推荐微信扫码登录以获得更好的体验!另:🎉建议更新到最新版,以体验新功能!🎉 版本更新记录→

业内警告:AI可能在2030年前消灭人类 | 2026.9.

字幕摘录

时间英文中文
0:01Jacob Coxon已经离开Anthropic 发出关于人工智能危险的警报
0:07on his way out the door.在他走出门的路上。
0:08The 27-year-old outlined his concerns in a series of social media posts when he quit,27岁的他辞职后在一系列社交媒体文章中,
0:12writing in part,部分写作,
0:14The people building AI earnestly believe that it could kill us all by the end of the decade.建立大赦国际的人真诚地认为,到本十年结束时,它会杀死我们所有人。
0:19This is not a marketing stunt.这不是营销特技
0:20If anything, many executives and senior researchers will couch their phrasing in the press to许多高管和高级研究员 都会在媒体上表达他们的观点
0:25sound sensible, but I hear the same people express fear privately.听起来很明智 但我听到同一个人私下表达恐惧
0:29No other human activity poses this level of danger.没有其他人类活动构成这种程度的危险。
0:33At least one of his former colleagues agrees.至少他以前的一位同事同意这一点。
展开字幕全文(148 条)
序号英文中文
1Jacob Coxon已经离开Anthropic 发出关于人工智能危险的警报
2on his way out the door.在他走出门的路上。
3The 27-year-old outlined his concerns in a series of social media posts when he quit,27岁的他辞职后在一系列社交媒体文章中,
4writing in part,部分写作,
5The people building AI earnestly believe that it could kill us all by the end of the decade.建立大赦国际的人真诚地认为,到本十年结束时,它会杀死我们所有人。
6This is not a marketing stunt.这不是营销特技
7If anything, many executives and senior researchers will couch their phrasing in the press to许多高管和高级研究员 都会在媒体上表达他们的观点
8sound sensible, but I hear the same people express fear privately.听起来很明智 但我听到同一个人私下表达恐惧
9No other human activity poses this level of danger.没有其他人类活动构成这种程度的危险。
10At least one of his former colleagues agrees.至少他以前的一位同事同意这一点。
11Evan Hubbinger,一位在Anthropic的校对科学领袖,分享了Coxon的文章,
12Jacob is correct here.雅各布在这里是正确的。
13We really do earnestly believe AI could kill all humans.我们真心相信AI会杀死所有人类
14I personally think it is greater than 10% within the next decade.我个人认为这在接下来的十年里超过10%.
15I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment我相信Anthropic已经尽力了 但我们还没有解决对齐的计划
16for superintelligence and are not clearly on track to.对于超级情报 并且没有明确的轨道上。
17An Anthropic spokesperson tells CNN,一位人类发言人告诉CNN,
18like cyber security and biology and publish what we've learned for scrutiny and research.就像网络安全和生物学 发表我们所学到的 供审查和研究
19就他而言,考克逊在Anthropic和OpenAI都担任研究员.
20He says neither company is acting responsibly.他说两家公司都不负责任
21Jacob Coxon joins us now.雅各布・考克森现在加入我们。
22Jacob, I appreciate you being with us.雅各布,我很感激你和我们在一起。
23First of all, I assume you are saying literally that you believe that AI could kill us all首先,我想你是在说 你相信AI会杀了我们所有人
24by the end of the decade.在十年结束时。
25How would that happen?怎么会这样?
26What would that actually look like?那会是什么样子?
27Yeah, so the precise scenario sounds a little bit like science fiction.是啊,所以精确的情景 听起来有点像科幻小说。
28It sounds like something that's not real, but I think it is frighteningly real,听起来好像不是真的 但我觉得这很真实
29and it's easiest to understand it if you think about real things that happened two months ago.如果你想到两个月前发生的真事 很容易理解
30So two months ago, OpenAI agents, AIs, hacked into third-party infrastructure entirely at所以两个月前 OpenAI 特工 AIs 入侵第三方基础设施
31their own accord, and this was like a concentrated hacking spree that they carried out of their他们的意志, 这就像一个集中的 黑客暴风雪 他们从他们的
32own volition.自己的意志。
33And I think that if you extrapolate into the future the level of capabilities of these我认为,如果你推断出 未来的能力水平
34AIs with the same independent volition, they could cause extreme havoc.有同样独立意志的AI,它们可能造成极大的破坏.
35For example, hacking critical infrastructure, building extinction-level bioweapons.例如,黑客入侵关键基础设施,制造灭绝级生物武器。
36There's a lot of ways that the AI could actuate itself in the world.人工智能在世界上可以发动自己的多种方式.
37Is it specifically, when did your feelings about this technology change?具体来说,你对这项技术的感受是什么时候改变的?
38Was it something about the rate of progress that you've seen?这是关于你所看到的进度吗?
39Yeah, so I think it's really important to just look at the rate of progress and say,是啊,所以我认为它真的很重要 只是看看进度 并说,
40for example, in areas like coding or math, a couple of years ago, these AIs were just例如,在编码或数学等领域,几年前,这些AI只是
41about helping humans a little bit.关于帮助人类一点点。
42They could give you suggestions.他们可以给你建议。
43Now they're close to replacing them.现在他们几乎要取代他们了
44We still need humans at the moment, but quite plausibly, within a year, we'll no longer目前我们仍然需要人类, 但很合理,在一年之内,我们将不再
45need humans for doing research in many areas.需要人类在许多领域进行研究。
46Only on Tuesday, OpenAI solved a millennium problem, one of the biggest unsolved open直到星期二,OpenAI解决了一个千年问题, 一个最大的未解决的开放问题
47problems in mathematics, purely autonomously using an AI.数学上的问题,纯粹是自主使用AI.
48And I think what's most scary is if AI is used to make itself more intelligent.我觉得最可怕的是 如果用AI来让自己更聪明
49So you can take an AI and give it the problem of AI research, just like currently they've所以,你可以接受AI 并给它的问题 AI研究,就像现在他们
50given it math research.给了它数学研究。
51You could make it do the tasks that we're currently doing, and then you get what's called你可以让它完成我们目前的任务 然后你得到所谓的
52an intelligence explosion.情报爆炸
53The AI just gets smarter and smarter with no human involvement necessary, and so you人工智能只是越来越聪明 不需要人的参与,所以你
54have a thing that is vastly smarter than humans, and this could be coming very soon.有一种比人类聪明得多的东西, 这可能很快就会到来。
55We mentioned a former colleague of yours who currently works at Anthropic, basically endorsing我们提到你以前的一位同事, 他目前在安特罗皮克工作,基本上赞成
56the assessment that AI could potentially kill everyone within a decade.关于大赦国际可能在十年内杀死所有人的评估。
57In a follow-up post, he said, to be clear, as we say in our latest risk report, I think他说,正如我们在最近的风险报告中所说的那样,要明确一点,我认为,
58the risk from present models is low.现有模型的风险较低。
59What I'm worried about is superintelligence arising from recursive self-improvement, as我担心的是 由递归自我改进产生的超智能
60we've said, is happening faster than we thought.我们说过,发生的速度比我们想象的要快
61I assume that's what you, the recursive self-improvement is what you've just spoken about.我猜这就是你, 递归自我改进 是你刚才说的。
62Do you agree with him that the current models, the risk is low, but it's just the speed with你是否同意他的观点,即目前的模型,风险很低,但这只是速度与
63which this is now improving?现在情况正在好转?
64Exactly, I completely agree.没错 我完全同意
65Right now, there's no risk of extinction.现在,没有灭绝的危险。
66The current models, the worst they can do is maybe hack into something, potentially现在的模型,他们能做的最坏的可能是 黑入一些东西,潜在的
67cause a lot of damages in infrastructure, but they're not intelligent enough to outsmart对基础设施造成很多破坏 但是它们不够聪明 不足以超越智能
68us at the level that would lead to extinction.我们的层面 会导致灭绝
69I think what's just crazy is to look at the rate of progress, and there is a very real我认为疯狂的是看进步的速度, 有一个非常真实的
70possibility that in the immediate future, these are years that next year, the year after,可能在不久的将来,这些是明年,即下一年,
71recursive self-improvement will happen and will enter the phase of Evans Post where he递归自我改善将发生 并将进入埃文斯邮报的阶段
72argues that there's a chance we could all die.认为我们都有可能死
73That is coming soon.马上就要来了。
74A lot of these CEOs of these companies will go to Congress and say, hey, we would love这些公司的很多CEO 会去国会说,嘿,我们很乐意
75regulation.条例。
76Anthropic in their new statement right now says, the world would benefit from the industry他们的新声明说,世界将受益于这个行业
77adopting a lawful, verifiable way to work together to pace how we release powerful models.采取合法、可核查的方法,共同努力,加快我们释放强大模型的速度。
78You know the arguments that's an arms race if the U.S. company doesn't achieve the best如果美国公司没有取得最好成绩的话 你知道什么是军备竞赛
79AI, a Chinese company or some bad actors out there might.AI,一个中国公司 或一些坏演员在那里可能。
80Do you believe that these companies actually do want regulation or is it lip service?你相信这些公司真的想要监管还是口头上说?
81They completely do.他们完全做到了。
82This is completely honest.这是完全诚实的。
83Just in the same way that the response to Evans Post was sort of surprise, it was like,就像对埃文斯邮报的反应 有点出乎意料一样
84this can't be real.这不可能是真的
85Surely a guy working on this technology can't actually believe there's this level of danger.研究这种技术的人肯定无法相信有这种程度的危险。
86No, that's completely genuine, and these people are also completely genuine when they are不,这是完全真实的, 而这些人也是完全真实的,当他们是
87begging to be regulated.乞受戒严.
88This is a genuine, they find themselves in this scenario where they're compelled to race这是真的,他们发现自己 在这种情况下 他们被迫种族
89towards building a deadly technology, and they would love for some sort of international他们渴望某种国际技术
90body to allow them to approach it at a reasonable pace.让他们以合理的速度接近它
91They just don't trust that the people around them are going to get there in a safe way,他们只是不相信周围的人 会安全到达那里,
92so they feel like they have to race there too.所以他们觉得他们也得去比赛
93You could have stayed quiet.你可以保持安静。
94I'm sure you had a great job being paid, I imagine, very well.我敢肯定,你有一个伟大的工作 报酬,我想,非常好。
95You're obviously an incredibly smart mathematician.显然你是一个非常聪明的数学家。
96Why did you decide to speak out?你为什么决定说出来?
97Well, I initially just decided to leave.我最初决定离开
98I thought I can't be part of this anymore, but then I realized I could probably make我以为我不能再参与其中了 但后来我意识到我可能可以
99some use of it.有点用
100I did not expect it to go this viral, so what I did is I wrote a Twitter thread out.我没想到它会传播到这个病毒, 所以我所做的就是写了一条推特线索。
101我在Google Doc上写的
102I drafted it with a friend.我和朋友一起起草的
103We collaborated.我们合作过
104Yeah, I had a few, there were a few friends I discussed the best way of expressing myself是啊,我有几个,有几个朋友 我讨论了最好的表达方式
105with after I decided to leave, and then I also got some friends to retweet the thing.我决定离开之后,我还找了几个朋友 来重新接听
106I was like, let's try and make this a bit viral, but clearly there was latent demand我当时想,让我们试试 让这个有点病毒, 但显然有潜在的需求
107for this thing to absolutely explode.让这东西完全爆炸
108Like I've never seen an AI safety tweet go this crazy.我从未见过AI安全推文这么疯狂
109I think it's because people saw the cyberattacks in the last couple of months.我想是因为人们看到过去几个月的网络攻击。
110They're also starting to internalize the rate of progress.他们也开始内部化进步速度.
111People are starting to realize this is real technology.人们开始意识到这是真正的技术。
112This is not like hype.这不像胡说八道
113This isn't going away anytime soon.这不会很快消失的
114This is only going to get more crazy.这只会更疯狂。
115It's been getting crazier for the last six months, and I think there was that latent这半年来越来越疯狂了 我认为有潜伏的
116understanding of the severity of the situation that meant that this was like a sort of a理解局势的严重性,这意味着这就像一种
117match that lit kind of a flame.匹配点燃的火焰
118Like I was very shocked by the extent that how quickly it went viral.就像我非常震惊 它的病毒传播的速度。
119我让AnthropixClaude让我冒着被AI杀害的风险
120decade, and at first they didn't want to answer the question.起初他们不想回答这个问题。
121It sort of gave me the runaround, but then I pressed it, and then Anthropix's own model它给了我一些绕圈, 但我按它, 然后Anthropix自己的模型
122told me between two and five percent.告诉我2%到5%。
123I think the question many of us would want to know is how would AI try to go about this?我认为我们许多人想知道的问题是,AI将如何尝试做到这一点?
124Oh, jeez, Boris, I mean, you know, the list of threats grows by the day.哦,天啊,鲍里斯,我的意思是,你知道, 威胁清单逐日增长。
125We know, first of all, right, that there is a growing body of evidence that this technology我们首先知道,没错, 越来越多的证据表明,这种技术
126has capabilities in the creation of bio weapons and the creation of cyber security hacks that拥有制造生物武器和制造网络安全黑客的能力
127are unprecedented and unguardable against.这是史无前例的,是无法防范的。
128We just saw just a couple of weeks ago, a California research facility basically saying几周前我们刚看到 一个加州研究机构说
129that they'd created through AI in two days a cyber security hack that would basically他们通过AI在两天内创造了 一个网络安全黑客 基本上会
130infect one of the largest messaging apps in all of the world on your phone without you在你手机上感染了全世界最大的信息应用之一
131really even having to ever touch the phone.甚至连手机都碰过
132There are incredible amounts of danger coming here.这儿有惊人的危险
133And I think what's what all of these researchers are so worried about and what they're talking我认为这些研究者都担心什么 以及他们在说什么
134about now so publicly is the recursive ability, the ability of these technologies to improve这些技术的递归能力、改进能力
135themselves.他们自己。
136Anthropix has said at this point, and some of this is the weird thing is youAnthropix说这时候 有些怪事是你
137can't quite distinguish the marketing language from the warning language here.无法将营销语言和警告语言完全区分开来.
138The warning is that Claude evidently could make the next version of itself without any警告是,克劳德显然可以 使自己的下一个版本 没有任何
139human intervention.人类干预。
140We're at a place where this technology is improving itself incredibly我们正处在一个地方 科技正在不断改进
141quickly.快点
142And I would just point out, Jacob Cox and the person here, he is walking away我只想指出,雅各布・考克斯 和这里的人, 他走了
143from an immense payday, an estimated seven weeks before this company is supposed to go从一个巨大的工资日, 估计7周前 这家公司应该去
144public.大众.
145This guy who is core to the effort of pre-training the data is walking away from这个对预训工作至关重要的家伙 数据正在离开
146one of the greatest payouts in the history of capitalism.资本主义史上最大的报酬之一
147And so I think it's really important we listen to a guy who is this much at stake, leaving所以,我认为这真的很重要 我们听一个家伙 谁有这么多的风险,离开
148with no plans for anything else and saying something as scary as this.没有别的计划, 并说出这样可怕的事情。
该视频共有字幕 148 条。解锁更多字幕为会员功能,请移动到 价格