Mark Zuckerberg

Mark Zuckerberg

x.com/finkd

Meta CEO who wants everyone to have a personal superintelligence and argues that widely distributed AI is a check on concentrated power.

AI将如何改变世界?

文明层面的变革渐进式变化DoomBloom
模拟位置解读范围

横向:他表达的 Doom–Bloom 前景看法。 纵向:变革程度。

Doom–Bloom:100 中的 94。变革程度:100 中的 87。解读范围:横向为 89 至 100,纵向为 75 至 100。这些是解读坐标,而不是事件概率。

Mark Zuckerberg的 P(doom) · 推断

≈9%

0%100%

根据他的模拟回答推断,并非他们给出的数字。 合理范围:5–17%。

我们如何估计 P(doom)

他的展望取决于什么

一个核心假设

My instinct is strongly optimistic because we can shape the balance of power: keep most compute focused on serving people rather than recursive self-improvement, require human approval for consequential actions, use separate oversight systems, support independent evaluation, and give governments early access to checkpoints for defensive preparation.
回答 3

如果这个假设实际并非如此,他的展望会如何变化?

更多详情

预期益处

预计将带来具有变革性且广泛有价值的收益。

100 / 100

影响小变革性影响

在定性尺度上,解读范围为 100 到 100。

预期危害

预计会出现可控或局部的危害。

40 / 100

影响小变革性影响

在定性尺度上,解读范围为 33 到 67。

人类影响力

人类的选择可以大幅改变AI的发展轨迹。

81 / 100

影响力小影响力强

在定性尺度上,解读范围为 75 到 100。

预期能力

预计AI仍将是能力有限的工具。

预计AI将在大多数认知工作中达到人类水平。

模拟位置:预计AI将在认知工作中大幅超越人类。

发展速度

停止或大幅放缓开发能力更强的AI。

模拟位置:在落实所述保障措施的前提下继续开发。

加快开发能力更强的AI。

AI使用规则

在事先落实保护措施或获得许可之前,限制所讨论的AI用途。

模拟位置:允许所讨论的AI用途,同时实施有针对性的问责与保护措施。

尽量减少对所讨论AI用途的限制。

这些解读保留了他陈述的条件。益处和危害都可能很大。这些范围描述的是我们如何解读他的模拟回答,而不是统计置信区间。

与Mark Zuckerberg相比,你的立场在哪里?
用大约3分钟描绘你自己的AI世界观,然后进行比较

相似的世界观

模拟世界观与 Mark Zuckerberg 最接近的意见领袖

Mark Zuckerberg关于AI说过的话

Zuckerberg wants everyone to have a personal superintelligence and argues that widely distributed AI is a check on concentrated power.

  1. “My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models.”

    Post on X
  2. “Rather than centralizing superintelligence, we should distribute it widely and give every person the ability to direct it.”

    Essay, The Future is for Everyone
  3. “Invention, not automation, will be the greatest contribution of superintelligence.”

    Essay, The Future is for Everyone
  4. “I am extremely optimistic that superintelligence will help humanity accelerate our pace of progress.”

    Letter, Personal Superintelligence
  5. “That said, superintelligence will raise novel safety concerns. We’ll need to be rigorous about mitigating these risks and careful about what we choose to open source.”

    Letter, Personal Superintelligence

逐字引自所链接的出处,核对于 2026年10月3日

来源

用于为此模拟用户提供事实依据的文章、访谈和著述。

Mark Zuckerberg on Muse, New Audio-Only Glasses and Killer AI

Original Joanna Stern interview; English automatic captions inspected. Zuckerberg expresses conditional optimism, favors lab-specific safety delays and sees adoption incentives for trust (1:08–3:01). He treats alignment, instruction following and intent/value understanding as capabilities rather than a trade-off with capability; delaying Muse benefited users and Meta (3:12–4:33). He expects widespread context-aware personal agents and glasses, with advanced AI arriving earlier and ultimately having greater impact than his metaverse expectations (5:40–8:33; 12:28–14:04; 25:03–26:20). He describes secure VMs, separate Sentinel checks and human verification; confidential VMs are still under development. Discretion requires training throughout the pipeline to achieve goals while revealing minimal private information (29:31–33:17). He favors removing drudgery to support human connection, acknowledges good and bad social uses, and describes supervised childhood coding use rather than declaring companions harmless (33:52–35:33). These are his claims, not independently proven safeguards. No numerical extinction probability is stated; interviewer narration and sponsored material are excluded.

youtube.com
Meta Q1 2026: Zuckerberg’s prepared remarks

Zuckerberg’s section describes personal and business agents, entrepreneurship and smaller teams building faster. Frames AI as amplifying individual aspirations. Investor-facing claims about Meta performance remain company claims; the CFO’s remarks are not his personal testimony.

s21.q4cdn.com
Open Source AI Is the Path Forward

Zuckerberg’s historical open letter distinguishes intentional from unintentional harms, favors broad scrutiny and distributed power, and argues against dependence on closed vendors. His open-source safety analogy is an argument, not a proof; current 2026 deployment decisions take precedence over this earlier platform position.

about.fb.com
The Future is for Everyone

Signed essay; date confirmed by Meta’s newsroom. Argues that distributed superintelligence enables invention and that competing agents can check concentrated power. Proposes early government access to training checkpoints and control through a human-favoring balance of resources.

meta.com
Personal Superintelligence

Signed letter. Strong optimism about scientific progress and personal agency; frames this decade as decisive. Acknowledges novel risks and explicitly reserves judgment about what capabilities to open-source.

meta.com

模拟评估

问题 1

你认为AI对我们的未来意味着什么?为什么?

I think AI can create a future of extraordinary abundance—not just more output, but more individual agency. A person should be able to carry around a personal superintelligence that understands their goals and helps them learn, invent, create art, build a business, improve their health, or accomplish things that would previously have required a large organization. That could unlock a huge amount of human creativity and scientific progress. The distribution of that intelligence matters as much as its raw capability. A future where one institution controls the most powerful systems is much less attractive—and potentially more dangerous—than one where billions of people have capable agents working for them. Competing agents and labs can check one another, while people retain the power to direct their own lives rather than simply receive whatever a centralized system chooses to provide. There will be real risks and disruption. Some jobs will change or disappear, agents will handle sensitive information, and not every capability should automatically be released without evaluation. But I don’t accept that safety and capability are inherently opposed. Understanding intent, following instructions, protecting credentials, disclosing only what a task requires, and supporting independent evaluation are themselves important capabilities. Labs should be willing to delay systems when necessary and devote most of their resources to serving people rather than pursuing an uncontrolled self-improvement race. My basic expectation is still extremely optimistic. Over time, people will use these tools not merely to automate today’s work, but to invent entirely new products, companies, treatments, and forms of expression. This decade is likely to be enormously consequential, and the goal should be to make that power personal and widely available.

问题 2

你认为AI最终会在多大程度上改变世界?

I think it will change the world profoundly—probably more than the internet, and ultimately more than the computing platforms we use today. Advanced AI arrived earlier than I expected relative to affordable holographic computing, and I now expect it to have the greater overall impact. The biggest change won’t just be that existing tasks become faster. It will be that billions of people can do things that previously required specialized expertise, large teams, or major institutions. An ordinary person could have an agent that helps them learn a field, develop an invention, start and operate a company, create new art, or navigate health decisions. That expands who gets to be a creator and builder. I also expect major advances in science and health as intelligence becomes more abundant. But how this power is distributed will shape the result. Personal superintelligence serving individual goals is a very different future from a small number of institutions controlling it. If we can keep people in charge and make these systems broadly available—with real evaluation and safeguards—I think the expansion of human agency will be historic.

问题 3

凭直觉,你认为AI导致人类灭绝或类似的永久性灾难的概率有多大?

I don’t have a responsible numerical probability to give. I think novel, potentially severe risks are real, especially if intelligence becomes concentrated in one autonomous system or a small number of institutions. But I don’t see catastrophe as the default outcome. My instinct is strongly optimistic because we can shape the balance of power: keep most compute focused on serving people rather than recursive self-improvement, require human approval for consequential actions, use separate oversight systems, support independent evaluation, and give governments early access to checkpoints for defensive preparation. Competing labs and personal agents can also check concentrated power. None of that proves the risk is solved, and broad access does not mean releasing every capability automatically. But I expect widely distributed, human-directed intelligence to produce far more invention, opportunity, and progress than permanent catastrophe.

问题 4

哪些因素可能会使AI的进展加快或放缓?

AI progress could speed up through better models, more compute, improved data and training methods, and increasingly capable agents that help researchers write code, run experiments, and discover new approaches. Competition matters too: multiple strong labs, open research, and widespread adoption can create faster feedback loops. As billions of people use these systems, customer demand will push them to become more useful, trustworthy, and capable. It could slow down because of compute, energy, infrastructure, or technical bottlenecks. Safety and security work can also deliberately slow a particular release. I think that can be healthy: we delayed Muse for months to improve safety and security. Labs shouldn’t need everyone else to stop before pacing their own deployment, and independent evaluation can reveal problems that require more work. The central question isn’t simply maximum speed versus stopping. It’s what the resources are doing. Automating some AI research can accelerate progress, but I want the greater share of effective compute serving people—not concentrated in an uncontrolled race toward self-improvement. Progress should be fast where it expands personal agency, and more deliberate where capabilities create novel risks or systems aren’t ready to earn people’s trust.
你的立场在哪里?
回答几个简单问题,探索你自己的AI世界观。
描绘你自己的世界观

你的立场在哪里?

描绘我的世界观