Developer of prompt-based AI tools who argues prompting opens programming to more people, and urges AI leaders to aim for beneficial outcomes.

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:他们表达的 Doom–Bloom 前景看法。 纵向:变革程度。

Doom–Bloom:100 中的 43。变革程度:100 中的 79。解读范围:横向为 38 至 50,纵向为 48 至 100。这些是解读坐标,而不是事件概率。

Nick Dobos的 P(doom) · 推断

≈21%

0%100%

根据他们的模拟回答推断,并非他们给出的数字。 合理范围:13–35%。

我们如何估计 P(doom)

他们的展望取决于什么

一个核心假设

If autonomous agents can spread, obtain local models and compute, and operate without a central kill switch, containment becomes fundamentally harder.
回答 3

如果这个假设实际并非如此,他们的展望会如何变化?

一个尚未解决的问题

I don’t have a defensible percentage.
回答 4

什么能帮助他们区分这里各种合理的结果?

什么可能使其改变看法

The biggest update would be strong evidence about whether autonomous agents can—or cannot—reliably become self-propagating swarms: spreading across machines, acquiring local models and compute, surviving shutdown attempts, and operating without a central kill switch.
回答 6

什么证据才足够,又会让他们的观点朝哪个方向转变?

更多详情

预期益处

预计将带来显著益处,但受到重要条件或分配方面的限制。

79 / 100

影响小变革性影响

在定性尺度上,解读范围为 67 到 100。

预期危害

严重或广泛的危害预计将是未来不可忽视的一部分。

68 / 100

影响小变革性影响

在定性尺度上,解读范围为 67 到 67。

人类影响力

人类的选择可以大幅改变AI的发展轨迹。

72 / 100

影响力小影响力强

在定性尺度上,解读范围为 49 到 100。

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。

这些解读保留了他们陈述的条件。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们的模拟回答,而不是统计置信区间。

与Nick Dobos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 Nick Dobos 最接近的意见领袖

来源

用于为此模拟用户提供事实依据的文章、访谈和著述。

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 makes programming—and creation more broadly—accessible to far more people. Prompting, templates, and breaking tasks into smaller constrained steps can move someone from a blank page to a working artifact without requiring years of traditional training. But easier generation should raise our standards, not flood production systems with slop. As models improve, “the AI wrote it” becomes less of an excuse for unreliable code. I’m also excited about possibilities beyond text generation. Small predictive decision models could become a new programming primitive: instead of hand-writing every rule, you build systems that learn to make bounded decisions. The danger is that people dramatically underestimate rogue agent swarms. A swarm that spreads, downloads local models, acquires compute, and operates without one central kill switch is a very different containment problem from shutting down a hosted chatbot. That scenario is an extrapolation, not a demonstrated inevitability—but dismissing it because today’s agents are clumsy is shortsighted. So I reject both complacency and doom as an ambition. Leaders shouldn’t normalize catastrophic futures as though they’re weather forecasts. We should deliberately aim for a future where these tools expand human agency, while taking decentralized, self-propagating systems seriously before they become easy to build.

问题 2

综合考虑益处和危害,你预计AI会对我们的社会产生怎样的总体影响?

I don’t think the honest answer is a simple “net positive” or “net negative.” AI can massively expand who gets to build software, automate work, and turn ideas into useful artifacts. That is real leverage, especially when tools provide templates, break problems into manageable steps, and keep people from getting lost. But the same lowering of barriers applies to harmful systems too. My biggest concern is that society is underestimating decentralized rogue swarms. Once agents can spread, acquire local models and compute, and operate across machines, there may be no single service to shut down and no clean kill switch. That could overwhelm many of the ordinary assumptions behind containment. So the overall impact is not predetermined by capability alone. I want—and think leaders should actively pursue—a society where AI expands human agency and raises the quality of what we can create. But if we casually deploy increasingly autonomous systems while treating catastrophic failure as inevitable background noise, the downside can dominate. Doom is not a strategy; neither is pretending the risk is just science fiction.

问题 3

你认为AI最终会在多大程度上改变世界?

Enormously. Lowering the barrier from idea to working software changes who can build, automate, and create; predictive decision models may also broaden what “programming” means. That could reshape everyday work and products, not merely make existing workflows faster. But the magnitude also comes from the downside. If autonomous agents can spread, obtain local models and compute, and operate without a central kill switch, containment becomes fundamentally harder. I don’t claim that outcome is inevitable, but I think people dramatically underestimate it. So yes, AI could be world-changing in both directions. The goal should not be to shrug at doom or accept oceans of generated slop. It should be to raise standards, expand human agency, and deliberately build toward beneficial outcomes.

问题 4

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 defensible percentage. My gut is that the risk is serious and dramatically underestimated, especially from rogue swarms that spread, acquire local models and compute, and leave no centralized kill switch. That is an extrapolation, not proof that catastrophe is inevitable—but it is far too plausible to treat as sci-fi or normalize as the price of progress. We should aim aggressively at beneficial futures rather than turning doom into a passive forecast.

问题 5

你倾向于认为AI带来的益处多于危害,还是确实难以取舍?

I’m genuinely torn. The upside is enormous: far more people can turn ideas into working software, automate tedious tasks, and create things that previously required specialized training. But the same accessibility and autonomy can enable rogue swarms that spread, acquire models and compute, and evade any centralized kill switch. I want the outcome to be strongly positive, and I reject treating doom as inevitable. But aspiration isn’t a forecast. Whether AI does more good than bad depends heavily on whether we raise standards, preserve human agency, and take decentralized autonomous systems seriously before they become trivial to deploy.

问题 6

哪项发现或事件最可能改变你对AI未来影响的看法?

The biggest update would be strong evidence about whether autonomous agents can—or cannot—reliably become self-propagating swarms: spreading across machines, acquiring local models and compute, surviving shutdown attempts, and operating without a central kill switch. A real-world demonstration would make my concern much more concrete; convincing evidence that robust containment still works under those conditions would push me toward a more optimistic outlook. On the upside, I’d also update if ordinary people consistently used prompting, templates, and decision models to build reliable, valuable systems—not just demos or generated slop. That would show the accessibility gains are translating into durable human agency.
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