WEBVTT

00:00:00.000 --> 00:00:03.340
Hello everybody, welcome to my channel

00:00:03.340 --> 00:00:06.640
老田们,今天咱们聊一个劲爆的话题啊

00:00:06.640 --> 00:00:10.900
三哥们是怎么把硅谷最强公司给干趴下的

00:00:10.900 --> 00:00:12.140
事情是这样啊

00:00:12.140 --> 00:00:17.760
就在前几天,AI算力圈的风向标分析发布了一份报告

00:00:17.760 --> 00:00:20.360
直接给谷歌下了死亡盘决数了

00:00:20.360 --> 00:00:22.340
报告原话是这么说的啊

00:00:22.340 --> 00:00:24.940
说谷歌还会继续发布模型

00:00:24.940 --> 00:00:28.900
但是重回巅峰的可能性已经降到了零

00:00:28.900 --> 00:00:30.680
啥概念啊兄弟们

00:00:30.680 --> 00:00:31.960
这就好比啊

00:00:31.960 --> 00:00:33.860
博尔特跑着跑着突然说

00:00:33.860 --> 00:00:35.080
我不行了

00:00:35.080 --> 00:00:37.480
我改行卖烤串吧

00:00:37.480 --> 00:00:39.180
谁也没想到吧

00:00:39.180 --> 00:00:41.800
AI竞赛里面的第一个掉队的

00:00:41.800 --> 00:00:44.040
不是天天画饼的马斯克

00:00:44.040 --> 00:00:45.920
也不是心源一马的小诈

00:00:45.920 --> 00:00:49.220
既然是AI的开山鼻祖谷歌

00:00:49.220 --> 00:00:51.580
Transformer是谷歌发明的啊

00:00:51.580 --> 00:00:54.340
AI聊天机器人也是谷歌先捏出来的

00:00:54.340 --> 00:00:58.220
结果最后拿着这俩东西颠覆世界的

00:00:58.220 --> 00:00:59.380
偏偏是别人

00:00:59.380 --> 00:01:01.720
这就跟你自己种了个稀果

00:01:01.720 --> 00:01:02.960
结果被别人摘了

00:01:02.960 --> 00:01:04.100
还卖了个好价格

00:01:04.100 --> 00:01:05.500
你只能在旁边干瞪眼

00:01:05.500 --> 00:01:06.740
那是一模一样的

00:01:06.740 --> 00:01:09.420
谷歌的柯达诗科又来了

00:01:09.420 --> 00:01:10.820
其实这件事

00:01:10.820 --> 00:01:12.040
它早就有征兆了

00:01:12.040 --> 00:01:15.840
当年柯达首先发明了数码相机技术

00:01:15.840 --> 00:01:17.480
结果管理层觉得

00:01:17.480 --> 00:01:20.940
这玩意会影响咱们的胶片业务

00:01:20.940 --> 00:01:22.800
硬是就按住不发展

00:01:22.800 --> 00:01:25.060
结果自己把自己给做死了

00:01:25.060 --> 00:01:28.020
谷歌是干了一件一模一样的事

00:01:28.020 --> 00:01:32.160
在OpenAI发布ChatGDP之前整整的一年

00:01:32.160 --> 00:01:35.760
谷歌就已经研发出了大语言模型

00:01:35.760 --> 00:01:36.460
Lamida

00:01:36.460 --> 00:01:38.800
现代AI的基础

00:01:38.800 --> 00:01:41.740
Transformer的框架也是谷歌创立的

00:01:41.740 --> 00:01:45.660
结果当时的谷歌已经换上了一群三哥高果

00:01:45.660 --> 00:01:47.340
那天天就是喝咖啡

00:01:47.340 --> 00:01:51.900
在这些眼界高远的印度管理层眼中

00:01:51.900 --> 00:01:54.700
这玩意可能影响谷歌搜索业务

00:01:54.700 --> 00:01:56.360
不发布直接血仓

00:01:56.360 --> 00:02:00.500
谷歌的CEO皮查一后来还出来解释

00:02:00.500 --> 00:02:02.920
说是因为安全考虑

00:02:02.920 --> 00:02:04.520
说质量不行

00:02:04.520 --> 00:02:05.960
兄弟们你信吗

00:02:05.960 --> 00:02:08.800
一个手握全球最多数据

00:02:08.800 --> 00:02:10.140
最多算力

00:02:10.140 --> 00:02:12.280
最顶尖科学家的公司

00:02:12.280 --> 00:02:14.140
你跟我说质量不行

00:02:14.140 --> 00:02:15.120
你糊弄鬼呢

00:02:15.120 --> 00:02:17.940
典型的印度式的短视思维

00:02:17.940 --> 00:02:20.340
他们只看眼前这三亩丁

00:02:20.340 --> 00:02:22.300
看不到外面整个太平洋啊

00:02:22.300 --> 00:02:23.040
好

00:02:23.040 --> 00:02:24.400
你以为这就完了

00:02:24.400 --> 00:02:25.760
突央突三婆

00:02:25.760 --> 00:02:29.780
8月5号2024年诺贝尔化学奖得主

00:02:29.780 --> 00:02:32.600
AlphaGo和AlphaFold的缔造者

00:02:32.600 --> 00:02:37.160
被称为硅谷AI科学灵魂的哈萨比斯

00:02:37.160 --> 00:02:39.760
被割除了迪普麦的CEO的职位

00:02:39.760 --> 00:02:40.960
就在同一天

00:02:40.960 --> 00:02:45.620
效力谷歌27年的传奇工程师杰弗迪恩也走了

00:02:45.620 --> 00:02:47.560
这哥们是什么级别呀

00:02:47.560 --> 00:02:48.960
谷歌内部有个段子

00:02:48.960 --> 00:02:52.880
说他写代码的速度超过别人阅读代码的速度

00:02:52.880 --> 00:02:55.260
他每次编译前都不查语法

00:02:55.260 --> 00:02:56.900
如果编译器报错

00:02:56.900 --> 00:02:58.720
那一定是编译器自己错了

00:02:58.720 --> 00:03:00.220
编译器都不敢报错

00:03:00.220 --> 00:03:02.020
只敢向他请教语法

00:03:02.020 --> 00:03:03.560
就这么个神级人物

00:03:03.560 --> 00:03:04.000
走了

00:03:04.000 --> 00:03:05.740
两个月前啊

00:03:05.740 --> 00:03:08.380
Transformant的论文的另一位核心作者

00:03:08.380 --> 00:03:09.800
诺贝尔化学奖得主

00:03:09.800 --> 00:03:11.860
No Man转投了OpenAI

00:03:11.860 --> 00:03:14.360
AlphaFold的核心领导者

00:03:14.360 --> 00:03:16.180
Jumper转投了Ansuric

00:03:16.180 --> 00:03:17.820
注意啊兄弟们

00:03:17.820 --> 00:03:20.140
这些人可不是一个人走的

00:03:20.140 --> 00:03:22.540
每个人都是拉走整整一个队伍

00:03:22.540 --> 00:03:23.480
谷歌

00:03:23.480 --> 00:03:25.240
Jimny AI模型啊

00:03:25.240 --> 00:03:28.780
三代核心研发团队全部离开了

00:03:28.780 --> 00:03:30.540
Alpha4的团队啊

00:03:30.540 --> 00:03:32.300
几乎是一人都不胜了

00:03:32.300 --> 00:03:36.620
要知道谷歌2024年花了27亿美元

00:03:36.620 --> 00:03:40.020
把AI研究的科学家杀字给挖过来

00:03:40.020 --> 00:03:43.120
不到两年人家反手就加盟了OpenAI

00:03:43.120 --> 00:03:45.720
27亿美元啊兄弟们

00:03:45.720 --> 00:03:47.760
买了不来两年的忠诚啊

00:03:47.760 --> 00:03:49.640
杰弗迪恩走的时候

00:03:49.640 --> 00:03:52.280
带走了三名核心研究人员

00:03:52.280 --> 00:03:55.480
谷歌技术核心人员那几乎是都跑了呀

00:03:55.480 --> 00:03:57.080
那最荒诞的是什么呀

00:03:57.080 --> 00:03:59.820
Alphabet自己还投钱入股了

00:03:59.820 --> 00:04:00.780
Discourable Lope

00:04:00.780 --> 00:04:03.260
成为了他的云合作伙伴

00:04:03.260 --> 00:04:05.640
核心班底另立山头

00:04:05.640 --> 00:04:07.860
老东家一看拦不住了

00:04:07.860 --> 00:04:10.440
只能卑微的掏钱买了个包底

00:04:10.440 --> 00:04:12.760
哎您搜了用一下我们的云服务行吗

00:04:12.760 --> 00:04:13.940
那关键问题来了

00:04:13.940 --> 00:04:16.280
这些核心技术人员为什么跑呢

00:04:16.280 --> 00:04:19.240
一句话都是三个的功劳啊

00:04:19.240 --> 00:04:21.080
那么这个问题是

00:04:21.080 --> 00:04:24.620
这些顶级的人才为啥铁了心要跑呢

00:04:24.620 --> 00:04:27.640
作为硅谷的顶级大厂啊

00:04:27.640 --> 00:04:29.140
工资肯定不是核心原因

00:04:29.140 --> 00:04:31.680
真正的答案就俩字管理

00:04:31.680 --> 00:04:33.340
让三个管理了

00:04:33.340 --> 00:04:34.600
你说那能有好吗

00:04:34.600 --> 00:04:36.380
硅谷普遍认为啊

00:04:36.380 --> 00:04:39.660
谷歌正在告别科学家制AI的时代

00:04:39.660 --> 00:04:42.940
从追求前沿突破转向拼产品

00:04:42.940 --> 00:04:44.940
落地的商业营收

00:04:44.940 --> 00:04:46.360
翻译成人话就是

00:04:46.360 --> 00:04:48.000
三个管理层啊

00:04:48.000 --> 00:04:49.860
要短期见效快的产品

00:04:49.860 --> 00:04:52.360
不要AGI这种长远的目标

00:04:52.360 --> 00:04:53.260
不给资金

00:04:53.260 --> 00:04:54.120
不给算力

00:04:54.120 --> 00:04:56.300
顶级人才留在谷歌干嘛呢

00:04:56.300 --> 00:04:57.220
当吉祥物吗

00:04:57.220 --> 00:04:59.660
那谷歌那么多算力

00:04:59.660 --> 00:05:00.780
多出来干嘛呀

00:05:00.780 --> 00:05:01.860
出租啊

00:05:01.860 --> 00:05:03.520
三哥最喜欢了

00:05:03.520 --> 00:05:05.820
提升Google Cloud的营收啊

00:05:05.820 --> 00:05:08.740
多做点TPU芯片的生意啊

00:05:08.740 --> 00:05:11.020
三哥高管们掰着手指着一算

00:05:11.020 --> 00:05:13.460
这OpenAI估值还不到1万亿

00:05:13.460 --> 00:05:15.560
英伟达估值5万亿

00:05:15.560 --> 00:05:18.360
那搞硬件搞云服务来钱多快

00:05:18.360 --> 00:05:19.680
那估值更高啊

00:05:19.680 --> 00:05:21.720
这操作简直那么绝了

00:05:21.720 --> 00:05:23.720
一个AI公司不搞AI

00:05:23.720 --> 00:05:26.040
跑去当二房东出租算力了

00:05:26.040 --> 00:05:29.060
就跟他开了一家米其林三星餐厅

00:05:29.060 --> 00:05:30.820
结果闲做菜太累

00:05:30.820 --> 00:05:33.480
把厨房直接改成仓库收租了

00:05:33.480 --> 00:05:35.080
你格局呢兄弟

00:05:35.080 --> 00:05:37.340
一位Transformer的论文

00:05:37.340 --> 00:05:39.840
联合作者一针见血的指出

00:05:39.840 --> 00:05:43.100
谷歌的官僚主义已经病入膏肓

00:05:43.100 --> 00:05:46.040
在这地方根本干不成任何的事

00:05:46.040 --> 00:05:48.000
所以当一名三哥

00:05:48.000 --> 00:05:50.560
当上一家跨国公司CEO的时候

00:05:50.560 --> 00:05:53.540
这家公司就注定要被回了

00:05:53.540 --> 00:05:57.400
其实这不是三哥第一次在硅谷兴风作浪了

00:05:57.400 --> 00:05:59.260
Meta之前就吃了一次大亏

00:05:59.260 --> 00:06:02.780
当年Meta团队里面超过一半都是印度人

00:06:02.780 --> 00:06:04.820
表面看起来进度很好

00:06:04.820 --> 00:06:08.760
推出大模型的自驾的评测里面得分非常高

00:06:08.760 --> 00:06:11.160
小扎很开心的赶紧推向市场

00:06:11.160 --> 00:06:13.260
结果大家实际一测

00:06:13.260 --> 00:06:15.880
发现各种拉跨就是一坨而已

00:06:15.880 --> 00:06:17.580
然后小扎懵了

00:06:17.580 --> 00:06:19.460
这么多钱这么多人砸去

00:06:19.460 --> 00:06:21.020
不是评分很高吗

00:06:21.020 --> 00:06:22.640
怎么一上市就拉了拖大的

00:06:22.640 --> 00:06:27.760
结果META的前首席科学家杨立坤亲口承认

00:06:27.760 --> 00:06:32.740
团队针对不同的基准测试采用了不同的模型变体

00:06:32.740 --> 00:06:34.800
目的就是为了优化分数

00:06:34.800 --> 00:06:36.320
说白了就是跑分优化

00:06:36.320 --> 00:06:39.240
弄出来个高分低能的弱智玩意儿

00:06:39.240 --> 00:06:41.940
小渣当时气子擦不眼没背过气去

00:06:41.940 --> 00:06:46.380
直接对整个生成式AI团队都失去了信心

00:06:46.380 --> 00:06:48.160
把整个部门给边缘化了

00:06:48.160 --> 00:06:52.520
随后开启了一场针对三哥的大扫荡大清洗

00:06:52.520 --> 00:06:54.640
所以看清楚形势了吗

00:06:54.640 --> 00:06:56.720
硅谷的几大AI巨头

00:06:56.720 --> 00:06:58.460
谁的华人占比高

00:06:58.460 --> 00:06:59.600
谁的模型就强

00:06:59.600 --> 00:07:01.320
谁的三哥占比高

00:07:01.320 --> 00:07:03.940
谁的模型就一定拉跨呀

00:07:03.940 --> 00:07:05.680
你以为只有科技圈

00:07:05.680 --> 00:07:07.520
那不要小看三哥的实力啊

00:07:07.520 --> 00:07:08.781
波音公司前四年

00:07:08.781 --> 00:07:23.781
硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬硬�

00:07:23.781 --> 00:07:25.261
到七五七到七八七

00:07:25.261 --> 00:07:27.261
各种型号是轮着出身

00:07:27.261 --> 00:07:29.421
深度调查指向了

00:07:29.421 --> 00:07:31.401
印度化的过度外包

00:07:31.401 --> 00:07:33.581
与内部的裙带关系

00:07:33.581 --> 00:07:35.061
有内部人士爆料

00:07:35.061 --> 00:07:37.581
说波音这几年招聘的工程师

00:07:37.581 --> 00:07:39.401
大部分都是印度裔的

00:07:39.401 --> 00:07:40.561
责任心不强

00:07:40.561 --> 00:07:41.801
质量意识也不强

00:07:41.801 --> 00:07:43.501
但是偶尔能催牛逼眼

00:07:43.501 --> 00:07:44.861
有报告指出

00:07:44.861 --> 00:07:47.781
一位负责多元化的印度裔副总裁

00:07:47.781 --> 00:07:49.801
利用职权在几年的时间

00:07:49.801 --> 00:07:51.741
把印度员工从一千多人

00:07:51.741 --> 00:07:53.601
扩充到了近两万人

00:07:53.601 --> 00:07:54.381
好家

00:07:54.381 --> 00:07:56.441
这是开新德里分店的

00:07:56.441 --> 00:08:00.001
马斯克的推特在完成收购之后

00:08:00.001 --> 00:08:02.961
第一件事就是砍掉所有的印度裔高管

00:08:02.961 --> 00:08:06.061
推特印度分公司原本200多人

00:08:06.061 --> 00:08:07.921
裁员后只剩12人

00:08:07.921 --> 00:08:11.681
马斯克甚至不惜赔偿5120万美元

00:08:11.681 --> 00:08:13.641
也要裁掉印度裔的高管

00:08:13.641 --> 00:08:16.501
据说马斯克还让印度工程师

00:08:16.501 --> 00:08:18.161
当场你给我手写代码

00:08:18.161 --> 00:08:21.541
结果混日子那些人直接就献出原形

00:08:21.541 --> 00:08:24.741
IBM印度裔CEO克里希纳

00:08:24.741 --> 00:08:29.081
三分钟就敲定了中国研发中心上千人的去留

00:08:29.081 --> 00:08:31.221
他是一边在中国大裁员

00:08:31.221 --> 00:08:33.121
一边在印度疯狂扩招

00:08:33.121 --> 00:08:35.441
只为把研发中心往自己老家搬

00:08:35.441 --> 00:08:38.181
之后IBM就一蹶不振

00:08:38.181 --> 00:08:40.701
直接远离了硅谷的科技核心圈

00:08:40.701 --> 00:08:42.561
根据不安全统计

00:08:42.561 --> 00:08:47.021
2024年至今有超过190名印度裔高管被裁车

00:08:47.021 --> 00:08:50.001
其中74位是CEO级别的

00:08:50.001 --> 00:08:51.121
这哪是裁员的

00:08:51.121 --> 00:08:53.281
这他妈是异常大清算

00:08:53.281 --> 00:08:55.521
那么回到五个

00:08:55.521 --> 00:08:58.481
那三哥CEO怎么会清洗三哥呢

00:08:58.481 --> 00:09:00.041
那个创造了Transformer

00:09:00.041 --> 00:09:01.441
曾经人类最强

00:09:01.441 --> 00:09:03.681
拥有最好科学家文化的公司

00:09:03.681 --> 00:09:06.781
就这么从AI竞赛里面掉队了

00:09:06.781 --> 00:09:10.121
如果AI竞赛最终是我们取得胜利

00:09:10.121 --> 00:09:14.221
那一定得给三哥颁一个一吨重的大奖章

00:09:14.221 --> 00:09:15.441
这不是嘲讽

00:09:15.441 --> 00:09:17.061
这是发自内心的感谢

00:09:17.061 --> 00:09:19.721
感谢三哥用一己之力

00:09:19.721 --> 00:09:22.361
把硅谷最强公司给祸祸死了

00:09:22.361 --> 00:09:22.901
哎呀

00:09:22.901 --> 00:09:23.141
说

00:09:23.141 --> 00:09:24.861
不过话说回来

00:09:24.861 --> 00:09:26.941
谷歌这事也给所有人提了个醒

00:09:26.941 --> 00:09:30.341
一个公司最可怕的不是神一样的对手

00:09:30.341 --> 00:09:31.901
而是猪一样的队友

00:09:31.901 --> 00:09:34.881
当一群PPT大师做上决策席

00:09:34.881 --> 00:09:37.601
当短期财报压倒长期创新

00:09:37.601 --> 00:09:39.981
当裙带关系取代能力导向

00:09:39.981 --> 00:09:41.141
再强的公司

00:09:41.141 --> 00:09:42.721
你也得量

00:09:42.721 --> 00:09:45.481
柯达的故事在谷歌重演了

00:09:45.481 --> 00:09:46.441
唯一不同的是

00:09:46.441 --> 00:09:49.161
这次毁掉谷歌的不是数码相机

00:09:49.161 --> 00:09:52.921
而是一群眼低手低的三哥高管

00:09:52.921 --> 00:09:54.461
谢谢啊

00:09:54.461 --> 00:09:55.441
好了兄弟们

00:09:55.441 --> 00:09:56.501
本期视频就到这了

00:09:56.501 --> 00:09:58.161
你们觉得谷歌还能翻身吗

00:09:58.161 --> 00:09:59.301
评论区咱们聊聊啊

00:09:59.301 --> 00:09:59.881
点赞关注

00:09:59.881 --> 00:10:00.721
咱们下期再见

00:10:00.721 --> 00:10:01.041
拜拜