去年11月突然推出ChatGPT震惊世界的OpenAI，在整整一年后以闪电解职CEO再次震惊世界。
有不少人以为这拿的是乔布斯的剧本，错了，这其实是天才科学家奋力一击的故事。
OpenAI的灵魂人物不是CEO Sam Altman，而是其创始人之一、首席科学家Ilya Sutskever。
我们先看看水面上的冰山一角，然后再潜入水下看看合订本。
OpenAI的董事会共6人。 数字化转型网www.szhzxw.cn
三人是公司的高管：
Sam Altman，OpenAI前CEO，犹太人，原Y Combinator孵化器的总裁。Sam Altman的更多故事见《不懂新冠的AI大佬不是亿万富翁》《美国防部长指导新冠防疫，我们该如何纪念吴老师》。
Greg Brockman，OpenAI前董事会主席兼总裁，美版支付宝Stripe的前CTO，《不懂新冠的AI大佬不是亿万富翁》里提到的天才青年、Stripe的CEO Patrick Collison的副手。Sam Altman曾投资过Stripe。所以，Greg Brockman算是Sam Altman的小弟，属于工具人。
Ilya Sutskever，OpenAI创始人之一，首席科学家。
三名外部独立董事：
Adam D’Angelo，美版知乎Quora的CEO，目前也在公司内部搞类似ChatGPT的大模型Poe。
Tasha McCauley，一个女性创业者，前GeoSim CEO，好莱坞明星囧瑟夫的老婆。
Helen Toner，乔治敦大学一个研究中心的主管。 数字化转型网www.szhzxw.cn
首席科学家Ilya Sutskever先和三位独立董事谈话，说服了他们，凑够了四票。然后在Sam Altman和Greg Brockman两人不在场的情况下，开了一次董事会，做出决定炒了Sam Altman。
然后11月16日周四晚上，Ilya Sutskever发短信给Sam Altman，约在周五中午谈话。周五中午，双方连入了Google在线视频会议，Ilya Sutskever一方四人都在。他们简短地通知Sam董事会决议，说他已经被炒了。然后结束会议。
稍后，Ilya Sutskever再发短信给Greg Brockman要求打一个quick call简短的电话。双方随即连入Google视频会议。Ilya Sutskever通知Greg他已被解除董事长职务，以及Sam Altman也已被解雇。
整个过程闪电般地结束，非常干净迅速，没有一点拖泥带水。
随后，Ilya Sutskever在OpenAI召开全体员工会议，宣布董事会决定。CTO Mira Murati 成为临时CEO。
在全体员工会议上，有人问：这是一场政变吗？
Ilya Sutskever回答：你可以这么说。但这只是“the board doing its duty董事会在尽自己的责任”。
但是，事情还没完。
不到24小时，OpenAI的最大几个投资人，包括投了100亿美元的微软、Thrive Capital、老虎基金（Tiger Global Management）、红杉资本，都对OpenAI施予巨大压力，要求让Sam Altman回归复职。
OpenAI董事会受到了巨大的压力。 数字化转型网www.szhzxw.cn
这些投资人为什么想让Sam Altman回归？因为他们在OpenAI身上下了很大的赌注。
微软，真金白银投了100亿美元，自己能否再次成为科技界的霸主，全靠OpenAI了。
Thrive Capital、老虎基金、红杉资本，不但投资了OpenAI，期待其成为AI领域的万亿规模的霸主，从而自己能狂赚一笔，而且还投了一大堆在OpenAI的生态体系内的创业公司，就等着OpenAI的应用市场一开，就能吹出一堆AI小独角兽。
这就像苹果的应用商店一开门，手机app就变成了风口上的猪。最近六七年，美国大部分新兴的科技公司都是做手机app的。
所以，这些投资人们都疯狂了，迫不及待地呼喊着：再来一次！再来一次！
那些在OpenAI的生态体系下存活的创业公司们更是惊恐，要是你OpenAI发生的变动不符合自己业务的发展方向，自己公司就死翘翘了。所以，有人声嘶力竭地在社交媒体上批评咒骂Ilya Sutskever，要他赶紧滚蛋，把OpenAI交给Sam Altman。为啥，阻碍别人发财，犹如杀人亲生父母。
那为什么这些人觉得Sam Altman能实现他们的想法，而首席科学家Ilya Sutskever不能实现他们的想法？
这就要从Ilya Sutskever的老师Geoffrey Hinton说起。Geoffrey Hinton教授的更多故事见《贫穷和阶级滑落成常态，新冠AI再暴击，大神荐社会主义》。 数字化转型网www.szhzxw.cn
Geoffrey Hinton，英国人，1947年出生。从小喜欢木工。但是为了谋生而去学习计算机，1978年获得爱丁堡大学的人工智能博士学位后，去了美国继续在学术界耕耘，后来因为对里根时代的政治理念破灭而移居加拿大，之后长期担任多伦多大学的计算机科学教授。
在人工智能领域，他经过长期研究后，选择的是神经网络的技术路线。
但是，这个路线受限于当时计算机计算能力小、数据量小，出成果一直都很慢，因此一直不是热门方面，并不被外界看好。因此，喜欢看排名、追求风口的博士生都直接无视了这个方向，和教授套瓷都不往这个方向靠。
Hinton教授自己说，到了1993年，46岁的他已经彻底死心了。研究进展缓慢，也没什么钱，生活也不太如意，中年两次丧偶，独自抚养两个孩子。
他就是希望他的博士生们能继续坚持神经网络的方向，也许100年后会有突破。
历史的进程比他预料得快得多。
10多年后，随着计算机的计算能力的高速提升，人类可获得的数据量的高速增长，神经网络方向的研究进展开始逐渐苏醒。 数字化转型网www.szhzxw.cn
这个时候，Hinton教授招了两个博士生，Alex Krizhevsky和Ilya Sutskever。
这两位博士生都是俄罗斯出生的，他们没有像其它寻常人一样选择其它热门领域，而是投奔了冷门的Hinton教授门下，显示出了他们非同寻常的个性。果然，这一点在未来的发展中也得到了验证。他们也各自做出了非同寻常的决定。
他们三人在2012年的时候完成了一个新型神经网络的开发，叫卷积神经网络，命名为AlexNet。我非常怀疑这是用其中一位博士生Alex Krizhevsky的名字来命名的，但是我没有证据。
这个AlexNet取得了重大突破，对图片识别的能力大大超过了人工智能其它技术路线的技术。当时有一项人工智能大赛ImageNet，其它技术的识别错误率大约是30%，而AlexNet的错误率提高了差不多一倍，达到15.3%，从而在2012年9月30日一举拿下ImageNet大赛的冠军。
余凯博士当时在百度领衔人工智能研发工作，立刻意识到了这项突破的重要性，于是联系Hinton教授希望合作，先提供100万美元研究经费。
随后几家美国公司也找上门来，同样希望合作，甚至提出希望独家合作。
于是，Hinton师徒三人就成立了一家皮包公司DNN Research，公司只有他们三个人，没有业务，没有产品，没有收入。让大家自由出价，价高者得。
参加这场竞拍的有四家公司，百度、Google、微软、刚刚成立一年的英国人工智能公司DeepMind。
余凯博士代表百度首先报价1200万美元。 数字化转型网www.szhzxw.cn
然后价格持续上升，随之变成了Google和微软的对决。
两家都志在必得，都愿意继续加价。
但是Hinton教授说，够了够了，你们太疯狂了，我们就3个人，你们竟然愿意炒到4400万美元。
当时Hinton师徒觉得，Google更有道德一些，毕竟“不作恶”写进了公司章程。那就接受Google的招安吧。
他们没想到谷歌与美国国防部进行的人工智能研究合作项目Project Maven在2018年曝光，员工们抗议，拿出公司章程“不作恶”条款要求公司管理层解释。谷歌管理层很快将“不作恶”这条公司章程删除。
当时，Hinton师徒拿到钱后，打算三人平分，每人1466万美元。两位博士生纷纷坚持，Hinton老师您应该拿40%，我们各30%就好了。随后三人就高高兴兴去Google上班了。Google随之设立了一个新部门：Google Brain谷歌大脑。
Google和微软一看对方准备在人工智能领域投入这么大手笔，也开始了在这一领域的军备竞赛。
在那次竞拍后的一年多后的2014 年1月26日，Google以6.5亿美元把四位竞拍者中的DeepMind也收购了。当时DeepMind的员工数是50人。算下来，人均价格相当于Hinton师徒的88折。
随后，Google派了Google Brain即Hinton师徒三人去帮助DeepMind。
DeepMind很快取得举世瞩目的成绩。
2014年开始开发人工智能围棋软件 AlphaGo。2016年AlphaGo击败人类世界冠军李世石。
2019年，专门用来玩星际争霸的AlphaStar以10:1战胜人类职业玩家，在团队协作人工智能领域取得突破。
2018年，用于预测蛋白质折叠结构的AlphaFold取得突破，这个方向上的科学问题基本得到解决。其影响是，一些使用基因测序仪器对蛋白质测序然后构建结构的科研人员开始要转型。据说这是某女博士回国的原因之一，这个不了解，纯八卦。 数字化转型网www.szhzxw.cn
DeepMind的管理层认为自己的技术过于强大，不应该让私人公司独占，而应该属于全人类，这对整个人类社会都更有益。而在隶属于Google的这几年里，Google的表现并不能让DeepMind满意，于是这几年来，他们一直在寻求从Google中独立出来。
Hinton师徒三人也有类似的想法。
Hinton教授是2023年从Google退休。
博士生之一的Alex Krizhevsky在2017年从Google退休，不再涉足人工智能。单算他加入Google前卖公司拿到的1320万美元也够生活了。因为一般美国人，一辈子挣的钱不会超过100万美元。
博士生之一的Ilya Sutskever的变动更早一点，是在2015年。
Ilya Sutskever是1985年出生于俄罗斯，5岁随家人移居以色列，所以应该是犹太人，成年后又随家人移居加拿大，继而加入Hinton教授门下。 数字化转型网www.szhzxw.cn
Ilya Sutskever也是个生活很简单的人，和乔布斯、马斯克、扎克伯格类似，同样的衣服买一打轮着穿，吃喝尽可能简单，要能快速完成，家里家具越少越好，一个床垫就能睡觉。应酬、晚宴、party、舞会邀约很多，但是基本都不去。
所以，师徒三人组里，还干得动、还能干的，就剩下Ilya Sutskever了。
而硅谷一帮大佬，看着谷歌垄断了人工智能的人才，就想着要搞事情。你谷歌吃独食可不好，我们也要分一杯羹。
于是，2015年，硅谷一众大佬就把Ilya Sutskever约出来吃饭。这些人包括风投之王Peter Thiel，特斯拉的马斯克，当时还是孵化器Y Combinator总裁的Sam Altman，LinkedIn创始人Reid Hoffman，本次政变中也有出场的Stripe的CTO Grey Brockman。
在这个饭局上，大家谈妥了。
Ilya Sutskever从谷歌出来，围绕他打造一个属于他的人工智能实验室OpenAI。让AI普及化，让硅谷所有人都能用上AI，不要让谷歌垄断，造福全人类。这个OpenAI就设为非盈利组织。
为什么这个OpenAI要设为非盈利组织？ 数字化转型网www.szhzxw.cn
这涉及到硅谷一个知名的大悲剧。维基百科的创始人Jimmy Wales，设立维基百科的时候，对公司治理不熟悉，想着属于全人类的百科全书应该属于有公信力的机构，于是就把自己的公司设立为非盈利组织。这下坏了，由于公司治理法规的限制，非盈利组织只能靠捐款存活，融资困难，上市无望。因此，虽然维基百科发展迅速成为世界第一大百科网站，但是其创始人和高管层还是穷得响叮当。Jimmy Wales痛不欲生，多次痛哭搞砸了公司治理架构。那也是，其它的后辈小弟们搞的在线订酒店、在线打车、在线交友、在线实名交友、在线匿名交友、在线同性恋匿名交友、甚至在线黄色小网站什么的，统统火红火红，上市赚大钱，财富自由，自己却被这非盈利组织的架构困死。
所以，有了维基百科这个反面案例在，所有硅谷的人都不会使用非盈利机构这个架构。因为赚不了钱。
而2015年的OpenAI为什么要故意设为非盈利机构，就是因为当时想的是让OpenAI去专心搞研发，搞出来了服务全人类，大家一起用，避免像Google那样垄断。
刚满30岁的Ilya Sutskever非常认可这一使命，就干了。
然后，马斯克捐了1亿美元，其它人凑了3000万美元，合计1.3亿美元当启动经费。马斯克与Sam Altman同为OpenAI联席董事长。 数字化转型网www.szhzxw.cn
干活的，就Ilya Sutskever一个人，他再去招人组团队。
所以，当时的情形就是十个人围观，看着Ilya Sutskever一个人干活。
也没办法，当时也就Ilya Sutskever一个人懂这活是怎么干的。
当时定的是Ilya Sutskever在OpenAI的待遇是年薪190万美元，和他在谷歌时差不多。从谷歌辞职时，谷歌大力挽留，愿意将Ilya Sutskever的待遇提高到400万美元乃至500万美元，只要他愿意留下来。
Ilya Sutskever说不，我要去创立一个技术不被私营企业垄断的新平台。
所以说，对于Ilya Sutskever，当初2012年谷歌买Hinton师徒的公司时，生活简单的Ilya Sutskever认为他已经拿到了足够的金钱。2015年离开Google、拒绝Google的500万美元年薪去一个未来不确定的非盈利性质的初创小平台，完全不是金钱驱动，而是有着更远大的目标。
之后，OpenAI的发展并不是一帆风顺的。 数字化转型网www.szhzxw.cn
Ilya Sutskever的头发掉得也很快。
因为Hinton师徒选择的这条路，有点漫长，出成果都挺慢的。其它人在掌握了AlexNet架构后，提出了众多的改进路线。例如一种叫人类反馈强化学习(RLHF)。举个例子，人工智能目前会算1+1=2，但是不会算5*4=20。那人类反馈强化学习就会强迫人工智能先把5*4=20记下来，你不理解也不管，总之以后我问你5*4等于多少你必须回答等于20。其它人把这种方法一实施，短期的效果非常好，立刻在各项评测里将Hinton师徒的技术打败。Hinton师徒则认为这属于人工激素催肥，不合理，应该让人工智能慢慢学，自己理解，这样才会有真正的发展。
所以OpenAI从创立的2015年到2019年都进展缓慢。而马斯克等大佬捐的1.3亿美元就花完了。
怎么办？两位联席董事长发生了分歧。马斯克仍然坚持要继续非盈利的路线。而Sam Altman主张要转型，这样研发下去是无底洞，要转型走传统私营公司的模式，融资干活，然后盈利，然后上市，然后成为AI霸主。 数字化转型网www.szhzxw.cn
两方随后一拍两散，马斯克撒手。Sam Altman成为董事长。然后在OpenAI下面成立一个常规的私营子公司，由私营子公司去融资、赚钱，然后OpenAI维持非盈利的公司治理架构。
在2019年转型后，OpenAI开始大规模融资。7月份开始，微软注资10亿美元，其它风投公司Khosla Ventures、LinkedIn创始人的个人基金Reid Hoffman Foundation、Matthew Brown Companies等等竞相跟进。然后之后的每一年都进行一次或两次融资。在2023年年初的第七次融资时，微软大手笔一挥，投下100亿美元重注。这些年来，微软把能买的OpenAI的权利都买了，2020年买的是GPT3的基础技术许可授权、技术集成的优先授权，2021年买的是技术商业化授权，对外提供付费API和AI工具的授权。2023年100亿美元下去，则是获得了将ChatGPT整合进Bing搜索引擎的授权。所以微软的想法是，将OpenAI完全融入微软，方式可以再谈，可以全盘收购，可以是技术全部授权，可以是其它，总是，你的就是我的。
马斯克在最近这些年里，一直批评OpenAI的转向，他表示不理解，他当初投了1亿美元的非盈利机构OpenAI，怎么就变成了盈利性机构？他完全不同意这个发展方向。但是没办法，话语权不够大，董事会里票数不够多。其它人硬来，没办法。所以马斯克在2023年又成立了自己的AI公司xAI。
马斯克不是唯一反对OpenAI转向的人。低调的Ilya Sutskever也是。他的反抗没有出现在媒体里，但是出现在了OpenAI对微软的授权条款里。
在OpenAI的授权条款里，有一项不同寻常的规定：
OpenAI对微软的所有授权和商业条款，仅限于OpenAI实现AGI之前。一旦OpenAI实现了AGI，该技术不对微软开放，之前达成的所有授权和商业条款对AGI系统和AGI技术无效。
什么是AGI？AGI的定义是：通用人工智能，一个高度自治的系统，在工作中的能力超过人类。
而怎么判断有没有达到AGI呢？这由OpenAI的董事会单方面来决定。而目前，这个OpenAI的董事会，就是文章开头提到的那六人董事会。
也就是说，Ilya Sutskever试图通过这个条款握住外部股东的咽喉，一旦技术达到突破或对方有异动，立刻斩断其喉咙。 数字化转型网www.szhzxw.cn
这个条款不是秘密条款。硅谷的人都知道。而且在2023年10月份开始在硅谷有了公开讨论。
这显示出什么？显示出，微软可能要有动作，来消除这个条款造成的不确定性，来保证它能顺利把OpenAI吞到肚子里去，使自己能100%确定无疑成为人工智能领域的霸主。
微软可能的动作包括尽快全资收购OpenAI，改组OpenAI董事会以确保自己能控制董事会的投票，将Ilya Sutskever提走等等。
除了微软这一方的势力，第二方势力是投资基金，他们投入了天量的资金在人工智能赛道上，正等着暴富。
第三方势力是在OpenAI生态下进行开发的小公司们，他们公司的生死都依赖着OpenAI的进展。
第四方势力是Sam Altman。要注意，虽然他现在是OpenAI的CEO，但是他没有任何股权，唯一的投资收益是几年前通过Y Combinator在OpenAI天使轮投进去的一点点，对于个人来说收益可以说是聊胜于无。而通常来说，一个CEO持有公司10%的股份是合情合理的。那么，对于Sam Altman来说，如何将自己的利益变现，是个非常强烈的动机，毕竟，大家都看到了，将OpenAI盈利化以后，是个万亿美元级别的行业霸主，那么，以通常情况来期望，Sam Altman会期望获得1000亿美元的收益。
那么，问题来了，Sam Altman在没有OpenAI股份的情况下，要怎么才能将这1000亿美元的期望值变现呢？1000亿美元哦，没有人会愿意放手。其中一个办法是Sam Altman和股东勾兑，例如和微软勾兑，通过微软兑现。 数字化转型网www.szhzxw.cn
当前，Ilya Sutskever现在是以一人之力，对抗四大力量。
各位能在媒体上、社交平台上能看到支持Ilya Sutskever的声音吗？看不到，一点都看不到。相信大家能看到很多声音，都是想他死的。毕竟，挡人财路，犹如杀人亲生父母。
马斯克和Hinton老师倒是和他的理念一致，但是目前在事外，插不上手。
除了Ilya Sutskever握住四大势力的喉咙外这个核心原因外，Ilya Sutskever还有一个地方招人恨。就是Ilya Sutskever希望AGI通用人工智能的商业化能慢一点。这与其它势力的赶紧铺开赶紧暴富的想法背道而驰。
Ilya Sutskever认为，人工智能在2016年的时候已经出现超越人类智力的信号。当时在AlphaGo与李世石的第二局比赛中，第37步，走出了非常特殊的一步。这一步被人类评论员认为是水平奇差无比的败着。而结果呢，那一步是奠定整盘胜局的核心一步。这是人类历史上从来没有过的对弈思路。
所以，通用人工智能AGI必然很快能超过人类。当其它人还在苦苦思索如何建造一个能赶得上人类思维水平的AGI系统时，Ilya Sutskever已经领先一步，他认为，AGI必然很快能超过人类，所以应该从现在开始为面对超越人类智慧的AI作准备。
他认为，AGI和人类最合适的关系是父母和婴儿的关系，AGI是父母，人类是婴儿，AGI父母无时无刻不为婴儿人类着想，让婴儿人类活得更好。 数字化转型网www.szhzxw.cn
而现有的控制AGI的方法不适用于比人类更聪明的人工智能。因为当前的模型是假设人类可以可靠地评估人工智能系统的工作。随着人工智能系统的能力越来越强，人类不知道怎么评估它们，不知道在每一项改动后，它们是做得更好还是更差。这会导致AGI无法被控制。
因此，在2023年7月，Ilya Sutskever拉上同事Jan Leike，在OpenAI内部成立了一个小部门，叫Superalignment超级对齐。目标是制定一套故障安全程序来控制AGI技术，计划将OpenAI全公司的计算资源的五分之一分配给这个部门，并在四年内解决这个问题。然后，在有了这种控制技术后，才把AGI技术开放给人类使用。
硅谷很多人对Ilya Sutskever的这套“超级对齐”的思路嗤之以鼻，而且认为Ilya Sutskever会阻碍OpenAI的发展。四年以后才开放使用？不行，我要暴富，我要今年就暴富，我要今年就财富自由，我要明天就暴富，我要明天就财富自由。
所以，昨天传出消息，各大势力逼迫OpenAI董事会解散重组，让Sam Altman回归担任CEO。因为这最能保障他们的利益。而Sam Altman一回归，必然改组董事会，将Ilya Sutskever踢出，确保自己牢牢掌握董事会。
Ilya Sutskever接下来会如何呢？
这是一曲科学家创始人的搏击之歌。 数字化转型网www.szhzxw.cn
我倒是希望Ilya Sutskever能坚持下去，坚决不向四大势力妥协，坚决掌握董事会。
毕竟，其它人想要一个巨型印钞机，一个人工智能领域的霸主。
而Ilya Sutskever想要一个能像父母照顾婴儿一般来照顾人类的人工智能。
翻译：
Behind OpenAI’s coup is the lament of its scientist founder
OpenAI, which shocked the world with the sudden launch of ChatGPT in November last year, shocked the world again with the lightning dismissal of its CEO exactly one year later.
Many people think that this is Jobs’s script, wrong, this is actually the story of a genius scientist struggling to strike.
The soul of OpenAI is not CEO Sam Altman, but one of its founders and chief scientist Ilya Sutskever.
Let’s look at the tip of the iceberg above the water, and then dive underwater to look at the bound book.
OpenAI has a board of six people. 数字化转型网www.szhzxw.cn
Three are executives of the company:
Sam Altman, former CEO of OpenAI, is Jewish and former president of the Y Combinator incubator. More stories by Sam Altman are in “AI Bosses who don’t Understand COVID-19 are not Billionaires” and “US Defense Secretary directs COVID-19 Prevention, How should we remember Teacher Wu”.
Greg Brockman, the former chairman and president of OpenAI, the former CTO of Stripe, the American version of Alipay, the talented young man who does not understand the new Crown is not a billionaire, and the deputy of Patrick Collison, the CEO of Stripe. Sam Altman previously invested in Stripe. So, Greg Brockman is kind of Sam Altman’s little brother, tool man.
Ilya Sutskever, co-founder and Chief scientist of OpenAI.
Three external independent directors:
Adam D’Angelo, the CEO of Quora, is currently working on a ChatGPt-like large-scale model, Poe, within the company.
Tasha McCauley, a female entrepreneur, former GeoSim CEO, wife of Hollywood star Gordon-Levitt.
Helen Toner, director of a research center at Georgetown University.
Chief scientist Ilya Sutskever talked to the three independent directors and convinced them to get four votes. Then there was a board meeting without Sam Altman and Greg Brockman present, and the decision was made to fire Sam Altman.
Then on Thursday night, Nov. 16, Ilya Sutskever texted Sam Altman for an appointment to talk at noon on Friday. At noon on Friday, the two parties joined a Google online video conference, with Ilya Sutskever and all four of them there. They briefly informed Sam of the board’s decision that he had been fired. Then end the meeting. 数字化转型网www.szhzxw.cn
Later, Ilya Sutskever texted Greg Brockman to ask for a quick call. The two parties then connected to a Google video conference. Ilya Sutskever informed Greg that he had been fired as chairman and that Sam Altman had also been fired.
The whole process ended in lightning, very clean and quick, without a bit of mud.
Ilya Sutskever then held an all-hands meeting at OpenAI to announce the board’s decision. CTO Mira Murati became interim CEO.
At the all-hands meeting, someone asked: Is this a coup?
Ilya Sutskever: You could say that. But it was just “the board doing its duty.”
But that’s not the end of it.
In less than 24 hours, OpenAI’s largest investors, including $10 billion from Microsoft, Thrive Capital, Tiger Global Management, and Sequoia Capital, put intense pressure on OpenAI to bring back Sam Altman.
The OpenAI board has come under intense pressure. 数字化转型网www.szhzxw.cn
Why do these investors want Sam Altman back? Because they’re betting a lot on OpenAI.
Microsoft has invested $10 billion in real money, and whether it can become the hegemon of the technology world again depends on OpenAI.
Thrive Capital, Tiger Fund, Sequoia Capital, not only invested in OpenAI, expecting it to become the trillion-dollar overlords in the AI field, so that they can make a fortune, but also invested in a lot of startups within the OpenAI ecosystem, waiting for the OpenAI application market to open. I can blow a bunch of little AI unicorns.
It’s like when Apple’s app Store opens, the mobile app becomes a pig in the wind. In the last six or seven years, most of the new tech companies in the United States have been making mobile apps.
So the investors went crazy, screaming: Do it again! One more time!
Startups that survive in the OpenAI ecosystem are even more terrified that if you make changes to OpenAI that don’t match the direction of your business, your company will die. So, some people are screaming at Ilya Sutskever on social media, telling him to get the hell out of here and hand OpenAI over to Sam Altman. Why? Preventing someone from getting rich is like killing your own parents.
So why do these people think that Sam Altman can implement their ideas and Ilya Sutskever, the lead scientist, can’t? 数字化转型网www.szhzxw.cn
It starts with Ilya Sutskever’s teacher Geoffrey Hinton. Professor Geoffrey Hinton has more stories in “Poverty and Class fall as the norm, COVID-19 AI Strikes Again, God recommends Socialism”.
Geoffrey Hinton, British, born 1947. I loved woodworking since childhood. However, in order to make a living, he studied computer science. After obtaining a doctorate in artificial intelligence from the University of Edinburgh in 1978, he went to the United States to continue to work in academia. Later, he moved to Canada because of the disillusionment with the political ideas of the Reagan era, and then served as a professor of computer science at the University of Toronto for a long time.
In the field of artificial intelligence, after long-term research, he chose the technical route of neural networks.
However, this route was limited by the small computer computing power and small amount of data at that time, and the results have been very slow, so it has not been a popular aspect and is not optimistic about by the outside world. Therefore, doctoral students who like to see rankings and pursue tuyere directly ignore this direction, and professors do not rely on this direction.
Professor Hinton himself says that by 1993, at the age of 46, he had given up. Research progress is slow, there is little money, life is not very satisfactory, middle-aged twice widowed, raising two children alone.
He just hopes that his PhD students will keep going in the direction of neural networks, and maybe in 100 years there will be a breakthrough. 数字化转型网www.szhzxw.cn
History moved much faster than he had expected.
More than 10 years later, with the rapid improvement of computer computing power and the rapid growth of the amount of data available to human beings, the research progress in the direction of neural networks began to gradually awaken.
At this time, Professor Hinton recruited two PhD students, Alex Krizhevsky and Ilya Sutskever.
The two doctoral students, both Russian-born, showed their unusual personalities by choosing to join the less popular professor Hinton rather than other popular fields. Sure enough, this point has also been verified in future development. They have also made extraordinary decisions.
In 2012, the trio completed the development of a new type of neural network called convolutional neural network, named AlexNet. I strongly suspect it was named after one of the PhD students, Alex Krizhevsky, but I have no proof. 数字化转型网www.szhzxw.cn
This AlexNet has made a major breakthrough, and its ability to recognize images far exceeds that of other AI technology routes. At the time of ImageNet, an AI competition, the error rate of other technologies was about 30%, and AlexNet’s error rate was almost doubled to 15.3%, thus winning the ImageNet contest on September 30, 2012.
Dr. Yu Kai, who was leading the AI research and development work at Baidu at the time, immediately recognized the importance of this breakthrough and contacted Professor Hinton to cooperate, offering a $1 million research grant.
Subsequently, several American companies also came to the door, also hoping to cooperate, and even put forward the hope of exclusive cooperation.
So Hinton and his apprentice set up a leather-clad company, DNN Research, with just the three of them, no business, no products, and no revenue. Let everyone bid freely. The highest bidder wins.
Four companies participated in the auction: Baidu, Google, Microsoft and DeepMind, a year-old British artificial intelligence company. 数字化转型网www.szhzxw.cn
Dr. Kai Yu, on behalf of Baidu, first offered $12 million.
And then the price kept going up, and it became a battle between Google and Microsoft.
Both are determined to win, and both are willing to continue raising prices.
But Professor Hinton said, enough, enough, you guys are crazy, there’s only three of us, and you’re willing to bid $44 million.
At the time, Hinton felt that Google was more ethical, after all, “don’t be evil” was written into the company’s articles of association. Then take Google’s offer.
They did not expect that Google’s artificial intelligence research cooperation Project Maven with the U.S. Department of Defense was exposed in 2018, and employees protested, taking out the “don’t be evil” clause of the company’s charter to ask the company’s management to explain. Google’s management quickly removed the “Don’t be evil” clause. 数字化转型网www.szhzxw.cn
At that time, after Hinton’s master and apprentice got the money, they planned to split it between the three people, $14.66 million each. The two doctoral students insisted that you should take 40%, Mr. Hinton, and we should each take 30%. Then the three happily went to work at Google. Google created a new division: Google Brain.
Google and Microsoft, seeing each other ready to invest so much in the field of artificial intelligence, have also begun an arms race in this field.
More than a year after that auction, on January 26, 2014, Google also acquired DeepMind, one of the four bidders, for $650 million. DeepMind had 50 employees at the time. That works out to a 88 percent discount per person to Hinton’s master and Apprentice.
Subsequently, Google sent Google Brain, Hinton’s three mentors, to help DeepMind.
DeepMind quickly achieved worldwide success.
In 2014, he began to develop the artificial intelligence Go software AlphaGo. In 2016, AlphaGo defeated the human world champion Lee Sedol. 数字化转型网www.szhzxw.cn
In 2019, AlphaStar, which is designed to play Starcraft, beat human professional players 10:1, making a breakthrough in the field of teamwork artificial intelligence.
In 2018, AlphaFold, which is used to predict the folding structure of proteins, made a breakthrough, and scientific problems in this direction were basically solved. The effect is that some researchers who use genetic sequencing machines to sequence proteins and then build their structures are starting to make the switch. It is said that this is one of the reasons for a female doctor to return to China, this is not understood, pure gossip.
DeepMind’s management believes its technology is too powerful and should not be reserved for private companies, but should belong to all of humanity, which is better for human society as a whole. In the past few years of belonging to Google, Google’s performance was not satisfactory to DeepMind, so they have been seeking independence from Google for several years.
The Hinton trio had similar ideas.
Professor Hinton will retire from Google in 2023.
One of the PhD students, Alex Krizhevsky, retired from Google in 2017 and is no longer involved in artificial intelligence. Even the $13.2 million he made from selling the company before joining Google was enough to live on. Because the average American doesn’t make more than a million dollars in their lifetime.
Ilya Sutskever, one of the PhD students, made the move earlier, in 2015.
Ilya Sutskever was born in Russia in 1985, moved to Israel with her family at the age of five, so she is believed to be Jewish, and moved to Canada with her family as an adult, where she joined Professor Hinton. 数字化转型网www.szhzxw.cn
Ilya Sutskever is also a very simple life, and Jobs, Musk, Zuckerberg similar, the same clothes to buy a dozen rounds to wear, eat and drink as simple as possible, to be able to quickly complete, the less furniture at home the better, a mattress can sleep. There are a lot of invitations to parties, dinners, parties, and dances, but they are basically not going.
Therefore, Ilya Sutskever was the only one of the three masters who could still work and was still capable.
And a bunch of big guys in Silicon Valley, watching Google monopolize artificial intelligence talent, want to do something. It’s not good for you, Google. We need a piece of the pie.
So, in 2015, the Silicon Valley bigotry asked Ilya Sutskever out to dinner. These include venture capital king Peter Thiel, Tesla’s Elon Musk, Sam Altman, who was then the president of Y Combinator, LinkedIn founder Reid Hoffman, Grey Brockman, Stripe’s CTO, also appeared in the coup.
At this dinner, we agreed.
Ilya Sutskever came out of Google to build OpenAI, his own artificial intelligence lab. Make AI universal, so that everyone in Silicon Valley can use AI, do not let Google monopoly, for the benefit of all mankind. This OpenAI is a non-profit organization. 数字化转型网www.szhzxw.cn
Why is OpenAI a non-profit organization?
It involves one of Silicon Valley’s most famous tragedies. Jimmy Wales, the founder of Wikipedia, was not familiar with corporate governance when he set up Wikipedia, thinking that an encyclopedia belonging to all mankind should belong to an institution with credibility, so he set up his company as a non-profit organization. This is bad. Because of corporate governance regulations, non-profit organizations can only survive on donations, and financing is difficult, and there is no hope of listing. As a result, although Wikipedia has rapidly grown to become the world’s largest encyclopedia site, its founders and executives are still desperately poor. Jimmy Wales broke down and cried a lot and messed up the corporate governance structure. That is also, other younger brothers engaged in online hotel booking, online taxi, online dating, online real name dating, online anonymous dating, online gay anonymous dating, and even online yellow small websites, all red, listed to make a lot of money, wealth freedom, but they were trapped by the non-profit organization structure.
So, with Wikipedia as a negative example, no one in Silicon Valley is using the nonprofit structure. Because there’s no money.
The reason why OpenAI in 2015 was deliberately set up as a non-profit institution was because it wanted to let OpenAI concentrate on research and development, make it serve all mankind, everyone use it together, and avoid monopolization like Google.
Ilya Sutskever, who just turned 30, was so impressed with the mission that she did it.
Musk then donated $100 million and others chipped in $30 million, for a total of $130 million. Musk is co-chairman of OpenAI with Sam Altman. 数字化转型网www.szhzxw.cn
Ilya Sutskever was the only one doing the work, and he went to recruit a team.
So, there were ten people watching Ilya Sutskever work alone.
I couldn’t help it. Ilya Sutskever was the only one who knew how the job was done.
At the time, it was agreed that Ilya Sutskever’s salary at OpenAI would be $1.9 million a year, similar to his salary at Google. When he resigned from Google, Google made a strong offer to raise Ilya Sutskever’s package to $4 million or even $5 million if he was willing to stay.
Ilya Sutskever said no, I’m going to create a new platform where technology is not monopolized by private companies.
So for Ilya Sutskever, who lived a simple life when Google bought Hinton’s company in 2012, he thought he was getting enough money. Leaving Google in 2015, turning down a $5 million annual salary from Google to work on a small non-profit startup with an uncertain future, was not motivated by money at all, but by a larger goal.
Since then, OpenAI’s development has not been smooth sailing.
Ilya Sutskever’s hair is falling out fast, too. 数字化转型网www.szhzxw.cn
Because the road Hinton has chosen is a little long, and the results are quite slow. Others, having mastered the AlexNet architecture, suggested a number of ways to improve it. One example is something called human feedback reinforcement learning (RLHF). For example, AI can currently calculate 1+1=2, but not 54=20. Then human feedback reinforcement learning will force the AI to first write down 54=20, you don’t care if you don’t understand, anyway, when I ask you what 5*4 is, you have to answer that it is equal to 20. When others implemented this method, the short-term results were very good, immediately beating Hinton’s technique in various tests. Hinton believes that this is artificial hormone fattening, unreasonable, should let the artificial intelligence slowly learn, understand themselves, so that there will be real development.
As a result, OpenAI progressed slowly from its inception in 2015 to 2019. The $130 million donated by Musk and other big names will be spent.
What to do? The two co-chairmen disagreed. Musk remains adamant that he will continue on the nonprofit path. And Sam Altman advocates to transform, so that research and development is bottomless, to transform the model of traditional private companies, financing work, and then profit, and then go public, and then become AI overlord. 数字化转型网www.szhzxw.cn
The two sides then split up and Musk walked away. Sam Altman became chairman. Then a regular private subsidiary is set up under OpenAI, and the private subsidiary raises money and makes money, and OpenAI maintains a non-profit corporate governance structure.
After its transformation in 2019, OpenAI began to raise money on a large scale. Starting in July, Microsoft invested $1 billion, and other venture capital firms Khosla Ventures, the Reid Hoffman Foundation, the personal fund of the founder of LinkedIn, and Matthew Brown Companies followed suit. And then every year after that, one or two rounds of financing. In the seventh round of financing in early 2023, Microsoft made a big bet of $10 billion. Over the years, Microsoft has bought all the OpenAI rights it can buy, bought the basic technology license of GPT3 in 2020, the priority license of technology integration, bought the technology commercialization license in 2021, and authorized to provide paid apis and AI tools. The next $10 billion in 2023 is the license to integrate ChatGPT into the Bing search engine. So Microsoft’s idea is to fully integrate OpenAI into Microsoft, in a way that can be negotiated, can be acquired, can be fully licensed technology, can be other, and always, what’s yours is mine.
Musk has been critical of OpenAI’s turn in recent years, saying he did not understand how OpenAI, a non-profit organization he had invested $100 million in, could become a for-profit organization. He completely disagrees with this direction of development. But there’s not enough voice, not enough votes on the board. The others came hard, we had no choice. So Musk set up his own AI company, xAI, in 2023.
Musk is not alone in his opposition to OpenAI’s pivot. So is the low-key Ilya Sutskever. His defiance did not appear in the press, but in the terms of OpenAI’s license to Microsoft.
In OpenAI’s license terms, there is an unusual provision: 数字化转型网www.szhzxw.cn
All OpenAI licenses and commercial terms to Microsoft are limited until OpenAI implements AGI. Once OpenAI implements AGI, the technology is not open to Microsoft, and all previously agreed licensing and commercial terms are void for AGI systems and AGI technologies.
What is AGI? The definition of AGI is: General artificial intelligence, a highly autonomous system that is capable of doing more than humans.
And how do you know if you’ve reached AGI? It is up to OpenAI’s board of directors to decide unilaterally. At present, the OpenAI board is the six-member board mentioned at the beginning of this article.
In other words, Ilya Sutskever tries to grasp the throat of external shareholders through this clause, and once the technology reaches a breakthrough or the other party has a change, it will immediately cut its throat. 数字化转型网www.szhzxw.cn
This clause is not a secret clause. Everyone in Silicon Valley knows that. And in October 2023, there was a public discussion in Silicon Valley.
What does that show? It shows that Microsoft may have to move to eliminate the uncertainty caused by this clause, to ensure that it can successfully swallow OpenAI into its stomach, so that it can be 100% sure of becoming the dominant player in the field of artificial intelligence.
Possible moves by Microsoft include acquiring full ownership of OpenAI as soon as possible, shaking up OpenAI’s board to ensure it can control board votes, and removing Ilya Sutskever.
In addition to the Microsoft side of the force, the second side of the force is the investment fund, they have invested a huge amount of money in the artificial intelligence circuit, is waiting to get rich.
The third party is the small companies that develop in the OpenAI ecosystem, and their companies depend on the progress of OpenAI. 数字化转型网www.szhzxw.cn
The fourth force is Sam Altman. It should be noted that although he is now the CEO of OpenAI, he does not have any equity, and his only investment income is a small amount of money he invested in OpenAI angel round through Y Combinator a few years ago, which is better than nothing for an individual. In general, it makes sense for a CEO to own 10% of the company. So, for Sam Altman, how to monetize their own interests, is a very strong motivation, after all, we all see that after the OpenAI profitability, is a trillion-dollar industry leader, then, in the usual situation to expect, Sam Altman will expect to get $100 billion in revenue.
So, the question is, how can Sam Altman monetize this $100 billion expectation without having a stake in OpenAI? $100 billion. No one wants to let go. One way is for Sam Altman to swap with shareholders, for example with Microsoft, and cash out through Microsoft.
At present, Ilya Sutskever is now one man against four forces.
Can you see voices in support of Ilya Sutskever in the media and on social platforms? No, not at all. I believe you can see a lot of voices, all want him dead. After all, standing in someone’s way is like killing your own parents. 数字化转型网www.szhzxw.cn
Mr. Musk and Mr. Hinton share his philosophy, but for now, they can’t get in the way.
In addition to the core reason that Ilya Sutskever holds the throat of the four powers, there is another aspect of Ilya Sutskever that people hate. It is Ilya Sutskever who wants the commercialization of AGI general AI to be slower. This runs counter to other forces’ desire to get rich quickly.
Ilya Sutskever believes that AI has shown signs of surpassing human intelligence in 2016. At that time, in the second game between AlphaGo and Lee Sedol, the 37th step, made a very special step. This step is considered by human commentators to be a failure of extraordinary level. And as it turned out, that move was the key to winning the whole set. This is a way of playing chess that has never been done in human history.
Therefore, general artificial intelligence AGI is bound to surpass humans soon. While others are still struggling to figure out how to build an AGI system that can match the level of human thinking, Ilya Sutskever is already one step ahead, arguing that AGI is bound to surpass humans soon, so it is time to start preparing for AI that will surpass human intelligence.
He believes that the most appropriate relationship between AGI and humans is the relationship between parents and babies, AGI is a parent, humans are babies, AGI parents think about baby humans all the time, so that baby humans live better. 数字化转型网www.szhzxw.cn
And existing methods of controlling AGI do not work for AI that is smarter than humans. Because current models assume that humans can reliably evaluate the work of AI systems. As AI systems become more capable, humans don’t know how to evaluate them, whether they’re doing better or worse after every change. This results in an AGI that cannot be controlled.
So, in July 2023, Ilya Sutskever teamed up with colleague Jan Leike to form a small division within OpenAI called Superalignment. The goal is to develop a fail-safe program to control AGI technology, allocate one-fifth of OpenAI’s company-wide computing resources to this division, and solve the problem within four years. Then, with this control technology in place, AGI technology is opened up for human use.
Many in Silicon Valley scoffed at Ilya Sutskever’s “super-alignment” approach and thought that Ilya Sutskever would hinder the development of OpenAI. Four years later? No, I want to be rich, I want to be rich this year, I want to be rich this year, I want to be rich tomorrow, I want to be rich tomorrow.
So, yesterday, news broke that various forces forced OpenAI’s board to dissolve and reorganize, bringing back Sam Altman as CEO. Because it’s in their best interest. As soon as Sam Altman returns, he will inevitably shake up the board, outing Ilya Sutskever and ensuring he has a firm grip on the board.
What happens next for Ilya Sutskever?
It’s a fight song by a scientist founder.
I wish Ilya Sutskever would stick it out, not compromise with the four powers, and take control of the board. 数字化转型网www.szhzxw.cn
After all, others want a giant money printing machine, a hegemon in artificial intelligence.
And Ilya Sutskever wants an AI that can take care of humans the way parents take care of their babies.
本文由数字化转型网（www.szhzxw.cn）转载而成，来源于华龙网；编辑/翻译：数字化转型网宁檬树。
免责声明: 本网站(https://www.szhzxw.cn/)内容主要来自原创、合作媒体供稿和第三方投稿，凡在本网站出现的信息，均仅供参考。本网站将尽力确保所提供信息的准确性及可靠性，但不保证有关资料的准确性及可靠性，读者在使用前请进一步核实，并对任何自主决定的行为负责。本网站对有关资料所引致的错误、不确或遗漏，概不负任何法律责任。
本网站刊载的所有内容(包括但不仅限文字、图片、LOGO、音频、视频、软件、程序等) 版权归原作者所有。任何单位或个人认为本网站中的内容可能涉嫌侵犯其知识产权或存在不实内容时，请及时通知本站，予以删除。
