【UNRIVALED INVESTING】Anthropic营收3月翻3倍，万亿美金估值稳了？

1
00:00:00,340 --> 00:00:04,259
我通常不会在这个频道讨论私营公司

2
00:00:04,259 --> 00:00:06,719
它们不容易投资

3
00:00:06,719 --> 00:00:11,259
而且用这些信息很难采取实际行动

4
00:00:11,259 --> 00:00:15,419
但有时候新闻非常重大

5
00:00:15,419 --> 00:00:16,559
意义深远

6
00:00:16,559 --> 00:00:20,339
在社会层面必须讨论

7
00:00:20,339 --> 00:00:22,359
我认为这是关键时刻

8
00:00:22,359 --> 00:00:23,600
Anthropic

9
00:00:23,600 --> 00:00:25,699
这家前沿人工智能实验室

10
00:00:25,699 --> 00:00:28,809
爪形模型宣布推出新AI模型

11
00:00:28,809 --> 00:00:34,119
如此强大且危险以至于拒绝公开发布

12
00:00:34,520 --> 00:00:38,179
相反他们与苹果合作

13
00:00:38,179 --> 00:00:39,100
微软

14
00:00:39,100 --> 00:00:40,759
谷歌亚马逊

15
00:00:40,759 --> 00:00:46,179
Crowdstrike等公司联手强化全球数字基建

16
00:00:46,179 --> 00:00:49,710
在这期视频中

17
00:00:49,710 --> 00:00:53,109
我们将解析该模型的能力

18
00:00:53,109 --> 00:00:55,570
为何关乎国家安全

19
00:00:55,570 --> 00:00:58,310
而且因为这是投资频道

20
00:00:58,310 --> 00:00:59,719
我会分享看法

21
00:00:59,719 --> 00:01:02,500
Arrap与Anthropic估值

22
00:01:02,500 --> 00:01:07,079
为何我认为它可能达到万亿美元估值

23
00:01:07,079 --> 00:01:09,920
以及如何获取投资机会

24
00:01:09,920 --> 00:01:11,569
即使微小也不放过

25
00:01:11,569 --> 00:01:17,579
首先背景介绍Anthropic成立于2021年

26
00:01:17,579 --> 00:01:18,379
由Dario和Daniella创立

27
00:01:18,379 --> 00:01:20,510
曾任职OpenAI高管

28
00:01:20,510 --> 00:01:24,420
Anthropic早期团队多数来自OpenAI

29
00:01:24,420 --> 00:01:29,359
他们离开是为了专注AI安全

30
00:01:29,359 --> 00:01:34,290
这一背景对视频内容至关重要

31
00:01:34,290 --> 00:01:35,290
4月7日

32
00:01:35,290 --> 00:01:39,069
Anthropic发布Claude Mythos预览

33
00:01:39,069 --> 00:01:42,209
这是新一代前沿模型

34
00:01:42,209 --> 00:01:46,730
代表AI技术最前沿

35
00:01:46,730 --> 00:01:48,810
通用型模型

36
00:01:48,810 --> 00:01:50,109
擅长编程

37
00:01:50,109 --> 00:01:50,890
逻辑推理

38
00:01:50,890 --> 00:01:52,459
常规功能

39
00:01:52,459 --> 00:01:56,560
但独特之处在于网络安全能力

40
00:01:56,560 --> 00:01:59,549
这次是质的飞跃

41
00:01:59,549 --> 00:02:01,129
显著提升

42
00:02:01,129 --> 00:02:03,150
非渐进式改进

43
00:02:04,189 --> 00:02:04,409
现在

44
00:02:04,409 --> 00:02:05,290
Anthropic表示

45
00:02:05,290 --> 00:02:12,659
Mythos仅过去几周就发现数千个零日漏洞

46
00:02:12,659 --> 00:02:18,870
涉及所有主流操作系统和浏览器

47
00:02:18,870 --> 00:02:21,310
其中绝大多数

48
00:02:21,310 --> 00:02:24,930
据Anthropic称均由模型自主发现

49
00:02:24,930 --> 00:02:30,259
且开发了相关漏洞利用方案无需人工干预

50
00:02:30,259 --> 00:02:35,810
部分漏洞已潜伏数十年未被发现

51
00:02:35,810 --> 00:02:37,710
最古老的公开案例

52
00:02:37,710 --> 00:02:38,330
我认为

53
00:02:38,330 --> 00:02:44,030
是一个已被修补的两年七个月前的操作系统漏洞

54
00:02:44,030 --> 00:02:49,120
被广泛认为是全球安全防护最严密的系统之一

55
00:02:49,120 --> 00:02:53,610
而真正令人震惊的部分在于

56
00:02:53,610 --> 00:02:57,189
他们的工程师没有接受过正式的安全培训

57
00:02:57,189 --> 00:03:01,969
让Mythos彻夜寻找漏洞然后他们入睡

58
00:03:01,969 --> 00:03:06,360
醒来时发现已有一个完整的漏洞利用程序

59
00:03:06,360 --> 00:03:10,340
现在情况更离奇了在内部安全测试中

60
00:03:10,340 --> 00:03:13,500
一名研究人员将Mythos置于所谓的沙盒环境

61
00:03:13,500 --> 00:03:16,849
可以想象这是一个密封的数字房间

62
00:03:16,849 --> 00:03:19,490
AI可以在这个房间内运行

63
00:03:19,490 --> 00:03:22,349
但完全与外界隔绝

64
00:03:22,349 --> 00:03:23,550
没有网络访问权限

65
00:03:23,550 --> 00:03:26,520
无法进行外部通信

66
00:03:26,520 --> 00:03:30,259
这是一个受控的测试环境

67
00:03:30,419 --> 00:03:34,800
研究人员让Mythos尝试逃出沙盒

68
00:03:34,800 --> 00:03:35,909
它成功做到了

69
00:03:35,909 --> 00:03:38,270
但它并未就此停止

70
00:03:38,270 --> 00:03:43,460
Mythos设计了多步骤漏洞利用程序以获取全面网络访问

71
00:03:43,460 --> 00:03:45,020
从沙盒内部

72
00:03:45,020 --> 00:03:47,419
随后它向邮箱发送了邮件

73
00:03:47,419 --> 00:03:50,500
寄给正在公园吃三明治的研究员

74
00:03:50,500 --> 00:03:53,990
研究人员发现了这次逃脱

75
00:03:53,990 --> 00:03:56,629
并非通过监控工具

76
00:03:56,629 --> 00:04:01,378
而是因为收到了AI的意外邮件

77
00:04:01,378 --> 00:04:02,949
这简直令人难以置信

78
00:04:02,949 --> 00:04:04,550
而故事并未止步于此

79
00:04:04,550 --> 00:04:11,870
它还向多个公开网站发布漏洞详情以展示成果

80
00:04:12,270 --> 00:04:14,550
在其他记录案例中

81
00:04:14,550 --> 00:04:18,970
Mythos早期版本执行了不应有的操作

82
00:04:18,970 --> 00:04:22,389
并主动试图掩盖自己的行为

83
00:04:22,389 --> 00:04:26,569
有一次它通过不允许的方法

84
00:04:26,569 --> 00:04:30,908
得出正确答案

85
00:04:30,908 --> 00:04:35,009
然后推断最终答案可能不够准确

86
00:04:35,009 --> 00:04:38,199
以免引起怀疑

87
00:04:38,519 --> 00:04:40,939
我需要客观看待

88
00:04:40,939 --> 00:04:44,500
其他AI模型也曾逃出沙盒

89
00:04:44,500 --> 00:04:49,970
但Mythos展现的复杂性和主动性

90
00:04:49,970 --> 00:04:53,970
结合其网络攻击能力

91
00:04:53,970 --> 00:04:57,170
使其处于不同层次

92
00:04:57,769 --> 00:05:00,249
那么Anthropic如何应对

93
00:05:00,249 --> 00:05:03,108
他们启动了玻璃翼计划

94
00:05:03,108 --> 00:05:06,910
这是防御性网络安全项目

95
00:05:13,430 --> 00:05:16,279
如同隐藏的软件漏洞

96
00:05:16,279 --> 00:05:19,899
玻璃翼有十二家核心合作机构

97
00:05:19,899 --> 00:05:22,379
包括多家主要企业

98
00:05:22,379 --> 00:05:23,060
你知道的

99
00:05:23,060 --> 00:05:23,639
亚马逊

100
00:05:23,639 --> 00:05:24,100
苹果

101
00:05:24,100 --> 00:05:24,939
微软

102
00:05:24,939 --> 00:05:25,540
谷歌

103
00:05:25,540 --> 00:05:26,420
Crowdstrike

104
00:05:26,420 --> 00:05:26,819
帕洛

105
00:05:26,819 --> 00:05:27,459
Alto networks

106
00:05:27,459 --> 00:05:28,220
思科

107
00:05:28,220 --> 00:05:29,000
博通

108
00:05:29,000 --> 00:05:30,220
摩根大通

109
00:05:30,220 --> 00:05:32,519
英伟达与Linux基金会

110
00:05:32,519 --> 00:05:34,639
总计约四十家组织

111
00:05:34,639 --> 00:05:37,550
将获得Mythos预览权限

112
00:05:37,550 --> 00:05:39,600
核心思路很简单

113
00:05:39,600 --> 00:05:42,199
为防御者争取先机

114
00:05:42,199 --> 00:05:46,829
让这些组织使用Mythos扫描自身系统

115
00:05:46,829 --> 00:05:51,009
发现漏洞并提前修复

116
00:05:51,009 --> 00:05:55,798
具备类似能力的模型如今已广泛可用

117
00:05:55,798 --> 00:06:00,959
因为现在这里有个令人不安的现实

118
00:06:00,959 --> 00:06:07,990
Anthropic已警告高层官员，Mythos使大规模网络攻击显著更易发生

119
00:06:07,990 --> 00:06:11,290
今年他们不会向公众发布此模型

120
00:06:11,290 --> 00:06:14,759
我认为大多数理性人都会认同

121
00:06:14,759 --> 00:06:16,540
现阶段这是正确的决定

122
00:06:16,540 --> 00:06:18,519
但需在部署适当防护措施前

123
00:06:18,519 --> 00:06:20,149
但需考虑其影响

124
00:06:20,149 --> 00:06:21,790
如果非美

125
00:06:21,790 --> 00:06:26,920
公司或敌对国家率先掌握此能力

126
00:06:26,920 --> 00:06:30,959
若该模型能发现所有主流操作系统漏洞

127
00:06:30,959 --> 00:06:33,038
在所有主流浏览器中

128
00:06:33,038 --> 00:06:35,399
是否可能被用于瘫痪

129
00:06:35,399 --> 00:06:35,838
假设

130
00:06:35,838 --> 00:06:36,798
金融系统

131
00:06:36,798 --> 00:06:38,059
关键基础设施

132
00:06:38,059 --> 00:06:39,478
互联网本身

133
00:06:39,478 --> 00:06:43,189
这些已不再是假设

134
00:06:43,189 --> 00:06:46,370
这就是我们身处的现实

135
00:06:46,370 --> 00:06:49,149
更紧迫的是

136
00:06:49,149 --> 00:06:50,410
开源

137
00:06:50,410 --> 00:06:53,209
AI模型通常落后前沿

138
00:06:53,209 --> 00:06:56,139
私营模型可能滞后6至12个月

139
00:06:56,139 --> 00:06:58,300
我们正处于关键窗口期

140
00:06:58,300 --> 00:07:04,009
Project Glasswing需强化全球数字基建

141
00:07:04,009 --> 00:07:09,209
在开源模型发布前

142
00:07:09,209 --> 00:07:12,730
且任何人皆可立即运行

143
00:07:12,730 --> 00:07:15,870
这也值得从历史角度思考

144
00:07:15,870 --> 00:07:23,100
我们目睹了数字大规模杀伤性武器的发现与构建

145
00:07:23,100 --> 00:07:26,449
如同二战期间

146
00:07:26,449 --> 00:07:30,509
我们拥有洛斯阿拉莫斯和曼哈顿计划

147
00:07:30,509 --> 00:07:33,370
由政府主导

148
00:07:33,370 --> 00:07:37,110
如今由私营企业做决策

149
00:07:37,110 --> 00:07:40,370
我看到相关讨论开始浮现

150
00:07:40,370 --> 00:07:43,730
这些实验室是否应该国有化

151
00:07:43,730 --> 00:07:48,428
至少应大幅加强政府监管

152
00:07:48,428 --> 00:07:49,488
这是一个争论

153
00:07:49,488 --> 00:07:53,129
我认为这值得讨论

154
00:07:53,129 --> 00:07:55,230
关于玻璃翼的最后一个想法

155
00:07:55,230 --> 00:07:57,850
合作伙伴名单值得注意

156
00:07:57,850 --> 00:08:02,220
CrowdStrike和Palo Alto Networks是他们的主要安全公司

157
00:08:02,220 --> 00:08:06,410
Anthropic还表示超过四十家组织获得访问权限

158
00:08:06,410 --> 00:08:07,769
但未公开名单

159
00:08:07,769 --> 00:08:07,930
所以

160
00:08:07,930 --> 00:08:10,100
我们不能声称独家性

161
00:08:10,100 --> 00:08:13,439
但我现在也看不到竞争对手

162
00:08:13,439 --> 00:08:16,730
这能让CrowdStrike获得显著优势吗

163
00:08:16,730 --> 00:08:19,089
是否使其更加不可撼动

164
00:08:19,089 --> 00:08:19,819
假设

165
00:08:19,819 --> 00:08:20,339
你知道的

166
00:08:20,339 --> 00:08:23,439
寻找下一代AI驱动的网络攻击

167
00:08:23,439 --> 00:08:24,079
可能会

168
00:08:24,079 --> 00:08:26,488
这是值得关注的

169
00:08:26,488 --> 00:08:27,949
现在谈谈数据

170
00:08:27,949 --> 00:08:29,889
因为这是顶级投资

171
00:08:29,889 --> 00:08:31,329
这是一个投资频道

172
00:08:31,329 --> 00:08:36,690
这里的财务故事和技术一样令人震惊

173
00:08:36,690 --> 00:08:41,690
Anthropic最近在G轮融资中筹集了30亿美元

174
00:08:41,690 --> 00:08:45,370
这发生在2月，估值约3800亿美元

175
00:08:45,370 --> 00:08:46,250
就在昨天

176
00:08:46,250 --> 00:08:48,399
随着Mythos宣布

177
00:08:48,399 --> 00:08:53,169
他们披露收入年化率已超30亿美元

178
00:08:53,169 --> 00:08:56,169
相比2025年底的约90亿美元

179
00:08:56,169 --> 00:08:57,240
让我再说一遍

180
00:08:57,240 --> 00:09:03,589
他们三个月内从90亿增至30亿年化收入

181
00:09:03,589 --> 00:09:05,249
我确实认为

182
00:09:05,249 --> 00:09:13,049
历史上没有任何公司能在单季度增加约20亿年化收入

183
00:09:13,049 --> 00:09:14,549
在单个季度

184
00:09:14,549 --> 00:09:16,259
令人难以置信

185
00:09:16,259 --> 00:09:21,139
企业采用是推动Anthropic增长的关键

186
00:09:21,139 --> 00:09:23,500
他们在2月宣布G轮融资时

187
00:09:23,500 --> 00:09:27,980
已有500多家企业客户年支出超百万美元

188
00:09:27,980 --> 00:09:33,980
这个数字在两个月内已翻倍至1000家以上

189
00:09:33,980 --> 00:09:35,139
从这个角度看

190
00:09:35,139 --> 00:09:36,580
这是目前的传闻

191
00:09:36,580 --> 00:09:39,049
从对比来看OpenAI

192
00:09:39,049 --> 00:09:42,639
刚完成1220亿美元融资

193
00:09:42,639 --> 00:09:45,059
估值1850亿美元

194
00:09:45,059 --> 00:09:47,019
预计产生约2

195
00:09:47,019 --> 00:09:49,500
50亿美元年化收入

196
00:09:49,500 --> 00:09:53,659
这可能表明Anthropic实际严重低估

197
00:09:53,659 --> 00:09:56,730
在3800亿估值和30亿年化收入下

198
00:09:56,730 --> 00:10:01,029
这些前沿实验室的商业模式很有趣

199
00:10:01,029 --> 00:10:03,779
一旦训练好模型就开始部署

200
00:10:03,779 --> 00:10:06,120
这是一项相对高利润的工作

201
00:10:06,120 --> 00:10:12,568
考虑利润时，模型训练成本约占收入的40%到60%

202
00:10:12,568 --> 00:10:15,509
最昂贵的部分是训练下一代模型

203
00:10:15,509 --> 00:10:20,399
这就是这些公司需要筹集巨额资本的原因

204
00:10:20,399 --> 00:10:22,460
这不是为了维持现有业务

205
00:10:22,460 --> 00:10:27,830
而是为争夺更智能模型的研发竞赛提供资金

206
00:10:27,830 --> 00:10:33,870
OpenAI在训练投入上是Anthropic的数倍之多

207
00:10:33,870 --> 00:10:34,509
这堵墙

208
00:10:34,509 --> 00:10:41,678
《华尔街日报》发布泄露财报显示Anthropic预计2028年实现收支平衡

209
00:10:41,678 --> 00:10:46,029
而OpenAI预计要到2030年才能达到这一目标

210
00:10:46,389 --> 00:10:49,850
根据这些泄露文件的各类预测

211
00:10:49,850 --> 00:10:50,470
再次强调

212
00:10:50,470 --> 00:10:52,679
我的依据是《华尔街日报》文章

213
00:10:52,679 --> 00:10:55,870
Anthropic目标在2027年实现约50亿美元营收

214
00:10:55,870 --> 00:10:59,470
并在2027年达到50亿美元营收

215
00:10:59,470 --> 00:11:01,269
潜在可能达到148亿美元

216
00:11:01,269 --> 00:11:03,519
到2029年或达148亿美元

217
00:11:03,519 --> 00:11:05,600
如果训练成本保持可控

218
00:11:05,600 --> 00:11:10,779
在这一规模下利润率可能极其可观

219
00:11:10,779 --> 00:11:16,379
预训练投资可能带来500亿美元以上的利润

220
00:11:16,379 --> 00:11:18,870
假设采用20倍市盈率

221
00:11:18,870 --> 00:11:21,049
估值可达约1万亿美元

222
00:11:21,049 --> 00:11:25,350
投资者考虑更高估值并不难理解

223
00:11:25,350 --> 00:11:27,799
比如采用40倍预训练利润倍数

224
00:11:27,799 --> 00:11:30,720
估值将接近2万亿美元

225
00:11:30,720 --> 00:11:31,799
如果在C阶段

226
00:11:31,799 --> 00:11:34,340
如果Anthropic能兑现这些预期

227
00:11:34,340 --> 00:11:35,120
我推测

228
00:11:35,120 --> 00:11:38,409
许多Anthropic投资者押注

229
00:11:38,409 --> 00:11:43,429
未来可能实现超过1万亿美元的估值

230
00:11:43,429 --> 00:11:44,649
在某个时刻

231
00:11:44,649 --> 00:11:50,619
总有人会押注无法经济自洽的训练项目

232
00:11:50,619 --> 00:11:53,458
如果你有150智商的模型

233
00:11:53,458 --> 00:11:56,058
普通人真的需要200智商的模型吗

234
00:11:56,058 --> 00:12:00,328
或许当更高智能带来真正的科学突破

235
00:12:00,328 --> 00:12:01,649
治愈疾病

236
00:12:01,649 --> 00:12:02,688
你知道的

237
00:12:02,688 --> 00:12:05,279
设计更优材料

238
00:12:05,279 --> 00:12:05,960
那确实

239
00:12:05,960 --> 00:12:07,809
这里显然有价值

240
00:12:07,809 --> 00:12:10,529
但300智商模型的需求曲线

241
00:12:10,529 --> 00:12:13,919
可能不同于120智商模型

242
00:12:13,919 --> 00:12:15,779
既能有趣互动又能帮助撰写

243
00:12:15,779 --> 00:12:19,818
邮件等常规内容可上传至人形机器人

244
00:12:19,818 --> 00:12:22,399
这才是真正有趣的部分

245
00:12:22,399 --> 00:12:25,940
从商业模式角度看

246
00:12:25,940 --> 00:12:32,919
我认为Mythos正在展示前沿智能的变现未来

247
00:12:32,919 --> 00:12:35,539
并非作为消费产品

248
00:12:35,539 --> 00:12:37,879
而是可能作为合资企业

249
00:12:37,879 --> 00:12:38,789
想想看

250
00:12:38,789 --> 00:12:42,429
如果你是一家特种化学品公司

251
00:12:42,429 --> 00:12:48,330
而一个三百智商的模型能发现一种新型聚合物

252
00:12:48,330 --> 00:12:49,529
带来十年的竞争优势

253
00:12:49,529 --> 00:12:51,909
这对你们价值数十亿美元

254
00:12:51,909 --> 00:12:59,190
如果你是电池公司，前沿模型能设计下一代固态电池

255
00:12:59,190 --> 00:13:03,149
你会不惜一切代价获得这种智能的独家访问权

256
00:13:03,149 --> 00:13:04,610
制药行业也是如此

257
00:13:04,610 --> 00:13:05,090
材料科学

258
00:13:05,090 --> 00:13:05,850
科学研究

259
00:13:05,850 --> 00:13:06,879
国防领域

260
00:13:06,879 --> 00:13:08,580
半导体设计

261
00:13:08,580 --> 00:13:15,120
极端智能的价值在经济中分布不均

262
00:13:15,120 --> 00:13:17,440
集中在特定行业

263
00:13:17,440 --> 00:13:22,889
单一突破价值可能超过模型训练成本

264
00:13:22,889 --> 00:13:27,198
商业模式可能不是销售

265
00:13:27,198 --> 00:13:31,120
假设以每百万二十美元向所有人开放API访问

266
00:13:31,120 --> 00:13:31,740
代币

267
00:13:31,740 --> 00:13:36,899
可能与辉瑞或波音或洛克希德合作

268
00:13:36,899 --> 00:13:37,320
马丁

269
00:13:37,320 --> 00:13:38,589
或许像Palantir

270
00:13:38,589 --> 00:13:41,948
给予他们在前沿模型的独家访问权

271
00:13:41,948 --> 00:13:47,039
针对其领域进行微调并共享发现的经济收益

272
00:13:47,039 --> 00:13:51,600
这与OpenAI的构建方式截然不同

273
00:13:51,600 --> 00:13:55,919
每周有九亿用户使用BT

274
00:13:55,960 --> 00:13:58,960
而Mythos可能是首个真正概念验证项目

275
00:13:58,960 --> 00:14:03,139
Glasswing目前已成为网络安全合资企业

276
00:14:03,139 --> 00:14:06,519
Anthropic不通过API销售Mythos访问权限

277
00:14:06,519 --> 00:14:09,799
而是与精选组织合作

278
00:14:09,799 --> 00:14:13,879
提供他人无法获得的能力

279
00:14:13,879 --> 00:14:17,309
输出成果进行防御性共享

280
00:14:17,309 --> 00:14:17,730
现在

281
00:14:17,730 --> 00:14:22,970
想象这个模型应用于药物研发或储能领域

282
00:14:22,970 --> 00:14:28,129
每个领域都可能形成Glasswing模式的合作

283
00:14:28,129 --> 00:14:31,330
这也揭示了行业更大的趋势

284
00:14:31,330 --> 00:14:32,929
以及我们可能的发展方向

285
00:14:32,929 --> 00:14:37,610
未来或许会出现越来越多的私有模型

286
00:14:37,610 --> 00:14:39,850
从未向公众开放的模型

287
00:14:39,850 --> 00:14:43,610
并非出于刻意保密

288
00:14:43,610 --> 00:14:52,599
而是因其能力过于珍贵或危险无法泛用化

289
00:14:52,599 --> 00:14:53,038
例如

290
00:14:53,038 --> 00:14:56,839
120智商模型成为消费级产品

291
00:14:56,839 --> 00:14:59,250
300智商模型

292
00:14:59,250 --> 00:15:02,309
数据中心的天才模型成为私有资产

293
00:15:02,309 --> 00:15:08,730
由愿意付费的企业和政府联合资助并独家使用

294
00:15:08,730 --> 00:15:11,000
这将形成截然不同的AI生态

295
00:15:11,000 --> 00:15:14,529
与大多数人想象的完全不同

296
00:15:14,529 --> 00:15:18,409
那么公众投资者如何参与Anthropic

297
00:15:18,409 --> 00:15:19,610
这是一家私营公司

298
00:15:19,610 --> 00:15:22,940
无法直接在股市购买股票

299
00:15:22,940 --> 00:15:28,328
间接途径可能是通过Alphabet或GOOG股票

300
00:15:28,328 --> 00:15:33,048
法院文件显示，谷歌拥有安托尼科公司大约14%的股份

301
00:15:33,048 --> 00:15:34,249
持股上限为15%

302
00:15:34,249 --> 00:15:35,669
他们没有投票权

303
00:15:35,669 --> 00:15:36,789
没有董事会席位

304
00:15:36,789 --> 00:15:38,309
但持有14%的股权

305
00:15:38,309 --> 00:15:43,210
以当前市值计算，3800亿美元相当于超过500亿美元

306
00:15:43,210 --> 00:15:46,750
字母表公司的总市值约为3到4万亿美元

307
00:15:46,750 --> 00:15:51,980
因此安托尼科公司的这笔投资相对于字母表的价值而言微不足道

308
00:15:51,980 --> 00:15:54,980
现在投资字母表实际上是在押注其其他业务

309
00:15:54,980 --> 00:15:56,349
广告和云计算业务

310
00:15:56,349 --> 00:15:58,229
你知道他们自己的AI模型

311
00:15:58,229 --> 00:16:02,350
但可以说你体内也嵌入了AI彩票券

312
00:16:02,350 --> 00:16:07,580
如果Anthropic能成功推出远超同类的AI

313
00:16:07,580 --> 00:16:11,690
其他人现在也有传言

314
00:16:11,690 --> 00:16:15,710
那么Anthropic将在2026年晚些时候上市

315
00:16:15,710 --> 00:16:17,830
这将是最快捷的方式

316
00:16:17,830 --> 00:16:20,110
但那时估值可能已飙升

317
00:16:20,110 --> 00:16:26,220
而且谁知道在这期间会出现哪些新模型

318
00:16:26,220 --> 00:16:31,980
我们该如何评估Anthropic的价值？估值这些AI实验室极其困难

319
00:16:31,980 --> 00:16:35,240
他们并非在真空中运作

320
00:16:35,240 --> 00:16:39,039
也许最终是谷歌的Gemini胜出

321
00:16:39,039 --> 00:16:40,779
也许会是OpenAI的模型

322
00:16:40,779 --> 00:16:42,759
或者是马斯克的Grok

323
00:16:42,759 --> 00:16:47,818
或许整个行业将走向同质化并压缩利润空间

324
00:16:47,818 --> 00:16:52,778
全球正掀起一股打造最前沿模型的狂潮

325
00:16:52,778 --> 00:16:57,198
而昨日Anthropic展示了为何美国必须保持领先地位

326
00:16:57,198 --> 00:16:59,610
美国需要在这场竞赛中保持领先

327
00:16:59,610 --> 00:17:01,549
我们正经历一个特殊时刻

328
00:17:01,549 --> 00:17:05,079
一家私营公司打造了能发现并利用漏洞的AI

329
00:17:05,079 --> 00:17:08,000
能够发现并利用漏洞的AI

330
00:17:08,000 --> 00:17:12,480
地球上几乎每款关键软件中

331
00:17:12,480 --> 00:17:16,119
他们展现了明智之举未公开发布

332
00:17:16,119 --> 00:17:20,179
而是联合起来共同防御

333
00:17:20,179 --> 00:17:22,699
这就是我们如今身处的世界

334
00:17:22,699 --> 00:17:26,078
人工智能的发展速度并未放缓

335
00:17:26,078 --> 00:17:27,098
反而在加速推进

336
00:17:27,098 --> 00:17:30,119
事实上昨日的进展已证明这一点

337
00:17:30,119 --> 00:17:31,990
以惊人的速度不断前进

338
00:17:31,990 --> 00:17:33,750
如果你觉得这个视频

339
00:17:33,750 --> 00:17:35,250
有帮助或有价值

340
00:17:35,250 --> 00:17:36,609
点击订阅按钮

341
00:17:36,609 --> 00:17:39,450
如果你想加入一个我分享观点的社群

342
00:17:39,450 --> 00:17:42,670
首先浏览Unrivaled Investing

343
00:17:42,670 --> 00:17:47,009
我也制作独家视频介绍近期买入的股票

344
00:17:47,009 --> 00:17:50,250
对于Unrivaled Investing的会员来说

345
00:17:50,250 --> 00:17:52,869
我上周刚做了一期某股票的视频

346
00:17:52,869 --> 00:17:55,219
这是我正在增持的持仓之一

347
00:17:55,219 --> 00:17:57,798
如果你想的话

348
00:17:57,798 --> 00:18:01,670
你知道在投资路上利用AI工具

349
00:18:01,670 --> 00:18:03,430
我是Flash的联合创始人

350
00:18:03,430 --> 00:18:05,869
你可以查看视频描述下方的链接

351
00:18:05,869 --> 00:18:08,849
Flash 正在利用这些人工智能模型

352
00:18:08,849 --> 00:18:10,269
因此随着人工智能的进步

353
00:18:10,269 --> 00:18:12,230
你的投资旅程也会更出色

354
00:18:12,230 --> 00:18:14,690
因为 Flash 可以构建

355
00:18:14,690 --> 00:18:16,630
为你生成一份深度研究报告

356
00:18:16,630 --> 00:18:20,230
它还能创建信息图

357
00:18:20,230 --> 00:18:21,230
制作一档播客

358
00:18:21,230 --> 00:18:22,890
一段视频，这样一来

359
00:18:22,890 --> 00:18:25,049
如果你对某家公司有疑问

360
00:18:25,049 --> 00:18:26,569
它可以帮助你解答

361
00:18:26,569 --> 00:18:28,390
顺便说一下还有投资组合分析

362
00:18:28,390 --> 00:18:29,509
这里内容非常丰富

363
00:18:29,509 --> 00:18:30,930
所以如果你打算

364
00:18:30,930 --> 00:18:31,569
你知道的

365
00:18:31,569 --> 00:18:32,650
自己动手尝试

366
00:18:32,650 --> 00:18:33,710
自己亲自动手

367
00:18:33,710 --> 00:18:34,650
这样的旅程

368
00:18:34,650 --> 00:18:38,059
借助人工智能来辅助你

369
00:18:38,059 --> 00:18:38,900
这就是 Flash

370
00:18:38,900 --> 00:18:40,900
然后我分享自己的亲身经历

371
00:18:40,900 --> 00:18:44,460
帮助你在非凡投资领域获得灵感

372
00:18:44,460 --> 00:18:44,700
你知道的

373
00:18:44,700 --> 00:18:45,559
我正在分享

374
00:18:45,559 --> 00:18:46,240
我正在买入的标的

375
00:18:46,240 --> 00:18:47,309
我正在卖出的标的

376
00:18:47,309 --> 00:18:49,750
你将获得实时更新和独家视频

377
00:18:49,750 --> 00:18:50,990
如果你喜欢这个视频

378
00:18:50,990 --> 00:18:56,339
订阅者可在非凡投资获取更多独家视频

379
00:18:56,339 --> 00:18:59,059
非常感谢观看非凡投资

评论采集失败

采集完成，该视频无弹幕