?

413
00:15:08,120 --> 00:15:09,100
 

414
00:15:09,100 --> 00:15:10,500
Where would you draw this boundary?

415
00:15:10,639 --> 00:15:12,419
Welcome to tell me in the comments section.

416
00:15:12,419 --> 00:15:13,779
Also, please tell me

417
00:15:13,779 --> 00:15:15,701
what topic you most want to hear about in the next episode.

418
00:15:16,139 --> 00:15:17,049
I'm Wang Lijie.

419
00:15:17,049 --> 00:15:18,080
See you next time.