1
00:00:02,720 --> 00:00:19,639
Speaker 1: Bloomberg Audio Studios, Podcasts, radio News. Hello and welcome to

2
00:00:19,680 --> 00:00:21,960
Speaker 1: another episode of The Odd Lots Podcast.

3
00:00:22,040 --> 00:00:24,480
Speaker 2: I'm Joe Wasental and I'm Tracy Alloway.

4
00:00:24,720 --> 00:00:26,880
Speaker 1: Tracy, I love being in Hong Kong. I love I

5
00:00:27,000 --> 00:00:29,040
Speaker 1: hear so much. I love it here so much. I

6
00:00:29,080 --> 00:00:30,800
Speaker 1: would like come here a few times a year if

7
00:00:30,840 --> 00:00:31,160
Speaker 1: we could.

8
00:00:31,440 --> 00:00:32,400
Speaker 2: I'm sure you would.

9
00:00:32,560 --> 00:00:36,320
Speaker 3: I've lived here for like, I guess, almost four years,

10
00:00:36,400 --> 00:00:40,640
Speaker 3: so it's kind of weird coming back. But Hong Kong

11
00:00:40,680 --> 00:00:43,880
Speaker 3: has a lot of pluses, like great food, great weather

12
00:00:44,080 --> 00:00:47,200
Speaker 3: for most of the year, beaches. I once heard someone

13
00:00:47,240 --> 00:00:50,159
Speaker 3: describe it as Manhattan meets Maui, which I think is

14
00:00:50,240 --> 00:00:51,519
Speaker 3: like pretty accurate.

15
00:00:51,760 --> 00:00:53,120
Speaker 1: Oh it's so nice.

16
00:00:53,680 --> 00:00:56,200
Speaker 3: The weather's shade, nugget, it's not great right now.

17
00:00:56,320 --> 00:00:58,520
Speaker 1: Like this week, we've come during a I guess it's

18
00:00:58,520 --> 00:00:59,960
Speaker 1: a monsoon season, right.

19
00:01:00,080 --> 00:01:03,000
Speaker 3: Yeah, it's the rainy season. But oh well, I like thunderstorms,

20
00:01:03,000 --> 00:01:03,920
Speaker 3: so I'm enjoying it.

21
00:01:04,040 --> 00:01:07,039
Speaker 1: Yeah, I'm enjoying it too. Anyway. One thing that has

22
00:01:07,160 --> 00:01:09,680
Speaker 1: changed since the last time you were in Hong Kong

23
00:01:10,400 --> 00:01:13,080
Speaker 1: you loved in twenty twenty two, Yeah, we weren't doing

24
00:01:13,080 --> 00:01:14,240
Speaker 1: as many AI episodes.

25
00:01:14,480 --> 00:01:18,000
Speaker 3: No, we definitely weren't. In fact, So I remember one

26
00:01:18,080 --> 00:01:20,280
Speaker 3: of the big stories when I was here in Hong

27
00:01:20,400 --> 00:01:24,399
Speaker 3: Kong in twenty twenty was China's tech crackdown, right.

28
00:01:24,319 --> 00:01:26,360
Speaker 1: That's right, that's right, right, And like there.

29
00:01:26,280 --> 00:01:28,960
Speaker 3: Was all this concern about whether or not the crackdown

30
00:01:29,080 --> 00:01:32,720
Speaker 3: was going to destroy China's entrepreneurial spirit. I'm doing air

31
00:01:32,840 --> 00:01:35,800
Speaker 3: quotes on a podcast. I don't know why, but fast

32
00:01:35,840 --> 00:01:42,360
Speaker 3: forward six years and there's entrepreneurism basically everywhere, and we

33
00:01:42,440 --> 00:01:45,800
Speaker 3: talk a lot about how China is producing all these

34
00:01:45,880 --> 00:01:47,440
Speaker 3: new AI models.

35
00:01:47,640 --> 00:01:51,320
Speaker 1: Okay, can I like say, like, I know very little.

36
00:01:51,680 --> 00:01:53,720
Speaker 1: I mean, I know very little about AI, but I

37
00:01:53,760 --> 00:01:56,600
Speaker 1: know even less about Chinese AI. But here are some

38
00:01:56,720 --> 00:02:01,120
Speaker 1: of my general impressions, which is, it seems like there's

39
00:02:01,200 --> 00:02:04,320
Speaker 1: so many open source Okay, so I know they're largely

40
00:02:04,360 --> 00:02:07,720
Speaker 1: open source. It seems like every random company you see,

41
00:02:07,760 --> 00:02:11,160
Speaker 1: like some toothpaste company, and they'll have produced an LLM.

42
00:02:11,240 --> 00:02:14,560
Speaker 1: So I'm very curious, like how they're making money on it.

43
00:02:15,200 --> 00:02:16,960
Speaker 1: I also get the impression and like, do you know

44
00:02:17,000 --> 00:02:19,720
Speaker 1: the heads of American AI labs speak in these like

45
00:02:19,760 --> 00:02:23,560
Speaker 1: sort of quasi mystical terms, et cetera. It doesn't feel

46
00:02:23,840 --> 00:02:26,080
Speaker 1: quite the same here where it feels like a bit

47
00:02:26,120 --> 00:02:28,920
Speaker 1: more of like yet another technology. But I'm glad you

48
00:02:28,960 --> 00:02:31,360
Speaker 1: brought up the point about the tech crackdown because at

49
00:02:31,400 --> 00:02:34,000
Speaker 1: the time, the whole story was like, oh, there needs

50
00:02:34,040 --> 00:02:36,160
Speaker 1: to be less focus on sort of digital tech and

51
00:02:36,200 --> 00:02:39,280
Speaker 1: more focused on card tech, which has been done extremely

52
00:02:39,680 --> 00:02:43,440
Speaker 1: that's been an extraordinarily successful endeavor. And then my last

53
00:02:43,440 --> 00:02:46,320
Speaker 1: impression though, is that since the release of JADGPT in

54
00:02:46,400 --> 00:02:49,440
Speaker 1: late twenty twenty two, that was the moment it's like, no,

55
00:02:49,520 --> 00:02:53,760
Speaker 1: we really have to also compete on sort of this

56
00:02:53,919 --> 00:02:57,720
Speaker 1: next era of software and sort of consumer facing tech

57
00:02:57,880 --> 00:02:59,280
Speaker 1: tech breakthroughs. Yeah.

58
00:02:59,320 --> 00:03:02,360
Speaker 3: Oh, for all that AI scene in China feels much

59
00:03:02,360 --> 00:03:06,639
Speaker 3: more utilitarian to me. It's more about like the big companies,

60
00:03:06,720 --> 00:03:09,520
Speaker 3: the ten cents, the Ali Baba is sort of using

61
00:03:09,680 --> 00:03:15,160
Speaker 3: AI for their existing business models rather than this existential thing,

62
00:03:15,200 --> 00:03:17,960
Speaker 3: which it is in the US, where like AI is

63
00:03:18,160 --> 00:03:19,720
Speaker 3: the business that's just.

64
00:03:19,680 --> 00:03:23,040
Speaker 1: It, right, Yeah, that's exactly. AI is sort of weird,

65
00:03:23,120 --> 00:03:24,880
Speaker 1: like it sort of sits in the middle of what

66
00:03:24,919 --> 00:03:27,560
Speaker 1: you would call like software and hard tech because we

67
00:03:27,600 --> 00:03:30,919
Speaker 1: consume it through the browser, right, sort of the same way,

68
00:03:31,080 --> 00:03:33,600
Speaker 1: or in many cases through the browser the same way

69
00:03:33,639 --> 00:03:35,840
Speaker 1: that we would go to an Amazon or online gaming

70
00:03:35,960 --> 00:03:38,360
Speaker 1: or something like that. But it's clearly, you know, it's

71
00:03:38,400 --> 00:03:41,840
Speaker 1: a scientific endeavor, and so it's sort of is this

72
00:03:41,960 --> 00:03:44,280
Speaker 1: blend And then you have to figure China's so far

73
00:03:44,320 --> 00:03:46,000
Speaker 1: ahead of the US when it comes to things like

74
00:03:46,120 --> 00:03:49,320
Speaker 1: robotics and evs and batteries, and one thing I don't

75
00:03:49,360 --> 00:03:52,520
Speaker 1: know anything about is the degree to which that melding

76
00:03:52,680 --> 00:03:57,280
Speaker 1: of hardware capabilities with AI capabilities, how that influences the

77
00:03:57,360 --> 00:04:00,000
Speaker 1: direction of the developments of the AI.

78
00:04:00,440 --> 00:04:03,680
Speaker 3: Yeah, I'm also very interested in, like the capital stack

79
00:04:03,800 --> 00:04:06,280
Speaker 3: for Chinese companies, because over in the US, we all

80
00:04:06,320 --> 00:04:09,160
Speaker 3: know that people are flinging money at anything with the

81
00:04:09,200 --> 00:04:13,240
Speaker 3: word AI in it, but in China it's very different.

82
00:04:13,520 --> 00:04:15,920
Speaker 3: I get the impression that it's like much harder to

83
00:04:16,040 --> 00:04:19,760
Speaker 3: raise enormous sums of capital, and so I'm very curious

84
00:04:19,800 --> 00:04:24,360
Speaker 3: how that limited capital actually influences the development of these

85
00:04:24,400 --> 00:04:25,280
Speaker 3: models and the tech.

86
00:04:25,600 --> 00:04:27,440
Speaker 1: I think it's safe to say that both of us

87
00:04:27,720 --> 00:04:31,960
Speaker 1: have a lot of impressions. Yes, right, impression Manytimes that

88
00:04:32,000 --> 00:04:33,760
Speaker 1: its intro is like, I get the impression, but I

89
00:04:33,760 --> 00:04:36,360
Speaker 1: actually have no idea. So that is a good reason

90
00:04:36,640 --> 00:04:39,120
Speaker 1: to actually bring in our guest, someone who is more

91
00:04:39,160 --> 00:04:42,600
Speaker 1: than quote impressions unquote about the AI tech scene. We're

92
00:04:42,640 --> 00:04:44,960
Speaker 1: going to be speaking to someone whose newsletter I'm a

93
00:04:44,960 --> 00:04:47,280
Speaker 1: big fan of and everyone should read. Are going to

94
00:04:47,320 --> 00:04:51,039
Speaker 1: be speaking to the perfect guest, Grace Shao. She's an independent,

95
00:04:51,080 --> 00:04:55,120
Speaker 1: a researcher, and she has a great substack called AI Prome,

96
00:04:55,200 --> 00:04:57,760
Speaker 1: and she joined us here in our Hong Kong office.

97
00:04:57,760 --> 00:05:00,120
Speaker 1: So Grace, thank you so much for coming on od Lave.

98
00:05:00,360 --> 00:05:02,040
Speaker 2: Thank you so much for having me, Joe and Tracy.

99
00:05:02,200 --> 00:05:05,159
Speaker 1: How did we do on our impression? Quote?

100
00:05:05,360 --> 00:05:07,320
Speaker 2: Those are pretty accurate impressions? I think?

101
00:05:07,320 --> 00:05:10,560
Speaker 3: Okay, good, that was the episode. Like, let's start at

102
00:05:10,600 --> 00:05:13,600
Speaker 3: a basic level. So the big impression, the one that

103
00:05:13,640 --> 00:05:17,919
Speaker 3: everyone knows is Chinese models are open source versus the

104
00:05:18,000 --> 00:05:23,360
Speaker 3: closed frontier models of the US. Why did it develop

105
00:05:23,440 --> 00:05:23,840
Speaker 3: that way?

106
00:05:24,720 --> 00:05:28,640
Speaker 4: Yeah, I think people like to think of these myscal reasons,

107
00:05:29,080 --> 00:05:31,480
Speaker 4: but really it was a very pragmatic business reason to

108
00:05:31,520 --> 00:05:34,160
Speaker 4: start with. To start with, a lot of the labs

109
00:05:34,160 --> 00:05:38,520
Speaker 4: have cited that, you know, for Western companies or Western

110
00:05:38,560 --> 00:05:41,680
Speaker 4: developers to trust them, they needed to open source their

111
00:05:41,720 --> 00:05:44,600
Speaker 4: models to build that trust and credibility. So in many ways,

112
00:05:44,640 --> 00:05:48,200
Speaker 4: it's a branding decision. Then on top of that, I

113
00:05:48,240 --> 00:05:50,640
Speaker 4: think you can see it as a philosophical drive. You know,

114
00:05:50,720 --> 00:05:53,680
Speaker 4: the founder of deep Seagalmen Fung, has openly said he

115
00:05:53,720 --> 00:05:56,400
Speaker 4: wants open source his most frontier research to really help

116
00:05:56,400 --> 00:05:59,040
Speaker 4: propel the whole industry as a whole, and that kind

117
00:05:59,040 --> 00:06:02,039
Speaker 4: of R and D sharing has now formed a layer

118
00:06:02,480 --> 00:06:05,000
Speaker 4: for the whole ecosystem where each of the labs kind

119
00:06:05,040 --> 00:06:08,320
Speaker 4: of integrate each other's kind of breakthroughs. You know, you

120
00:06:08,360 --> 00:06:10,840
Speaker 4: see them congratulate each other even on X when they

121
00:06:10,880 --> 00:06:13,600
Speaker 4: have new models announced, so you can say it's a

122
00:06:13,640 --> 00:06:17,000
Speaker 4: bit more collegial. I wouldn't say they're not competing though, however,

123
00:06:17,400 --> 00:06:20,239
Speaker 4: because of the compute constraint, they're faced with, talent constraint

124
00:06:20,279 --> 00:06:23,040
Speaker 4: and the capital constraint even mentioned, they are a lot

125
00:06:23,080 --> 00:06:25,440
Speaker 4: more conscious with where they want to put their money,

126
00:06:25,600 --> 00:06:27,279
Speaker 4: where they want to put their time in R and

127
00:06:27,320 --> 00:06:30,200
Speaker 4: D and all of that forms the basis of a

128
00:06:30,240 --> 00:06:32,160
Speaker 4: strong open source ecosystem.

129
00:06:32,760 --> 00:06:36,920
Speaker 1: Is the culture as pro sharing and pro open source

130
00:06:37,279 --> 00:06:41,600
Speaker 1: as it was even two years ago now. The Deep

131
00:06:41,640 --> 00:06:45,080
Speaker 1: Seek moment was right around Trump's inauguration in early twenty

132
00:06:45,120 --> 00:06:47,680
Speaker 1: twenty five, so about a year and a half ago.

133
00:06:48,480 --> 00:06:52,760
Speaker 1: Since then, has the culture stayed the same or has

134
00:06:52,880 --> 00:06:56,640
Speaker 1: that sort of competition bug, that intense competition bug that

135
00:06:56,680 --> 00:07:01,240
Speaker 1: we know among American AI labs, has it spread to

136
00:07:01,320 --> 00:07:02,640
Speaker 1: the Chinese labs at all.

137
00:07:03,400 --> 00:07:07,520
Speaker 4: I think the sharing is an unintentional result rather than

138
00:07:07,720 --> 00:07:10,440
Speaker 4: you know, an intentional effort in the beginning to even

139
00:07:10,480 --> 00:07:13,680
Speaker 4: start with. They are for sure extremely competitive, and you know,

140
00:07:13,720 --> 00:07:17,200
Speaker 4: we all know the word involution, so like China Ai

141
00:07:17,520 --> 00:07:20,280
Speaker 4: is do adding as well, that means like there's evolution

142
00:07:20,360 --> 00:07:23,640
Speaker 4: in this ecosystem as well. However, I think bringing up

143
00:07:23,640 --> 00:07:26,360
Speaker 4: deep sea deep sea plays a very interesting role in

144
00:07:26,400 --> 00:07:30,000
Speaker 4: the whole ecosystem. Like you mentioned, V three, propel the

145
00:07:30,040 --> 00:07:32,840
Speaker 4: whole industry forward. Everyone kind of start taking China Ai

146
00:07:32,920 --> 00:07:36,160
Speaker 4: more seriously. You know, it brought a lot of interests

147
00:07:36,200 --> 00:07:39,520
Speaker 4: from investors globally back into the internet companies that Tracy mentioned,

148
00:07:39,560 --> 00:07:41,280
Speaker 4: you know, project that there was a bit of a

149
00:07:41,320 --> 00:07:45,560
Speaker 4: slump for three to five years. However, you know, Jiarpoolzai

150
00:07:45,640 --> 00:07:47,880
Speaker 4: is now public listed in Hong Kong, Mini Max is

151
00:07:47,880 --> 00:07:50,520
Speaker 4: pubably listed in Hong Kong. Moonshot is you know, in

152
00:07:50,600 --> 00:07:54,840
Speaker 4: preparation to go public next year. They are competing with

153
00:07:54,920 --> 00:07:58,800
Speaker 4: each other to capture market share, to capture developer mind share.

154
00:07:59,200 --> 00:08:01,400
Speaker 4: But deep Sea plays an interesting role. I want to

155
00:08:01,400 --> 00:08:05,200
Speaker 4: bring it back to DEEPCV four, so V four, you know,

156
00:08:05,320 --> 00:08:07,800
Speaker 4: on the surface, you know, people said, Okay, it wasn't

157
00:08:07,840 --> 00:08:11,240
Speaker 4: as maybe impressive on evails and performance, They didn't catch

158
00:08:11,320 --> 00:08:12,400
Speaker 4: up with the most front of your labs in the

159
00:08:12,480 --> 00:08:16,800
Speaker 4: US whatnot, Right, But it was a very interesting move

160
00:08:16,920 --> 00:08:19,600
Speaker 4: because what I heard from researchers on the ground in

161
00:08:19,640 --> 00:08:24,160
Speaker 4: Beijing was that the lab actually delayed their release for

162
00:08:24,200 --> 00:08:26,440
Speaker 4: about three to four months because they wanted to re

163
00:08:26,520 --> 00:08:30,120
Speaker 4: engineer a lot of the inference onto Huawei. So I'm

164
00:08:30,120 --> 00:08:34,600
Speaker 4: not saying this completely replaces in video or Kuda, not

165
00:08:34,760 --> 00:08:37,120
Speaker 4: at all, because if you ask any developers, they still

166
00:08:37,160 --> 00:08:40,400
Speaker 4: want to use Kuda if they can. However, it was

167
00:08:40,440 --> 00:08:43,800
Speaker 4: the first effort to really I kind of like did

168
00:08:43,840 --> 00:08:46,200
Speaker 4: one for the team, like they kind of like hunt them.

169
00:08:46,480 --> 00:08:49,800
Speaker 3: It's supposed to be like a signal basically, yeah, we're

170
00:08:49,840 --> 00:08:52,480
Speaker 3: doing this all on like a Chinese stack. Yeah.

171
00:08:52,520 --> 00:08:54,520
Speaker 4: They were like, look, guys, like you can actually do this,

172
00:08:54,679 --> 00:08:58,320
Speaker 4: and they became a shared foundation layer for China's model ecosystem. So,

173
00:08:58,640 --> 00:09:02,160
Speaker 4: because again everything is open source and open weight, other

174
00:09:02,320 --> 00:09:05,120
Speaker 4: labs were able to study what they did to actually

175
00:09:05,280 --> 00:09:09,120
Speaker 4: start inferencing on Huawei stack, and I think that was

176
00:09:09,160 --> 00:09:12,200
Speaker 4: the first step, whether it's signaling or actually you know,

177
00:09:12,440 --> 00:09:17,440
Speaker 4: very pragmatic reason to start shifting some reliance on you know,

178
00:09:17,520 --> 00:09:18,880
Speaker 4: the China AI.

179
00:09:18,679 --> 00:09:20,600
Speaker 2: Stack aside from deep Sea.

180
00:09:20,679 --> 00:09:24,520
Speaker 3: Can you kind of describe the differences or what China

181
00:09:24,640 --> 00:09:27,679
Speaker 3: is trying to do on the actual frontier side, because

182
00:09:27,679 --> 00:09:29,480
Speaker 3: there are there are some I.

183
00:09:29,720 --> 00:09:31,520
Speaker 4: Think if you really have to have to look at

184
00:09:31,559 --> 00:09:34,840
Speaker 4: the ecosystem, we can kind of put side the big

185
00:09:34,880 --> 00:09:37,600
Speaker 4: tech for now, but looking at the maybe the foremost

186
00:09:37,640 --> 00:09:42,200
Speaker 4: relevant startup labs Deep Seak, Moonshot, who has Kimmi, Zia

187
00:09:42,280 --> 00:09:46,319
Speaker 4: who has GLM, and then Mini Max. They are still

188
00:09:46,360 --> 00:09:50,720
Speaker 4: probably the most committed to frontier research. However, because the

189
00:09:51,040 --> 00:09:53,760
Speaker 4: constraint we mentioned that they face, whether it's compete when

190
00:09:53,800 --> 00:09:57,160
Speaker 4: it is capital or even friendly talent, they have decided

191
00:09:57,280 --> 00:09:59,959
Speaker 4: out of necessity to basically each focus on a different

192
00:10:00,240 --> 00:10:03,640
Speaker 4: vertical in capturing a different kind of business share. So

193
00:10:04,160 --> 00:10:08,080
Speaker 4: Zi is very focused on coding capabilities. So if anything,

194
00:10:08,120 --> 00:10:10,440
Speaker 4: their gl AND plan is much more similar to maybe

195
00:10:10,440 --> 00:10:12,840
Speaker 4: what you think of Claude cloud co wor cloud code

196
00:10:12,840 --> 00:10:15,760
Speaker 4: et cetera, a codex that kind of product. And then

197
00:10:15,760 --> 00:10:20,679
Speaker 4: you look at Minimax, they're really focused on the multimodality capabilities, Moonshot,

198
00:10:20,720 --> 00:10:23,560
Speaker 4: they're really focused on agents and deep sea.

199
00:10:23,679 --> 00:10:24,000
Speaker 2: Again.

200
00:10:24,200 --> 00:10:27,200
Speaker 4: Really is just focused on pushing the frontier and trying

201
00:10:27,240 --> 00:10:30,520
Speaker 4: to play catch up and push the Chinese ecosystem as

202
00:10:30,559 --> 00:10:31,440
Speaker 4: fast as possible.

203
00:10:32,160 --> 00:10:34,960
Speaker 1: It's really crazy to look at some of the ones

204
00:10:35,080 --> 00:10:38,040
Speaker 1: that they have already gone public here and just put

205
00:10:38,080 --> 00:10:42,240
Speaker 1: in So Minimax is public and in US dollar terms,

206
00:10:42,800 --> 00:10:46,040
Speaker 1: it's a twenty billion dollar company. I mean there are

207
00:10:46,160 --> 00:10:49,319
Speaker 1: people in the US who have done nothing but publish

208
00:10:49,360 --> 00:10:52,520
Speaker 1: a paper on archive dot org, who do not even

209
00:10:52,559 --> 00:10:56,160
Speaker 1: have a product yet, who have probably VC backed applied

210
00:10:56,280 --> 00:10:59,400
Speaker 1: valuations of our twenty billion dollars. How do they make money?

211
00:10:59,480 --> 00:11:02,480
Speaker 1: You know again and open source? Okay? Like in these

212
00:11:02,520 --> 00:11:05,600
Speaker 1: four models that you name, do they have different thoughts

213
00:11:05,720 --> 00:11:08,920
Speaker 1: on how they plan to make money or different business models?

214
00:11:09,480 --> 00:11:09,800
Speaker 2: Yeah?

215
00:11:09,880 --> 00:11:14,160
Speaker 4: So China's VC space in general has not been that vibrant,

216
00:11:14,200 --> 00:11:17,080
Speaker 4: frankly since Internet crackdown and a lot of USD funds

217
00:11:17,080 --> 00:11:19,840
Speaker 4: did exit you know, three to five years ago, with

218
00:11:19,920 --> 00:11:22,439
Speaker 4: Saquoia being maybe the most like high profile, right, like

219
00:11:22,480 --> 00:11:25,800
Speaker 4: we all remember that. Now, people forget even in twenty

220
00:11:25,800 --> 00:11:27,640
Speaker 4: twenty two, a lot of these labs that we just

221
00:11:27,679 --> 00:11:30,560
Speaker 4: talked about, they were struggling to even raise money, you know,

222
00:11:30,640 --> 00:11:33,240
Speaker 4: raise capital, and a lot of them spun out of

223
00:11:33,320 --> 00:11:36,920
Speaker 4: academic institutions. You know, you mentioned their value anywhere roughly

224
00:11:36,920 --> 00:11:39,120
Speaker 4: between twenty to thirty billion right now, but they went

225
00:11:39,160 --> 00:11:41,880
Speaker 4: public between like six to eight billion. That's like kind

226
00:11:41,920 --> 00:11:46,280
Speaker 4: of tiny compared to American valuations. Right now, however, they

227
00:11:46,320 --> 00:11:49,800
Speaker 4: are actually making money, you know, the publicly disclosed information

228
00:11:49,960 --> 00:11:52,320
Speaker 4: I think from mini max and DRUP who indicates that

229
00:11:52,440 --> 00:11:56,200
Speaker 4: they were making just as much like in their last month.

230
00:11:56,280 --> 00:11:58,559
Speaker 4: They've made the same amount of money last month as

231
00:11:58,559 --> 00:12:01,679
Speaker 4: they did last year essentially, and their end of your

232
00:12:02,000 --> 00:12:04,719
Speaker 4: AR projection is anywhere between one to one point two

233
00:12:04,760 --> 00:12:05,560
Speaker 4: billion right now.

234
00:12:05,960 --> 00:12:07,240
Speaker 2: So they are making money.

235
00:12:07,240 --> 00:12:10,600
Speaker 4: And how well, just because their open source doesn't mean

236
00:12:10,600 --> 00:12:12,760
Speaker 4: they don't make money. I think people forget. You know,

237
00:12:12,920 --> 00:12:15,880
Speaker 4: we had open source softwares before as well. People are

238
00:12:15,880 --> 00:12:19,880
Speaker 4: paying for managed services, and when you're paying for an

239
00:12:19,880 --> 00:12:24,240
Speaker 4: API through that AI or minimax whatnot, you basically don't

240
00:12:24,240 --> 00:12:25,600
Speaker 4: have to self host, you don't have to get your

241
00:12:25,600 --> 00:12:27,280
Speaker 4: own GPU. You don't have to get you figure out

242
00:12:27,280 --> 00:12:29,160
Speaker 4: your gown compute. You don't have to figure out your

243
00:12:29,200 --> 00:12:31,960
Speaker 4: own guardrails or deployment, your security or monitoring whatnot.

244
00:12:32,040 --> 00:12:36,679
Speaker 1: Right, So, just to be clear, you can self host

245
00:12:36,760 --> 00:12:40,360
Speaker 1: all of these models, but for the most part, they

246
00:12:40,400 --> 00:12:43,200
Speaker 1: do offer the inference part of the stack, and that

247
00:12:43,440 --> 00:12:44,960
Speaker 1: is a profit center for them.

248
00:12:45,280 --> 00:13:02,679
Speaker 2: Yes, exactly, got it all right.

249
00:13:02,720 --> 00:13:07,360
Speaker 3: So there seem to be two major constraints on Chinese AI.

250
00:13:07,520 --> 00:13:09,720
Speaker 3: Maybe energy is a constraint as well, we should talk

251
00:13:09,760 --> 00:13:14,679
Speaker 3: about that. But there's the capital issue, so not as

252
00:13:14,760 --> 00:13:18,480
Speaker 3: much capital available or people aren't flinging it at AI

253
00:13:18,600 --> 00:13:20,840
Speaker 3: companies the way they are in the US. And then secondly,

254
00:13:20,880 --> 00:13:23,760
Speaker 3: there are the export controls on chips and we talked

255
00:13:23,760 --> 00:13:26,280
Speaker 3: about that a little bit, but can you describe how

256
00:13:26,360 --> 00:13:31,120
Speaker 3: those controls are actually I guess influencing the development of

257
00:13:31,200 --> 00:13:34,439
Speaker 3: the models themselves and like I guess optimization.

258
00:13:35,120 --> 00:13:37,480
Speaker 4: So some of these big tech are actually even buying

259
00:13:37,480 --> 00:13:40,160
Speaker 4: out the contracts that data centers have with some of

260
00:13:40,160 --> 00:13:42,600
Speaker 4: these labs, and you know they are taking over compute.

261
00:13:42,600 --> 00:13:47,720
Speaker 4: So these labs essentially are now optimizing for the highest

262
00:13:47,800 --> 00:13:50,319
Speaker 4: quality inference amount, if that makes sense. They don't actually

263
00:13:50,400 --> 00:13:53,400
Speaker 4: have enough even supply, they don't have enough compute power

264
00:13:53,400 --> 00:13:56,600
Speaker 4: to even like meet the demand that's coming through. So

265
00:13:56,720 --> 00:13:59,720
Speaker 4: that's on like how they're servicing clients, how they're changing

266
00:13:59,760 --> 00:14:02,120
Speaker 4: I guess, or how they're optimizing O their training is

267
00:14:02,120 --> 00:14:04,680
Speaker 4: that a lot of them are really focused on the

268
00:14:04,720 --> 00:14:08,400
Speaker 4: post training. And this goes back to so you know

269
00:14:08,440 --> 00:14:10,480
Speaker 4: how open ai has like three buckets where they check

270
00:14:10,520 --> 00:14:12,800
Speaker 4: money at there's like the R and D, there's pre training,

271
00:14:12,800 --> 00:14:15,240
Speaker 4: this post training R and D. A lot of times

272
00:14:15,280 --> 00:14:17,480
Speaker 4: a lot of money is spent, but say like one

273
00:14:17,520 --> 00:14:19,680
Speaker 4: out of ten things stick, but you need a lot

274
00:14:19,680 --> 00:14:22,560
Speaker 4: of compute and resource some people to be figuring out

275
00:14:22,600 --> 00:14:25,520
Speaker 4: where to go. For a lot of these labs in China,

276
00:14:25,720 --> 00:14:28,840
Speaker 4: they frankly don't have that luxury. So they've even given

277
00:14:28,840 --> 00:14:31,400
Speaker 4: me a metaphor and said it's kind of like knowing

278
00:14:31,440 --> 00:14:34,840
Speaker 4: what the answer to the homeworks and working backwards. So

279
00:14:34,880 --> 00:14:37,320
Speaker 4: they will wait till the frontier labs to come out

280
00:14:37,360 --> 00:14:40,600
Speaker 4: with where the right direction is for the next frontier model,

281
00:14:40,800 --> 00:14:43,240
Speaker 4: and they will work backwards and actually focus all their

282
00:14:43,280 --> 00:14:46,680
Speaker 4: resources on post training. So with post training, they will

283
00:14:46,680 --> 00:14:49,880
Speaker 4: optimize a lot of times the data they collect. For example,

284
00:14:50,000 --> 00:14:53,720
Speaker 4: if a data provider like Mercore provides a very very

285
00:14:53,800 --> 00:14:57,560
Speaker 4: niche set of data set for like an open AI whatnot,

286
00:14:58,240 --> 00:15:01,040
Speaker 4: Maybe they will charge them ten twenty mille dollars. The

287
00:15:01,160 --> 00:15:05,200
Speaker 4: Chinese lab will weigh out that exclusivity contract three to

288
00:15:05,240 --> 00:15:07,600
Speaker 4: six months time, let's say, and then pay a fraction

289
00:15:07,720 --> 00:15:09,520
Speaker 4: if not like a tenth of that price the same

290
00:15:09,600 --> 00:15:11,880
Speaker 4: data set, and that kind of plays into that like

291
00:15:11,960 --> 00:15:14,480
Speaker 4: six to nine month leg that we hear about as well.

292
00:15:14,680 --> 00:15:18,120
Speaker 3: That's really interesting. Let's talk about energy then, because the

293
00:15:18,200 --> 00:15:21,240
Speaker 3: story in the US is that electricity is really the

294
00:15:21,240 --> 00:15:24,280
Speaker 3: big constraint on AI use, and you know, you got

295
00:15:24,320 --> 00:15:27,400
Speaker 3: to find a data center that has an electricity hookup,

296
00:15:27,440 --> 00:15:29,480
Speaker 3: and it has to be reliant and all of that.

297
00:15:30,000 --> 00:15:32,560
Speaker 3: It seems to be in short supply. Is it a

298
00:15:32,640 --> 00:15:34,640
Speaker 3: similar story in China?

299
00:15:34,720 --> 00:15:38,240
Speaker 4: Honestly, energy is probably not the biggest bottleneck right now

300
00:15:38,280 --> 00:15:40,880
Speaker 4: in China. And I think people like to say, well,

301
00:15:40,920 --> 00:15:43,320
Speaker 4: some people like to say, oh, somehow the Chinese government

302
00:15:43,360 --> 00:15:46,880
Speaker 4: had foresight on the AI boom, driving like the energy consumption,

303
00:15:46,960 --> 00:15:50,680
Speaker 4: but definitely not. I think people forget that China's economic

304
00:15:50,760 --> 00:15:52,920
Speaker 4: growth over the last three to four decades also meant

305
00:15:53,040 --> 00:15:56,200
Speaker 4: a rise of urbanization and a lot of the cities

306
00:15:56,240 --> 00:15:58,080
Speaker 4: that you know, we are visiting these days, like at

307
00:15:58,120 --> 00:16:00,720
Speaker 4: least westerns are visiting, like Beijing, China High Engine with

308
00:16:00,760 --> 00:16:03,800
Speaker 4: all these robots and evs or whatnot. These were all

309
00:16:03,920 --> 00:16:08,240
Speaker 4: really urbanized within the last two three decades. And because

310
00:16:08,240 --> 00:16:10,720
Speaker 4: of that, the grid is very new, and because of that,

311
00:16:10,920 --> 00:16:13,720
Speaker 4: the government already foresaw that there was going to be

312
00:16:13,880 --> 00:16:17,000
Speaker 4: a increase in energy demand, and you know, so a

313
00:16:17,040 --> 00:16:20,880
Speaker 4: lot of the energy plants, you know, the solar plants,

314
00:16:20,960 --> 00:16:23,760
Speaker 4: hydro plants, whatnot, were actually built out in you know,

315
00:16:23,800 --> 00:16:28,000
Speaker 4: anticipation for that. Now obviously this has coincided with now

316
00:16:28,040 --> 00:16:30,960
Speaker 4: the AI boom, and it's really helped out beyond that.

317
00:16:31,040 --> 00:16:33,920
Speaker 4: You know, China has an advantage in the fact that

318
00:16:33,960 --> 00:16:37,840
Speaker 4: they can actually drive top down mandates and provincial governments

319
00:16:37,880 --> 00:16:40,840
Speaker 4: will follow suit. This is something quite unique to China

320
00:16:40,920 --> 00:16:43,640
Speaker 4: because it's not like decided about each state. So when

321
00:16:43,680 --> 00:16:46,160
Speaker 4: they pushed out the East State to West compute where

322
00:16:46,320 --> 00:16:49,240
Speaker 4: it's basically a top down initiative where they built a

323
00:16:49,360 --> 00:16:54,200
Speaker 4: ton of renewable energy for cheap in rural mountainous areas

324
00:16:54,280 --> 00:16:59,280
Speaker 4: in guat Zo Province like even Sinzong, Inner Mongolia Suchwan

325
00:16:59,360 --> 00:17:03,720
Speaker 4: you know, those were like very easily executed, frankly, and

326
00:17:03,720 --> 00:17:06,840
Speaker 4: then ninety percent of the population actually sit on the

327
00:17:06,880 --> 00:17:10,760
Speaker 4: eastern coastal lines, like we think about Beijing, Tianjing, Shanghai,

328
00:17:10,760 --> 00:17:13,040
Speaker 4: sh Engine, that's all on east, so that's where the

329
00:17:13,160 --> 00:17:17,240
Speaker 4: data comes from. So that kind of optimization has also

330
00:17:17,320 --> 00:17:20,000
Speaker 4: really helped them, you know, with the low that is

331
00:17:20,359 --> 00:17:22,159
Speaker 4: the demand right now, I want to.

332
00:17:22,119 --> 00:17:24,680
Speaker 1: Get back to something you said. So first of all,

333
00:17:24,840 --> 00:17:28,920
Speaker 1: just to clarify, you mentioned companies like Mercore that sell

334
00:17:29,040 --> 00:17:33,360
Speaker 1: proprietary data that they are able to collect and manufacture

335
00:17:33,440 --> 00:17:36,080
Speaker 1: in various ways. Then they sell it to an open

336
00:17:36,119 --> 00:17:38,720
Speaker 1: aira So a company like Merkcore will hire a bunch

337
00:17:38,800 --> 00:17:41,640
Speaker 1: of people to say, build power points, and then they'll

338
00:17:41,640 --> 00:17:43,720
Speaker 1: collect the data on how they do that, and that

339
00:17:43,840 --> 00:17:46,399
Speaker 1: is fresh data that they can sell. So those have

340
00:17:47,160 --> 00:17:50,679
Speaker 1: exclusivity windows, after which they can then sell them to anyone.

341
00:17:50,960 --> 00:17:54,240
Speaker 4: I'm not saying Marcos specifically, but supposedly there are these

342
00:17:54,280 --> 00:17:57,080
Speaker 4: data providers that do the cell and they have exclusivity windows,

343
00:17:57,119 --> 00:17:59,520
Speaker 4: and then the Chinese labs kind of weigh that out

344
00:17:59,600 --> 00:18:02,679
Speaker 4: so they can pay like maybe a million dollars versus

345
00:18:02,720 --> 00:18:04,399
Speaker 4: like time and folk same data side.

346
00:18:04,480 --> 00:18:07,840
Speaker 1: So This gives to something generally speaking, which is that

347
00:18:08,080 --> 00:18:12,640
Speaker 1: people are around the world correctly like quite impressed by

348
00:18:12,960 --> 00:18:16,200
Speaker 1: how high quality the Chinese models are, even if they're behind.

349
00:18:16,680 --> 00:18:18,359
Speaker 1: But then you have things like that, and then you

350
00:18:18,440 --> 00:18:21,720
Speaker 1: also have accusations from the likes of Anthropic that they're

351
00:18:21,720 --> 00:18:25,640
Speaker 1: distilling models and that they're finding ways to collect the

352
00:18:25,680 --> 00:18:31,560
Speaker 1: outputs of American models for training. So then you could say, well, yeah, sure,

353
00:18:31,760 --> 00:18:34,040
Speaker 1: that's like, it's great this open source model and it

354
00:18:34,040 --> 00:18:36,800
Speaker 1: can stay close to the edge. But then the encounter

355
00:18:36,920 --> 00:18:40,359
Speaker 1: is that they can only be so advanced because there's

356
00:18:40,480 --> 00:18:44,399
Speaker 1: this extremely capital intensive closed source model in the US

357
00:18:44,880 --> 00:18:49,359
Speaker 1: that's really establishing the frontier, and that these Chinese companies

358
00:18:49,359 --> 00:18:52,160
Speaker 1: wouldn't be anywhere near where they were if they weren't

359
00:18:52,200 --> 00:18:54,600
Speaker 1: sort of I guess you would say drafting off the

360
00:18:54,640 --> 00:18:55,479
Speaker 1: American labs.

361
00:18:56,040 --> 00:18:58,600
Speaker 4: Yeah, I think the compute constraint on the capital constraint

362
00:18:58,640 --> 00:19:01,840
Speaker 4: Israel and frankly, like no one's hiding that or pretending

363
00:19:01,880 --> 00:19:04,040
Speaker 4: that that's not an issue for them right now, like

364
00:19:04,119 --> 00:19:07,040
Speaker 4: deep Sea has openly said they even were struggling right

365
00:19:07,080 --> 00:19:09,919
Speaker 4: like they needed more compute. I think on the distillation

366
00:19:10,400 --> 00:19:14,080
Speaker 4: allegations or accusation it is quite interesting. Like recently, I've

367
00:19:14,080 --> 00:19:16,560
Speaker 4: been thinking about this a lot and thinking about what

368
00:19:16,600 --> 00:19:19,000
Speaker 4: it means for distillation and what it means for the

369
00:19:19,160 --> 00:19:22,280
Speaker 4: models to catch up. Right. So, there was this one

370
00:19:22,600 --> 00:19:25,080
Speaker 4: quote from y'all showing you who is a Google Deep

371
00:19:25,119 --> 00:19:29,480
Speaker 4: Mind researcher. He said, there're smart distillation and dumb disolation.

372
00:19:29,880 --> 00:19:32,840
Speaker 4: Dumb distillation is something I think most of us were

373
00:19:32,880 --> 00:19:35,280
Speaker 4: frankly non technical think about. It's like, okay, you take

374
00:19:35,320 --> 00:19:38,119
Speaker 4: like a thousand queries, you take the answers of whatever

375
00:19:38,200 --> 00:19:40,080
Speaker 4: claud gives you, right, and then you kind of force

376
00:19:40,160 --> 00:19:43,520
Speaker 4: copy that into your said model, and then you forcefully

377
00:19:43,720 --> 00:19:46,760
Speaker 4: make them basically like get the exact same answer. Smart

378
00:19:46,760 --> 00:19:49,960
Speaker 4: distillation is like you using the frontier model almost as

379
00:19:50,000 --> 00:19:52,720
Speaker 4: a partner to help you with the judgment, for the

380
00:19:52,800 --> 00:19:56,400
Speaker 4: valuation and even the data labeling itself. So you're using

381
00:19:56,480 --> 00:19:59,359
Speaker 4: it as almost a teacher for your own model. It

382
00:19:59,480 --> 00:20:03,000
Speaker 4: guides it little bit versus really copy pasting the answer,

383
00:20:03,080 --> 00:20:06,240
Speaker 4: if that makes sense, And that part of it is

384
00:20:06,320 --> 00:20:09,960
Speaker 4: frankly not that unethical or like you know that frown

385
00:20:10,080 --> 00:20:12,399
Speaker 4: upon right now, because that is what enterprises do when

386
00:20:12,400 --> 00:20:14,879
Speaker 4: they're fine to me, So it's all kind of a

387
00:20:14,880 --> 00:20:16,200
Speaker 4: bit of a mercury area.

388
00:20:15,960 --> 00:20:16,480
Speaker 2: To be honest.

389
00:20:16,560 --> 00:20:19,119
Speaker 3: Okay, so you mentioned data, just then talk to us

390
00:20:19,119 --> 00:20:22,440
Speaker 3: about what the Chinese data set actually looks like, because

391
00:20:22,480 --> 00:20:25,320
Speaker 3: I imagine, if you're a ten cent I mean you've

392
00:20:25,320 --> 00:20:28,160
Speaker 3: got we chat, right, that must be a whole load

393
00:20:28,200 --> 00:20:31,479
Speaker 3: of data on which to actually like build your AI.

394
00:20:32,119 --> 00:20:34,000
Speaker 3: But on the other hand, I imagine, like there are

395
00:20:34,000 --> 00:20:38,359
Speaker 3: some restrictions around the Internet. Obviously, what does it actually

396
00:20:38,400 --> 00:20:39,000
Speaker 3: look like here?

397
00:20:39,520 --> 00:20:41,280
Speaker 2: So I actually split that into two parts.

398
00:20:41,320 --> 00:20:44,560
Speaker 4: On the data itself, people often think China is so

399
00:20:44,800 --> 00:20:47,840
Speaker 4: data intensive and you just have mass amount of data

400
00:20:47,840 --> 00:20:51,080
Speaker 4: to use for AI training. However, actually people forget again

401
00:20:51,280 --> 00:20:55,640
Speaker 4: China's enterprise build out or you know whatever. The knowledge

402
00:20:55,640 --> 00:20:59,960
Speaker 4: work economy is very new and not as sophisticated frankly

403
00:21:00,080 --> 00:21:02,800
Speaker 4: as American ecosystem or the Western ecosystem, if you have

404
00:21:02,840 --> 00:21:05,560
Speaker 4: to put it that way. So data is often unstructured,

405
00:21:05,840 --> 00:21:08,480
Speaker 4: and data thus a lot of the specific needs for

406
00:21:08,680 --> 00:21:11,040
Speaker 4: you know, the kind of training we're seeing today is

407
00:21:11,080 --> 00:21:14,320
Speaker 4: not as vibrant, or the data ecosystem is not as

408
00:21:14,359 --> 00:21:18,359
Speaker 4: sophisticated as what American data providers kind of can provide,

409
00:21:18,400 --> 00:21:21,320
Speaker 4: such as mercre like we just mentioned now on the

410
00:21:21,359 --> 00:21:24,240
Speaker 4: big tech side, it's bit interesting. So I'm glad you

411
00:21:24,280 --> 00:21:27,280
Speaker 4: brought up Tencent, because Tencent actually just announced last week

412
00:21:27,320 --> 00:21:30,439
Speaker 4: that they are working in the works of creating an

413
00:21:30,480 --> 00:21:32,880
Speaker 4: agent that can be plugged into we Chat. This has

414
00:21:32,920 --> 00:21:35,560
Speaker 4: been very controversial and it has actually had a lot

415
00:21:35,560 --> 00:21:40,760
Speaker 4: of pushback even internally, because WeChat's product manager, Alan Jung

416
00:21:40,880 --> 00:21:44,520
Speaker 4: has been famously or notoriously known to be kind of

417
00:21:44,560 --> 00:21:46,840
Speaker 4: hard to work with if you want to push something

418
00:21:46,840 --> 00:21:49,720
Speaker 4: with a WeChat because he's so protective of that user experience.

419
00:21:50,000 --> 00:21:53,480
Speaker 4: It's his baby, right, And Tencent, like you said, has

420
00:21:53,600 --> 00:21:56,959
Speaker 4: we Chat, which is a super app, has more than

421
00:21:57,000 --> 00:22:00,720
Speaker 4: one point four billion MAU globally, like mostly one point

422
00:22:00,720 --> 00:22:02,840
Speaker 4: three million people in China and the Chinese ask for

423
00:22:03,040 --> 00:22:07,280
Speaker 4: globally or people who work with China. That is immense value.

424
00:22:07,600 --> 00:22:11,919
Speaker 4: But the risk and compliance risk of potentially of agent

425
00:22:12,000 --> 00:22:16,119
Speaker 4: going rogue within that chatbot, or that of agent going

426
00:22:16,200 --> 00:22:19,240
Speaker 4: rogue in executing you know, whether it's a purchase whatnot,

427
00:22:19,320 --> 00:22:22,120
Speaker 4: is that risk is very high. So they've been working

428
00:22:22,200 --> 00:22:25,320
Speaker 4: on that. And on top of that, Tencent itself has

429
00:22:25,400 --> 00:22:28,119
Speaker 4: been lagging behind compared to other big tech in their

430
00:22:28,160 --> 00:22:31,280
Speaker 4: own proprietary models, and they've really been trying to really

431
00:22:31,280 --> 00:22:31,959
Speaker 4: play catch up.

432
00:22:32,240 --> 00:22:33,360
Speaker 2: They actually last.

433
00:22:33,200 --> 00:22:36,480
Speaker 4: Year poached someone from open Ai who is a researcher

434
00:22:36,560 --> 00:22:38,520
Speaker 4: called y'all shoam U as well has the same name

435
00:22:38,560 --> 00:22:40,800
Speaker 4: as the other researcher we just mentioned, to lead this

436
00:22:40,840 --> 00:22:43,800
Speaker 4: whole initiative, and their whole goal is to basically build

437
00:22:43,840 --> 00:22:47,560
Speaker 4: a ten cent agent native model, and that is their

438
00:22:47,600 --> 00:22:49,600
Speaker 4: biggest goal because end of the day, like you said

439
00:22:49,600 --> 00:22:52,719
Speaker 4: in the very beginning, their goal is to optimize their

440
00:22:52,760 --> 00:22:56,600
Speaker 4: existing businesses already and bring AI to the mass consumer

441
00:22:56,720 --> 00:22:57,520
Speaker 4: as fast as they can.

442
00:22:57,880 --> 00:23:01,520
Speaker 1: You know, you mentioned poaching a research from open Ai,

443
00:23:01,800 --> 00:23:04,600
Speaker 1: and it's like the way I see it, AR will

444
00:23:04,600 --> 00:23:07,720
Speaker 1: definitely be built by the Chinese. The question is whether

445
00:23:07,760 --> 00:23:10,080
Speaker 1: it be built by the Chinese working the American labs

446
00:23:10,160 --> 00:23:12,520
Speaker 1: or whether it will be built by Chinese working and

447
00:23:12,640 --> 00:23:15,840
Speaker 1: Chinese labs. Has there been a gathering of steam of

448
00:23:15,880 --> 00:23:20,440
Speaker 1: researchers from that had been working at American labs going

449
00:23:20,920 --> 00:23:23,320
Speaker 1: to Chinese labs or are they still sort of one

450
00:23:23,359 --> 00:23:24,359
Speaker 1: off and somewhat rare.

451
00:23:24,920 --> 00:23:27,000
Speaker 4: I think even during the Internet era, we saw a

452
00:23:27,000 --> 00:23:29,920
Speaker 4: lot of Chinese nationals or Chinese I think people returning

453
00:23:29,920 --> 00:23:32,919
Speaker 4: to China right. I think this it's easy to blanket

454
00:23:32,960 --> 00:23:37,399
Speaker 4: statement as geopolitical headwinds. People are scared, but realistically, I

455
00:23:37,440 --> 00:23:40,280
Speaker 4: think most people are just trying to take care of

456
00:23:40,320 --> 00:23:42,800
Speaker 4: their families and live a good life, right, yea. I

457
00:23:42,840 --> 00:23:44,800
Speaker 4: hate to sound so crass about it, but you know,

458
00:23:44,920 --> 00:23:49,200
Speaker 4: sometimes it's what your package look like and over generalize.

459
00:23:49,240 --> 00:23:52,359
Speaker 4: I've heard from many researchers say, look, if my wife

460
00:23:52,400 --> 00:23:54,320
Speaker 4: is a lawyer in China, my wife is a nurse

461
00:23:54,400 --> 00:23:56,840
Speaker 4: in China, my wife is a teacher in China, that

462
00:23:56,960 --> 00:24:00,000
Speaker 4: kind of employment opportunity is very, very hard to actually

463
00:24:00,119 --> 00:24:01,000
Speaker 4: you transfer to.

464
00:24:00,960 --> 00:24:01,760
Speaker 2: A new market.

465
00:24:01,880 --> 00:24:03,800
Speaker 4: They're like, if I can get a similar package and

466
00:24:03,840 --> 00:24:06,600
Speaker 4: a growth opportunity in one of the big labs in China,

467
00:24:06,920 --> 00:24:09,399
Speaker 4: I would pick that over living in the US. And

468
00:24:09,440 --> 00:24:11,160
Speaker 4: on top of that, I think that something is lost

469
00:24:11,200 --> 00:24:14,359
Speaker 4: in the nuance is my parents immigrated to North America

470
00:24:14,440 --> 00:24:17,160
Speaker 4: thirty plus years ago. It was a very clean cut,

471
00:24:17,359 --> 00:24:21,080
Speaker 4: like quality of life. It's just like objectively better in

472
00:24:21,200 --> 00:24:23,920
Speaker 4: any city in North America compared to any city in China.

473
00:24:24,440 --> 00:24:27,280
Speaker 4: Now that's kind of a personal debate, right, because it

474
00:24:27,320 --> 00:24:29,800
Speaker 4: depends on what you really value. If you want to

475
00:24:29,840 --> 00:24:32,240
Speaker 4: be close to the city center. You want that fast paced,

476
00:24:32,280 --> 00:24:37,520
Speaker 4: like techno evy futuristic lifestyle, China actually gives that to you.

477
00:24:37,560 --> 00:24:38,960
Speaker 4: And then on top of that, if you want to

478
00:24:38,960 --> 00:24:41,399
Speaker 4: be close to your family, it's a very personal reason.

479
00:24:41,520 --> 00:24:44,080
Speaker 4: So I've met a lot of research I actually decided

480
00:24:44,160 --> 00:24:47,040
Speaker 4: to come back to China or this part of the

481
00:24:47,080 --> 00:24:49,320
Speaker 4: world simply because they wanted to do it for personal

482
00:24:49,359 --> 00:24:50,080
Speaker 4: family reasons.

483
00:24:50,920 --> 00:24:53,119
Speaker 3: Are they paid as much as they are in the US?

484
00:24:53,240 --> 00:24:55,479
Speaker 3: Because we get headlines all the time about you know,

485
00:24:55,520 --> 00:24:59,480
Speaker 3: so and so is joining whatever company, and people treat

486
00:24:59,560 --> 00:25:04,320
Speaker 3: that new whose like sports stars, right, like teams trading

487
00:25:04,800 --> 00:25:07,160
Speaker 3: their best players. Is it a similar thing here?

488
00:25:07,680 --> 00:25:09,960
Speaker 4: I think you definitely get less of that sports star

489
00:25:10,400 --> 00:25:14,960
Speaker 4: vibe or mentality. Here. They're still getting paid like hafty amounts.

490
00:25:15,040 --> 00:25:18,240
Speaker 4: How much they don't usually disclose, but at least even

491
00:25:18,240 --> 00:25:21,159
Speaker 4: in the Internet era, like a byte dance product manager

492
00:25:21,280 --> 00:25:24,040
Speaker 4: can make just as much as a meta product manager. Similarly,

493
00:25:24,440 --> 00:25:27,360
Speaker 4: if you're like an average AI researcher, you're probably making

494
00:25:27,440 --> 00:25:31,000
Speaker 4: a similar amount. Although the star star players, like the

495
00:25:31,040 --> 00:25:33,800
Speaker 4: ones that are signing one hundred million bonuses, I don't

496
00:25:33,840 --> 00:25:37,080
Speaker 4: know if we had anything like that big like in China,

497
00:25:37,640 --> 00:25:40,440
Speaker 4: but look, they made their money with the IPOs, they

498
00:25:40,480 --> 00:25:44,280
Speaker 4: made the money recently with all this AI boom. It's

499
00:25:44,320 --> 00:25:46,720
Speaker 4: just on a maybe slightly smaller scale. Doesn't mean that

500
00:25:46,720 --> 00:25:49,720
Speaker 4: they're not making much more than the frankly average person.

501
00:25:50,040 --> 00:25:52,320
Speaker 1: Tracy, Can I say something that might be sort of

502
00:25:52,359 --> 00:25:56,280
Speaker 1: a sacrilege for a podcast host to say? Uh, Okay,

503
00:25:57,520 --> 00:26:01,240
Speaker 1: I'm only speaking for myself here, not necessarily speaking for

504
00:26:01,280 --> 00:26:04,399
Speaker 1: the team, But it occurred to me like, I'm not

505
00:26:04,560 --> 00:26:08,240
Speaker 1: really sure if I'd be that interested in, say, getting

506
00:26:08,280 --> 00:26:11,240
Speaker 1: the CEO of like an American AI lab on the

507
00:26:11,280 --> 00:26:13,160
Speaker 1: podcast as a guest. I don't know what I would

508
00:26:13,160 --> 00:26:15,840
Speaker 1: ask them, because like, like do I really want to

509
00:26:15,880 --> 00:26:20,280
Speaker 1: hear like Sam Altman or Dabi Sasabi's or whatever like

510
00:26:20,480 --> 00:26:23,399
Speaker 1: the future of work and all this stuff or like

511
00:26:23,600 --> 00:26:26,760
Speaker 1: these all the big you know, I love doing AI episodes.

512
00:26:27,160 --> 00:26:30,240
Speaker 1: I just feel like at that level I would rather

513
00:26:30,359 --> 00:26:32,919
Speaker 1: talk to us or like someone in actually the engine

514
00:26:33,000 --> 00:26:35,679
Speaker 1: room rather than this sort of big picture a person

515
00:26:35,720 --> 00:26:39,320
Speaker 1: who may have some degree of AI psychosis and just

516
00:26:39,359 --> 00:26:42,760
Speaker 1: like as on a speaks in like the biggest generalities.

517
00:26:42,960 --> 00:26:46,600
Speaker 3: Okay, well, now Sam Altman needs to like invite himself

518
00:26:46,600 --> 00:26:49,280
Speaker 3: on the show just to test your your commitment to

519
00:26:49,400 --> 00:26:51,080
Speaker 3: not having AI CEOs.

520
00:26:51,359 --> 00:26:53,199
Speaker 1: No, I would, I would do it, But I like,

521
00:26:53,480 --> 00:26:56,080
Speaker 1: let's just agree here that if we ever get one

522
00:26:56,080 --> 00:26:59,200
Speaker 1: of the really big like lab CEOs, let's just ask

523
00:26:59,240 --> 00:27:03,280
Speaker 1: the very like sort of mundane questions about operations and now,

524
00:27:03,400 --> 00:27:05,800
Speaker 1: like what are we all going to do? And you know,

525
00:27:05,960 --> 00:27:07,959
Speaker 1: is what is the meaning of life going to be

526
00:27:08,040 --> 00:27:09,800
Speaker 1: when we don't have jobs? Because I'm so sick of

527
00:27:09,840 --> 00:27:13,439
Speaker 1: those conversations. They may be important at some point, but Grace,

528
00:27:13,480 --> 00:27:17,960
Speaker 1: I'm sort of curious from your perception, like it does

529
00:27:18,040 --> 00:27:20,639
Speaker 1: feel like the heads of the American AI labs have

530
00:27:20,840 --> 00:27:24,520
Speaker 1: some degree of AI psychosis themselves. Either they talk about

531
00:27:24,920 --> 00:27:27,800
Speaker 1: all white color employment is going to disappear, or that

532
00:27:27,800 --> 00:27:30,720
Speaker 1: they're going to build a monster that if done wrong,

533
00:27:31,000 --> 00:27:33,800
Speaker 1: is going to be out of control and that they're

534
00:27:33,840 --> 00:27:36,359
Speaker 1: not you know, it'll escape, that it'll escape the sandbox.

535
00:27:36,720 --> 00:27:41,679
Speaker 1: Is there the same sort of existential discourse in the

536
00:27:41,760 --> 00:27:43,040
Speaker 1: Chinese AI community?

537
00:27:43,600 --> 00:27:48,040
Speaker 4: Yeah? I think to start with in the AI community themselves,

538
00:27:48,200 --> 00:27:50,600
Speaker 4: I would say people are a lot more pragmatic and

539
00:27:50,640 --> 00:27:52,760
Speaker 4: I think recently I was talking to Nathan Lambert, who

540
00:27:52,760 --> 00:27:55,160
Speaker 4: is open source researcher who just came to China visited

541
00:27:55,160 --> 00:27:57,520
Speaker 4: all the labs. He said, Look, I was shocked to

542
00:27:57,560 --> 00:28:00,840
Speaker 4: see majority of labs are so as in like a

543
00:28:00,880 --> 00:28:04,399
Speaker 4: lot of the researchers are still students, a lot of

544
00:28:04,400 --> 00:28:07,040
Speaker 4: them are interns, and the core research teams are maybe

545
00:28:07,080 --> 00:28:09,720
Speaker 4: led by a handful of people. And then these people

546
00:28:09,760 --> 00:28:13,360
Speaker 4: are academics by training. So maybe they're a bit less commercial.

547
00:28:13,560 --> 00:28:14,160
Speaker 2: Maybe you can.

548
00:28:14,080 --> 00:28:17,720
Speaker 4: Say they're less like sophisticated to make manimally the market,

549
00:28:17,760 --> 00:28:20,560
Speaker 4: whatever you want to call them. So I definitely feel

550
00:28:20,600 --> 00:28:23,120
Speaker 4: like there's less of that kind of psychosis or high

551
00:28:23,280 --> 00:28:24,679
Speaker 4: level narrative going around.

552
00:28:25,000 --> 00:28:26,480
Speaker 2: However, I would.

553
00:28:26,200 --> 00:28:30,560
Speaker 4: Say that there is obviously anxiety from the public in

554
00:28:30,600 --> 00:28:34,399
Speaker 4: some degree, not as much of a back pushback. But

555
00:28:35,240 --> 00:28:37,880
Speaker 4: recently there was a very interesting court case in Hongjo,

556
00:28:38,080 --> 00:28:40,640
Speaker 4: which is you know, home to Ali Baba and a

557
00:28:40,680 --> 00:28:44,560
Speaker 4: lot of these al abs. Basically, a company try to

558
00:28:45,240 --> 00:28:48,480
Speaker 4: like lay off a person saying you are being replaced

559
00:28:48,480 --> 00:28:51,080
Speaker 4: with AI, and the court literally rules say that is

560
00:28:51,120 --> 00:28:54,040
Speaker 4: not allowed and no company can use AI as an

561
00:28:54,080 --> 00:28:57,000
Speaker 4: excuse to lay off or replace or even cut short

562
00:28:57,280 --> 00:29:00,440
Speaker 4: their contract time. So that was a really swift reaction

563
00:29:00,520 --> 00:29:04,040
Speaker 4: from regulators, and I think it really did serve as a.

564
00:29:04,040 --> 00:29:05,560
Speaker 2: Calming factor for the public.

565
00:29:05,960 --> 00:29:08,280
Speaker 4: Obviously, I also think I want to preface the fact

566
00:29:08,280 --> 00:29:10,680
Speaker 4: that the knowledge worker economy makes.

567
00:29:10,480 --> 00:29:12,840
Speaker 2: Up less of the whoverall.

568
00:29:12,280 --> 00:29:15,400
Speaker 4: Economy in China as well, so that kind of fear

569
00:29:15,560 --> 00:29:19,280
Speaker 4: maybe doesn't feel as imminent, but that conversation is being had.

570
00:29:19,600 --> 00:29:21,800
Speaker 4: But I do think in Asia in general, not only

571
00:29:21,800 --> 00:29:24,840
Speaker 4: in China. You look at South Korea, Singapore, all these

572
00:29:24,880 --> 00:29:28,560
Speaker 4: countries are approaching AI in a very pragmatic way, and

573
00:29:28,680 --> 00:29:30,959
Speaker 4: the tire moms are trying to train up the kids

574
00:29:31,000 --> 00:29:34,000
Speaker 4: to be AI native, the students are trying to train

575
00:29:34,040 --> 00:29:36,480
Speaker 4: themselves up to be AA native. People are preparing for

576
00:29:36,520 --> 00:29:39,480
Speaker 4: the future versus pushing back on the future.

577
00:29:39,920 --> 00:29:43,280
Speaker 1: That's interesting. You know, in the US, obviously, companies announced

578
00:29:43,320 --> 00:29:46,640
Speaker 1: that they're laying people off, and they cite AI even

579
00:29:46,640 --> 00:29:49,360
Speaker 1: frequently when there's no evidence that AI had anything to

580
00:29:49,400 --> 00:29:51,400
Speaker 1: do with it. So it's interesting they're not allowed to

581
00:29:51,440 --> 00:29:55,239
Speaker 1: do it. In defense of American AI labs, most of

582
00:29:55,280 --> 00:29:58,040
Speaker 1: them lose a lot of money, and yet they actually

583
00:29:58,120 --> 00:30:01,200
Speaker 1: do spend, at least the one spend there quite a

584
00:30:01,240 --> 00:30:06,440
Speaker 1: decent amount on so called like alignment research, safety research,

585
00:30:06,520 --> 00:30:10,080
Speaker 1: making sure the AIS don't go rogue et cetera. How

586
00:30:10,080 --> 00:30:13,160
Speaker 1: big of a part of the Chinese labs, how much

587
00:30:13,200 --> 00:30:15,080
Speaker 1: did they spend on I guess what they would you

588
00:30:15,120 --> 00:30:17,200
Speaker 1: would put but the American labors would put into the

589
00:30:17,200 --> 00:30:18,160
Speaker 1: safety bucket.

590
00:30:18,360 --> 00:30:20,360
Speaker 4: I do have to say, I'm not a policy expert,

591
00:30:20,400 --> 00:30:22,080
Speaker 4: so I don't work with a lot of the safety

592
00:30:22,120 --> 00:30:25,880
Speaker 4: people as much. But there are organizations in China that

593
00:30:26,000 --> 00:30:29,360
Speaker 4: are definitely working, like the regulators as well as the

594
00:30:29,680 --> 00:30:33,959
Speaker 4: private sector working together. And for some context, in China,

595
00:30:34,080 --> 00:30:37,040
Speaker 4: there are various moving parts in the government. There's MIT,

596
00:30:37,240 --> 00:30:41,640
Speaker 4: the CAC, et cetera. These agencies basically some are to

597
00:30:41,680 --> 00:30:45,840
Speaker 4: propel economic development. So in this case AI diffusion, the

598
00:30:45,880 --> 00:30:49,280
Speaker 4: whole idea of AI plus AI plus every single sector

599
00:30:49,320 --> 00:30:52,640
Speaker 4: you can think of, some act more like a guardrail

600
00:30:53,000 --> 00:30:55,480
Speaker 4: as a protector, so they are working.

601
00:30:55,200 --> 00:30:55,920
Speaker 2: Hand in hand.

602
00:30:56,000 --> 00:31:00,000
Speaker 4: And on top of that, every single AI jen AI

603
00:31:00,040 --> 00:31:03,080
Speaker 4: application as well as LM company have to go through

604
00:31:03,120 --> 00:31:07,320
Speaker 4: the National Registry in China, so they actually disclose what

605
00:31:07,520 --> 00:31:10,320
Speaker 4: is being trained, what is you know, the potential risk

606
00:31:10,680 --> 00:31:13,560
Speaker 4: that said, I think right now, you know, no one

607
00:31:13,680 --> 00:31:17,120
Speaker 4: really knows what the real impact of AI will be

608
00:31:17,160 --> 00:31:20,560
Speaker 4: on economy, but you know definitely that fear mongering narrative

609
00:31:20,720 --> 00:31:36,040
Speaker 4: is not mainstream in China.

610
00:31:37,960 --> 00:31:41,000
Speaker 3: How would you describe I guess the approach of the

611
00:31:41,080 --> 00:31:44,720
Speaker 3: Chinese government to AI in general, because it feels like

612
00:31:44,800 --> 00:31:47,640
Speaker 3: the trade off for maybe not being on the cutting

613
00:31:47,760 --> 00:31:51,200
Speaker 3: edge with frontier models is well, you're further along with

614
00:31:51,640 --> 00:31:54,560
Speaker 3: sovereign AI and the government maybe has like a better

615
00:31:54,640 --> 00:31:57,360
Speaker 3: handle on what the labs are actually doing.

616
00:31:58,200 --> 00:32:01,520
Speaker 4: The Chinese government probably sees this in a much more

617
00:32:01,600 --> 00:32:05,959
Speaker 4: pragmatic way. You know, just like how there was an

618
00:32:05,960 --> 00:32:10,040
Speaker 4: Internet plus policy fifteen years ago, there's now an AI

619
00:32:10,080 --> 00:32:13,200
Speaker 4: plus policy. When deep seak moment took off, you know,

620
00:32:13,240 --> 00:32:16,760
Speaker 4: there was a frenzy of private companies even in home appliances,

621
00:32:16,800 --> 00:32:19,760
Speaker 4: trying to embed deep seat, integrate deep sea. I'm just like,

622
00:32:20,200 --> 00:32:22,520
Speaker 4: what is a AI vacuum going to do for you?

623
00:32:22,640 --> 00:32:25,720
Speaker 4: Or AI like electric truth rush for you? It was wild, right,

624
00:32:26,120 --> 00:32:28,520
Speaker 4: but the government picked that up. And I think what

625
00:32:28,640 --> 00:32:30,920
Speaker 4: the Chinese government, going back to what we talked about earlier,

626
00:32:30,960 --> 00:32:33,920
Speaker 4: is that they have the advantage of having the ability

627
00:32:33,920 --> 00:32:36,640
Speaker 4: to push things down from top down and at the

628
00:32:36,800 --> 00:32:40,200
Speaker 4: very high level, they're seeing AI as an economic driver

629
00:32:40,760 --> 00:32:45,600
Speaker 4: to propel maybe efficiency, to address some of the labor

630
00:32:45,600 --> 00:32:48,480
Speaker 4: shortage that is coming as the population could continues to

631
00:32:48,520 --> 00:32:51,520
Speaker 4: age and decline. It also addresses a lot of issues

632
00:32:51,560 --> 00:32:53,520
Speaker 4: where a lot of the young people don't want to

633
00:32:53,560 --> 00:32:56,760
Speaker 4: work in manufacturing roles. They want to be automated actually,

634
00:32:56,920 --> 00:32:59,000
Speaker 4: and they want to work in urban areas. So they

635
00:32:59,040 --> 00:33:02,800
Speaker 4: have that now, and then each of the provincial governments

636
00:33:02,880 --> 00:33:05,760
Speaker 4: will take that as kind of like a KPI. They're like,

637
00:33:05,760 --> 00:33:08,880
Speaker 4: all right, let's go ask like vcs essentially and go

638
00:33:09,560 --> 00:33:13,240
Speaker 4: find you know, future deep seeks and fund them. However,

639
00:33:13,320 --> 00:33:15,720
Speaker 4: how much these companies want to take government money is

640
00:33:15,760 --> 00:33:18,560
Speaker 4: a different discussion. A lot of them will then even

641
00:33:18,720 --> 00:33:22,840
Speaker 4: support them by providing you know, infrastructure like buildings, offices,

642
00:33:23,000 --> 00:33:25,720
Speaker 4: even like some of them heard like dormitory for these

643
00:33:25,760 --> 00:33:28,840
Speaker 4: young entrepreneurs, and then give them money and capital try

644
00:33:28,840 --> 00:33:31,840
Speaker 4: things out. There's also these AI pilot zones being rolled

645
00:33:31,840 --> 00:33:34,200
Speaker 4: out across the country. I think now about eleven or

646
00:33:34,200 --> 00:33:36,640
Speaker 4: twelve of them, where you know, people can try out

647
00:33:36,680 --> 00:33:40,480
Speaker 4: new AI products. I met with the largest AI developer

648
00:33:40,520 --> 00:33:42,960
Speaker 4: community founder in China a couple of weeks ago, and

649
00:33:43,000 --> 00:33:45,760
Speaker 4: she was saying, there's more than a couple hundred thousand

650
00:33:45,840 --> 00:33:48,800
Speaker 4: developers in this ecosystem and they are working with.

651
00:33:48,840 --> 00:33:50,680
Speaker 2: Regulators and the private sector.

652
00:33:50,920 --> 00:33:52,920
Speaker 4: So if say bitdowns have a new product, they might

653
00:33:52,960 --> 00:33:54,680
Speaker 4: go to them first and say can you try this out?

654
00:33:55,040 --> 00:33:58,320
Speaker 4: And then they will report and debug and see what's happening,

655
00:33:58,560 --> 00:34:01,600
Speaker 4: and then tell the whoever low government that's funding them

656
00:34:01,600 --> 00:34:04,560
Speaker 4: with providing with the infrastructure, say, hey, this product might

657
00:34:04,600 --> 00:34:06,640
Speaker 4: come out. Do you want to be part of it?

658
00:34:06,720 --> 00:34:08,160
Speaker 4: Do you want to give it money? Do you want

659
00:34:08,200 --> 00:34:10,920
Speaker 4: to provide it with whatever resource you want to? So

660
00:34:11,000 --> 00:34:14,360
Speaker 4: there is kind of this like cohesive ecosystem where they

661
00:34:14,480 --> 00:34:18,360
Speaker 4: kind of all dance together. How much these companies actually

662
00:34:18,440 --> 00:34:22,840
Speaker 4: want to take state money, I think that's debatable. However,

663
00:34:23,000 --> 00:34:26,000
Speaker 4: as AI and robot has become morem when sensitive and

664
00:34:26,040 --> 00:34:30,080
Speaker 4: being recognized as not only an economic driver but potentially

665
00:34:30,120 --> 00:34:34,160
Speaker 4: a military use or geopolitical I guess talking point. At

666
00:34:34,160 --> 00:34:36,920
Speaker 4: this point, it is becoming more and more nationalized, not

667
00:34:36,960 --> 00:34:39,560
Speaker 4: only in China but globally in the US and so on.

668
00:34:40,440 --> 00:34:44,240
Speaker 1: Robotics is obviously an area where China is just straight

669
00:34:44,320 --> 00:34:47,120
Speaker 1: up ahead of the United States, or at least according

670
00:34:47,160 --> 00:34:49,720
Speaker 1: to all the videos on my Twitter and Instagram feed

671
00:34:49,880 --> 00:34:53,920
Speaker 1: of humanoid robots and so forth. How much does you

672
00:34:53,920 --> 00:34:56,680
Speaker 1: know when we were all kids when we thought of

673
00:34:56,800 --> 00:34:59,480
Speaker 1: like AI, I think we thought of robots, right, we

674
00:34:59,520 --> 00:35:02,480
Speaker 1: thought about fifteen one thousand or some version of it.

675
00:35:02,760 --> 00:35:04,600
Speaker 1: And now when we think of AI, most people think

676
00:35:04,600 --> 00:35:07,600
Speaker 1: of chatbots. But that's just one aspect of AI. For

677
00:35:07,840 --> 00:35:11,160
Speaker 1: the advanced labage, whether we're talking about the deep sea

678
00:35:11,280 --> 00:35:14,040
Speaker 1: so then Mini Max's or GLM and so forth, like,

679
00:35:14,800 --> 00:35:18,200
Speaker 1: are they actively working hand in hand with some of

680
00:35:18,239 --> 00:35:21,919
Speaker 1: the unit trees and advanced robotics companies to figure out

681
00:35:22,160 --> 00:35:25,320
Speaker 1: how you can actually have that the true AI robot

682
00:35:25,360 --> 00:35:26,000
Speaker 1: of the future.

683
00:35:26,520 --> 00:35:30,480
Speaker 4: Yeah, I think China, having been the manufacturing hub of

684
00:35:30,560 --> 00:35:33,200
Speaker 4: like literally everything in the sun over the last like

685
00:35:33,239 --> 00:35:37,480
Speaker 4: three decades, has definitely an advantage of owning all the

686
00:35:37,480 --> 00:35:40,040
Speaker 4: supply chain, right. And it's like not only just owning

687
00:35:40,080 --> 00:35:42,520
Speaker 4: the supply chain, but you know, there are literally regions

688
00:35:42,560 --> 00:35:46,040
Speaker 4: where that whole supply from raw material to like the

689
00:35:46,280 --> 00:35:49,359
Speaker 4: result end product from OEM is all within like say,

690
00:35:49,400 --> 00:35:53,120
Speaker 4: fifty kilometers of each other. So what we're seeing is

691
00:35:53,320 --> 00:35:56,920
Speaker 4: a lot of American investors and entrepreneurs coming into China

692
00:35:56,960 --> 00:35:58,799
Speaker 4: to kind of get a sense of that. And like

693
00:35:58,880 --> 00:36:01,759
Speaker 4: you mentioned, because we've been so fixed in software, I

694
00:36:01,800 --> 00:36:05,520
Speaker 4: think China having a very strong hardware background is now

695
00:36:05,600 --> 00:36:08,520
Speaker 4: thinking about how can we actually integrate the software into

696
00:36:08,560 --> 00:36:12,200
Speaker 4: the hardware. How ready that is to the mass market.

697
00:36:12,320 --> 00:36:15,040
Speaker 4: I frankly don't think it's really there yet. So recently

698
00:36:15,040 --> 00:36:18,080
Speaker 4: I just met with some robotic companies. They actually can't

699
00:36:18,080 --> 00:36:20,560
Speaker 4: just plug in a Mini Max. You know, that's like

700
00:36:20,880 --> 00:36:24,279
Speaker 4: for them, they need to actually get physical data. They

701
00:36:24,360 --> 00:36:26,400
Speaker 4: This is where like you know, now all the hype

702
00:36:26,440 --> 00:36:29,239
Speaker 4: is on world models physical AI. You know, that is

703
00:36:29,280 --> 00:36:32,960
Speaker 4: a complete different set of kind of technology essentially, where

704
00:36:33,560 --> 00:36:37,800
Speaker 4: without the three D data that these models need right now,

705
00:36:38,120 --> 00:36:41,160
Speaker 4: the bottomneck right now is that you know, these hardware

706
00:36:41,200 --> 00:36:44,560
Speaker 4: is these humanoids, quadrupids, dogs, whatever you want to call them,

707
00:36:44,719 --> 00:36:46,800
Speaker 4: they cannot be powered by lms.

708
00:36:47,040 --> 00:36:47,880
Speaker 2: That's number one.

709
00:36:48,080 --> 00:36:51,520
Speaker 4: Number two is despite that China being very strong on hardware,

710
00:36:51,840 --> 00:36:54,200
Speaker 4: the bottleneck is actually a lot of times in the

711
00:36:54,280 --> 00:36:57,880
Speaker 4: integration as well as the battery solutions. You know, you

712
00:36:57,920 --> 00:37:01,000
Speaker 4: think of China having very strong battery solutions, but most

713
00:37:01,040 --> 00:37:03,920
Speaker 4: of these gadgets can't last more than like say two hours,

714
00:37:04,400 --> 00:37:06,640
Speaker 4: and there's no one that's really come out with a

715
00:37:06,680 --> 00:37:09,840
Speaker 4: better solution so far. What I've seen the most creative

716
00:37:09,840 --> 00:37:11,879
Speaker 4: things so far is like you know, those glasses you wear,

717
00:37:11,960 --> 00:37:14,600
Speaker 4: like the metaglasses, they kind of die within two hours.

718
00:37:14,840 --> 00:37:18,000
Speaker 4: But China, like I Fly Tech or Rocket that's kind

719
00:37:18,040 --> 00:37:22,840
Speaker 4: of a newer player startup, they created these battery capsules

720
00:37:22,840 --> 00:37:25,960
Speaker 4: where you can just like stick on to your glasses.

721
00:37:26,239 --> 00:37:29,359
Speaker 4: It's very lightweight, doesn't really affect your user experience, and

722
00:37:29,400 --> 00:37:31,600
Speaker 4: that's actually able to kind of extend it by a

723
00:37:31,640 --> 00:37:34,719
Speaker 4: few hours. So to go back to your question, is

724
00:37:34,800 --> 00:37:36,680
Speaker 4: China trying to do physical AI?

725
00:37:36,840 --> 00:37:37,400
Speaker 2: Definitely?

726
00:37:37,680 --> 00:37:39,920
Speaker 4: What is their edge? I think it's still in manufacturing.

727
00:37:40,080 --> 00:37:42,359
Speaker 4: Is their software good enough? I don't think anyone really

728
00:37:42,360 --> 00:37:44,239
Speaker 4: has good enough software right now as of now.

729
00:37:44,680 --> 00:37:46,040
Speaker 2: Wait, I'm just going to press you on this.

730
00:37:46,200 --> 00:37:49,600
Speaker 3: So if we fast forward ten years, like, what would

731
00:37:49,640 --> 00:37:53,279
Speaker 3: you say is most likely to be China's comparative advantage.

732
00:37:53,360 --> 00:37:57,600
Speaker 3: Is it like the cheap, open source super optimized models.

733
00:37:57,880 --> 00:38:01,920
Speaker 3: Is it software AI software that's like integrated with like

734
00:38:02,040 --> 00:38:05,680
Speaker 3: industry and existing business or is it robotics and the

735
00:38:05,719 --> 00:38:07,200
Speaker 3: sort of hardware side of AI.

736
00:38:07,840 --> 00:38:09,440
Speaker 2: I think any.

737
00:38:11,360 --> 00:38:13,000
Speaker 3: But we want to challenge here.

738
00:38:14,360 --> 00:38:16,040
Speaker 2: A lot of these companies that exist ten years ago

739
00:38:16,120 --> 00:38:16,960
Speaker 2: or not even five years.

740
00:38:17,040 --> 00:38:20,800
Speaker 3: That's fair, okay, I can shorten the timeframe in three years.

741
00:38:20,880 --> 00:38:24,200
Speaker 4: All right, So don't chase me down if I'm wrong

742
00:38:24,239 --> 00:38:27,480
Speaker 4: in three years. But I think, you know, there's two parts.

743
00:38:27,520 --> 00:38:29,480
Speaker 4: One is I think I agree with the hardware side.

744
00:38:29,560 --> 00:38:32,400
Speaker 4: China is definitely going to have I think, more breakthroughs

745
00:38:32,680 --> 00:38:35,000
Speaker 4: and have a lot of edge. Not only is the

746
00:38:35,040 --> 00:38:38,719
Speaker 4: supply chain all domestically there, I think something overlooked by

747
00:38:38,760 --> 00:38:41,120
Speaker 4: people is the fact that a lot of the know

748
00:38:41,200 --> 00:38:45,200
Speaker 4: how is also there, and that's not easy to transfer overnight.

749
00:38:45,239 --> 00:38:47,879
Speaker 4: You know Patrick McGee's book recently in his Apple book

750
00:38:47,960 --> 00:38:50,399
Speaker 4: saying how Apple tried to move this whole supply chain

751
00:38:50,440 --> 00:38:53,040
Speaker 4: to India. The biggest bottom neck is actually these like

752
00:38:53,120 --> 00:38:58,080
Speaker 4: highly skilled laborist jobs that actually are so technical that

753
00:38:58,160 --> 00:39:01,880
Speaker 4: cannot be even trained in one generation. It took decades

754
00:39:01,920 --> 00:39:04,960
Speaker 4: to really train up the local community labor force whatnot.

755
00:39:05,320 --> 00:39:08,320
Speaker 4: So that's still there now because of that ecosystem.

756
00:39:08,600 --> 00:39:09,360
Speaker 2: A lot of these.

757
00:39:09,320 --> 00:39:13,400
Speaker 4: Robots home appliance is whatnot. These tech gadgets are produced

758
00:39:13,520 --> 00:39:16,239
Speaker 4: at less than fifty percent of the costs of where

759
00:39:16,280 --> 00:39:18,279
Speaker 4: you could produce that anywhere else in the world. They

760
00:39:18,320 --> 00:39:22,399
Speaker 4: are also extremely innovative. I've talked to people at ev

761
00:39:22,520 --> 00:39:26,000
Speaker 4: companies just for example, to ship out a new model

762
00:39:26,440 --> 00:39:31,279
Speaker 4: from ideation to production to hitting the floors that takes

763
00:39:31,320 --> 00:39:34,640
Speaker 4: maybe less than fifteen months. Wow, But for traditional OEM

764
00:39:34,680 --> 00:39:36,600
Speaker 4: like way for at least three to five years. Right,

765
00:39:36,840 --> 00:39:41,360
Speaker 4: So there's the hardware side. I think another very underappreciated

766
00:39:41,440 --> 00:39:43,920
Speaker 4: fact on the Chinese open source model is that people

767
00:39:44,000 --> 00:39:46,920
Speaker 4: don't realize. So a year ago when I spoke to

768
00:39:47,120 --> 00:39:50,080
Speaker 4: startups as semlic Valley, they were the most cost conscious,

769
00:39:50,080 --> 00:39:55,680
Speaker 4: frankly less compliance conscious and gives very little the care

770
00:39:55,680 --> 00:39:59,120
Speaker 4: about geopolitics. They were building on top of QUID. Now

771
00:39:59,200 --> 00:40:02,040
Speaker 4: it's actually all a lot of Ai Native American enterprises

772
00:40:02,080 --> 00:40:05,239
Speaker 4: building on Chinese open source because we're seeing headlines on

773
00:40:06,040 --> 00:40:09,960
Speaker 4: the ROI is not like showing it's extremely expensive for

774
00:40:10,000 --> 00:40:16,000
Speaker 4: these token maxing projects whatnot. So Harvey Cursor, they've talked

775
00:40:16,000 --> 00:40:19,759
Speaker 4: about using a hybrid model where they will build majority

776
00:40:19,760 --> 00:40:22,440
Speaker 4: on GLM or KIMMI, but kind of like what we

777
00:40:22,440 --> 00:40:26,399
Speaker 4: talked about earlier where they use like Opus to act

778
00:40:26,440 --> 00:40:29,120
Speaker 4: as a judge or a guidance. So I think that's

779
00:40:29,160 --> 00:40:32,759
Speaker 4: something where we're continue to see and these companies are

780
00:40:32,840 --> 00:40:35,560
Speaker 4: generating a lot of revenue, and going back to the

781
00:40:35,600 --> 00:40:37,520
Speaker 4: fact that like a lot of them don't even have

782
00:40:37,640 --> 00:40:40,279
Speaker 4: enough capability to support the demand that's coming through.

783
00:40:40,600 --> 00:40:42,879
Speaker 1: That's such a fascinating idea, and it makes a lot

784
00:40:42,920 --> 00:40:46,520
Speaker 1: of sense that at the application layer that probably a

785
00:40:46,560 --> 00:40:48,799
Speaker 1: lot of different models can go into it.

786
00:40:48,880 --> 00:40:49,040
Speaker 4: You know.

787
00:40:49,160 --> 00:40:50,480
Speaker 1: By the way, I was talking to someone out a

788
00:40:50,520 --> 00:40:53,560
Speaker 1: dinner recently and he said he thinks that AI writing

789
00:40:53,600 --> 00:40:56,799
Speaker 1: will get a lot better when AI is embedded in

790
00:40:56,920 --> 00:41:00,799
Speaker 1: humanoid robots, because then the will have this sort of

791
00:41:00,920 --> 00:41:03,560
Speaker 1: groundedness in the real world where I have no idea

792
00:41:03,680 --> 00:41:05,960
Speaker 1: this is true, he said. The reason his theory was

793
00:41:06,000 --> 00:41:08,040
Speaker 1: that the reason why AI writing is still so weird

794
00:41:08,120 --> 00:41:11,759
Speaker 1: is because it's in this disembodied data centers, and that is,

795
00:41:12,080 --> 00:41:14,759
Speaker 1: as soon as they're really in robots, then they'll have

796
00:41:15,040 --> 00:41:17,200
Speaker 1: a sort of real world groundedness. A right, have one

797
00:41:17,239 --> 00:41:20,880
Speaker 1: last question, you know, after open Claw came out, I

798
00:41:20,920 --> 00:41:24,879
Speaker 1: started seeing again my entire Twitter feed and answergram feed.

799
00:41:25,000 --> 00:41:27,400
Speaker 1: I have done this to myself, but all I do

800
00:41:27,440 --> 00:41:31,040
Speaker 1: all day is consume Chinese propaganada. But I started seeing

801
00:41:31,040 --> 00:41:33,520
Speaker 1: all these videos, like all these grandmothers and stuff like

802
00:41:33,840 --> 00:41:36,279
Speaker 1: setting up their claws and stuff like that. And I

803
00:41:36,320 --> 00:41:38,400
Speaker 1: saw these videos and I said to myself, I just

804
00:41:38,440 --> 00:41:40,719
Speaker 1: don't believe this. I think this is fake news. I

805
00:41:40,760 --> 00:41:43,759
Speaker 1: do not actually believe that there's all these eighty year

806
00:41:43,800 --> 00:41:47,040
Speaker 1: old Grammas or whatever really excited about setting up their

807
00:41:47,080 --> 00:41:49,800
Speaker 1: open claw or whatever. Are those real? Like, what's the

808
00:41:49,880 --> 00:41:50,399
Speaker 1: deal with that?

809
00:41:51,120 --> 00:41:51,279
Speaker 3: Yeah?

810
00:41:51,320 --> 00:41:53,160
Speaker 2: I think that was definitely a bit of a hype.

811
00:41:53,320 --> 00:41:53,839
Speaker 1: I knew it.

812
00:41:54,040 --> 00:41:56,799
Speaker 4: No, No, but I will say there were Grammas lining

813
00:41:56,920 --> 00:41:57,680
Speaker 4: up to get it done.

814
00:41:57,800 --> 00:42:01,520
Speaker 2: Okay, I'll answer this twofold on the kind of the

815
00:42:01,600 --> 00:42:02,560
Speaker 2: surface is.

816
00:42:02,640 --> 00:42:08,120
Speaker 4: I think Chinese aunties, uncles whatnot. They are just much

817
00:42:08,160 --> 00:42:11,440
Speaker 4: more open to technology because you know, you go around,

818
00:42:11,480 --> 00:42:14,160
Speaker 4: whether it's by force or by nature. You know, you

819
00:42:14,200 --> 00:42:17,920
Speaker 4: can't really navigate modern Chinese life without being on alipay.

820
00:42:18,040 --> 00:42:18,799
Speaker 1: No, you really can't.

821
00:42:18,800 --> 00:42:22,080
Speaker 3: You can't like buy Starbucks in Beijing without like having

822
00:42:22,160 --> 00:42:22,720
Speaker 3: we paid.

823
00:42:22,800 --> 00:42:25,640
Speaker 4: After COVID, I went back to I think Shanghai for

824
00:42:25,680 --> 00:42:28,120
Speaker 4: the first time, and I was sitting at a restaurant

825
00:42:28,200 --> 00:42:30,680
Speaker 4: just like waiting for someone to help me order food.

826
00:42:30,840 --> 00:42:33,200
Speaker 4: No one came because they're just like, why are you

827
00:42:33,360 --> 00:42:35,480
Speaker 4: so Like, why are you a caveman? Don't you know

828
00:42:35,520 --> 00:42:38,680
Speaker 4: how to like scan the QR code on your table? Okay,

829
00:42:38,680 --> 00:42:43,160
Speaker 4: So that Tangent's side, I think the overall optimism around.

830
00:42:43,239 --> 00:42:47,799
Speaker 4: Technology is very different from the West because in the

831
00:42:47,880 --> 00:42:50,920
Speaker 4: last twenty thirty years, a lot of rural eras in

832
00:42:51,000 --> 00:42:56,080
Speaker 4: China literally could not access resources, information, goods, whatever that,

833
00:42:56,480 --> 00:42:59,680
Speaker 4: you know, like big cities could not until these super

834
00:42:59,719 --> 00:43:02,319
Speaker 4: apps came about. So a lot of people don't have

835
00:43:02,360 --> 00:43:04,280
Speaker 4: TVs in their homes and they live in a village

836
00:43:04,280 --> 00:43:06,759
Speaker 4: and there may be annual household income is like one

837
00:43:06,800 --> 00:43:09,480
Speaker 4: thousand dollars, but they will have a smartphone and that

838
00:43:09,640 --> 00:43:12,560
Speaker 4: smartphone will be able to actually enable them to get

839
00:43:12,680 --> 00:43:16,200
Speaker 4: micro loans to purchase goods, you know, to help their

840
00:43:16,280 --> 00:43:20,279
Speaker 4: kids access information online whatever that is. So technology is

841
00:43:20,480 --> 00:43:24,440
Speaker 4: very much kind of accepted and respected and actually like

842
00:43:24,480 --> 00:43:26,759
Speaker 4: big tech is loved, like if you work for one

843
00:43:26,800 --> 00:43:29,520
Speaker 4: of big tech, you are like a pride of the family.

844
00:43:30,000 --> 00:43:33,279
Speaker 4: So there's that very culture aspect of it. Then is

845
00:43:33,360 --> 00:43:36,080
Speaker 4: like going back to the super apps. So I think

846
00:43:36,120 --> 00:43:39,120
Speaker 4: the open Claw frenzy was interesting because some people say

847
00:43:39,200 --> 00:43:42,400
Speaker 4: it was the first agent that you know, Chinese people

848
00:43:42,440 --> 00:43:44,320
Speaker 4: could get their hands on, like a Western agent that

849
00:43:44,360 --> 00:43:47,160
Speaker 4: they can get their hands on because you know, anthropic

850
00:43:47,200 --> 00:43:49,279
Speaker 4: and open it doesn't actually operated in China. You can't

851
00:43:49,280 --> 00:43:53,520
Speaker 4: access that. So when Tensan and Ali Baba try to

852
00:43:53,960 --> 00:43:58,719
Speaker 4: embed open claw products into their own like series of

853
00:43:58,800 --> 00:44:02,560
Speaker 4: products or business whatever offerings. It got people really excited,

854
00:44:02,960 --> 00:44:06,239
Speaker 4: and because of these super app models that they have,

855
00:44:06,719 --> 00:44:10,000
Speaker 4: it was a very natural way for people to access them.

856
00:44:10,000 --> 00:44:12,759
Speaker 4: There's a functional adjacency to you know, the search bar

857
00:44:13,160 --> 00:44:15,560
Speaker 4: and then opening up an open claw and then trying

858
00:44:15,600 --> 00:44:18,320
Speaker 4: to like run like you know, manage your mini programs

859
00:44:18,320 --> 00:44:21,200
Speaker 4: within ten cent we chat and then try to order something.

860
00:44:21,239 --> 00:44:23,600
Speaker 4: So all of that kind of took off. But that said,

861
00:44:24,080 --> 00:44:28,080
Speaker 4: actually the Chinese government again regulators acted very swiftly in

862
00:44:28,160 --> 00:44:31,680
Speaker 4: the beginning. I think local government, like we see try

863
00:44:31,719 --> 00:44:36,040
Speaker 4: to even encourage local businesses to embrace open claw. But

864
00:44:36,239 --> 00:44:39,839
Speaker 4: the Beijing government immediately said, guys, actually be very very

865
00:44:39,840 --> 00:44:43,120
Speaker 4: careful of your data privacy. Your security banks didn't do

866
00:44:43,120 --> 00:44:45,560
Speaker 4: this as so you shouldn't do this. Be mindful of

867
00:44:45,920 --> 00:44:48,600
Speaker 4: what you're doing with this technology. And then the big

868
00:44:48,640 --> 00:44:50,719
Speaker 4: tech kind of rolled back a bit of their marketing

869
00:44:51,000 --> 00:44:54,880
Speaker 4: and you can see, actually hilariously there were advertisements for

870
00:44:55,080 --> 00:44:58,840
Speaker 4: helping these aunties and uncles how to uninstall these open claw.

871
00:44:58,640 --> 00:45:01,719
Speaker 2: On their gadget. So anyway, that's a bit of a

872
00:45:01,840 --> 00:45:02,839
Speaker 2: kind of background on that.

873
00:45:03,239 --> 00:45:05,880
Speaker 1: All right, Graceha, thank you so much for coming on

874
00:45:05,920 --> 00:45:07,960
Speaker 1: odd lot. It's great to connect with you here in

875
00:45:08,000 --> 00:45:11,200
Speaker 1: Hong Kong. And we'll have you back in three years, no,

876
00:45:11,280 --> 00:45:14,720
Speaker 1: hopefully before then. We'll have you back at least certainly

877
00:45:14,719 --> 00:45:17,080
Speaker 1: in three years to see how your predictions held up.

878
00:45:17,320 --> 00:45:30,880
Speaker 2: Thanks so much, Tracy.

879
00:45:30,920 --> 00:45:33,600
Speaker 1: That was fun. There was a lot of interesting ideas

880
00:45:33,719 --> 00:45:36,840
Speaker 1: in a fairly short conversation. But one thing specifically and

881
00:45:36,840 --> 00:45:39,920
Speaker 1: then that like sort of stands out to me is

882
00:45:40,000 --> 00:45:44,279
Speaker 1: thinking about some of these application companies, how much it

883
00:45:44,360 --> 00:45:46,959
Speaker 1: makes sense for them to sort of Yeah, they'll use

884
00:45:47,080 --> 00:45:49,759
Speaker 1: like a state of the art American closed model for

885
00:45:49,840 --> 00:45:53,319
Speaker 1: like some of the work, but then other just you know,

886
00:45:53,680 --> 00:45:56,960
Speaker 1: almost as capable models underneath, so you have like a

887
00:45:57,200 --> 00:46:00,600
Speaker 1: legal AI app like a Harvey or something. Sure, some

888
00:46:00,640 --> 00:46:03,680
Speaker 1: of these open source models work well for some tasks

889
00:46:03,719 --> 00:46:06,640
Speaker 1: within that context, And how much it makes sense to

890
00:46:06,719 --> 00:46:10,240
Speaker 1: sort of combine them under one app layer.

891
00:46:10,480 --> 00:46:14,560
Speaker 3: Yeah, Well, you don't need the cutting edge model for everything, right,

892
00:46:14,840 --> 00:46:17,960
Speaker 3: but like if you can get some of the cutting

893
00:46:18,080 --> 00:46:22,000
Speaker 3: edge model combined with like the cheapness of the open

894
00:46:22,000 --> 00:46:24,959
Speaker 3: source thing, like that seems like a pretty good deal

895
00:46:25,200 --> 00:46:26,200
Speaker 3: for a lot of companies.

896
00:46:26,440 --> 00:46:30,200
Speaker 1: Yeah, definitely. I also think like, how is the massive

897
00:46:30,320 --> 00:46:34,279
Speaker 1: manufacturing edge that China has not going to just keep

898
00:46:34,320 --> 00:46:38,480
Speaker 1: compounding itself. I mean, this is like the multi trillion

899
00:46:38,520 --> 00:46:42,279
Speaker 1: dollar question of the entire world. But it really does.

900
00:46:42,400 --> 00:46:44,680
Speaker 1: I mean, when you think about, Okay, there's a sort

901
00:46:44,680 --> 00:46:49,160
Speaker 1: of presumably natural synergy, there's all this real world physical

902
00:46:49,280 --> 00:46:54,439
Speaker 1: data that Chinese manufacturing companies can theoretically get from their

903
00:46:54,560 --> 00:46:58,480
Speaker 1: various robotic vacuum cleaners and so forth, and then feed

904
00:46:58,520 --> 00:47:02,600
Speaker 1: those into certain model then we call AI. That really

905
00:47:02,640 --> 00:47:05,400
Speaker 1: does seem like a potential leg up that you know,

906
00:47:05,520 --> 00:47:09,520
Speaker 1: just done the data collection alone from all those physical things,

907
00:47:09,560 --> 00:47:12,040
Speaker 1: like a huge potential edge over the next several years.

908
00:47:12,120 --> 00:47:14,799
Speaker 3: Yeah, although it was very interesting, Grace was saying that

909
00:47:14,840 --> 00:47:17,600
Speaker 3: it's not as developed or as structured as it is

910
00:47:17,640 --> 00:47:20,520
Speaker 3: in the US, because I had thought the same thing, like,

911
00:47:20,560 --> 00:47:23,960
Speaker 3: if everyone's talking over we chat, if everyone's paying over

912
00:47:24,040 --> 00:47:27,280
Speaker 3: we pay, you must have oodles and oodles of data.

913
00:47:27,400 --> 00:47:31,239
Speaker 3: But yeah, that was interesting. The ecosystem thing, I think

914
00:47:31,360 --> 00:47:33,960
Speaker 3: is really important, and it just seems like really hard

915
00:47:34,040 --> 00:47:37,640
Speaker 3: to get an ecosystem kind of going from scratch. And

916
00:47:37,719 --> 00:47:41,000
Speaker 3: I remember maybe it was with Dan Wong, but someone

917
00:47:41,080 --> 00:47:44,000
Speaker 3: describing how like, if you go to shen Zen, you

918
00:47:44,040 --> 00:47:48,920
Speaker 3: can basically start like an entire company manufacturing a physical

919
00:47:49,040 --> 00:47:52,640
Speaker 3: thing because every single supplier is there. You just go

920
00:47:52,719 --> 00:47:57,040
Speaker 3: from like one storefront to another storefront to another storefront.

921
00:47:57,600 --> 00:48:00,000
Speaker 3: I don't think that can be replicated anywhere in the US.

922
00:48:01,040 --> 00:48:02,640
Speaker 1: I mean, this is a bit of a tangent. But

923
00:48:02,719 --> 00:48:06,279
Speaker 1: you know this is economists talk about quote agglomeration all

924
00:48:06,320 --> 00:48:09,880
Speaker 1: the time. The advantage is exactly that of having everything

925
00:48:09,960 --> 00:48:12,600
Speaker 1: there and then you build these deep networks. One thing

926
00:48:12,600 --> 00:48:14,360
Speaker 1: I find to be a little strange, and again this

927
00:48:14,520 --> 00:48:19,279
Speaker 1: is a tangent, is how San Francisco concentrated AI is,

928
00:48:19,760 --> 00:48:22,120
Speaker 1: even though at the American level it is sort of

929
00:48:22,160 --> 00:48:25,600
Speaker 1: pure like desk work right Like, it's not the sort

930
00:48:25,600 --> 00:48:29,959
Speaker 1: of we need the bolts manufacturer, we need the servos manufacturer,

931
00:48:30,040 --> 00:48:34,200
Speaker 1: we need robovision manufacturer, and yet it's still so agglomerated.

932
00:48:34,760 --> 00:48:37,040
Speaker 1: Or the fact that finance is so agglomerated in New

933
00:48:37,120 --> 00:48:40,279
Speaker 1: York City. I find that or conglomerated. I find that

934
00:48:40,320 --> 00:48:43,440
Speaker 1: to be a little odd. So, but yeah, I don't

935
00:48:43,600 --> 00:48:47,720
Speaker 1: conglomerated is a conglomerated I don't know whether it's conglomerated

936
00:48:47,800 --> 00:48:50,799
Speaker 1: or agglomerated. I'm just gonna say glomerated. We know that

937
00:48:50,880 --> 00:48:53,359
Speaker 1: we are certain industries in the US that are quote

938
00:48:53,360 --> 00:48:57,080
Speaker 1: conglomerated one way or another. But yeah, I did find

939
00:48:57,120 --> 00:49:00,400
Speaker 1: that to be And man, these like Mini Mac billion

940
00:49:00,440 --> 00:49:03,640
Speaker 1: dollar company that's like nothing compared to the valuations that

941
00:49:03,680 --> 00:49:06,719
Speaker 1: we see with American companies. Pretty wild stuff. Also, the

942
00:49:06,719 --> 00:49:09,440
Speaker 1: point we should do more, I mean, there's a million

943
00:49:09,520 --> 00:49:11,960
Speaker 1: we gotta do we have and there are plenty of

944
00:49:12,000 --> 00:49:14,160
Speaker 1: non a I things to do, so you can't just

945
00:49:14,239 --> 00:49:16,880
Speaker 1: keep saying we should do an episode on X or Y.

946
00:49:17,360 --> 00:49:20,880
Speaker 1: But data markets and like the idea of like, okay,

947
00:49:20,920 --> 00:49:26,719
Speaker 1: these Chinese companies can buy very pricey proprietary data after

948
00:49:26,760 --> 00:49:29,839
Speaker 1: some exclusivity window. There's some interesting stuffing to down there.

949
00:49:29,920 --> 00:49:32,680
Speaker 3: Yeah, the monetization was really interesting to hear.

950
00:49:32,800 --> 00:49:34,319
Speaker 2: Okay, shall we leave it there for now?

951
00:49:34,360 --> 00:49:35,120
Speaker 1: Let's leave it there.

952
00:49:35,360 --> 00:49:37,920
Speaker 3: This has been another episode of the All Thoughts podcast.

953
00:49:38,040 --> 00:49:39,080
Speaker 2: I'm Tracy Alloway.

954
00:49:39,160 --> 00:49:41,560
Speaker 3: You can follow me at Tracy Alloway.

955
00:49:41,320 --> 00:49:43,640
Speaker 1: And I'm sure wasn't Thal. You can follow me at

956
00:49:43,680 --> 00:49:47,160
Speaker 1: the Stalwart. Follow our guest Grace Show. She's at Grace

957
00:49:47,440 --> 00:49:51,120
Speaker 1: MZ show and check out her subtck ai prom Follow

958
00:49:51,160 --> 00:49:55,080
Speaker 1: our producers Carmen Rodriguez at Carman armand Dashel Bennett at Dashbot,

959
00:49:55,160 --> 00:49:58,880
Speaker 1: Calebrooks at Cale Brooks and Kevin Lozano at Kevin Lloyd

960
00:49:58,920 --> 00:50:02,040
Speaker 1: Lozano and from our Oddlots content. Go to Bloomberg dot

961
00:50:02,040 --> 00:50:04,600
Speaker 1: com slash odd loss for the daily newsletter and all

962
00:50:04,640 --> 00:50:06,839
Speaker 1: of our episodes, and you can chat about all these

963
00:50:06,880 --> 00:50:10,640
Speaker 1: topics twenty four seven in our discord Discord dot gg

964
00:50:10,680 --> 00:50:11,640
Speaker 1: slash odlines.

965
00:50:11,760 --> 00:50:13,920
Speaker 3: And if you like odd Lots, if you want us

966
00:50:13,960 --> 00:50:16,440
Speaker 3: to do an episode on data markets, then please leave

967
00:50:16,520 --> 00:50:20,240
Speaker 3: us a positive review on your favorite podcast platform. And remember,

968
00:50:20,280 --> 00:50:22,920
Speaker 3: if you are a Bloomberg subscriber, you can listen to

969
00:50:23,040 --> 00:50:26,359
Speaker 3: all of our episodes absolutely ad free. All you need

970
00:50:26,400 --> 00:50:29,200
Speaker 3: to do is find the Bloomberg channel on Apple Podcasts

971
00:50:29,280 --> 00:50:32,120
Speaker 3: and follow the instructions there. Thanks for listening.