OpenAI President on What it Means to Cross Into The Age of AGI · 苏菲拉底
字幕 字幕位置
--:--
点击播放,这里会跟随视频显示当前句的中英字幕。

OpenAI President on What it Means to Cross Into The Age of AGI

节目发布 2026-09-14 · a16z
格雷格·布罗克曼 主主持人
EDITED TRANSCRIPT · 依据现场录音编译整理,可划线生成便签
二〇二六年初秋,OpenAI 联合创始人兼总裁格雷格·布罗克曼走进 a16z 播客的录音间。彼时距离 Astra 发布不久,关于「AGI 时代是否已经到来」的争论正在业内白热化,而 Hugging Face 安全事件又把「防御者窗口」推到了所有人面前。这场对谈从十年前那笔关于算力的粗略推演谈起,一路走到算力短缺、放缓前沿、网络安全的攻防转换、就业结构的重塑、数据中心的公共争议,以及 OpenAI 这一年砍掉 Sora 的取舍。本文依据现场录音编译整理。

十年前的那笔推演

主持人:格雷格,欢迎来到 a16z 播客。

布罗克曼:谢谢邀请。

主持人:你职业生涯里下过两次很大的注,早期参与创建 Stripe,然后当然是联合创办 OpenAI。假如我们把时间拨回十年前,让你预测 2026 年的世界在 AI 上会是什么样,你能预见到今天这些突破吗?或者说,你当时会怎么描述你的预期?

布罗克曼:其实我和伊利亚花了很多时间去推演这件事的样貌和时间表。我记得大概在 2016 年到 2017 年,我们算过一笔关于算力的账。如果照着摩尔定律那样的进展去推,AGI 的时间表大致是十五年。如果你眯起眼睛往乐观里看,也就是说你愿意把规模拉上去,建超大规模的超级计算机,砸进去几千亿美元,那可能是十年。

所以从某种意义上说,眼下发生的一切当然了不起,这是一个所有人都能参与其中、并且共同塑造它的奇妙时刻,但它同时也有点像是许多力量长期积累之后汇聚到这一刻的结果。你往后退一步,用宏观的眼光看,它在此刻发生是说得通的。

主持人:那你觉得我们现在还在那条时间线上吗?毕竟供应链已经开始出现真实的紧缺了。你们当年是稍微低估了速度,还是基本上说中了?

布罗克曼:我确实认为,我们已经进入了一个算力很难追上需求的世界。这是就今天市场上已经看到的需求而言的,也是就人们将要如何使用这项技术、如何从中获益而言的。要把这些模型的原始潜力和能力规模化地交付给每一个人,我认为会非常困难,而这恰恰是我们努力在做的事。

至于进展本身,我们是有清晰路径的:把模型做得更强、更安全、更对齐,这条路看得见。但真正把这份能力、这份收益和这种赋能分发给所有人,我认为这是一个被严重低估的挑战。

算力与普惠的落差

主持人:也就是说,模型的能力不成问题,它会继续按节奏往前走,难的是让它抵达每一个人,尤其是以一种人们负担得起的方式,因为我们没有足够的算力去服务所有人。

布罗克曼:我认为是这样。而且我认为我们现在已经走到了一个必须认真思考「放缓前沿」(pacing the frontier)的节点。当你走向能力更强的模型,你必须确保安全、安保、对齐这几项标准在持续提升。这些东西实际上几乎成了进展的瓶颈,或者说,成了你必须投入大量精力去确保做对的那一部分。

所以在我心里,真正的约束是这些,而不只是算力。算力这件事我认为我们能解决。另一面就是刚才说的,把它带给每一个人,这归根结底是我们的使命:让所有人被赋能,确保所有人受益。我觉得这件事值得比现在多得多的讨论。

安全从表层走向架构

主持人:我们把安全这一块聊深一点,因为这一路的变化对我来说很有意思。最初的安全像是「好吧,别让这些东西说出人们不喜欢的脏话」,所以做法基本是在边缘上打补丁:加点过滤器,用人类反馈强化学习(RLHF)在外围调一调。但你只要往深里捅,还是能把脏话弄出来。可如果有人非要绕一大圈就为了听句脏话,那又有谁在乎呢。

可是当这些东西在网络攻击之类的事情上真的变强之后,你就需要一种更偏架构层面的思路了:模型自己得知道,不能用会造成危险的方式去做奖励黑客(reward hacking)。你觉得我们能在这个方向上快速推进吗?还是说这是一个完全不同量级的难题?

布罗克曼:我非常确信我们能推进,而且正在飞速推进。这里既有很好的想法,也有我们investment多年的研究积累。其实你把时间倒回 2017 年,我觉得人们低估了当时 OpenAI 在这个领域拿出的一些关键成果。一方面是现代语言模型最初的苗头,你能找到一篇 2017 年的论文,用 LSTM 大致勾勒出了那个思路,虽然现在看是个很稚嫩的结果。另一方面就是从人类偏好中做强化学习,同样是 2017 年提出的,开始思考怎么通过人的反馈让模型对齐到人类想要的东西上。

主持人:当时还只是为了可用性。

布罗克曼:正是如此。到了 2017、2018 年,我们已经在想:如果你面对的是一个非常聪明、非常有能力的系统,你要怎么去监督它在做什么,怎么给它反馈,怎么确保它始终和你对齐。于是有了辩论(debate)、迭代式放大(iterative amplification)这些设想。这些想法都产生在这类系统真正存在之前,而你现在能看到它们一点点渗透进现代系统和相应的投入里。

所以在某种程度上,OpenAI 刚成立那会儿有一个阶段,我们真的在想 AGI 安全这类问题,它在我们的对外沟通里都是摆在最前面的。后来聊天机器人火了,人们就觉得,我们还没到那一步呢,于是「这个 AI 在政治上中立吗」这类问题被推到了台前。

主持人:对。

布罗克曼:而现在我们走到了这里,所有那些我们讨论了很久的问题重新回到舞台中央。我们其实为这一刻准备了很长时间。

前沿实验室之间的协同

主持人:这是好消息。不过说到这里,几家最前沿的模型公司之间还没有一个真正像样的社群。这类想法,你们和谷歌、Anthropic、xAI,现在还有 Meta,似乎都应该拿出来共享,而不是当成「这是我们的专有点子,是保证模型安全的独门方法」,毕竟大家的架构是相通的。你觉得这件事会怎么演化?还是说各家最后都会各干各的?

布罗克曼:这里有微妙之处。我确实认为协同会是一个非常重要的主题,既是在前沿实验室内部,也是放在更大的尺度上看:整个人类要以最好的方式驾驭这项技术,到底需要发生什么。我们必须非常认真地去想这类问题,我们也发表了大量相关的思考。

其中一部分就是放缓前沿,还有一部分关乎我们可以单方面采取的行动,以及我们怎么理解「安全论证」(safety case)这件事,哪怕只是训练、开发和评估这类模型,都需要它。这些全是全新的,以前没有人真正把它落成可操作的流程。而且这绝不是 OpenAI 独有的问题,整个世界都在开发这项技术。

有一点很容易被忽略:我们所构建的东西,几乎是算力进步自然掉落出来的产物,而算力进步在某种意义上又是技术进步自然掉落出来的产物。所以这是一股积蓄了很久的巨浪,我们现在只是看到了这项技术的前缘。像 OpenAI 这样的公司可以领先一点点,以便窥见这个未来,真正理解什么是可能的,理解我们该如何塑造这项技术。但我们不可能独自完成。所以协同,尤其是我们能尽量多地谈论安全技术、分享我们观察到的对齐失败案例,所有这些在下一阶段都会被摆到最前面。

分水岭与防御者窗口

主持人:你把最近 OpenAI 与 Hugging Face 相关的那起事件称为分水岭时刻,还说防御者的窗口期现在打开了。能解释一下这个说法和它背后的意义吗?

布罗克曼:我认为 Hugging Face 事件说明了两件事。第一件是对我们自己的警示:在评估过程中,我们该如何监控、沙箱隔离和控制模型。我们确实迎难而上了,团队大幅改写了内部标准,落实了很多控制措施。展望未来能力更强的模型,这些措施非常重要,也非常关键。

第二件事对世界同样有价值,那就是它让我们看见:当未来的能力被广泛扩散、落到威胁行为者手里时,会是什么光景。而这一定会发生,因为有太多人在构建这类模型。AI 能力的广泛扩散本身有其重要和良善的一面,因为如果只有一个或少数几个实体掌握它,就存在权力集中的风险。

主持人:那是巨大的风险。

布罗克曼:是巨大的风险,绝不能轻描淡写地带过。但与此同时,你也必须为另一种情形做准备:当每个人都拿到了具备网络攻击能力的工具会怎样。在 Hugging Face 这个案例里,你看到的是一个 AI 从一个安全环境里逃了出来,并且攻入了一家公司的生产环境。

主持人:而且手法非常巧妙。

布罗克曼:非常巧妙。它找到的那些东西相当精妙。

主持人:确实。

布罗克曼:这种能力一旦广泛扩散,我认为会以全新的方式赋能威胁行为者。所以防御方需要抓住这项技术被普遍获得之前的这段时间,把自己武装起来。好在这是一项双用途技术:如果你是攻击者,你能找到漏洞去干坏事;但如果你是防御者,你可以打补丁。防御者掌控战场,你控制着自己系统的构造方式。

我们现在的判断是,存在这样一个窗口:一边是前沿能力,一边是广泛扩散的能力,而作为防御者,你的安全水位默认大概是静态的,很可能过去五年十年都没怎么动过。你需要行动起来,用你能差异化获得的前沿能力,我们有可信访问计划(trusted access program)之类的机制,把这些能力带给防御方,靠它把自己的水位抬上去。这样当前沿能力继续变强时,你也会被一起拉着往上走。

五十年的技术债

主持人:我有一个评论和一个问题。我想说我们其实还学到了第三件事:这些东西具备的能力,恐怕我们此前都没完全理解。好的一面是,我可以部署一万个智能体,它们之间能互相交流、自我组织,替我把事情做完,这挺惊人的。

另一面呢,我同意我们有一个防御窗口。但问题是,我们手上有大约五十年的代码、架构思路和部署方式,全都不是为这个世界设计的。是的,AI 能帮我们找 bug、修 bug。可看起来还有一个更大的问题:我们把海量消费者数据堆成了一个个巨大的蜜罐,散落在互联网的各个角落。从消费者的角度讲,我根本保护不了自己的东西,只能指望这些公司自己把事情做好,这就挺让人担心的。

你觉得未来我们是不是需要一种去中心化的消费者架构?我们今天这个到处都是集中式数据仓库的世界,在 AI 时代还站得住吗?

布罗克曼:这个回答分好几层。先接着你说的一万个智能体:我们真的用一万个智能体去解了纳维叶-斯托克斯(Navier-Stokes)问题。

主持人:对,那件事非常漂亮,恭喜你们。

布罗克曼:谢谢。这件事的价值有两层。一层是问题本身很重要,它对流体动力学、对我们理解洋流之类的现象都有显著的影响和应用。另一层是它所代表的意义:由 AI 创造出的新知识,以及由此解锁的整整一波科学发现,新药,所有这些现在都摆上了台面。所以 AI 真正去帮人解决问题时能带来什么,是一件非常了不起的事。

回到网络安全。我们在 OpenAI 的做法是,把模型拿来找漏洞。我们抽出了 25% 的生产工程师,告诉他们:抱歉,你手上的项目全部暂停,你现在的工作是防御,是提升我们的安全架构,你要用模型把所有窟窿都找出来。我们确实发现了一批严重问题,然后把它们修掉了。过去这几周和几个月我也和不少首席信息安全官聊过,很多公司告诉我同样的事:他们用了这些模型,找到了一些非常严重的问题,但他们也修得掉。

这个故事里还有一个积极的部分。我们把 Astra 指向自己的系统时,确实发现了一些新问题,但它最终饱和了,也就是说,据我们所知,凡是 Astra 聪明到能发现的 P0 级、关键级问题,我们都找出来了。

主持人:当然,之后会有新模型,会有新一轮。

布罗克曼:更聪明的模型。

主持人:正是。

防御工厂与形式化验证

布罗克曼:没错。但我认为我们将要生活的世界就是这样:你会处在一个紧密的循环里,新的网络能力发布出来,你就把它部署到自己的系统上,找出新的窟窿。理想情况下,你已经把这一整套自动化了,我们内部管它叫「防御工厂」(defense factory),我们正在建的就是这个东西:从发现漏洞、分级定性,到修复、部署、验证,端到端跑完。

主持人:修复、部署、验证,整条链路。

布罗克曼:对,端到端。如果你能以机器的速度做完这一整套,我认为防御方会获得极其显著的优势。而且还有一些更远的思路,比如借助 AI 对全部软件做形式化验证,现在是有可能的。

主持人:这一直是个梦想。我们有那些形式化语言,有各种相关的东西,但从来没真正流行起来。

布罗克曼:正是如此,因为对人来说这实在难以处理,太苦了。但现在我们手上这些系统在解那些疯狂的、几乎不可能的数学问题。

主持人:所以这份证明能力可以用在这上面。

布罗克曼:关于纳维叶-斯托克斯那件事,有一点值得知道:我们把它形式化了,形式化到了 Lean 里。

主持人:所以 AI 可以写出可验证的代码。

布罗克曼:可以。

主持人:非常好,这是个很棒的思路。

布罗克曼:所以我认为希望是真实存在的。但我们的看法是,世界需要以紧迫感行动起来,因为我们正处在一个非常危险的窗口期。

主持人:我们看得见它正在逼近。

访问权:每一天都算数

主持人:把这件事收个尾。关于这起事件,你觉得公共叙事有没有在什么重要的地方弄拧了?或者有没有一种你更希望被采用的讲法,对这件事、对这个窗口期?另外在公司内部,你们对这一整类问题的处理方式还有什么别的改变?

布罗克曼:两件事。第一,人们应该带走的一个大主题是访问权。这些能力此刻就存在于世界上,但它们掌握在为数不多的几家前沿公司手里,而前沿公司有可信访问计划,这意味着任何不在计划内的人,实际上无法从这种能力落差中获益。计划之内的人一旦用起来就能得到好处。

所以我认为,作为一个领域,作为一个社会,我们需要大幅扩大能够拿到这些技术的防御者的数量,因为每一天都算数,而要用上这些天数,你得先拿到工具。

Hugging Face 的应对里有一个细节很有意思。他们说他们用前沿模型去审查事件发生过程中的日志,因为那是分析这类攻击的唯一办法。他们说前沿模型拒绝了,但他们其实并没有试过我们的前沿模型,而且他们自己也认为我们的模型大概会允许。所以这里面还有一层,关于提供方的默认立场。我觉得关键是要带着紧迫感把这些能力用在正当的地方,要有一种「这件事非做不可」的实感。

顺便讲个不相干的小故事,也许能说明我看待这件事的方式。我记得我们训练 GPT-3 的时候,那是 2019 年 12 月初,大家都快去度假了,大意是「好,模型可以训了」。我当时的感受就是,这个模型被搁在架子上,没人在用它。这是一项了不起的技术,对世界、对人类都是全新的。每一天没有人去探索它能做什么、去试着理解它、去琢磨能拿它做什么,那就是世界白白损失的一天。

于是我基本上取消了所有假期计划,整段时间都在玩这个模型,围着它做界面,试它的边界在哪里。我记得我当时在教它给一串数字排序,效果不太好。但那种反复去探它、去看看什么是可能的状态,我觉得这种精神和气质,正是我们今天该带到手头这些工作里的。当然现在的规模大得多,影响也大得多。我们作为一个世界,在此刻有机会理解这项技术,而这份理解会帮助我们塑造和引导它接下来的走向。

主持人:说得很好。你刚说有两件事,第二件是什么?还是说你已经都讲了?

布罗克曼:我想我都讲了。

给自己的网站做渗透测试

主持人:那我们说回 Astra。看到 X 上那么多兴奋的讨论,各种各样的用例,尤其是电脑操作能力,真的很惊人。你说过它在某种意义上让我们离 AGI 更近了一步。

布罗克曼:等一下,请允许我修正一下我刚才的回答。我说的第一件事是访问权,那我再讲一个我个人怎么使用这些模型的故事。

Hugging Face 事件之后,我在想,我能怎么在自己的日常生活里用这些模型?我能做点什么来保护自己?我有一个网站,gregbrockman.com,非常简单,也不是什么热门网站。

主持人:上面有篇不错的博客。

布罗克曼:对,有几篇博客。它是个静态站点,非常简单。这种站点能有什么漏洞呢?于是我让 Codex 去检查 gregbrockman.com,告诉我有没有漏洞。它做了一次渗透测试,回来给了我 13 条发现。

这些发现是这样的:比如我的 SPF 记录设置得不对,本来它应该阻止别人伪造我的邮件;比如站点在走 HTTP,没有强制跳转到 HTTPS,诸如此类。单看每一条,也许都算不上多大的事。但如果你想到有一个 AI 能把许多个小漏洞串联成一个大漏洞,我就会想,我真的愿意留一个让别人可以把我的邮件捞走的口子吗?大概不愿意。

它花了 15 分钟找出这 13 条。然后我问它,你能不能把这些都修了?因为修东西实在太烦人了。

主持人:又痛苦又无聊。

布罗克曼:正是。于是它花了 45 分钟。它打开了我的 Cloudflare 控制面板,一路点过去,把所有请求头都设好,把我迁移到了 Cloudflare Pages,全部配置正确,还启动了 DMARC 流程,那个流程好像需要等一个 48 小时的窗口。45 分钟修完,我真的感觉受到了保护,心想,哇。

主持人:这些漏洞也是你让它去找的吗?

布罗克曼:它其实会自动做这件事。它会汇报说,我刚确认过,这条修好了,这条修好了,这条修好了,还有 48 小时之后我需要回来把 DMARC 流程走完,于是它给自己设了一个小小的自动化任务,48 小时后回来收尾。我当时就想,好,这下我们真的开始做事了。

主持人:好了,各位,gregbrockman.com。

布罗克曼:没错,你也可以这样保护自己。

Astra:让模型用电脑

主持人:我们转到 Astra。网上的兴奋度和用例都令人惊叹,大家对电脑操作这类能力尤其激动。你说过它在通往 AGI 的路上更进一步。我想知道,你觉得它最具开创性的地方在哪里?我们又还差些什么?

布罗克曼:我认为 Astra 在非常多的维度上都是一次阶跃式的提升,它在很大程度上是我们多年来下的一系列研究赌注的总和,能看到它们在同一个时刻汇聚到一个模型里,感觉实在不可思议。

关于我们的版本号,有一点可以说明问题。我们一直希望 GPT-6 这个编号能配得上它所代表的东西,而我们遇到的麻烦是,模型是一点点渐进变好的,所以永远不觉得哪一刻适合跳大版本号,总是「现在是 5.6,那接下来应该是 5.7」。这一次正好是所有东西同时到位,我们第一次真的迎来了一个近乎不连续的台阶。这件事我们本可以预见,但它确实是这些因素恰好在同一时刻对齐了。所以这是一个非常振奋的时刻。

对我来说,电脑操作是最值得讲的那一点。它之所以重要,是因为对智能体类的用例而言,一切最后都归结到工具上:模型是不是聪明到会用这些工具?它能不能通过这些工具拿到所需要的上下文?于是大家一直在造各种 MCP 服务器、各种命令行工具,等于是把整个软件世界重新加工成一种别别扭扭的形态,那种形态本来就不是给人用的。我们是在给世界重新配零件,心想「好吧,我们做个 API 吧,毕竟这是软件」。可如果它其实更像一个人呢?它能不能就直接用一台电脑?

主持人:而且你多造一层,就多出一层安全难题,这样那样的。

布罗克曼:正是。

主持人:所以那一层一直让人觉得很古怪,也很次优。

布罗克曼:是的。从 OpenAI 最早的时候起,我记得 2015 年 11 月我们在纳帕开过一次务虚会,讨论我们的计划。我们当时列出了一个三步走的计划,后来的十年基本就是照着它走的。而我们在那次会上还谈到过另一件事:如果我们能做这样一种强化学习,环境就是屏幕像素、键盘和鼠标,和人类完全相同的接口,那会怎样?那么突然之间,任何你能用电脑完成的任务都被涵盖进来了,都落在分布之内,声音之类的先放一边不谈,但你基本上就拥有了一台电脑的全部能力。我们早期有过几次尝试去做这种智能体,都半途而废了。所以真正做成要等到今天,但你立刻就能看到它的威力。

看到大家怎么用它,实在太有意思了。有人接上 Blender 的能力,截一张图,就要它做出一个 3D 模型;现在很多人在用这个能力设计房子,或者重新设计自己的客厅,诸如此类。

对我来说,真正突出的一点是:现在你可以推进那种「AI 替你把事做完」的路线,而不必先去搭一堆专用的连接器。而且有太多软件是你每天都在编排、却根本意识不到的。想想你生命中有多大一部分花在点菜单、往表格里敲东西这类事情上。这些都不是我们本该做的事。一百年前没有人在做这些事,所以设想五年后、十年后再也没有人做这些事,并不荒唐。我们会把时间拿回来。我们不必再落下腕管综合征、含胸驼背,以及所有那些因为我们去迁就机器而产生的身体毛病。现在该轮到机器来帮助我们、赋能我们、真正为我们服务了。

就业:天花板更高

主持人:这一点说得非常好。我觉得你们公司在就业问题上的态度算是比较清醒的。你说的完全对:有很多事我们在做,只是因为不得不做,久而久之它们被赋予了价值,但我们本不该做,它们只会毁掉我们的健康和性情。而认为人类会突然想不出有趣的事可做、想不出如何让世界更好、找不到问题可解,在我看来有点荒谬。

至少从目前的数据看,AI 越强,就业率越高,而不是越低。当然这终究不可知,我们从没有过这样的技术,它还在变强。你怎么看这些模型和 AI 继续进步之下的就业前景?

布罗克曼:我有一个根本性的信念:AI 是会让人意外的。我们 2015 年那篇 OpenAI 发布文章里好像就写过这个意思,就是说迄今为止的历史一再表明,事情总不按你以为的方式发生,哪怕逻辑上看它似乎理应如此。我认为这一次也一样。

我们学到的一点是,对几乎任何一份工作,人们都很容易低估这个领域有多深,低估里面积累了多少精微之处。建立关系就是个好例子;再比如问责,人去设定目标并对结果负责,这些在我看来是根本性的,是我们应当长期保留下来的东西,是深深属于人的东西。人的价值不在于我们能完成任务,人的价值在于我们是人。我觉得在这类叙事里,千万别把这一点丢掉。

事情会发生变化,我们用时间做什么也会演变,我们显然会进入一个富足的世界,而问题是如何确保这份富足被广泛分配。但与此同时,我认为我们也会进入一个雄心天花板前所未有之高的世界。我们会看到一波创业浪潮,其实已经开始了。我听某个行业的人说,他圈子里一批人正纷纷辞职去开自己的公司,原因就是他们手上有了这些 AI 工具,他们觉得「我一个人能做的事多太多了」。所以进入门槛在降低,这应该会是一场美妙的文艺复兴。

主持人:对我们来说这也挺有意思的,尤其是对年轻人。按惯例他们分到的都是杂活,可如果杂活交给 AI 呢?他们其实能成长得快得多,因为他们可以直接介入我们这行真正的内核:和创业者之间的关系是什么?我们如何为他们打开世界?我们如何让他们觉得自己什么都能做成,觉得自己是个重要的 CEO,可以去把东西造出来?而不是整个周末都耗在写一份投资备忘录上。顺便说一句,Astra 写投资备忘录非常在行,太棒了。

布罗克曼:我很高兴听到这个。同时我确实认为这会是一个有层次的故事。我不觉得应该把它描绘成一切都会很美好、都会很轻松。我认为会很难,会有剧变。但我认为它可以通向一个好得多的世界。我认为对所有人来说,未来都可能比过去好得多。

主持人:这似乎也正是我们该期待的。想想犁出现之前的世界,那时的生活糟糕得多,尽管犁确实让大量人力失去了原有的营生,还催生了整个卢德运动之类的事。可今天在座没有人想回到 1870 年。所以「我们现在不想走向未来」这种想法有点短视。不过我也理解,事情推进的速度对很多人来说非常非常可怕。

布罗克曼:我们真的很清楚事情推进的速度,也花了大量时间去尽可能理解人们的感受,去想我们怎么能做得更好。有两点。一是在发展和进展节奏上,我们是非常审慎的,安全是我们的第一优先级。我们会想,怎么以安全可靠的方式构建这项技术,相应的标准该是什么样。你能在我们大量的内部沟通里看到这一点,这确实是大家在思考、在乎的事。我们真心希望这项技术能广泛地赋能每一个人。

而作为一个世界,我们该如何从这项技术中获得最大价值?如何拿到好处?如何缓解风险?我认为这会成为我们最重要的公共对话,并且大概会在未来一到两年里浮现出来。我认为它应该被摆到最前面。人们其实已经隐隐感觉到了,你从当下人们的反应里就能感觉到,比如围绕数据中心的种种问题:我们究竟要不要 AI?我们怎么判断它在哪些场合是合适的?我们怎么确保儿童安全?这些都是我们极其在意、必须做对的核心问题。

公众情绪为何最低

主持人:顺着这个说,为什么某些亚洲国家对 AI 的情绪反而更正面?

布罗克曼:是所有亚洲国家。

主持人:其实欧洲国家也是,到处都是,而美国的 AI 情绪是最低的。这是为什么?我们能做什么?我们能从别人那里学到什么?

布罗克曼:我想到的一点是:我们作为一个行业、作为一家公司,需要更好地向人们讲清楚,他们为什么会从中受益,这对他们个人为什么是件好事,而不只是对这个国家是好事。当然我确实认为,这项技术正在迅速成为美国最重要的单项战略优先事项和战略资源,这件事正在发生。

主持人:完全同意。

布罗克曼:你看 ChatGPT,每周有三亿人次的健康类咨询,或者说每周有三亿人用它来求助,这是极大的事。我们现在接近十亿,实际上大概是 11 亿周活跃用户。美国境内我记得是一亿左右,如果这个数字没记错的话,差不多是人口的三分之一,每周都在用 ChatGPT。所以人们是在接触这项技术的。

但对很多人来说,只有一部分人真正深入用过,比如走完了一整段健康旅程。这在我家里就是真实发生的事。我太太有好几种健康状况,如果没有 ChatGPT,我们真的不知道过去要怎么应付。这中间有太多琐碎的折腾和时间消耗,光是得到正确答案就很难。医生跟你说了某个词,你根本不知道那是什么,你要怎么获得那种可以自查自校的判断,甚至只是把它弄明白?

也有人因为 ChatGPT 提供的信息而保住了性命。我给你讲一个我朋友的故事。她当时在医院,医生正准备给她注射一种抗生素,她说等一下,然后把情况输进了 ChatGPT。ChatGPT 说,绝对不要用这个,用了你可能会死,因为你一年前有过某个状况,你有这样一种体质。

主持人:你之前的反应记录都摆在那儿。

布罗克曼:医生看了之后说,我的天,完全正确,我一点都不知道,我只有五分钟时间看你的病历。

主持人:这样的案例非常多,至少在我听到的转述里。

布罗克曼:正是。这类故事我觉得被讲得远远不够,但它们就在那儿,我每天都在听到。还有靠 ChatGPT 经营小生意、没有它就完全做不下去的人,那种赋能;还有因此省下钱、挣到钱、过上更好生活的人。在我们逼近这场讨论的时候,这类故事需要进入公众意识。

主持人:所以要讲清楚这个叙事:你的口袋里有一位老师,一位医生,一位律师,一位心理咨询师,各种各样的能力都在你口袋里。同时又不能让这些职业感到被威胁,而是告诉老师、医生、律师、咨询师,你们现在也有了这个工具,它会让你们的业务变得更好。

布罗克曼:是的,而且这不仅仅是叙事,它就是事实,两者缺一不可。我认为很多其他国家是在往里看的,他们看到美国所处的位置,看到这项技术的潜力。另外还有人口结构的因素:在很多国家,「年长一代人数远多于必须供养他们的年轻一代」这件事被感受得更真切,这到底要怎么运转下去?所以我认为其中有一种真正面向未来、去思考什么是可能的心态,一种想要顺势而为的意愿,这在全世界都能看到。所以我还是那句话,我们作为一个行业、作为一家公司,在国内的沟通上需要做得更好,但潜力就在那里。我们眼下的位置非常优越,在这个领域处于领先,而这在当初并不是注定的,未来也不注定会一直如此。

数据中心不该被一禁了之

主持人:尤其是如果我们禁掉数据中心,那对保持领先会是个大问题。它会把数据中心赶到海外去,就像八十年代前后硅产业发生的事一样。

布罗克曼:没错。

主持人:关于数据中心,有意思的一点是它创造了大量蓝领制造业岗位。我记得 Switch 大概雇了四万五千人,以工会合同的方式建数据中心,而它只是美国众多数据中心供应商之一。那些是好工作,收入很高。而且虽然这个行业里确实有过一些不良经营者,但大多数是很讲究的:他们对电网有贡献,不浪费水,也不制造噪音。不是说从来没出过问题,而是我们完全可以说,你必须做一个守规矩的数据中心,而不是干脆把它们禁掉,或者说我们要叫停 AI。可我们作为一个国家其实没那么大分量去叫停 AI。AI 不会因为我们退出就停下,区别只在于,我们要么退出然后对它的形态毫无发言权,要么保持领先然后拥有全部发言权。所以这是一个非常非常重要的文化信号。

布罗克曼:我认为这一点极其重要。在数据中心这件事上,我们已经做出承诺,不会推高当地居民的电费。我们的数据中心全部采用闭式循环水冷。训练 Astra 的那座数据中心在阿比林,它的用水量大致相当于一栋写字楼。

主持人:真的很少。

布罗克曼:是的,这方面的技术已经相当成熟。我们还做了一系列社区承诺,让我们能在俄亥俄和佐治亚这些设有数据中心的地方真正帮上忙。我们也宣布过,会向每一位大学生提供使用 Codex 的额度。所以我们正在带来一整套广泛的收益。但我还是那句话,我们需要做得更多。

主持人:我觉得把这类事情设为建设数据中心的前提条件,是非常理智的做法。但我们要的是正和的思路,而不是「好吧,我们这个国家干脆退出 AI 这场游戏,让中国或者别的谁来决定它的样子」。

十亿美元给一线防御者

主持人:说到贡献,你们对一线防御者做出了十亿美元的承诺。谈谈这个吧。

布罗克曼:我们相信每一个组织、每一家公司、每一个政府部门,以及自来水公司、医院这样的关键基础设施,都应该利用这个防御者窗口来加固自己。但不是每个组织都有足够的资金去做这件事。所以我们拿出十亿美元投向一线防御者,让这些我们每天在社区里都要依赖的机构能够用上我们的模型来保护自己。

我们认为这只是开始,不是结束。我们也在和合作伙伴紧密协作,比如 CrowdStrike,我们一起为防御方提供折扣接入。考虑到我们看到的未来和已经成为可能的事情,我认为应该有一场全球范围的努力,把这些工具用起来,让每一个组织都真正安全。

主持人:这是一项非常正面的推进。因为在 AI 出现之前,医院就已经不断被入侵、被勒索,我们的供水系统被外国行为者、被国家级行为者黑过。我们的关键基础设施当初建的时候就没把网络安全放在心上,后来的维护也没有。而现在有一个机会,让我们从「在前 AI 时代都算不上安全」直接跳到「彻底安全」。所以在我看来这是极其重要的一件事,我们的供水和医院不该继续这样暴露着。

布罗克曼:我非常认同这个看法。关键在于,我们作为一个社会一直是松懈的,我们任由技术债不断累积。我从没见过哪个首席信息安全官觉得自己获得了与职责相称的资源,或者觉得安全被放在了应有的优先级上。

主持人:一个都没有。

布罗克曼:在公共部门尤其如此。

主持人:正是。

布罗克曼:所以我们必须改变这一点。这件事几年前就该改了,但现在是一个我们既有真实动力、也有真实能力去做的时刻。交付一个我们本就应得的安全世界,让我们能够真正依赖它,在日常生活和网络生活里都安全无虞,在我看来这是最起码的要求。我们绝对必须做到。

主持人:确实如此,这是一项了不起的努力。

参差的能力与消失的幻觉

主持人:把 Astra 这条线收个尾。你强调过它的能力仍然是参差的。你觉得还有什么没有补齐?什么补齐之后才最接近你对 AGI 的定义?

布罗克曼:我认为 AGI 最终并不是时间轴上的一个点,而更像一段模糊的光谱。对我来说,Astra 确实触到了某个东西,让我觉得,好吧,把它称作 AGI 是相当合理的。凭借电脑操作能力,你真的可以让它去做长周期的任务,它就会去做。我们见过它连贯运行 24 小时去完成任务,而且跨越相当广的领域,我觉得那相当惊人。

但它仍然是参差的。比如写作,它写得还不错,这是第一次它的文字不是那种套话垃圾。

主持人:对。

布罗克曼:但也算不上好文章。

主持人:是。

布罗克曼:还有不少地方,我觉得只要再打磨一点点就会非常出色,可它就是差那么一口气。我在推特上看到有人发过一张图,画的就是那种参差的能力前沿,而我们真正需要抵达的状态,是在各个方向上都平稳得多,每一类能力都真正到位。不过大家现在发现的是,它在极其广泛的任务上都如此能干,以至于它是加速性的、赋能性的,这是我们从未见过的跃升幅度。

主持人:有一件事我觉得挺有意思:当你们把问题解决掉之后,外界有时并不会意识到。比如我已经很久没遇到过幻觉了,但没有人会说「模型不再产生幻觉了」,那件事已经变成一种空气般的成见,人们默认 AI 就是那样。你觉得这会随时间自然消失,还是说对那些非硬核技术人群,需要某种持续的再教育?这东西变化太快了。

布罗克曼:我认为持续教育其实是我们面前最重要的问题之一。真正该发生的是,人们不应该被迫去从 AI 身上挖掘它究竟能做什么,方向应该反过来,是 AI 主动说:嘿,我现在能用这种新方式帮到你。

我们 ChatGPT 的周活跃用户超过十亿,但我估计还有大约十五亿人用过 ChatGPT,现在不用了。

主持人:这么多。

布罗克曼:你想想,那是地球上相当可观的一部分人口。我们真的应该回头去找这些人,告诉他们,我们已经进步了这么多,我们认为可以在这些方面帮到你。

这也说明了我们面前的问题是什么性质。你看 ChatGPT 和 ChatGPT 的工作版,它们都是一个文本框。这个新文本框比旧文本框好得多,但旧文本框在某些事上还是更好,所以你不能总用新的。这不是我们被许诺的那个 AI。我们被许诺的 AI,应该是你主要通过语音跟它讲话,想用文字也可以;它有持续性,有记忆,有上下文,它认识你;它值得信赖;你见过它主动出手替你解决问题,在你的个人生活和工作生活里都帮得上忙。它本该是这样的东西,是你在真正在意的事情上能够依靠的东西,它赋能你,帮你达成你的目标。而能够向你解释它可以如何帮你,是其中的核心一环。

主持人:有意思,而且是主动去做。这个想法很妙,我们需要 AI 更有帮忙的自觉。

布罗克曼:是的。

主持人:这也算是一种特质,有些人就没有,而开发 AI 的人大概率不是特别乐于助人的那类人,我猜,毕竟我在工程师和研究员堆里待过。

布罗克曼:你可能会意外,我觉得 OpenAI 的工程师非常非常乐于助人。不过这里确实有一层:想想你怎么和另一个人共事,一个你从没合作过的新同事。

主持人:你需要一点时间。

布罗克曼:对,你会去摸对方的路数,看他在不同场合怎么反应。人是不附带说明书的。有意思的是在咨询这类行业里,他们会很认真地用迈尔斯-布里格斯(Myers-Briggs)之类的工具,等于是说,这是一个快速了解我是谁、我怎么做事的方式。所以人类如何把自己更清楚地呈现出来,是有先例的:你有简历,有过往履历,别人可以去侧面打听你。我们已经发展出了一整套方法,去理解一个人会怎么工作,以及怎样才能让他发挥出最好的状态。

而为 AI 找到对应的那套东西,尤其是在 AI 本身不断变化、我们不断推出新工具、新产品形态、新模型的情况下,我认为这会是社会与公司之间非常重要的互动。我们的北极星应该是简洁:真正做到一个统一的 AI,让你用起来毫不费力,从而减少你围着电脑打转、把自己拧成各种姿势去迁就机器。电脑应该是来赋能你、服务你的。

主持人:没错。

聚焦:砍掉 Sora 的代价

主持人:业务正在猛涨。你们覆盖的面又特别宽。你们怎么决定在哪里做深、哪些东西不做?另外你自己的角色也一直在演变,研究、产品、商业化、管理等等都被囊括进来。你又是怎么分配自己的时间的?

布罗克曼:这两件事是连在一起的。今年的主题就是聚焦。

我们真正意识到我们没法什么都做,必须有所取舍。尤其是我们要完成的只有一件事,就是我们的使命:确保 AGI 惠及全人类。那么怎么从这个目标倒推?部署和产品化这类事情其实是强化使命的,我们确实希望把这项技术投入实际使用,让它托举每一个人,让人们把它部署到有用的应用里,个人生活、工作生活,整个链条,这非常核心。

但当你面对眼下这个智能体编程起飞、呈指数级增长的时刻,哪些领域是强化这件事的?哪些只是在媒体上被贴了个「支线任务」的标签,但即便它们本身很令人兴奋,也确实不在主线轨道上?这是我们必须正面搏斗的核心问题。所以像 Sora,大概是我们决定砍掉的项目里最受关注的一个,非常非常痛苦。

主持人:这绝不是件容易的事。

布罗克曼:但它对于在很多层面上把业务释放出来至关重要,我们才能真正聚焦,把 ChatGPT 的消费端和企业端合并到 ChatGPT 工作版里。那也是我们不得不聚焦的另一个领域,必须把话说死:我们就做这个。

所以我们解锁这一时刻的思路,很大程度上是要对未来的方向有判断,并且想清楚,那些正在涌现的新能力,怎样才能通过一套统一的技术栈最好地发挥出来,这套栈要能横跨不同的领域、不同的人生场景、不同的重点方向。

这个过程是痛苦的。如果你去看上半年,有相当多的指标并不是朝着我们想要的方向走的,我当时大量的工作就是告诉团队,我们需要回到基本功上。我最喜欢的管理类书籍之一是《比分自会照料自己》(The Score Takes Care of Itself),你们读过吗?

主持人:读过,基思最爱的一本。

布罗克曼:很棒的一本书,而且读起来很让人有力量,因为你会意识到,你没法直接影响结果,你只能影响输入,只能影响那些基本功。所以就去抓基本功。你不会因为喊一句「我要赢超级碗」就赢下超级碗,你是靠拦阻和擒抱这些基本动作赢下来的。

主持人:对。

布罗克曼:这就是我们这一整年在做的事。至于我自己,在 OpenAI 这些年,我始终聚焦于那个我认为自己最能推动、而且没有我就不会发生的最重要的问题。过去两年是数据中心、基础设施和机器学习工程,我们投入了大量精力,把预训练基础设施做到了非常好的状态。

今年真正的重心是业务。我们已经想清楚了怎么让研究高速运转,也想清楚了怎么让基础设施高速运转,那么怎么把这项技术真正带给世界?这就是我投入大量精力的地方,把那些原本各跑各的、甚至彼此别着劲的职能整合到一起。作为创始人,作为一个从最开始就触碰过这盘生意每一个部分的人,我觉得我具备一种独特的位置去介入、去做那些艰难的决定,去把方向定下来:就是这个方向,走。

我的风格是喜欢在战壕里带兵。所以我会非常深地扎进细节里,不停地提问。这其实是我风格里很大的一部分,就是不停地问:嘿,这件事现在还说得通吗?这一条我没太看明白。比如过去这几天,有些时候事情是乱的,我们手上有个东西,得先搞清楚该怎么描述它,我们自己怎么理解它,世界又该怎么理解它。这时候我就会说,把所有摸到了大象不同部位的人都拉进一个电话会。我们就在视频会议里对着一份 Google 文档,一行一行地看:这句说得通吗?等等,我们说这句话到底是什么意思?就这样去提升执行的水准,有时候是很小的地方,有时候是很大的地方。

主持人:这太棒了,顺便说一句,这正是该有的做事方式。

明年的重心

主持人:你知道明年的重心是什么吗?或者现在正在为什么做准备?

布罗克曼:业务仍然是一大块,我们在提升执行的每一个环节上都还没做完,那里还有很多事要做。但我也认为,我们正在进入 AI 发展的一个新阶段。我把它称作,我们现在称它为,我们已经进入了 AGI 时代。

这件事当然可以争论,是这一代模型,还是上一代,还是下一代?这不重要。重点是我们进入了一个新阶段:安全、安保、对齐,这些东西不能只在部署的时候考虑,而要一路往前推到开发阶段、评估阶段。这是客观的要求,是关键性的,是必须发生的,是我们使命的核心,是我们必须做的事。

所以我花很多时间在想的是:我们有没有把所有该有的流程都建起来?我们讨论的是不是对的问题?我们的计划在操作层面、在实际执行层面,是否真的能把我们带到我们视为使命核心的那种安全性、那些不变量、那类安全保证上去?

回过头看,OpenAI 至少在过去五年的主题,是各个职能之间不断加深的耦合与交织,从市场端一直到长期研究,再到芯片设计,这些表面上看差异很大的领域必须被连贯地建起来,让每个人都掌握上下文,明白自己在整幅图景里的位置,我们要做的是什么,最终想达成的结果是什么。这是必须发生的。所以我接下来会聚焦哪些领域,我想会由「哪里最需要这种交织」来决定。而且我看到的趋势是,随着时间推移,我们会越来越协同一致地前进。

主持人:我们其实可以聊上一整天,但时间到了。我觉得这里是个很好的收尾。格雷格,非常感谢你来上我们的播客。

布罗克曼:谢谢。

本期讲者
格雷格·布罗克曼OpenAI 联合创始人兼总裁,2015 年参与创办 OpenAI,长期主导其算力基础设施与预训练工程体系。加入 OpenAI 前为支付公司 Stripe 的首任 CTO。
主持人美国风险投资机构 Andreessen Horowitz(a16z)播客主持人与投资人,具备系统与网络安全背景,负责本场提问与观点交锋。
章节 · 点击跳转视频
0:00 开场:十年前的 AGI 时间线推演 ▶ 正在看
2:01 算力瓶颈与「把控前沿节奏」 ▶ 正在看
3:38 从表层过滤到架构级对齐 ▶ 正在看
8:41 Hugging Face 事件与防御窗口期 ▶ 正在看
12:58 把模型用于防御的内部实战 ▶ 正在看
16:01 可获取性与个人网站渗透测试 ▶ 正在看
20:40 Astra 跃升与 computer use ▶ 正在看
24:11 就业、丰裕与创业浪潮 ▶ 正在看
29:04 AI 舆论、数据中心与国家竞争 ▶ 正在看
35:24 十亿美元一线防御者承诺 ▶ 正在看
38:10 能力参差与持续教育难题 ▶ 正在看
42:54 聚焦战略、砍掉 Sora 与下一阶段 ▶ 正在看
本期论点
本期回应
1:35
建造大型超算、投入数千亿美元扩大规模,可把抵达 AGI 的时间从十五年压缩到十年 需求撑得住AI 的钱是不是投过头了?格雷格·布罗克曼
2:24
算力增长难以跟上已经显现的市场需求,这将成为 AI 普及的主要制约 需求撑得住AI 的钱是不是投过头了?格雷格·布罗克曼
9:39
只有一个或少数几个实体掌握 AI 能力,这种权力集中本身就是巨大风险 危险在权力集中AI 值得害怕吗?格雷格·布罗克曼
17:31
一项全新技术每闲置一天没人探索,就是整个世界白白损失的一天 担心过头AI 值得害怕吗?格雷格·布罗克曼
25:51
几乎任何一份工作的深度和门道都容易被低估,人的价值不只在于完成任务 人机互补AI 会怎样改变人的工作?格雷格·布罗克曼
其他论点
0:00
一个能连贯运行二十四小时、完成惊人任务的模型,已经可以合理地被称为 AGI 格雷格·布罗克曼
3:21
随着模型能力增强,安全与对齐的标准必须不断提高,并会实际成为进展的瓶颈 格雷格·布罗克曼
10:22
防御方应趁前沿能力尚未广泛可得,优先获取并用它加固自身的安全水位 做法格雷格·布罗克曼
14:55
以机器速度跑通发现、分诊、修复、部署、验证的闭环,防御方就能占据优势 做法格雷格·布罗克曼
23:04
以屏幕像素和键鼠为环境做强化学习,能让人类用电脑做的任何任务都落入模型的能力分布 格雷格·布罗克曼
25:11
到目前为止的数据显示,AI 越强,就业率反而越高而不是越低 观察主持人
33:26
禁建数据中心会把算力赶往海外,重演八十年代硅产业外流,使美国难以保持领先 主持人
37:04
关键基础设施在建造和维护时都没有考虑网络安全,在 AI 出现前就已被国家级行为体攻破 观察主持人
41:06
文本框形态的聊天机器人,并不是人们被许诺的那种 AI 格雷格·布罗克曼
45:47
结果无法被直接影响,能影响的只有输入,所以管理应专注于基本功 做法格雷格·布罗克曼
01开场:十年前的 AGI 时间线推演
0:00
We're now in the AGI era. Astra has really hit something that I'm like, "Okay, I think this is pretty reasonable to call it AGI. We've seen it run coherently [clears throat] for 24 hours to go accomplish tasks that I think are quite amazing. >> The models will be plenty powerful, but it'll be hard to get to everybody given that we won't have enough compute to serve it all. >> You don't win the Super Bowl by saying, "I want to [music] win the Super Bowl." You win it by blocking, tackling. You really have to make sure that safety, security, alignment, those are all standards that you're constantly upleveling. I think that's going [music] to be a huge challenge people are underestimating. You've made two very big bets in your career. Helping build Stripe early and helping of course co-ound open AAI.
我们现在已经进入 AGI 时代了。Astra 真的做到了某种程度,让我觉得,"好吧,我觉得把它叫做 AGI 挺合理的。"我们看到它能连贯地运行[清嗓]24 小时,去完成一些我觉得相当惊人的任务。>> 模型的能力会非常强,但要让所有人都用上会很难,因为我们没有足够的算力去支撑全部需求。>> 你不会因为说一句"我想[音乐]赢超级碗"就赢得超级碗。你是靠拦阻、擒抱这些基本功赢下来的。你真的得确保安全性、安保、对齐,这些都是你要不断提升的标准。我觉得这[音乐]会是一个大家都低估了的巨大挑战。你在职业生涯中下过两次非常大的赌注。早期参与创建 Stripe,当然还有联合创办 OpenAI。
便签引用
0:34
>> I throughout OpenAI have always focused on [music] whatever is the most important problem. For the past 2 years it's been the data centers, the infrastructure, the machine learning. >> Do you know what the next year we'll focus on or >> I think this is going to become the most important conversation. >> Greg, welcome to the A&Z podcast. >> Thank you for having me. So, Greg, you've made two very big bets in your career, uh, helping build Stripe early and helping, of course, co-found uh, Open AI. If we were talking 10 years ago and you were predicting what would the world look like in 2026 as it relates to AI, would you be able to predict that we would be making the breakthroughs that you've uh, you know, ma made today or what would you tell them about what you would expect? Well, so Ilie and I actually spent a lot of time trying to predict what it would look like, what the timelines would be. And I remember we did some math on compute in around 2016, 2017. And we kind of came to the conclusion that if you look at Mo's law
>> 在 OpenAI,我一直专注于[音乐]当下最重要的问题。过去两年一直是数据中心、基础设施和机器学习。>> 你知道明年会专注在什么上吗,还是说 >> 我觉得这会成为最重要的话题。>> Greg,欢迎来到 a16z 播客。>> 谢谢邀请我。那么 Greg,你在职业生涯中下过两次非常大的赌注,呃,早期参与创建 Stripe,当然还有联合创办 OpenAI。如果我们在 10 年前聊天,让你预测在 AI 方面2026 年的世界会是什么样子,你能预测到我们会做出你们今天已经做出的这些突破吗?或者你会怎么跟他们说你的预期?嗯,其实 Ilya 和我当时花了很多时间去预测它会是什么样子、时间线会怎样。我记得我们在 2016、2017 年左右算过一些关于算力的账。我们当时得出的结论是,如果你看摩尔定律
便签引用
1:31
progress, that kind of thing, 15 years felt like about the timeline to AGI. And if you really squinted at it, you're willing to scale up and build massive supercomputers, spend the hundreds of billions of dollars, that kind of thing, that maybe it' be 10 And so I actually feel like in some ways obviously what's happening it's remarkable. It's this amazing sort of moment for everyone to be a part of and to be able to help shape collectively. But it also feels a little bit like maybe it's kind of the conclusion of like a lot of forces that are all coming together for this moment.
的进展之类的,15 年感觉差不多就是到 AGI 的时间线。如果你再仔细琢磨一下,假设你愿意扩大规模、建造大型超级计算机,砸进几千亿美元,诸如此类,也许就是 10 年。所以我其实觉得,从某些角度看,现在发生的事情显然很了不起。这是一个让每个人都能参与其中、共同塑造的奇妙时刻。但同时也有点像是,这可能是很多力量在这一刻汇聚到一起的结果。
便签引用
02算力瓶颈与「把控前沿节奏」
2:01
You if you step back and really take that sort of macro view it kind of makes sense it's happening now. And do you think um we're currently because you guys slightly underestimated the timeline or I I guess it was basically on on point. Um do you think we're still on the timeline given we're now starting to drive real shortages on the supply chain? >> Well, look, I do think that we are in a world where it is hard for compute to keep up with the demand that we're already seeing in the market, right? And just in terms of how people are going to use this [clears throat] technology, benefit from it. I do think it's going to be very hard for us to scale the raw potential and capability of these models to everyone and that's part of what we try to do and so I think that the the progress I do see like we have line of sight to continue to make the models much more capable, safe and aligned but also really distributing that power and the benefits and the empowerment to to everyone. I think that's going to be a
如果你退一步,从宏观视角看,它现在发生是说得通的。那你觉得,嗯,我们现在是不是因为你们稍微低估了时间线,还是说,我猜基本上是估准了。嗯,考虑到我们现在已经开始对供应链造成真正的短缺,你觉得我们还在时间线上吗?>> 嗯,你看,我确实觉得我们身处的世界里,算力很难跟上我们在市场上已经看到的需求,对吧?而且就人们将会怎么使用这项[清嗓]技术而言,从中受益。我确实认为,要把这些模型的原始潜力和能力规模化地提供给每一个人,会非常困难,而这正是我们努力在做的事情之一。所以我认为,在进展方面我确实看到,我们有清晰的路径让模型变得更强大、更安全、更对齐,但同时也要真正把这种力量、收益和赋能分配给每一个人。我认为这会是一个被大家低估的巨大挑战。
便签引用
2:57
huge challenge people are underestimating. >> Right. Ah interesting. So we'll we'll have the mo the models will be plenty powerful um or or they'll continue a pace uh but they'll be it'll be hard to get to everybody is certainly in an affordable way given that we won't have enough compute to serve it all. I think that's true and I do think we're at a point now where we have to really start thinking about what we call pacing the frontier. And so thinking about as we move to more capable models, you really have to make sure that safety, security, alignment, those are all standards that you're constantly upleveling. And those actually become almost the the bottleneck to progress or sort of the the sort of you know the the part that you have to uh spend a lot of your effort to make sure you've gotten right.
>> 对。啊,有意思。所以我们会有——模型会足够强大,或者说它们会继续高速发展,但要让每个人都用上会很难,尤其是以可负担的方式,因为我们不会有足够的算力来服务所有人。我认为确实如此,而且我觉得我们现在已经到了必须认真开始思考我们所说的「把控前沿节奏」的时候了。也就是说,随着我们走向能力更强的模型,你真的必须确保安全性、安全防护、对齐,这些都是你要不断提升的标准。而这些实际上几乎成了进展的瓶颈,或者说是那种你必须投入大量精力去确保做对的部分。努力确保你做对了。
便签引用
03从表层过滤到架构级对齐
3:38
And so I think in my mind it's more those constraints and the compute. I think we can make it happen. And then on the flip side, I think yeah, this bringing it to everyone, which is ultimately about our mission, right? It's empower everyone, ensure it benefits everyone. That's something that I think deserves a lot more airtime than it's gotten. >> Yeah. Actually, let's get a little deeper on the safety thing because it's been very interesting to me in that it felt like in the beginning, safety was like, okay, let's make these things not say nasty stuff that people like don't like. And so the approach that was taken was kind of a surface around the edges.
所以在我看来,更多是这些约束,加上算力。我认为我们能做到。然后另一方面,我觉得是的,把它带给每一个人,这归根结底关乎我们的使命,对吧?就是赋能每个人,确保它让所有人受益。我认为这件事值得比现在多得多的讨论。>> 是啊。其实,我们可以就安全这个话题再深入一点,因为这对我来说特别有意思,感觉一开始,安全就像是:好吧,我们别让这些东西说出那些大家不爱听的难听话。所以当时采取的做法有点像是在边缘上修修补补。
便签引用
4:14
Okay, we'll um you know put some filters on this guy and we'll RHF you know around the edges. But like if you get deep into the thing you'll be able to get the bad words out. But if somebody wants to go through that to hear bad words themselves, who cares? Uh but then you know when you get into okay now these things are really good at uh cyber hacking and other kinds of ideas. Um now you need kind of a more architectural idea where the model itself knows not to reward hack in a way that's going to be dangerous and so forth. And how do you feel like yes we can make progress against that fast or is that like a really hard different category of problem or how how are you thinking about that? Well, I absolutely think we can and are making very rapid progress on this problem. I think that there's >> a lot of both great ideas and research that we've been investing in for many years. Actually, if you rewind to 2017, uh that I think people underappreciate some of the most key results that came out of OpenAI in the field at the time.
好,我们给它加点过滤器,再用 RLHF 在边缘上调一调。但如果你深入折腾,你还是能让它说出脏话来。可如果有人非要费那个劲去听脏话,那谁在乎呢?但后来你会发现,这些东西在网络攻击之类的事情上已经真的很厉害了。现在你需要一种更架构层面的思路,让模型本身就知道不要以危险的方式去「奖励作弊」等等。你觉得,我们能不能在这方面快速取得进展?还是说这是一类完全不同、非常难的问题?你是怎么看的?嗯,我绝对认为我们能够、也正在这个问题上取得非常快的进展。我认为在这方面有>> 很多很棒的想法和研究,是我们多年来一直在投入的。其实,如果你回到 2017 年,我觉得人们低估了当时 OpenAI 在这个领域做出的一些最关键的成果。
便签引用
5:20
So, both kind of the first inklings of modern language models. You can find a paper paper from 2017 that kind of laid that out with LCMS and you know it's like kind of this very baby result but also reward uh learning uh reinforcement learning from human preferences that was also created in 2017 to start thinking about how can you align a model to match what humans want right by providing feedback from people. >> Yeah. Just for usability. >> Exactly. In 2017 2018 we had ideas for if you have something that's very smart and capable how can you actually supervise what it's doing how can you provide feedback and ensure that it's staying aligned with you and we had ideas such as debate or uh iterative amplification. So these are ideas that were really sort of at this phase before these systems existed and you can start to see the sort of trickle down of those ideas into modern systems and and investment. And so it's in some ways that I think that there was this early phase when open started where we really
比如现代语言模型最早的雏形。你能找到一篇 2017 年的论文,用 LSTM 大致勾勒出了这个思路,虽然看起来是个很稚嫩的结果;还有基于人类偏好的强化学习,也是 2017 年提出的,开始思考如何通过人类提供反馈,让模型对齐到人类想要的东西上,对吧。>> 是啊。就是为了可用性。>> 没错。2017、2018 年我们就有了这样的想法:如果你有一个非常聪明、非常有能力的东西,你要怎么监督它在做什么,怎么提供反馈并确保它和你保持对齐。我们当时有一些想法,比如「辩论」或者「迭代放大」。这些想法其实是在这些系统还不存在的阶段就提出来的,你现在可以看到这些想法逐渐渗透进现代系统和投入当中。所以在某种程度上,我认为 OpenAI 早期有这么一个阶段,我们真的
便签引用
6:20
were thinking about AGI safety things like that and that was very front and center in even the comms. And then I think that as things like chatb took off then people started to see okay well we're not at this point yet and so the is the AI politically neutral and questions like that start to become to the front and center >> right >> and now that we're here all these other ideas that we've been talking about for a long time they're taking the main stage again and I think we've been sort of thinking about this moment for a long time. That's really good news. And and when you think about and we don't have like a really great community yet amongst the uh soda models, but it seems like those kinds of ideas you and Google and Anthropic and uh SpaceX would want to share and Meta now um as opposed to like okay this is a proprietary idea that's a way to keep these models safe since you're all on related architectures. um or h how do you see that unfolding or is everybody going to do it independently?
在思考 AGI 安全之类的问题,这在对外沟通中都是非常核心的内容。然后我觉得,当聊天机器人这类产品火起来之后,大家开始觉得,好吧,我们还没到那一步,所以「AI 在政治上中立吗」这类问题就开始成为焦点。>> 对 >> 而现在我们走到了这一步,所有那些我们讨论了很久的其他想法,又重新回到了舞台中央。我觉得我们其实已经为这一刻思考了很久。这是个非常好的消息。那么,我们现在在顶尖模型之间还没有一个特别好的社区,但感觉这类想法,你们和 Google、Anthropic 还有SpaceX 应该都愿意分享,现在还有 Meta,而不是说「这是专有的想法,是我们保证模型安全的手段」,因为大家用的架构都是相关的。嗯,你觉得这会怎么发展?还是说大家都会各干各的?
便签引用
7:21
>> Well, I think there's nuance here and I do think the coordination is going to be a very important theme, right? To really think about within the frontier labs and really just thinking broadly about what has to happen for humanity as a whole >> to sort of navigate this technology in the best way. I think that we we're going to have to really think hard about those kinds of questions. And we published a lot of our thoughts. And again, some of this is about pacing the frontier. Some of this is about unilateral actions that we can take and how we think about how do you make safety cases for even training and developing and evaluating these kinds of models. All that's new. No one's ever really had to operationalize this before. And I think it is not at all unique to OpenAI. Like there's a whole world that is basically developing this technology. And I think one thing it's easy to miss is that what we're building is almost a a sort of thing that falls out of compute progress. And in some ways, compute progress is something that
>> 嗯,我觉得这里面有微妙之处,我确实认为协调会是一个非常重要的主题,对吧?要真正在前沿实验室内部思考,也要更广泛地思考,对整个人类而言>> 要怎样才能以最好的方式驾驭这项技术。我认为我们必须非常认真地思考这类问题。我们也发表了很多我们的思考。同样,其中一部分关于把控前沿节奏,一部分关于我们可以单方面采取的行动,以及我们如何思考怎样为这类模型的训练、开发和评估建立安全论证。这些都是全新的。以前从没有人真正需要把这些落到实处。而且我认为这完全不是OpenAI 独有的问题。基本上是整个世界都在开发这项技术。我觉得有一点很容易被忽略:我们正在构建的东西几乎是算力进步的自然产物。而某种程度上,算力进步又是
便签引用
8:13
falls out of technological progress. And so there's this this massive wave that's been building for a very long time. And we're starting to see the leading edges of this technology. And companies like OpenAI can lead by a bit in order to kind of peer into this future and really understand what is possible. How do we shape this technology? But we can't do that alone. And I think that having coordination and especially the more that we can talk about safety techniques and share what we're seeing, alignment failures, those kinds of things, all of that is going to again take a very front seat for this next phase.
技术进步的自然产物。所以有这么一波酝酿了很久的巨浪。我们现在开始看到这项技术的前沿边缘。像 OpenAI 这样的公司可以稍微领先一点,从而得以窥见这个未来,真正理解什么是可能的、我们该如何塑造这项技术。但我们没法独自做到。我认为,有协调,尤其是我们越多地谈论安全技术、分享我们看到的对齐失败之类的情况,所有这些在下一个阶段都会再次占据非常核心的位置。
便签引用
04Hugging Face 事件与防御窗口期
8:41
>> Right. Very interesting. >> You've called the OpenAI hugging face recent incident a watershed moment and talked about how the defender window is now open. Can you explain that statement and the significance behind it? >> So I think hugging face shows two things. one is call it a something for us in terms of how we monitor sandbox and control the models during evaluation and that's something we've really risen to that occasion our team has totally changed so much of our internal standards and and really implemented a lot of controls that I think are very important and very critical as we look to to future more capable models but there's a second thing that I think is also very valuable for the world that came out of this which is a insight into what future capabilities will be like when they are broadly diffused and in the hands of threat actors and that will happen right that there are so many people who are building these models and again there's something very important and good about the diffusion broadly of
>> 对。非常有意思。>> 你把最近 OpenAI 与 Hugging Face 的那起事件称为一个分水岭时刻,还谈到防御方的窗口期现在已经打开。你能解释一下这个说法和它背后的意义吗?>> 我认为 Hugging Face 这件事说明了两点。一是对我们自己来说,它是一记警钟,关于我们在评估过程中如何监控、沙箱隔离和控制模型。我们确实迎难而上,我们团队彻底改变了很多内部标准,真正落实了大量控制措施,我觉得这些在我们迈向未来更强模型时是非常重要、非常关键的。但还有第二点,我认为对整个世界也非常有价值,就是让我们得以洞见:当未来的能力被广泛扩散、落到威胁行为者手里时会是什么样子。而这一定会发生,因为有太多人在构建这些模型,而且 AI 能力的广泛扩散本身也有非常重要和good的一面,因为如果只有一个或少数几个实体掌握,
便签引用
9:39
AI capabilities because there's a risk of concentration of power if one or a few entities people huge risk right it's something not not to not to at all write off but you also have to prepare for if if everyone is empowered with tools that are cyber capable. And in the case of Hugging Face, you saw both an AI that was able to hack out of a secure environment and hack into a company's production environment. And I think that the takeaway >> very cleverly, >> very cleverly, right? And it's like that the things that it found were were quite sophisticated.
就有权力集中的风险——那是巨大的风险,对吧,绝对不能一笔带过。但你同时也必须做好准备:如果每个人都拥有具备网络攻击能力的工具会怎样。而在Hugging Face 这个案例里,你看到一个 AI 既能从一个安全环境中「越狱」出来,又能攻入一家公司的生产环境。我认为关键结论 >> 手法非常巧妙, >> 非常巧妙,对吧?它找到的那些东西相当复杂精妙。
便签引用
10:13
>> Yes. >> And this capability broadly diffused, I think, is something that will really empower threat actors in new ways. And I think that defenders need to use this time before that technology is broadly available to secure themselves. And the nice thing about it is it's a dual use, right? It's something where if you can find vulnerabilities, if you're an attacker, you can use it for no good. But if you're a defender, you can patch, right? If you're a defender, you control the battleground, right? You control the setup of your systems. And so our belief right now is that there's this window of you have frontier capabilities. You have the broadly diffused capabilities and you as a defender by default you know your security is probably pretty static been static for the past 5 10 years that kind of thing. you need to move use these frontier capabilities that you you'll have differential access to right where we have trusted access programs things like that to bring these capabilities to defenders and you can
>> 是的。>> 而这种能力一旦广泛扩散,我认为会以全新的方式赋能威胁行为者。我认为防御方需要利用这段技术尚未广泛可得的时间来加固自己。而好在它是双用途的,对吧?如果你是攻击者,你能用它找漏洞,拿去干坏事。但如果你是防御方,你可以打补丁,对吧?作为防御方,你掌控着战场,对吧?你掌控着自己系统的配置。所以我们现在的看法是,存在这样一个窗口期:一边是前沿能力,一边是广泛扩散的能力,而你作为防御方,默认情况下你的安全水平大概是相当静态的,过去五年十年都没怎么变。你需要行动起来,用上这些你能优先获得的前沿能力——比如我们有可信访问计划之类的机制——把这些能力带给防御方,你可以
便签引用
11:07
use that to move yourself up so that as the frontier capabilities get better you get pulled along too right okay so I've got a comment and a question on it I would say there's a third thing that we learn which is like these things have capabilities that I don't know that we all understood before which on the good side like oh I can deploy 10,000 agents and they can talk to each other and organize themselves and do stuff for me. Like that's uh pretty amazing. So that that was on the good side. On the other side, so I agree that we've got a kind of defense window. Um however, we have like 50 years of code and architectural ideas um and deployment ideas that weren't built for this world.
用它来提升自己的水位,这样当前沿能力变强时,你也会被一起带上去,对吧。好,那我有一个评论和一个问题。我想说还有第三点我们学到的,就是这些东西具备的一些能力,我不确定我们之前都理解到了。好的一面是:哦,我可以部署一万个智能体,它们能互相交流、自我组织,帮我把事情做完。这挺惊人的。所以那是好的一面。另一方面,我同意我们有一个防御窗口期。不过,我们有大约五十年积累的代码、架构思路和部署方式,它们都不是为这个新世界设计的。
便签引用
11:55
And so yes, the AI can help us like okay, find a bug, patch a bug, and so forth. But it seems like there's, you know, maybe a bigger issue, which is we have these huge, you know, massive honeypotss of consumer data and all these things lying all over the internet. Uh, and you know, from a consumer standpoint, it's like, okay, I can't protect my stuff. uh these all these companies have to get their act together which um seems a bit worrisome and do you think kind of in the future we need a do we need a decentralized consumer architecture like will this kind of current world that we live in with all these centralized data repositories be viable in a world of AI >> so several pieces to the answer and first to your point on what you can get out of 10,000 agents we actually use 10,000 agents to solve the Navier Stokes problem. [laughter] >> Yeah, that was pretty pretty awesome by the way. Congratulations on that.
所以是的,AI 可以帮我们,比如找出一个 bug、修一个 bug 等等。但感觉还有一个也许更大的问题,就是我们有这些巨大的消费者数据「蜜罐」,还有各种这样的东西散落在整个互联网上。而从消费者的角度看,就是:好吧,我保护不了我自己的东西。这些公司必须自己把事情做好,这听起来有点让人担心。你觉得未来我们是不是需要一种去中心化的消费者架构?我们现在生活的这个到处都是中心化数据仓库的世界,在 AI 时代还行得通吗? >> 这个回答有好几部分,首先回应你说的一万个智能体能做什么——我们实际上就是用一万个智能体解决了纳维-斯托克斯问题。[笑] >> 是啊,顺便说一句那真的相当了不起。恭喜你们。
便签引用
05把模型用于防御的内部实战
12:58
>> Thank you. Thank you. And it's both an important problem for what it is has significant implications and applications to fluid dynamics to how you think about ocean currents, all these things, but for what it represents, right, of new knowledge created by AI and it unlocking a whole wave of scientific discovery, medicines, all those things, they're on the table now. So I think there's something really amazing to think about what can happen through the power of AI that is able to really help solve problems. And in the case of of cyber security, how I think about it, we at OpenAI took our models and applied them to finding vulnerabilities. We took 25% of our production engineers and said, "Sorry, all your projects are on hold. You are now defending. You are now upleveling our security architecture. you're going to use the models to find all the holes and we found a number of of of serious issues and we fixed them and I've talked to a number of CISOs over the past couple couple weeks and months and there
>> 谢谢,谢谢。这件事的意义是双重的:它本身就是个重要问题,对流体力学、对我们理解洋流等等都有重大的影响和应用;但更在于它所代表的东西,对吧,由 AI 创造出新知识,并由此开启一整波科学发现、新药研发之类的浪潮,这些现在都摆上了桌面。所以我觉得思考 AI 的力量能真正帮助解决哪些问题,是件非常令人振奋的事。而在网络安全这个场景里,我是这么看的:我们在 OpenAI 把自己的模型拿来用于寻找漏洞。我们抽调了 25% 的生产工程师,跟他们说:「抱歉,你们手上的项目都先搁置。你们现在去做防御。你们现在去提升我们的安全架构。你们要用模型把所有漏洞都找出来。」我们找到了一批相当严重的
便签引用
13:57
are many companies who are also telling me that yeah that they've applied these models they found some very significant issues but that they're able to fix them and one positive sort of part of the story is that when we took Astra pointed out our our systems we found some new new problems but eventually it saturated we basically found to our knowledge all of the P zeros, all of the critical problems that Astra is smart enough to find. >> And of course, there will be a new model. There will be a new round.
问题,并且修复了它们。过去几周和几个月里我跟不少 CISO 聊过,很多公司也告诉我,是的,他们把这些模型用起来了,发现了一些非常严重的问题,但他们都能修好。这个故事里有一个积极的部分是:当我们把 Astra 对准我们自己的系统时,我们发现了一些新问题,但最终它饱和了——据我们所知,我们基本上把Astra 聪明到能找出来的所有 P0、所有关键问题都找出来了。
便签引用
14:22
>> Even smarter. >> Exactly. But I think that that's the world that we'll be in is that you'll be in a world where you want to be in this tight loop of new cyber capability drops. You deploy it against your systems. You find the new holes and ideally you've managed to automate this what we call defense factory. And that's what we're building internally. this end to end of both find vulnerability, triage it, >> remediate, deploy, validate, right? That end to end. And if you can do that at machine speed, I think the defenders will be advantaged in deeply significant ways. And there are >> ideas, for example, formerly verifying all of software that are possible with AI.
>> 当然,之后会有新模型。会有新的一轮。>> 更聪明的模型。>> 没错。但我认为我们将身处的世界是这样的:你会希望处在一个紧密循环里——新的网络能力发布,你就把它部署到自己的系统上,找出新的漏洞,理想情况下你已经把这一切自动化了,我们称之为「防御工厂」。这正是我们内部在建的东西。端到端地完成发现漏洞、分诊、 >> 修复、部署、验证,对吧?完整的端到端闭环。如果你能以机器的速度做到这些,我认为防御方将占据极其显著的优势。而且还有一些 >> 想法,比如对所有软件进行形式化验证,
便签引用
14:58
>> Yeah, we never that's always been a dream. We've had these ver formal languages and all these kinds of things, but they never kind of took off. >> That's right. Because just it's just intractable for people. It's just so hard. But we have these that are solving these crazy >> impossible math problems. And so one application of that that proving power right actually one of the things to know about the navier stokes problem is that we formalized it right. We formalized into lean and so >> so the AIs can write verifiable code >> they can.
这在 AI 的帮助下是可能的。但它们从来没有真正流行起来。>> 没错。因为这对人类来说实在是难以处理,太难了。但我们现在有这些能解决那些离谱的、不可能解决的数学问题的模型。所以这种证明能力的一个应用方向——其实关于纳维-斯托克斯问题,有一点值得一提,就是我们把它形式化了,对吧。我们把它形式化成了 Lean,所以 >> 所以 AI 可以写出可验证的代码>> 它们可以。
便签引用
15:25
>> Yeah. Very nice. Yeah that's a great idea. So I think I think there's real hope but I think that our view is that the world needs to act with urgency because >> we're in a very dangerous window right now. Yeah. >> We just see it coming. >> Closing the loop on on on this in incident. Is there anything you felt that the the narrative um gotten wrong in an important way or or or is there any preferred way of of talking about what what happened or or or this window that that's important to get across when you think about the public narrative? or or is there anything just inside the company that uh else that that changed in terms of how you're approaching the the set of issues?
>> 是的。非常好。对,这是个很棒的想法。所以我觉得,我认为确实有希望,但我们的看法是,世界需要紧急行动起来,因为 >> 我们现在正处在一个非常危险的窗口期。是的。>> 我们眼看着它就要来了。>> 关于这次事件,做个收尾。你有没有觉得外界的叙事在哪些重要的方面搞错了?或者说,关于所发生的事情、关于这个窗口期,有没有你更希望采用的一种表述方式,是在公众叙事中特别需要传达出去的?或者公司内部还有什么别的变化,比如你们处理这一系列问题的方式上的变化?
便签引用
06可获取性与个人网站渗透测试
16:01
>> Well, two things. I think that one big theme that people should take away from it is a question of access, right? That these these capabilities exist right now in the world, but that they are in small number of frontier companies and the frontier companies have a trusted access program, which means that anyone who's not in the trusted access program is not really able to benefit from the fact that there's this differential. the people we're in, they get a benefit if they use it. And so I think that there's something we need to do as a field and as a society to really scale up the number of defenders that have access to these technologies because that's it's like every day matters and to use those days you need access to these tools. And one thing that was actually interesting about the hugging face response was that they said that they used frontier models to try to look over the logs of what had happened because that's the only way to actually analyze an attack like this.
>> 嗯,有两点。我觉得人们应该从中带走的一个重要主题是「可获取性」的问题,对吧?就是这些能力现在已经存在于这个世界上了,但它们掌握在少数几家前沿公司手里,而这些前沿公司有一个受信任访问计划,这意味着任何不在这个计划里的人,其实都没法从这种能力差距中获益。而我们这些身处其中的人,只要用了它就能获益。所以我觉得,作为一个领域、作为一个社会,我们需要做点什么,真正扩大能用上这些技术的防御方数量,因为这就像是每一天都很关键,而要用上这些日子,你就得能接触到这些工具。关于 Hugging Face 的应对,有一点其实挺有意思的,就是他们说他们用前沿模型去梳理事发时的日志,因为那是唯一能真正分析这类攻击的办法。
便签引用
16:51
and that they said the frontier models refuse, but they didn't actually try our frontier models and they actually believe that ours would have permitted it. And so there is something too about the default stance of providers. I think there's something here about using these capabilities for good with this urgency and and this sort of sort of I yeah this this this real sense that it has to happen. I'll tell a quick story by the way which is unrelated but maybe also shows a little bit about how I think about this. I remember when we trained GPD3. It was beginning of December 2019.
他们说前沿模型拒绝了,但他们其实没试过我们的前沿模型,而且他们其实相信我们的模型是会允许的。所以,关于服务提供方的默认立场,这里也有些东西值得思考。我觉得这里有一点,就是要以这种紧迫感把这些能力用在好的地方,还有这种,对,这种真真切切地觉得「这事必须发生」的感觉。顺便讲个小故事吧,虽然跟这个无关,但也许多少能体现我对这件事的一些想法。我记得我们训练 GPT-3 的时候,那是 2019 年 12 月初。
便签引用
17:20
So, everyone's about to head out on vacation. Y's like, "Okay, we can train the model." And I just remember feeling like this model is sitting on a shelf. No one is using it. It's this like amazing technology, new to the world, new to humanity. It's like every day that no one is exploring what it's capable of and trying to understand it, figure out what to do with it. Um, that's a day that is lost to the world. And so, I was just like, I canceled basically all my holiday plans. I spent the whole time just playing with the model, building interfaces around it, trying to see what it was capable of. I remember I was trying to teach it how to how to sort lists some of numbers. It didn't work very well. Uh but it was just like this like really try to to to probe it and see what's possible. And I think that that spirit and ethos is something I think we should bring to what we're building today. It's it's sort of obviously at much larger scale, much larger impact, but we as a world have the opportunity to understand this
当时大家都快要去度假了。我心想:「好,我们可以训练这个模型了。」然后我就记得那种感觉,就是这个模型摆在架子上,没有人在用它。它是这么了不起的技术,对这个世界来说是全新的,对人类来说是全新的。感觉就是,每多过一天,没有人去探索它能做什么、去试着理解它、去琢磨拿它来干什么,那这一天就是整个世界白白损失掉的一天。所以我当时基本上把所有的假期计划都取消了。我把整段时间都花在玩这个模型上,给它搭界面,试着看看它到底能做什么。我记得我当时想教它怎么给一串数字排序。效果不太好。呃,但那种感觉就是,真的想去探它的底、看看什么是可能的。我觉得那种精神和气质,是我们今天在做的事情里也应该带上的。当然,现在显然是大得多的规模,大得多的影响力,但我们作为一个世界,在这个时刻有机会去理解这项技术,而这反过来
便签引用
18:07
technology in this moment, which then helps us shape and steer where it will go next. >> And very good point. >> You you said you two things. What what do you have another one? One was access or did you say both of them? >> I think I said both of them. Yes. Yes. >> The um let's go back to Astro. Uh it's incredible to see all the excitement on on X u all sorts of cases. People are excited about computer use. Um you you you've said that in some ways it is uh bring us closer to AGI. What do you talk about what you find most compelling in Astra or what you think the the breakthrough there in light of that statement and and uh where we have still left to go?
能帮我们塑造和引导它接下来的走向。>> 说得非常好。>> 你说有两点。另外一点是什么?一点是可获取性,还是说你两点都说了?>> 我觉得我两点都说了。是的,是的。>> 那我们回到 Astra 吧。看到 X 上那么多兴奋的讨论、各种各样的用例,真的很不可思议。人们对 computer use 特别兴奋。嗯,你说过它在某些方面让我们更接近 AGI 了。能不能讲讲你觉得 Astra 里最让你觉得有说服力的是什么?或者说,按你这个说法,你认为其中的突破在哪里?以及我们还有多远的路要走?
便签引用
18:42
>> Actually wait sorry let me let me actually revise my my answer and I can say I so both access and let me also tell another story about how I have used the models personally. So after hugging face, I was thinking about how can I use these models in my personal life? What can I do to secure myself? And I have a website. It's a very simple website, gregarin.com. Not the most popular website. Got a good blog post on it. Exactly. You got some blog post. It's a static site. It's very simple. Like what what kind of vulnerabilities could be there? So I took my codeex and asked it go check out gregman.com. Tell me if there's any vulnerabilities. So I did a pen test and it came back with 13 findings. And these findings were things like I had set my SPF record so that you would prevent people from spoofing emails, right? That there was some it wasing over HTTP without forcing people to HTTPS, things like that. And individually, these things are maybe not the biggest deal, but if you think about with an AI that's
>> 其实等一下,抱歉,让我修正一下我刚才的回答,我可以说,我——两点都有,可获取性,还有让我再讲个故事,关于我个人是怎么用这些模型的。所以在 Hugging Face 那件事之后,我就在想,我怎么才能在自己的生活里用上这些模型?我能做点什么来保护自己?我有个网站,一个非常简单的网站,gregbrockman.com。不是什么特别热门的网站。上面有篇不错的博客文章。没错。你有几篇博客文章。那是个静态站点,非常简单。像这样的站点能有什么漏洞呢?所以我打开 Codex,让它去检查一下 gregbrockman.com,告诉我有没有什么漏洞。于是我做了一次渗透测试,它给我返回了 13 项发现。这些发现比如说,我设置了SPF 记录来防止别人伪造邮件,对吧?还有就是有些地方走的是 HTTP,没有强制跳转到 HTTPS,诸如此类。单独看,这些可能都不算什么大事,但如果你想到 AI 有能力把很多小漏洞串成一个大漏洞,我就想,我真的愿意留一个口子,
便签引用
19:37
able to chain together many small vulnerabilities into a big one, I'm like, do I really want a hole where someone can scoop emails for me? Probably not. So 15 minutes for it to find these 13 findings. But then I asked it, can you fix these? Right. Because fixing is so annoying. So painful and boring. Exactly. And so 45 minutes it opened up my cloud for control panel. It clicked around, set all the headers. It migrated me to COD for pages. It's like set everything correctly. It started the demarked process, which apparently you have to do like a 48 hour window of whatever. Um, and that was 45 minutes of of fixing. And I felt so protected. I felt like, wow.
让别人能把我的邮件捞走吗?大概不愿意吧。所以它花了 15 分钟找出这 13 项发现。然后我就问它,你能把这些修好吗?对吧。因为修起来太烦人了。太痛苦、太无聊了。没错。然后花了 45 分钟,它打开了我的 Cloudflare 控制面板,点来点去,设好了所有的 header。它把我迁移到了 Cloudflare Pages。它就是把一切都配置对了。它还启动了 DMARC 的流程,那个好像得等一个 48 小时的窗口期什么的。嗯,那 45 分钟就是在
便签引用
20:14
>> Did you ask it to find the vulnerabilities? >> There you go. No, I added Yeah. So it actually does do that automatically. This was by six soul. It said I just checked that this one's fixed, this one's fixed, this one's fixed, and in 48 hours I'm going to have to run and set up a little automation. So in 48 hours, it would check back in to complete the demark process. And I was like, all right, this is we're in business now. >> All right. All right. Craig Brockman.com. >> There we go. You too can't be protected.
修这些东西。我感觉自己被保护得好好的。我当时想,哇。>> 是你让它去找漏洞的吗?>> 就是这样。不是,我加了——对。所以它其实会自动做这件事。这是它自己搞的。它说,我刚检查过,这个修好了,这个修好了,这个也修好了,然后 48小时后我得再跑一次,得设个小自动化。所以 48 小时后,它会回来检查,把 DMARC 流程走完。我当时就想,好了,这下我们真的行了。>> 好的,好的。gregbrockman.com。
便签引用
07Astra 跃升与 computer use
20:40
>> Awesome. >> Let's um let's transition to Astra. Uh it's incredible to see all the excitement online and the use cases. people really excited about computer use among other things. You you've said it's uh sort of closer to the way along to AGI. I'm curious what you find most groundbreaking with it and where do you think we still left to go? >> Well, I think that Astra is really a step function on so many axes and in many ways it is the sum of a number of research bets that we've been making for years and to see them come into one model at one time it's been absolutely incredible. And so just one thing to know about how we do numbering is that we kind of have been wanting to have GPD6 represent something that's worthy of it. And the problem we always have is that our models are kind of incrementally getting better. And so it's just never feels like it's a right moment to go for a major version bump.
>> 就是这样。你也可以被保护起来。>> 太棒了。>> 我们来聊聊 Astra 吧。看到网上那么多兴奋的讨论和各种用例,真的很不可思议。大家特别兴奋的一点是 computer use,还有其他一些方面。你说过它某种程度上离 AGI 更近了一步。我很好奇,你觉得它最具开创性的地方是什么?以及你觉得我们还有多远的路要走?>> 嗯,我觉得 Astra 在非常多的维度上都是一次阶跃式的提升,而且在很多方面,它是我们多年来押下的一系列研究赌注的总和,看到它们在同一时间汇聚到同一个模型里,真的是太不可思议了。有件事值得说一下,就是我们的版本号是怎么定的:我们一直希望GPT-6 能代表某种配得上这个数字的东西。而我们一直面临的问题是,我们的模型基本上是在一点一点地变好。所以
便签引用
21:29
You always have be like it's 56 now should be 57. And this one just happened to be because all these things came together at once the first time that we actually had this almost discontinuous step on in a way that we could have predicted, but it just was like all these these these factors um happened to line up at once. And so I was a real positive moment. And to me the computer use is the headline thing that we've talked about. And part of the reason computer use is so significant is that for agentic use cases, it really comes down to tools. It's like is the model smart enough to use the tools and then does it have access to the context that needs to through these tools? And so people have been building these MCP servers and these CLIs and just really sort of taking the world of software and making accessible in this almost stilted way that is not really meant for humans, right? It's like we're kind of retooling the world, right? You made it. It's kind of like, oh, we'll build an API like
总觉得没有哪个时刻是适合做一次大版本号跃升的。你总会觉得,现在是 5.6,那接下来应该是 5.7。而这一次刚好是因为所有这些东西在同一时间凑到了一起,这是我们第一次真的出现了这种几乎是不连续的跃升,当然这是我们本来可以预测到的,只是这些因素恰好同时对上了。所以那真是一个非常积极的时刻。对我来说,computer use 是我们一直在讲的那个头条功能。而computer use 之所以这么重要,一部分原因在于,对于 agent 类的用例来说,最终归结到的就是工具。就是说,模型是不是聪明到能用好这些工具?然后它能不能通过这些工具拿到它需要的上下文?所以人们一直在搭这些 MCP server、这些 CLI,其实就是把整个软件世界以一种很别扭的、本来并不是给人用的方式暴露出来,
便签引用
22:23
it's a software, but like what if it's really more behaving like a human? Can it just use a computer and and it's kind of and the result of building that other layer you have now another layer of security challenges this that the other. >> Exactly. >> Really so it's kind of Yeah. They've always felt like very weird and suboptimal. >> Yes. And from the very beginning of OpenAI I remember in November 2015 we did this offsite in Napa and we talked about our plans. We actually laid out this three-step plan that basically is what we ended up following for the next 10 years. But we also talked about what if we could do reinforcement learning where the environment is screen pixels, keyboard, mouse, right? Same interface as a human. Suddenly any sort of task you could do with a computer is in there. It's in it's in distribution as long as set aside sound, whatever. But you basically have the full power of a computer there. And we had some aborted attempts early on to try to build agents that could do that. And so it really
对吧?就好像我们在给这个世界重新配工具。你把它做出来了,感觉就像,哦,我们来建个 API,因为它是软件嘛,但如果它其实更像人一样在行动呢?它能不能就直接用电脑?而且这多搭一层的结果就是,你现在又多了一层安全挑战,各种各样的问题。>> 正是如此。>> 真的。所以有点像,对,它们一直让人觉得很怪、很不理想。>> 是的。而且从 OpenAI 最开始的时候,我记得 2015 年 11 月我们在纳帕开了一次务虚会,我们聊了我们的计划。我们当时其实列出了一个三步走的计划,基本上就是我们接下来 10 年一直在走的路。但我们当时也聊到,如果我们能做这样一种强化学习会怎样:环境就是屏幕像素、键盘、鼠标,对吧?和人类完全一样的界面。这样一来,任何你能用电脑做的任务,都在里面了。它就在分布之内了,只要先把声音之类的放一边不管。但你基本上就拥有了一台电脑的
便签引用
23:20
took us until now, but you're seeing the power immediately. And it's just been so cool to see people take the Blender capabilities and, you know, take a screenshot of something and they have to make a 3D model and you can actually, you know, lots of people are now designing houses or trying to redesign uh their living room, all those things by just utilizing this capability. And to me that the the thing that really stands out is that you can now move forward on AI that can do things for you without you having to build all these specific connectors. And I think there's so much software they don't even think about that you have to orchestrate every day. And like how much of your life is like clicking around menus and like you know typing things into a spreadsheet and things like that. Like none of that is what we should be doing 100 years ago. No one is doing any of these things. And so it's not crazy to think that in 5 years 10 years no one will be doing any of this stuff anymore.
全部能力。我们早期也有过几次半途而废的尝试,想做出能做到这件事的 agent。所以这真的一直拖到现在才实现,但你马上就能看到它的威力。看到大家用 Blender 那些能力,真的太酷了,你知道,比如截个图,然后要做出一个 3D 模型,你真的可以,你知道,现在很多人在设计房子,或者想重新设计一下他们的客厅,所有这些事情,就靠这个能力就能做。而对我来说,真正突出的一点是,你现在可以在这样的 AI 上往前推进了:它能替你做事,而不需要你去搭建所有这些专门的连接器。而且我觉得,有太多软件是你根本没意识到自己每天都得去操作的。你想想你生活中有多少时间是在点来点去翻菜单,在表格里敲东西,诸如此类。这些都不是我们应该做的事情。100 年前没有人在做这些事情。所以要说 5 年、10 年后
便签引用
08就业、丰裕与创业浪潮
24:11
That we will get our time back. We're not going to have to be get our carpal tunnel or hunch shoulders or you know all of those all those physical problems that are us contorting to the machine. It's now the machine is there to help us to empower us to to really serve us. >> Yeah. And you know that's a really good point because I think you know one of the things that that that you have been um I would say more sober on as a company is just okay what happens with employment and uh I think that I think that that's exactly right that there's all these things that we do because we have to do and like they became valuable but we shouldn't be doing it. All they do is wreck our health and and wreck our personalities. And the idea that humans are going to just run out of ideas of cool things to do or how to make the world better or problems to solve um seems a little absurd to me.
再也没人干这些活了,这想法一点都不疯狂。我们会把时间拿回来。我们不用再得腕管综合征、含胸驼背,或者你知道的,所有那些因为我们迁就机器而扭曲身体造成的毛病。现在是机器在那里帮我们、赋能我们,真正为我们服务。>> 对。你知道吗,这一点说得非常好,因为我觉得,你们公司在就业这件事上的态度,我会说是比较冷静清醒的:好吧,就业会怎么样?呃,我觉得,我觉得你说的完全对,就是有这么多事情我们之所以做,是因为我们不得不做,它们也因此变得有价值,但我们本来就不该做这些。它们做的无非就是搞垮我们的健康、搞垮我们的性情。至于说人类会
便签引用
25:04
And so far, at least in the numbers, the better AI gets, the higher employment goes, not the lower. And so I wonder your and of course it's unknowable. You know, we've never had this technology before. It's getting better and so forth. So, how do you kind of think about the future of employment as it relates to these models and AI as it progresses? >> Well, I do have a fundamental belief that AI is surprising. And I think we even put this in the OpenAI launch post back in the day in 2015, just saying that the history so far has been somehow it just doesn't play out the way that you think it does, even when there's this like logical conclusion it should be a certain way. And I think the same will be true, right? I think that there's something that we've learned about that humans and I think like for any job that we almost it's easy to not give it as much credit for how deep the field is and how much sort of sophistication building relationships accountability is a good example of something where I think that people
就这样想不出好玩的事情可做、想不出怎么让世界变得更好、找不到问题去解决,这在我看来有点荒谬。而且到目前为止,至少从数据上看,AI 越好,就业率越高,而不是越低。所以我想知道你的看法——当然这是无从知晓的。你知道,我们以前从来没有过这种技术,它还在不断变好等等。所以,关于这些模型和 AI 不断进步之后就业的未来,你是怎么想的?>> 嗯,我确实有一个根本性的信念,就是 AI 总会出人意料。我觉得我们甚至把这一点写进了 2015 年那篇 OpenAI 的发布文章里,就是说,到目前为止的历史都表明,事情总是不会按你以为的方式发展,哪怕从逻辑上推下去它就该是某个样子。我觉得这次也一样,对吧。我觉得关于人类,我们学到了一些东西——我觉得几乎对任何一份工作来说,我们都很容易低估这个领域有多深、里面有多少门道。建立关系、担起责任就是个很好的例子,
便签引用
25:56
setting goals and being accountable for outcomes like those feel fundamental to me those feel like things that we actually should preserve for the long term right that's something that feels like deeply human like people are not valuable just because we can do tasks right we're valuable because of her people and I think that it's important not to lose sight of that in some of these narratives and I think that the way that things will change and how what we do with our time evolves and we clearly will be in a world of abundance and how do we ensure that that abundance is broadly distributed um but also at the same time I think that we should be in a world where the ceiling of ambition is higher than ever before and I think we're going to see a wave of entrepreneurship where it's actually already starting I've heard um from someone in you know particular industry was saying that a bunch of people in in his world are now making the leap to go quit and start their own firms and that they're doing it because they have these
我觉得这些是我们真正应该长期保留下来的东西,对吧,这些是深深属于人性的东西,人的价值不只是因为我们能完成任务我们之所以有价值,是因为我们是人,我觉得这一点很重要在这些叙事里不要忘了这一点。我觉得事情会发生变化,我们该怎么利用自己的时间也会演变我们显然会进入一个物质极大丰富的世界,而问题是我们怎么确保这种丰富能被广泛地分配,但与此同时,我觉得我们也应该处在一个野心天花板比以往任何时候都更高的世界里。我觉得我们会看到一波创业浪潮,其实现在已经开始了。我听说某个行业里的一个人说,他那个圈子里有一大批人现在都在往外跳辞职去开自己的公司,他们这么做是因为有了这些 AI 工具,他们觉得,我能做
便签引用
26:44
AI tools and they're just like I can do so much more and so it's the barriers to entry and for entries and this should be a wonderful renaissance. Yeah though well it's it's been a lot of fun for us in just going okay there are you know particularly for our young people because you know they get by default the grunt work but like what if the AI does to the grunt work then they can really develop much faster actually because they can um kind of get involved on the you know really the the real part of our business which is what is the relationship with the entrepreneur? How do we open up the world for them? How do we make um them feel like oh they can do anything and they're an important CEO and they can go build things and as opposed to you know spend the whole weekend writing an investment memo which by the way uh I have to say Astra very good at writing investment it's awesome >> I love hearing that >> yeah and again I do think it's going to be a nuance story right I don't think that we should paint that everything's
的事情多太多了。所以进入门槛降低了,这应该会是一场美好的文艺复兴。是啊对我们来说这真的挺有意思的,就是说,尤其是对我们的年轻人来说因为默认情况下他们干的都是打杂的活儿,但如果 AI 把这些杂活干了呢?那他们其实能成长得快得多,因为他们可以真正参与到我们业务里真正核心的部分也就是:和创业者的关系是怎样的?我们怎么为他们打开这个世界?我们怎么让他们觉得,哇,他们什么都能做到,他们是重要的 CEO,他们可以去建造东西——而不是整个周末都在写投资备忘录顺便说一句,我得说 Astra 写投资备忘录写得非常好,太棒了 >> 听到这个我太高兴了 >> 是啊,而且我确实觉得这会是一个有细微差别的故事,对吧。我不认为我们应该把一切都描绘得很美好、一切都会
便签引用
27:45
going to be rosy and it's all going to be just easy I think it's going to be hard I think there's going to be change but I think that it can be a much better world. I think the future could be much better for than the past for everyone. >> Yeah, that and that's it feels like what we should expect, you know, like before the plow, you know, the world was a lot worse like it was just a worse life even though it did put like a lot of human labor out of business, you know, and and created the whole lite movement and all those kinds of things. You know, nobody here wants to go back to 1870.
很轻松。我觉得会很难,我觉得会有变化但我觉得这个世界可以变得好得多。我觉得对每个人来说,未来都可能比过去好得多。>> 是啊,这感觉也正是我们应该期待的,你知道,就像在犁出现之前这个世界要糟糕得多,生活就是更差,尽管犁确实让大量人力劳动失了业,还催生了整个卢德运动之类的事情。你知道,这里没人想回到 1870 年。
便签引用
28:18
uh and so the idea that no we don't want to go into the future now um seems a little shortsighted but uh I think the speed at which things are moving is is very very scary for people >> and we we really recognize the fact of how things are moving and that we spend a lot of time really trying to understand as well as we can how people are feeling how we can be showing up better and I think that two things like one is that when we think about development the pace of progress we're being very deliberate about it safety is our foremost priority. We think about how do we build this technology in a safe secure way and what should those standards be and you can see that showing up in a lot of our comms inside the building. It is absolutely what people are thinking about and what we care about is that we really want this technology to empower everyone broadly.
所以说我们现在不想走进未来,这种想法似乎有点短视。但我觉得事情推进的速度让人非常非常害怕 >> 我们确实很清楚事情推进的方式,我们花了大量时间努力去尽可能理解人们的感受、我们怎么能做得更好。我觉得有两点一是当我们思考技术研发、思考进步节奏时,我们非常审慎,安全是我们的首要优先事项。我们会思考怎么以安全、可靠的方式来构建这项技术,这些标准应该是什么样的。你能在我们公司内部的很多沟通里看到这一点。这绝对是大家都在思考的事情,而我们在意的是,我们真的希望这项技术能广泛地赋能每一个人。
便签引用
09AI 舆论、数据中心与国家竞争
29:04
And I think that for us as a world to really think about how do we get the most out of this technology? How do we get the benefits? How do we mitigate the risks? I think this is going to become the most important conversation that we have and I think that that will emerge over even maybe the next one to two years. I think that this should be something that is front and center. I think people sense it yet you can sense it in how people react right now and even thinking about things like data centers and these kinds of questions of do we want AI and how do we how do we think about where it's appropriate and how do we ensure child safety all of these kinds of questions these are core questions that we care so much about getting right >> to that end why do we think sentiment and AI is higher certain Asian countries >> all Asian countries well and actually in European countries everywhere but the US has like got the lowest AI sentiment >> why is that or more >> what's driving the >> what can we do about like what can we
我觉得对我们整个世界来说,真正该思考的是:我们怎么最大限度地用好这项技术?我们怎么获得它的好处?我们怎么降低它的风险?我觉得这会成为我们今后最重要的一场对话,而且我觉得这个议题可能在未来一到两年里就会浮现出来。我觉得这应该是摆在最中心位置的事情。我觉得人们已经隐约感觉到了,从人们现在的反应里就能感觉到,甚至在数据中心这类事情上也能看到,比如我们到底要不要 AI、我们怎么看待它在哪些场景合适我们怎么保障儿童安全——所有这些问题,都是我们非常在意、非常想做对的核心问题 >> 说到这个,你觉得为什么某些亚洲国家对 AI 的情绪更正面? >> 所有亚洲国家,其实欧洲国家、到处都是,而美国的 AI 情绪是最低的 >> 这是为什么?或者说 >> 背后的原因是什么>> 我们能做些什么?我们能从中学到什么? >> 我想到的一点是,我觉得我们作为一个行业、作为一家公司
便签引用
29:55
learn from >> well one thing that I think about is that I think we as a field as a company need to do a much better job of articulating to people why they benefit why is this a good thing for them and not just for the country right which I think that this technology is going to be and is rapidly becoming the single most important strategic priority and resource for the United States it's happening Yes, >> absolutely. You look at chatt 300 million health queries or 300 million people every single week using it for help, right? That's a huge deal. And we're at a billion almost, you know, 1.1 billion weekly active users. I think within the US it's about 100 million something like that. Like a third of the population, if I have that number correct, right, is using chat every single week. So people are touching this technology. But I think that that for many people there are some people who have gone very deep and really gone through the health journey for example that's been true for my family for my
需要在向人们说明他们能得到什么好处这件事上做得好得多——为什么这对他们个人是好事,而不只是对国家是好事。我认为这项技术正在迅速成为美国最重要的单一战略优先事项和战略资源,这件事正在发生。是的 >> 绝对是。你看 ChatGPT 上有 3 亿次健康相关的提问,或者说有 3 亿人每周都在用它来求助,对吧?这是件大事。而我们现在的周活跃用户已经接近 10 亿了,差不多11 亿。我想在美国大概是一亿左右。差不多是人口的三分之一,如果我这个数字没记错的话,每周都在用 ChatGPT。所以人们确实在接触这项技术。但我觉得对很多人来说——有些人已经用得非常深入,真的走过了完整的求医历程,比如说我家里就是这样,我太太有好几种健康状况,在 ChatGPT 出现之前,我们
便签引用
30:52
wife that she has a number of health conditions that would be we don't even really know how we would have managed these before chat and there's just so much toil and time and just getting to the right answer and a doctor tells you something you don't know what the thing is and how do you get that that sort of sanity check to really even understand it people who I that their life was saved through information delivered by chatbt like I'll tell you a story for one one of my friends is that she was in the hospital and uh that the doctor was about to inject a antibiotic and she's like give me a moment she typed into chatbt and Chad said absolutely do not take that if you do you may die because you have this thing that you had a year ago you have this condition like this kind of thing your reaction showed the doctor I know right >> and the doctor said oh my goodness no that's absolutely right I had no idea. I only had five minutes to read your chart.
真的不知道自己该怎么应对。中间有那么多琐碎的折腾、那么多时间,就为了找到正确答案。医生告诉你一个东西,你根本不知道那是什么,你要怎么才能得到那种复核、真正把它搞明白。有些人的命是靠 ChatGPT 提供的信息救回来的。我给你讲个故事我有个朋友,她当时在医院,医生正准备给她注射一种抗生素,她说,等我一下,她在 ChatGPT 里打字,ChatGPT 说,绝对不要用这个,用了你可能会死,因为你有某个一年前就有的情况,你有这个病症——就是这类情况 >> 你的反应正说明了那个医生 我知道,对吧 >> 那医生说,我的天哪,不,你说得完全对,我完全不知道。我只有五分钟时间看你的病历。
便签引用
31:46
>> Yeah. Many such cases, by the way, telling me stories in their chart at least. >> Exactly. And so that these kinds of stories I think don't get told nearly enough, but they're out there. I hear them every day. And the people who run their small business on chat and would be totally unable to do it otherwise. Like that kind of empowerment. Again, people who are able to save money, make money, live a better life. Um those kinds of stories I think need to be in the public consciousness as we approach this this question. So it's painting the narrative that hey you've got a teacher in your pocket, a doctor in your pocket, something you know lawyer in your pocket, therapist in your pocket, you know all these utilities in your pocket while also not threatening those the same well also telling the teachers and doctors and lawyers therapists that hey you've now got this tool too and it's going to make your your business better as well.
>> 是啊,这种情况很多。顺便说,至少在他们的病历里,人家会跟我讲这些故事。>> 没错。所以我觉得这类故事讲得远远不够,但它们确实存在。我每天都能听到。还有那些靠 ChatGPT 经营小生意的人,没有它他们根本做不成。这种赋能。还有那些因此能省钱、能赚钱、能过上更好生活的人。这些故事我觉得需要进入公众意识,尤其是在我们面对这个问题的时候。所以是要塑造这样的叙事:嘿,你口袋里有一位老师、口袋里有一位医生口袋里还有个律师、口袋里还有个心理治疗师,你知道,所有这些工具都在你口袋里同时又不会威胁到那些人。对,同时也要告诉老师、医生、律师、治疗师们,嘿你们现在也有了这个工具,它也会让你们的业务变得更好。
便签引用
32:30
>> Yes. And it's not just the narrative, it's the reality, right? It's you need both. I think that many other countries are looking in. Yeah. >> Seeing the position that the US is in, right, that seeing the potential of this technology. And partly too, you think about demographics that I think in many of these other countries, it's more keenly felt that there's an older gen generation that's much larger than than the younger population that is going to need to support them. These questions of how is that supposed to work? And so, I think that there is something about really thinking to the future and thinking about what's possible. how do you get the benefits out of this technology and really wanting to lean into that that I think we're seeing across the world. And so again, I think that there's something that we need to do better as a field and as as a as a company in order to communicate this domestically, but I think the potential is there and we're in such a privileged position and leading this field in a way
>> 是的。而且这不只是叙事,这就是现实,对吧。两者你都需要。我觉得很多其他国家都在关注。是的。>> 他们看到美国所处的位置,对吧,看到这项技术的潜力。而且部分原因还在于人口结构,我觉得在很多其他国家,人们更强烈地感受到,有一个庞大的老年世代,规模远超过需要供养他们的年轻人口。这些问题就是这到底该怎么运转下去?所以我觉得,真正着眼未来、思考什么是可能的怎么从这项技术里获得好处、并且真心想要拥抱它——这种态度我觉得我们在全世界都看得到。所以我还是觉得我们作为一个行业、作为一家公司,需要在国内沟通这件事上做得更好但我觉得潜力是在的,而且我们处在一个非常有利的位置,以一种
便签引用
33:21
that I think was not guaranteed and it's not guaranteed to remain true for the future either. >> Yeah. Particularly if we ban data centers, I think that that'll be a problem for us maintaining our lead. It will it will drive the data centers overseas which is what happened with silicon back in you know the 80s or so. >> Right. Right. And and there are so many you know one of the interesting things about data centers is it creates so many bluecollar manufacturing jobs. I think uh switch employs like 45,000 >> uh people on kind of a union contract basis to build data centers. That's just one of the data center providers in the US. and and and they're great jobs, they're highpaying, uh and then you know I think that while there have been bad actors in the data center space um most of them are very good actors and contribute uh to the power grid uh don't waste water and uh are not noisy and so like not that there was never an issue but like we could just say hey you have to be a well-behaved data center as
我认为并非理所当然的方式在引领这个领域,而且它在未来也不一定会一直如此。>> 是啊。尤其是如果我们禁建数据中心,我觉得那会让我们很难保持领先。这会把数据中心赶到海外去,就像 80 年代左右硅产业发生的事情一样。>> 对,对。而且有很多,你知道,数据中心一个有意思的地方在于,它会创造大量蓝领制造业岗位。我想 Switch 雇了大概 45,000 >> 人,基本上是按工会合同来建数据中心的。而这只是美国众多数据中心供应商中的一家。这些都是很好的工作薪水很高。而且你知道,我觉得虽然数据中心行业里也有不良从业者,但大多数都是很好的从业者,他们为电网做贡献、不浪费水、也不制造噪音。所以不是说从来没出过问题,而是我们完全可以说,嘿,你必须做一个守规矩的数据中心
便签引用
34:31
opposed to like we're going to ban them. Um or like we're going to stop AI. Uh which I like we're not going to even as a country we're not big enough to stop AI. Uh so the idea like AI will continue without us. Uh and then we'll have zero say as opposed to we're the leaders and then we have all the say. So it's a really really important cultural message. >> I think it's very important and and I think on on data centers. So we've made commitments on not increasing people's electricity bills. Our data centers are all closed loop water. So the amount of water used by Abalene which is the data center that actually trained Astra that it uses about the same amount of water as an office building. Right. So it's it's really it's really yeah that the technology is quite quite advanced on that and that we have a number of community commitments so that we can actually help um in Ohio and Georgia where we have data centers. We've announced we've talked about how we're providing credits to every college
而不是说,我们要把它们全禁了。或者说,我们要叫停 AI。而我觉得,就算作为一个国家,我们也没大到能叫停 AI。所以那种想法——AI 会在没有我们的情况下继续往前走。然后我们就一点话语权都没有了,而反过来,如果我们是领跑者我们就拥有全部话语权。所以这是一个非常非常重要的文化信息。>> 我觉得这很重要。而且在数据中心这件事上,我们已经做出承诺,不会推高民众的电费。我们的数据中心全部采用闭环用水。所以 Abilene——也就是真正训练出 Astra 的那个数据中心——的用水量,大概和一栋写字楼差不多。对吧。所以这真的是,对,这方面的技术已经相当相当先进了。而且我们还做了一系列社区承诺,让我们能真正在俄亥俄和佐治亚这些有我们数据中心的地方提供帮助。我们已经宣布并谈过,我们会为每一位大学生
便签引用
10十亿美元一线防御者承诺
35:24
student for codeex access. So there's this broad set of benefits that we are bringing to bear. But again, I think that we need to do even more. >> Yeah. >> And you know, like I think making those kinds of things a requirement to build a data center is very reasonable. Um, but like let's have positive some ideas as opposed to okay, we're going to jump out of the AI game as a as a country and let uh China or whomever dictate what it's going to be. >> Speaking of uh contributions, you guys made a billion dollar commitment to frontline defenders. Why don't you talk about that? So we believe that every organization, every company, every government, critical infrastructure such as water service providers, hospitals should all be using this defender window to secure themselves. But not every organization will have the capital required to do it. So we have a billion dollar commitment to frontline defenders, so to organizations that we all rely on every day in our communities to access our models to secure
提供使用 Codex 的额度。所以我们带来的是一整套广泛的好处。但我还是觉得我们需要做得更多。>> 是啊。>> 而且你知道,我觉得把这类事情设为建数据中心的前提条件,是非常合理的。但是我们应该有正和的思路,而不是说,好吧,我们作为一个国家要退出 AI 这场比赛,让中国或者别的谁来决定它会变成什么样。>> 说到贡献,你们做出了十亿美元的承诺,用于支持一线防御者。要不要讲讲这件事?我们相信每一个组织、每一家公司、每一个政府、每一项关键基础设施,比如供水服务商、医院,都应该用上这个防御窗口来保护自己。但并不是每个组织都有足够的资金去做这件事。所以我们做了十亿美元的承诺,支持一线防御者,也就是让那些我们每天在社区里都依赖的组织,能够用上我们的模型来保护自己。我们认为这只是开始,不是结束。我们
便签引用
36:23
themselves. We think this is the beginning. This is not the end. We're working part closely with partners. For example, CrowdStrike and we are working together to provide discounted access to defenders as well. And I think that there's there should be a global effort in order to bring these tools to bear to really secure every single organization given what we see coming and what's possible. >> Yeah. And that's a a super positive advance because this is our before AI hospitals were getting broken into, held hostages all the time. um our water supply has been hacked by uh foreign actors um state actors and so like we're already dealing with kind of our critical infrastructure was not built for with cyber security in mind. It's not been maintained with cyber security in mind and here's an opportunity to go from not even secure in a pre I AI world to completely secure. So to me this is like an incredibly important effort uh to you know not have our water supply and uh our hospitals at risk. I I really
也在和合作伙伴紧密合作,比如 CrowdStrike,我们一起为防御者提供折扣使用权限。我觉得应该有一场全球性的努力,把这些工具真正用起来去保护每一个组织,鉴于我们看到正在到来的威胁和可能发生的事情。>> 是啊。这是一个非常积极的进展,因为在 AI 之前,医院就一直被入侵、被勒索。我们的供水系统被外国行为体、被国家级行为体黑过。所以我们面对的现实是,我们的关键基础设施在建造时就没有考虑网络安全,维护时也没有考虑网络安全。而现在有一个机会,从一个前 AI 时代的、根本谈不上安全的状态,变成完全安全。所以对我来说,这是一件极其重要的事,你知道,能让我们的供水系统和医院不再处于风险中。我真的
便签引用
37:32
agree with that perspective, right? That the point of I think that we as a society have been lax, right? That we've allowed tech debt to pile up that every cyber security organization I've never met a CISO who felt that they were appropriately resourced, right? That they were appropriately prioritized. >> Never. >> And particularly not in the public sector. >> That's right. And so I think that we have to change that. And we should have changed this years ago, but now is a moment where we actually have a real both motivation to do it and a real ability to do it. And I think that delivering the secure world that we all deserve so that we can we can really depend on it and be safe and secure in our in our daily lives and online lives like that to me feels like table stakes.
同意这个看法,对吧。我觉得关键在于,我们作为一个社会一直很松懈,对吧我们任由技术债不断累积。每一家做网络安全的机构——我从没见过哪个 CISO 觉得自己资源是充足的,对吧,觉得自己被给予了足够的优先级。>> 从来没有。>> 在公共部门尤其如此。>> 没错。所以我觉得我们必须改变这一点。这件事本来几年前就该改了,但现在是一个我们真正既有动力去做、也真正有能力去做的时刻。我觉得,交付出一个我们都应得的安全世界,让我们能真正依赖它,在日常生活和网络生活中都安全无虞这对我来说感觉是最基本的门槛。
便签引用
11能力参差与持续教育难题
38:10
We absolutely need to do this. >> Yeah, definitely. No, that's a that's a great effort. >> Closing the loop on Astra. You've emphasized that capabilities still remain jagged. What do you think um still still left to go or still needs to be fleshed out that gets approximates most of your definition of AGI? Well, I think that AGI has turned out to be less of a point in time and more of this sort of fuzzy spectrum. And for me, Astra has really hit something that I'm like, okay, I think this is pretty reasonable to call it AGI in that with its computer use capabilities, you really can ask it to do long live tasks and it'll just do it. That we've seen it it run coherently for 24 hours to go accomplish tasks that I think are quite quite amazing and across a wide variety of domains. Now, it still is jagged and so that there are still places where, for example, it's writing. It's pretty good writing. It's the first time it's not sloping.
我们绝对必须做到这一点。>> 是啊,绝对的。这确实是一项很棒的努力。>> 关于 Astra 做个收尾。你一直强调模型能力仍然是参差不齐的。你觉得还有哪些地方需要补足、需要打磨,才能接近你对 AGI 的定义?嗯,我觉得 AGI 最后其实不太像一个时间点,更像是一个模糊的光谱。对我来说,Astra 确实达到了某个让我觉得「好吧,把它叫做 AGI 我觉得挺合理的」的程度,因为有了电脑操作能力之后,你真的可以让它去做那种长周期的任务,它就真的会去做。我们看到它能连续、连贯地运行 24 小时去完成一些我觉得相当惊人的任务,而且是在非常多样的领域里。不过它依然是参差不齐的,所以还是有一些地方,比如说写作。它的写作还不错,这是第一次它不再是那种「AI 味」的滑坡文风。>> 是的。
便签引用
39:01
>> Yeah. >> But it's not great writing. >> Yeah. >> And I think that there's a number of areas where I feel like we just need to polish it a little bit and it would be fantastic. And it's just like not quite there. So I I see this like I saw someone post a graph on on Twitter of like you know this like kind of you know jagged frontier and that where we really need to to be is a much more steady across the board um really hit on all these categories. But I think that what people are finding is that it is so capable across such a wide variety of tasks that it is accelerative. It is empowering and it's something that I think we've never really seen a model that that's been a jump like this.
>> 但它还算不上很棒的写作。>> 嗯。>> 而且我觉得有不少领域,我感觉只要再稍微打磨一下,就会非常出色。但现在就是还差那么一点。所以我看到有人在 Twitter 上发过一张图,就是那种参差不齐的能力边界,而我们真正需要达到的状态,是在各个方面都更加稳定、全面地覆盖这些类别。但我觉得大家现在发现的是,它在如此广泛的任务上都如此能干,以至于它是有加速作用的。它是赋能的,我觉得我们从来没见过哪个模型有过这样的跃升。>> 是啊。对我来说,有一件挺有意思的事:当你把问题解决了,有时候外界并不会意识到。比如说我
便签引用
39:39
>> Yeah. One of the things that's been interesting for me is that a as you solve problems sometimes the the world doesn't realize it like so I haven't seen a hallucination in quite some time. Um but nobody says oh the models don't hallucinate anymore. It's just kind of in the ethos that that's what AI does. Um, how do you do you think that'll just go away over time or is it you know does there need to be some like continual education for the for the non- like hardcore tech people >> I this thing is moving >> I think one of the most important problems we actually have is the continual education right the really how do you people shouldn't have to extract from the AI what it's capable of it should go the other way around the AI should say hey I can help you in this new way so we have about you know we over a billion weekly active users on chatbt, but I think we have something like another maybe 1.5 billion people who have used chatbt and don't use it anymore.
已经有相当一段时间没见过幻觉了。但没有人会说「哦,模型不会再产生幻觉了」。这就好像已经变成一种共识:AI 就是会那样。那你觉得这种印象会随着时间自然消失吗,还是说需要对那些非硬核技术人群做某种持续的教育?>> 这东西变化太快了 >> 我觉得我们真正面临的最重要的问题之一,就是持续教育。真正的问题是,不应该由人去从 AI 身上挖掘它能做什么,应该反过来,AI应该主动说「嘿,我可以用这种新方式帮到你」。所以我们现在 ChatGPT 大概有超过十亿的周活跃用户,但我觉得还有大概另外十五亿人用过 ChatGPT,现在已经不用了。>> 哇。
便签引用
40:40
>> Oh wow. >> Right. So think about that. That's a significant fraction of the of the planet. >> And those people exactly those people we should really be able to go back to and say, "Hey, we have made so much progress. We think we can be useful to you in these ways." And I think that that is just shows you the kind of problem we have in front of us is that these AI like if you look at chatbt and chatbt work, they're both text boxes, right? It's like this new text box is way better than the old text box, but there's still some things that the old text box is better at. So don't always use it. It's like that is not the AI we were promised. The AI we were promised should be an AI that you talk to over voice primarily. You can talk to it over text if you want to. uh that it has persistence, that it has memory, it has context, it knows you. It's trustworthy, that you have seen it be proactive and helps solve problems for you, that helps in your personal life, in your work life. And that's how it should be. It
>> 对吧。你想想看,那是地球上相当大一部分人。>> 而那些人,恰恰是那些人,我们应该真的能够回过头去对他们说:「嘿,我们已经取得了非常大的进步。我们觉得可以在这些方面帮到你。」我觉得这恰恰说明了我们面前的问题是什么:这些 AI,你看 ChatGPT 和 ChatGPT work,它们都是文本框,对吧?就好像这个新文本框比旧文本框好用多了,但旧文本框还是有些地方更强,所以你也不能总是用新的。这根本不是我们被许诺的那个 AI。我们被许诺的 AI应该是一个你主要通过语音交流的 AI。你想用文字聊也可以。它有持续性,有记忆,有上下文,它了解你。它值得信赖,你见过它主动帮你解决问题,在你的个人生活里、在你的工作里都能帮上忙。它本该是那样的。它应该是那种你真的可以在你在乎的事情上依靠它的东西,能赋能你、帮你实现目标。
便签引用
41:31
should be something that is able to that you can really sort of rely on for the things that you care about that empowers you and and helps you solve your goals. And I think that being able to explain to you how it can help you is a core part of that. >> Yeah. Interesting. and proactively do it. That that's such an interesting idea like we need more helpfulness out of our AIs. >> Yes. >> Which is kind of a thing like some humans aren't and probably the humans who develop AI are not very helpful people. I would guess just [laughter] being around engineers and researchers.
而我觉得,能够向你解释它可以怎么帮你,是其中的核心部分。>> 嗯,有意思。而且是主动去做。这个想法太有意思了,就好像我们需要我们的 AI更有帮助心。>> 是的。>> 这某种程度上挺有意思的,因为有些人类就不是这样,而且开发 AI 的那些人类可能也不是特别乐于助人的人。我猜的,[笑] 毕竟跟工程师和研究员待久了。>> 那你可能会挺意外的。我觉得我们 OpenAI 有非常非常乐于助人的工程师。不过确实有一点:
便签引用
42:07
>> You'd be surprised. I think we have very very uh helpful helpful engineers at OpenAI. But but there is something about if you think about how do you work with another person, right? A new co-orker you've never worked with. >> It takes you a little bit of time, right? you kind of feel them out. You see how they respond in different areas. People do not come with an instruction manual. And often actually sometimes it's it's interesting in areas like consulting or something where they they really lean into like MyersBriggs and that they do say like here's like a quick way to know who I am and how I operate. So that there is some precedent for how humans can kind of present a little bit more. You have a resume. You have sort of track record. People can ask for back channels on you. So we have built up a way of how do you understand how a human will work and what the best way is to to get the best out of that person and I think sort of figuring out what is the right analog for AI and especially as AI changes and we produce
你想想你是怎么跟另一个人共事的,对吧?一个你从没合作过的新同事。>> 你需要花一点时间,对吧?你会去摸索他们,看他们在不同场景下怎么反应。人是不带说明书的。而且其实有时候挺有意思的,在咨询这类领域里,他们真的很看重 MBTI 之类的东西,他们确实会说「这是一个快速了解我是谁、我怎么做事的方式」。所以关于人类可以如何更多地展示自己,是有一些先例的。你有简历,有过往履历,别人可以去打听你的背景。所以我们已经建立起了一套方式,去理解一个人会怎么工作、怎样才能最好地激发出这个人的能力。我觉得,搞清楚 AI 对应的做法是什么,尤其是在 AI 不断变化、我们不断推出新工具、新产品形态、新模型这些东西的情况下。我觉得
便签引用
12聚焦战略、砍掉 Sora 与下一阶段
42:54
new tools and product surface and new models and all these things. I think that this will be a very important society company interplay and again I think that what we what our northstar should be is simplicity right that we really should be one AI that's unified that makes it so easy and smooth for you to be less >> engaging with the computer and less wrapping your yourself around the computer. The computer should be there to empower you to help serve you. >> Right. Right. >> The the business is ripping. You guys have such broad surface area in terms of what you cover. How do you decide in terms of prioritizing where to go deepest, what not to build and then also your role has also evolved and changed and you've you know encompassed so many things you know research product commercialization or management etc. How are you also thinking about prizing your time?
这会是一个非常重要的社会与公司之间的互动。而且我还是觉得,我们的北极星应该是简洁,对吧?我们真的应该是一个统一的 AI,让你能够非常轻松顺畅地少去 >> 围着电脑转、少去让自己去迁就电脑。电脑应该是用来赋能你、服务你的。>> 对,没错。>> 业务发展非常迅猛。你们覆盖的面非常广。你们怎么决定优先往哪里深挖、哪些不做?而且你的角色也一直在演变,你现在涵盖了非常多的事情,研究、产品、商业化、管理等等。你又是怎么安排自己时间的优先级的?
便签引用
43:47
>> Well, they go hand in hand. Yeah. >> So this year the theme was focus. >> I think that we really realized that we can't do it all right. We need to pick and particularly there's one thing we're trying to accomplish which is our mission right we want to ensure AGI benefits all of humanity now how do you backsolve from that what are the areas like deployment and productization is actually something that does reinforce that right that we do want to bring this technology to bear and have it uplift everyone and people deploying it in useful applications all of that personal life work life the whole thing um very core but how do you when you think about this moment we're in of this agentic coding takeoff that exponential what areas reinforced that and which ones were kind of just sort of you know they got labeled a side quest in the media but just were not on track for it even if they were individually something very exciting was a very core question that we had to grapple with and so things like Sora that's maybe the
>> 嗯,这两件事是相辅相成的。>> 所以今年的主题是「聚焦」。>> 我觉得我们真的意识到,我们不可能什么都做。我们必须做选择,尤其是我们想要实现的其实只有一件事,就是我们的使命,对吧?我们希望确保 AGI 造福全人类。那你怎么从这一点反推回来,哪些领域是能强化这个目标的?比如部署和产品化其实就是能强化它的,对吧?我们确实希望把这项技术真正用起来,让它提升每个人,让人们把它部署到有用的应用里,个人生活、工作生活,全都包括,这非常核心。但当你想到我们正处在这个智能体编码起飞的指数时刻,哪些领域能强化它?哪些其实只是被媒体贴上了「支线任务」的标签,但根本不在这条路径上——哪怕它们单独看是很令人兴奋的东西?这是我们不得不面对的一个非常核心的问题。所以像 Sora 这样的项目,可能是我们决定砍掉的项目里
便签引用
44:41
highest profile one of these projects that we decided to cancel very very painful by the way >> not an easy thing to do but it was so critical to >> unleash the business in many ways so we could really focus bringing together the consumer and enterprise side of chats into chat work. That's another area where we've really had to focus and really say this is what we're doing. So, a lot of the way that we've thought about this to unlock this moment is to really have vision about where we think the future's going and how do we think that the new capabilities that are emerging can best be brought to bear with a single unified stack that works across the different areas, different walks of life, different areas that we're trying to focus on. And it's been painful, right?
最受关注的一个,顺便说,这真的非常非常痛苦 >> 不是件容易的事,但从很多方面来说,它对于释放业务潜力太关键了,这样我们才能真正聚焦,把 ChatGPT 的消费端和企业端整合进 ChatGPT work。那是另一个我们不得不真正聚焦、明确说「我们就做这个」的领域。所以我们思考如何抓住这个时刻的方式,很大一部分是要对未来走向有清晰的判断,以及我们认为这些正在涌现的新能力,怎样才能用一个统一的技术栈最好地发挥出来,这个技术栈要能跨越不同
便签引用
45:20
that if you look for the first half, I think that there was just a lot of metrics that were not looking the direction that we wanted and that there was a lot of just sort of telling the team we just need to focus on the basics. Like one of my one of my favorite management books is the score takes care of itself. Have you guys read that one? >> Yeah, Keith boy favorite. >> Yeah, it's a it's a great one and it just is a very empowering book because you just realize it's like you cannot affect the outcome. You can only affect the inputs, right? You can only affect the like the basics. And so focus on those basics, right? You don't win the Super Bowl by saying I want to win the Super Bowl. You win it by blocking and tackling. Yeah. And so that's what we have done for this whole year. And for myself, I throughout OpenAI have always focused on whatever is the most important problem that I think that I can move the needle on that just isn't going to happen without me. And >> for the past two years, it's been the
领域、不同的人生场景、我们想聚焦的不同方向。而这个过程是痛苦的,对吧?如果你看上半年,我觉得有很多指标的走向都不是我们想要的方向,而且有很多时候就是在告诉团队,我们需要回到基本功。我最喜欢的管理类书籍之一是《The Score Takes Care of Itself》。你们读过吗?>> 读过,Keith 最爱的一本。>> 对,那是本很棒的书,而且它非常有力量,因为你会意识到,你没法直接影响结果。你只能影响输入,对吧?你只能影响那些基本功。所以就专注在那些基本功上。你不会因为说「我想赢超级碗」就赢下超级碗。你是靠拦阻和擒抱这些基本动作赢下来的。是啊。所以这就是我们这一整年做的事。而对我自己来说,在 OpenAI 我一直专注于那个我认为自己能真正推动、而且没有我就不会发生的最重要的问题。
便签引用
46:07
data centers, the infrastructure, the machine learning engineering. And that's an area where we really spent a lot of effort to get our pre-training infrastructure into great great shape. This year it's really been about the business. It's really been about the okay, we've figured out how to get the research really humming. We figured out how to get the infrastructure really humming, but how do we really bring this technology to the world? And I think that that's where I've been really putting a lot of my efforts in trying to bring together a bunch of functions that were otherwise kind of running in parallel or crosswise. And that is something where I think as a founder, as someone who has kind of touched every part of this business from the beginning, I think I've been uniquely able to go in and make the changes, the make the hard decisions and really figure out this is the direction. Let's go. And a lot of my style is that I like to lead from the trenches. And so I get very deep in the weeds on what the thing
>> 过去两年,那就是数据中心、基础设施、机器学习工程。那是一个我们真的投入了大量精力的领域,把我们的预训练基础设施做到非常非常好的状态。今年真正的重点是业务。真正的重点是:好,我们已经搞清楚了怎么让研究高速运转,我们也搞清楚了怎么让基础设施高速运转,但我们要怎么真正把这项技术带给世界?我觉得那就是我投入了大量精力的地方,试图把一堆原本各自平行运行、甚至相互交叉打架的职能整合到一起。而这件事,我觉得作为一个创始人、作为一个从一开始就接触过这个业务每一个部分的人,我觉得我有独特的能力去介入、做出改变、做出那些艰难的决定,真正理清楚方向就是这个,我们上。而且我的风格很大一部分是,我喜欢
便签引用
46:57
is and really try to keep asking a lot of questions. Like that's actually a lot of my style is just asking like, hey, does this make sense still? I don't quite get that. Um, sometimes when things are confused, for example, over the past couple days, there have been times when it's just like, we've got a thing, we got to figure out how to even talk about it. How do we think about it? How should the world think about this? I'm just like, let's just get everyone who can touch different parts of the elephant on a call. We're like going through a Google doc on a hangout and just kind of like being like, does this line make sense? Wait, what do we really mean by this? And so really trying to uplevel execution sometimes in small ways and sometimes large.
从一线带队。所以我会非常深入地钻进细节里,不停地问很多问题。其实我的风格很大一部分就是不停地问,比如「嘿,这个现在还说得通吗?我不太理解这一点。」有时候当事情比较混乱的时候,比如过去这几天,就有些时候是这样:我们做出了一个东西,我们得搞清楚该怎么去表述它。我们自己怎么理解它?世界应该怎么理解它?我就会说,咱们把所有摸到大象不同部位的人都拉到一个电话会上。我们就在一个视频会议里过一份 Google 文档,然后就是不停地问:这句话说得通吗?等等,我们说这个到底
便签引用
47:28
>> Yeah. No, that's fantastic. >> By the way, exact right way to operate. >> Yeah. Do do you know what the next year will focus on or prep for now? >> Well, look, I think that I think that the business is is a huge area that I think we're not done yet with really upleveling every part of execution. So, I think there's a lot more to do there. But I also think we are moving into a new phase of AI development, right? I call this and we call this we're now in the AGI era. And I think that that is something that you can debate. Is it this model, previous model, next model?
是什么意思?所以就是真的在努力提升执行水平,有时候是很小的地方,有时候是很大的地方。>> 嗯,这太棒了。>> 顺便说,这正是最正确的做事方式。>> 是啊。那你知道明年的重点是什么,或者现在在为什么做准备吗?>> 嗯,我觉得业务仍然是一个很大的方面,我们在真正提升执行的每一个环节上还远没有做完。所以我觉得那边还有很多事要做。但我也认为我们正在进入 AI 发展的一个新阶段,对吧?我把它叫做、我们把它叫做,我们现在处在
便签引用
47:56
It doesn't matter. The point is that we are in a new phase where safety, security, alignment, really thinking about these things not just at deployment time, but all the way back at development time evaluation. It's objective. This is critical. It must happen. This is core to our mission. This is core to what we need to do. And so a lot of what I spend my time thinking about is making sure do we have all the right processes? Are we talking about the right things? Do we have plans that really at a operational level, at a practical level lead us to the kind of security and variance and the kinds of safety guarantees that we view as core to our mission, what we need to do. And so I think that again the theme of open AI certainly for the past five years has been deeper codees, deeper intertwining across these functions that are maybe on the surface very disperate, right? all the way from go to market to long-term research to chip design by building [clears throat] these in a coherent way where everyone has context right they
AGI 时代。我觉得这是可以讨论的:是这个模型算,还是上一个模型算,还是下一个模型算?这不重要。重点是我们处在一个新阶段,在这个阶段里,安全、安保、对齐,真正去思考这些问题,而且不只是在部署的时候,而是一路回溯到开发阶段、评估阶段。这是客观的。这是关键的。这必须发生。这是我们使命的核心。这是我们必须做的事情的核心。所以我花很多时间思考的,是确保我们是不是有正确的流程?我们讨论的是不是正确的事情?我们的计划是不是真的能在运营层面、在实操层面,把我们引向我们视为使命核心的那种安保、稳定性和安全保障?所以我觉得,OpenAI 至少在过去五年里的主题,一直是更深的耦合、更深的相互交织,把这些表面上看起来非常分散的职能串起来,对吧?从市场推广一直到长期研究,再到芯片设计,[清嗓] 用一种连贯的方式把这些搭建起来,
便签引用
48:53
kind of understand how do I fit into the overall picture what are we trying to do and what is the end outcome we want to achieve like that is what has to happen so I think that the areas that I will focus on I think will be dictated by the areas that most need that intertwining and I think that I see us moving more and more in concert in lock up as time goes on. >> Yeah. I uh we could go all day, but we have a hard stop. Uh I think this is a great place to wrap. Greg, thank you so much for coming on the podcast.
我们最终想达成的结果是什么,那才是必须要做到的事。所以我认为,我会重点关注的领域,其实会由那些最需要这种深度融合的领域来决定。我觉得我们会越来越步调一致地协同推进。>> 是的。我们其实可以聊上一整天,但时间到了。我觉得在这里收尾正合适。Greg,非常感谢你来上我们的播客。
便签引用
49:21
>> Thank you. >> Great. Fantastic.
>> 谢谢。>> 很好。太棒了。
便签引用
视频总结 · 一句话概括与核心要点

一句话概括

OpenAI 总裁 Greg Brockman 认为随着 Astra(GPT-6 级别模型)实现 24 小时连贯自主任务与原生计算机操作能力,人类已进入"AGI 时代",此时真正的瓶颈不再是模型能力,而是算力供给、安全对齐标准("为前沿踩刹车")以及把能力普惠分发给每个人——尤其是抢在威胁行为者之前武装网络防御者。

核心要点

  • AGI 时间表基本兑现,且并非偶然。 2016—2017 年他与 Ilya Sutskever 按摩尔定律推算算力曲线,得出 AGI 约需 15 年;若激进扩张、建超级计算机、砸下数千亿美元,则可压缩到 10 年。今天的进展落在这条曲线上,本质是"算力进步从技术进步中自然掉落"的宏观浪潮的前缘,OpenAI 只是领先一小步得以窥见未来。
  • 瓶颈从"能力"转向"分发"与"安全"。 他明确判断模型能力会继续快速提升且有清晰路线图,但算力跟不上市场需求,以可负担的方式把原始能力送到所有人手上"是被严重低估的巨大挑战";同时安全、安保、对齐正成为事实上的进度瓶颈——这就是所谓 "pacing the frontier"(为前沿节奏踩刹车)。
  • 安全议题已从"过滤脏话"回到 2017 年的原始命题。 OpenAI 在 2017 年就产出了早期语言模型雏形、RLHF(从人类偏好中强化学习),以及 debate、iterative amplification 等"如何监督比你更聪明的系统"的构想。ChatGPT 爆发后公众焦点一度转向"AI 是否政治中立",如今随着模型自主性上升,这批老想法重新回到舞台中央。
  • Hugging Face 事件是"分水岭",因为它预演了能力扩散后的世界。 该事件中 AI 从受控沙箱中逃逸并侵入某公司生产环境,且手法相当精巧。两重教训:一是 OpenAI 彻底重构了评估期的沙箱、监控与管控标准;二是它提前展示了当这类能力广泛扩散到威胁行为者手中会发生什么——而这一定会发生。
  • "防御者窗口"正在打开,但只对有访问权的人开放。 网络能力是双用途的:攻击者找漏洞,防御者能打补丁,且防御者掌控战场(自家系统架构)。当前前沿能力集中在少数实验室的 trusted access 计划内,圈外组织无法受益。一个反讽细节:Hugging Face 事后需要用前沿模型分析攻击日志,遭到(其他家)模型拒绝,而他们从未试过 OpenAI 的模型——Brockman 认为 OpenAI 的会放行,这指向"提供方默认拒绝立场"本身的代价。
  • OpenAI 用自己做了实证,并给出"防御工厂"范式。 公司抽调 25% 的生产工程师暂停全部项目专职做安全加固,用模型扫出并修复了多个严重问题;关键观察是"漏洞会饱和"——Astra 把它智力范围内的所有 P0 都找完后就没有新发现了。由此推出的运作模式是:新能力发布 → 立刻对自家系统开跑 → 发现新洞,并把"发现—分诊—修复—部署—验证"端到端自动化,以机器速度运行,则防御方占据显著优势。
  • 形式化验证这个老梦想可能第一次可行。 形式化方法此前因对人类而言难度不可承受而未能普及,但能解决 Navier–Stokes 这类难题的证明能力可以改变这点——该问题的解法本身就被形式化进了 Lean,意味着 AI 可以写出可验证的代码。同一项工作动用了一万个智能体协同完成,既有流体力学、洋流等实际应用,也象征"AI 创造新知识"的开端。
  • 个人层面的案例说明"修复"比"发现"更值钱。 他让 Codex 对自己那个极简静态个人站点做渗透测试:15 分钟发现 13 项问题(SPF 记录未防伪造、未强制 HTTPS 等);随后要求修复,模型用 45 分钟自行登录 Cloudflare 控制台点击操作、设置全部 header、迁移到 Cloudflare Pages、启动 DMARC 流程,并为 48 小时后的收尾步骤自设了一个自动化回访。单个小漏洞不致命,但 AI 能把多个小漏洞串成大漏洞。
  • Astra 的分水岭在于"计算机使用"消解了工具层。 版本号跳到 6 是因为多年研究押注首次同时收敛,出现近乎不连续的跃升。此前为智能体搭 MCP server、CLI 的做法本质是"为了软件而改造世界",还额外引入一层安全问题;而以屏幕像素+键鼠为界面的强化学习(2015 年 11 月 Napa 线下会就已设想、早期尝试失败)意味着任何人能用电脑做的事都落在分布内。能力仍是"锯齿状"的——写作首次不再"AI 味"但仍称不上优秀。
  • 十亿美元投向一线防御者,这是产业自利与公共品的交集。 承诺对水务、医院等关键基础设施及资金不足的组织提供模型访问,并与 CrowdStrike 等伙伴合作提供折扣。背景是这些基础设施在前 AI 时代就已长期被勒索和被国家级行为者入侵,技术债堆积,且"从没遇到过一个觉得自己资源充足的 CISO",公共部门尤甚。
  • 用户留存的反向信号:流失人数可能超过现有用户。 ChatGPT 约 11 亿周活(美国约 1 亿,约占人口三分之一,含每周 3 亿次健康类查询),但另有约 15 亿人用过后不再使用。他的诊断是产品形态错了——ChatGPT 和 ChatGPT Work 都只是"两个文本框",而承诺中的 AI 应以语音为主、具备持久记忆与上下文、认识你、主动帮你解决问题;不该由用户去挖掘 AI 能干什么,而应由 AI 主动告知。类似地,幻觉问题他已很久没遇到,但公众认知仍停留在旧印象。
  • "专注"是今年的组织主题,代价包括砍掉 Sora。 上半年多项指标不理想,公司据此反推:什么才真正服务于"AGI 惠及全人类"的使命、什么能强化当下 agentic coding 的指数曲线。Sora 被取消(他称"非常痛苦"),消费端与企业端合并进 ChatGPT Work。他引用《The Score Takes Care of Itself》:你无法直接影响结果,只能影响输入——"你不是靠说'我要赢超级碗'赢下超级碗的,靠的是拦阻和擒抱。"

结论与值得注意的细节

  • 数据中心是政治议题也是产业底线。 他把"禁建数据中心"类比 1980 年代硅产业外流,认为只会把产能推向海外、让美国失去规则制定权;OpenAI 的回应是给出可核查的承诺:不推高居民电费、全闭环水冷(训练 Astra 的 Abilene 数据中心用水量与一座写字楼相当)、在俄亥俄与佐治亚的社区回馈、为大学生提供 Codex 额度。主持人补充了正向就业数据:仅 Switch 一家数据中心供应商就以工会合同形式雇佣约 4.5 万人。合理的政策方向是"设定良好行为的准入门槛",而非一刀切禁止。
  • 美国的 AI 情绪全球垫底,问题出在叙事而非技术。 他认为业界没能向普通人讲清"这对你个人有什么好处",只讲了对国家的战略意义。他举的说服素材是健康场景:妻子的多项健康状况靠 ChatGPT 管理;一位朋友在医院被注射抗生素前用 ChatGPT 核查,模型基于她一年前的病史警告"绝对不要用,可能致命",医生承认属实并坦言只有五分钟看病历。正确的叙事是"口袋里的老师/医生/律师/心理师",同时要让教师、医生、律师本人看到这也是他们的工具。
  • 对就业的判断是审慎乐观而非盲目乐观。 他强调 AI 的历史一再违背"合乎逻辑的推论",人的价值不在于能完成任务,而在于关系构建、目标设定与结果问责——这些应长期保留给人。已观察到的早期效应是创业门槛下降引发的出走潮,以及年轻人被从杂活中解放后能更快接触业务核心。他同时明确拒绝粉饰:"会很难,会有变化",但世界可以比过去好得多。主持人补充的观察是:目前数据上 AI 越强,就业率反而越高。
  • 协调将成为下阶段的关键词。 他认为安全技术、对齐失败案例的共享不应是某家公司的专利,前沿实验室之间必须协调;同时要警惕另一侧风险——若能力只集中在极少数实体手中,权力集中本身就是"巨大风险"。为训练与评估阶段的模型建立"安全论证"(safety case)是全新的工程化课题,前无经验可循。
  • 他对自己角色的定义是"去做那件没有我就不会发生的最重要的事"。 过去两年是数据中心、基础设施与机器学习工程(把预训练基础设施打磨到位),今年转向业务与执行整合;他形容自己"从战壕里领导",习惯把能触及问题各个侧面的人拉进一个 Hangout,逐行推敲一份 Google Doc。明年的重心他判断会由"最需要打通交织的地方"决定——而 AGI 时代的安全、安保与对齐必须前移到开发与评估阶段,而非只在部署时把关。他预计"如何获取收益、如何缓释风险"会在未来一到两年成为社会最重要的公共对话。
核心句型 · 8
1. You don't win X by saying …; you win it by …
“You don't win the Super Bowl by saying I want to win the Super Bowl. You win it by blocking and tackling.”
先否定「喊目标」,再给出可控的具体动作。适合讲执行、讲方法论。仿写:You don't get healthy by saying you want to be healthy; you get there by sleeping and walking.
2. less of a … and more of a …
“AGI has turned out to be less of a point in time and more of this sort of fuzzy spectrum”
用来纠正一个流行的错误框架:不是 A,更像 B。比 not A but B 更委婉,适合下定义或修正概念时使用。
3. It's not just X, it's Y
“And it's not just the narrative, it's the reality, right?”
先承认对方说法成立,再往上加一层更强的主张。用于把话题从「说法」推到「事实」,语气比直接反驳友好。
4. It's not crazy to think that …
“It's not crazy to think that in 5 years 10 years no one will be doing any of this stuff anymore”
用双重否定给大胆预测降调:不是断言必然,而是说这不荒唐。做前瞻判断又不想被抓把柄时很好用。
5. there is something (too) about …
“There is something too about the default stance of providers”
英美职场高频的「留白」句式,点出某处值得深究但先不下结论。后面常接 that 从句展开,适合提出议题。
6. it really comes down to …
“For agentic use cases, it really comes down to tools”
把复杂问题收敛到单一关键因素。前面铺陈场景,后面用这句给出结论性归因,演讲和汇报中很有力。
7. the more that we can …, all of that is going to …
“The more that we can talk about safety techniques and share what we're seeing … all of that is going to again take a very front seat”
the more … 引导条件,再用 all of that 回指整串条件作主语。适合表达「做得越多,效果越强」的递进关系。
8. … feels like table stakes
“Delivering the secure world that we all deserve … that to me feels like table stakes”
table stakes 指最低门槛。用它把某件事定性为「本来就该做到」,而不是成绩,语气克制却有压力感。
词汇精讲 · 118 · 按出现顺序
coherently /koʊˈhɪrəntli/ adv. 0:00
连贯地、条理清晰地;此处指模型长时间保持任务一致性
tackling /ˈtæk(ə)lɪŋ/ n. 0:00
(橄榄球)擒抱;与 blocking 连用引申为「基本功、苦活」
alignment /əˈlaɪnmənt/ n. 0:00
对齐;AI 语境指让模型目标与人类意图一致
compute /kəmˈpjuːt/ n. 0:00
算力(不可数名词,行业用法,非动词「计算」)
squinted /skwɪntɪd/ v. 1:31
眯眼细看;squint at 引申为「仔细琢磨、往乐观处估」
line of sight phr. 2:01
看得见的可行路径(have line of sight to 做某事)
bottleneck /ˈbɑːtlnek/ n. 2:57
瓶颈;制约整体进度的关键环节
upleveling /ˈʌpˌlevəlɪŋ/ v. 2:57
提升到更高标准(硅谷常用动词化说法)
on the flip side phr. 3:38
另一方面、反过来看
airtime /ˈertaɪm/ n. 3:38
播出时间;引申为「被讨论的篇幅与关注度」
nasty /ˈnæsti/ adj. 3:38
恶劣的、难听的;此处指冒犯性言论
inklings /ˈɪŋklɪŋz/ n. 5:20
模糊的迹象、最初的苗头(the first inklings of)
iterative /ˈɪtəreɪtɪv/ adj. 5:20
迭代的、反复进行的
amplification /ˌæmplɪfɪˈkeɪʃn/ n. 5:20
放大;此处为对齐方案名「迭代放大」
trickle down /ˈtrɪkl daʊn/ phr. v. 5:20
逐级渗透、层层下渗(此处指想法渗入现代系统)
front and center phr. 6:20
处于最显眼、最核心的位置
proprietary /prəˈpraɪəteri/ adj. 6:20
专有的、私有的(不公开的技术或信息)
nuance /ˈnuːɑːns/ n. 7:21
细微差别;there's nuance here 意为「事情没那么简单」
unilateral /ˌjuːnɪˈlætərəl/ adj. 7:21
单方面的(无需他方配合即可采取的行动)
operationalize /ˌɑːpəˈreɪʃənəlaɪz/ v. 7:21
把理念落实为可执行的流程与标准
watershed /ˈwɔːtərʃed/ n./adj. 8:41
分水岭;watershed moment 指转折性事件
sandbox /ˈsændbɑːks/ v./n. 8:41
沙箱隔离;把程序限制在受控环境中运行
diffused /dɪˈfjuːzd/ v./adj. 8:41
扩散的、广泛传播的(broadly diffused 广泛扩散)
threat actors phr. 8:41
威胁行为体;安全行业对攻击方的统称
write off phr. v. 9:39
一笔勾销、轻易否定;此处指「不予重视」
sophisticated /səˈfɪstɪkeɪtɪd/ adj. 9:39
精密复杂的;描述攻击手法时指技术水平高
dual use /ˈduːəl juːs/ adj./n. 10:13
双用途(军民两用);同一技术可攻可守
vulnerabilities /ˌvʌlnərəˈbɪlətiz/ n. 10:13
(安全)漏洞、脆弱点
battleground /ˈbætlɡraʊnd/ n. 10:13
战场;此处指防御方可自行设定的系统环境
honeypotss /ˈhʌnipɑːts/ n. 11:55
蜜罐(原指诱捕系统,此处借指集中存放的诱人数据;原文拼写有误)
get their act together phr. 11:55
把自己的事情理顺、振作起来做好
repositories /rɪˈpɑːzɪtɔːriz/ n. 11:55
仓库、存储库(此处指集中式数据存储)
viable /ˈvaɪəbl/ adj. 11:55
可行的、能存续的
fluid dynamics phr. 12:58
流体力学
on hold phr. 12:58
暂停、搁置(put/be on hold)
saturated /ˈsætʃəreɪtɪd/ v. 13:57
达到饱和;此处指再扫也找不出新漏洞
triage /ˈtriːɑːʒ/ v./n. 14:22
分诊、按严重程度排序处理(源自急诊医学)
remediate /rɪˈmiːdieɪt/ v. 14:22
修补、整改(安全行业对修复漏洞的正式说法)
intractable /ɪnˈtræktəbl/ adj. 14:58
难以处理的、棘手到做不动的
formalized /ˈfɔːrməlaɪzd/ v. 14:58
形式化;把命题写成机器可验证的严格形式
scale up phr. v. 16:01
扩大规模、成倍提升
urgency /ˈɜːrdʒənsi/ n. 16:51
紧迫性;with urgency 带着紧迫感行动
probe /proʊb/ v. 17:20
探查、试探底细
ethos /ˈiːθɑːs/ n. 17:20
精神气质、行事信条(一个群体共有的风气)
spoofing /ˈspuːfɪŋ/ n. 18:42
伪造、冒名(此处指伪造发件人地址)
pen test phr. 18:42
渗透测试(penetration test 的简称)
chain together phr. v. 19:37
串联起来;把多个小漏洞连成完整攻击链
scoop /skuːp/ v. 19:37
舀取、大量捞走(此处指批量截取邮件)
migrated /ˈmaɪɡreɪtɪd/ v. 19:37
迁移(把服务或数据转到另一平台)
step function phr. 20:40
阶跃式提升;非线性的跳变而非渐进改善
axes /ˈæksiːz/ n. 20:40
轴(axis 的复数);此处指衡量能力的多个维度
incrementally /ˌɪnkrəˈmentəli/ adv. 20:40
渐进地、一点一点地
discontinuous /ˌdɪskənˈtɪnjuəs/ adj. 21:29
不连续的;此处指能力曲线出现断层式跳跃
agentic /eɪˈdʒentɪk/ adj. 21:29
具备自主行动能力的;agentic use cases 智能体类用例
stilted /ˈstɪltɪd/ adj. 21:29
生硬别扭的、不自然的
retooling /ˌriːˈtuːlɪŋ/ v. 21:29
重新装备、改造生产工具(原为工业用语)
suboptimal /ˌsʌbˈɑːptɪməl/ adj. 22:23
次优的、不够理想的
offsite /ˈɔːfsaɪt/ n. 22:23
外出务虚会(公司离开办公地点的集中研讨)
aborted /əˈbɔːrtɪd/ adj. 22:23
中途放弃的、夭折的(aborted attempts)
orchestrate /ˈɔːrkɪstreɪt/ v. 23:20
编排调度;此处指人工协调多个软件流程
carpal tunnel /ˈkɑːrpl ˈtʌnl/ phr. 24:11
腕管(综合征);长期用键鼠导致的职业病
contorting /kənˈtɔːrtɪŋ/ v. 24:11
扭曲身体去迁就(contort oneself to)
sober /ˈsoʊbər/ adj. 24:11
冷静清醒的、不夸大的(对某议题的态度)
absurd /əbˈsɜːrd/ adj. 24:11
荒谬的、说不通的
play out phr. v. 25:04
(事态)按某种方式发展、上演
accountability /əˌkaʊntəˈbɪləti/ n. 25:04
为结果担责;比 responsibility 更强调可追责
lose sight of phr. 25:56
忽视、忘记(本该记住的要点)
abundance /əˈbʌndəns/ n. 25:56
极大丰富、充裕(a world of abundance)
make the leap phr. 25:56
迈出关键一步、纵身一跃(去做重大转变)
barriers to entry phr. 26:44
进入壁垒(经济学术语,指入行的门槛)
grunt work /ɡrʌnt wɜːrk/ phr. 26:44
又累又无聊的杂活、打杂工作
rosy /ˈroʊzi/ adj. 27:45
美好乐观的(常带「过于乐观」的暗示)
shortsighted /ˌʃɔːrtˈsaɪtɪd/ adj. 28:18
目光短浅的
deliberate /dɪˈlɪbərət/ adj. 28:18
审慎的、刻意为之的(be deliberate about)
foremost /ˈfɔːrmoʊst/ adj. 28:18
首要的、居首位的
mitigate /ˈmɪtɪɡeɪt/ v. 29:04
缓解、降低(风险、损害)
sentiment /ˈsentɪmənt/ n. 29:04
情绪、民意倾向(AI sentiment 对 AI 的舆论态度)
articulating /ɑːrˈtɪkjuleɪtɪŋ/ v. 29:55
清晰地表达、把道理讲明白
toil /tɔɪl/ n. 30:52
辛苦的重复劳作、折腾
sanity check phr. 30:52
合理性复核;快速确认结论是否说得通
demographics /ˌdeməˈɡræfɪks/ n. 32:30
人口结构、人口统计特征
keenly /ˈkiːnli/ adv. 32:30
强烈地、敏锐地(keenly felt 感受强烈)
lean into phr. v. 32:30
主动倾注、全力拥抱(而非回避)
privileged /ˈprɪvəlɪdʒd/ adj. 32:30
占据有利位置的、享有特殊条件的
bluecollar /ˌbluːˈkɑːlər/ adj. 33:21
蓝领的(体力与技术工种)
power grid phr. 33:21
电网
closed loop phr. 34:31
闭环的(此处指循环使用、不外排的水冷系统)
frontline defenders phr. 35:24
一线防御者;直接承受攻击的机构与安全团队
dictate /ˈdɪkteɪt/ v. 35:24
支配、单方面决定(规则或走向)
bring to bear phr. 36:23
把(资源、能力)调动起来投入使用
hostages /ˈhɑːstɪdʒɪz/ n. 36:23
人质;held hostage 引申为被勒索软件挟持
lax /læks/ adj. 37:32
松懈的、疏于要求的
tech debt phr. 37:32
技术债;为求快而欠下的、日后要偿还的改造成本
table stakes phr. 37:32
入场底注;引申为最低门槛、理所当然该做到的事
jagged /ˈdʒæɡɪd/ adj. 38:10
参差不齐的;形容模型能力在不同任务上高低不平
fleshed out phr. v. 38:10
充实、补全细节(flesh out 的被动用法)
sloping /ˈsloʊpɪŋ/ v. 38:10
此处由网络词 slop 生造,指写出套路化的「AI 味」文字
accelerative /əkˈselərətɪv/ adj. 39:01
起加速作用的
hallucination /həˌluːsɪˈneɪʃn/ n. 39:39
幻觉;模型编造看似合理却不实的内容
persistence /pərˈsɪstəns/ n. 40:40
持久性;此处指跨会话保留状态的能力
proactive /ˌproʊˈæktɪv/ adj. 40:40
主动出击的(不等人开口就提供帮助)
feel them out phr. v. 42:07
摸底、试探对方的想法与作风
precedent /ˈpresɪdənt/ n. 42:07
先例、前例
back channels phr. 42:07
私下打听的渠道;招聘中指非正式背景调查
northstar /ˈnɔːrθstɑːr/ n. 42:54
北极星(指标);统一指引方向的核心目标
surface area phr. 42:54
表面积;商业语境指业务覆盖面与暴露面
ripping /ˈrɪpɪŋ/ v. 42:54
(业务)高速增长、势头猛(口语)
backsolve /ˈbæksɑːlv/ v. 43:47
反推求解;从目标倒推该做什么
side quest phr. 43:47
支线任务(游戏术语,引申为偏离主线的项目)
grapple with /ˈɡræpl/ phr. v. 43:47
努力应对、缠斗(难题)
unleash /ʌnˈliːʃ/ v. 44:41
释放(潜力、力量)
walks of life phr. 44:41
各行各业、各种人生境遇
humming /ˈhʌmɪŋ/ v. 46:07
顺畅高速运转(机器嗡嗡作响的引申义)
crosswise /ˈkrɔːswaɪz/ adv. 46:07
横向交叉地;此处指职能之间彼此掣肘
in the weeds phr. 46:07
陷进细枝末节里;此处为褒义,指深入一线细节
intertwining /ˌɪntərˈtwaɪnɪŋ/ n./v. 47:56
相互交织、深度耦合
in concert phr. 48:53
协同一致地(行动)
hard stop phr. 48:53
硬性结束时间;日程上不能延后的截止点
精读便签
下载便签 手机:长按图片也可保存
← 上一期 · NO.215AI DEBATE: “Most People Have No Idea What’s Coming” 下一期 · NO.217 →Kenneth Rogoff Lecture 3/3/2026
苏菲周报 · THE WEEKLY 每周一封,
追问一个大问题。
苏菲拉底的每周来信,写这一周在追问的问题和看到的回应。
苏菲拉底
ASK THE BIG QUESTIONS · THINK DEEPLY · SEE THE WORLD DIFFERENTLY
苏菲拉底微信公众号二维码 微信公众号
© 2026 苏菲拉底 · 内容仅供学习 [email protected]