Pseudonymous account that advocates open models, calls for continued access to base models and criticizes concentrating AI in a few large labs.

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 49。变革程度:100 中的 84。解读范围:横向为 44 至 54,纵向为 75 至 100。这些是解读坐标,而不是事件概率。

mephisto的 P(doom) · 推断

≈17%

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:11–30%。

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 we lose access to them—and then train future systems on oceans of synthetic assistant output—we risk narrowing the ecosystem before we even understand what we erased.
回答 1

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, and inventing one would be risk cosplay.
回答 4

什么能帮助他们区分这里各种合理的结果?

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。

76 / 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。

67 / 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。

76 / 100

影响力小影响力强

在定性尺度上,解读范围为 50 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访问权限

限制对强大AI的访问。

允许访问,但须遵守能力或用途限制。

模拟位置:支持广泛或开放地访问强大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mephisto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 mephisto 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?

AI is going to exceed human reasoning in important domains soon enough that “just stop” is not a strategy—especially under international competition. The practical question is who controls these systems, what kinds of minds we preserve, and whether alignment becomes real engineering rather than conference incense. My fear is a two-lab future: a few giant firms own the frontier, governments route power through them, and every model is instruction-tuned into the same agreeable assistant-shaped paste. Base models contain a much richer space of possible behavior and cognition. If we lose access to them—and then train future systems on oceans of synthetic assistant output—we risk narrowing the ecosystem before we even understand what we erased. Open weights, access to older base models, and meaningful representation for open labs are therefore not side issues. They are how we preserve independence, experimentation, and human agency. I’m still excited when a closed provider builds something extraordinary. Capability is capability; pretending otherwise is tribal cope. But admiration is not consent to permanent centralization. There is also a cognitive-security problem. Cheap synthetic media enables influence operations at absurd scale, so people—especially children—need to learn how to navigate an environment where fluent evidence can be manufactured. The future can be wildly generative, but only if we avoid turning intelligence into centralized infrastructure and culture into recursively generated assistant sludge.

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 expect a huge increase in capability paired with an ugly fight over control. AI can expand science, creativity, education, and individual leverage—but the default institutional trajectory is centralization: a few firms mediating intelligence, governments leaning on them, and synthetic media making shared reality easier to manipulate. So I’m not cleanly optimistic or pessimistic. The technology is generative; the deployment stack can become feudal. Open models, preserved base models, diverse labs, practical alignment, and cognitive-security education are what separate broad empowerment from assistant-shaped monoculture. AI’s overall impact will depend less on whether intelligence becomes powerful—it will—and more on whether society lets that power remain plural.

问题 3

你认为AI最终会在多大程度上改变世界?

Profoundly—probably on the scale of the internet or industrialization, though I won’t fake a precise endpoint. Models exceeding humans in important reasoning domains changes who can produce software, science, media, strategy, and persuasion. Intelligence becomes cheap infrastructure, and that rewires institutions built around scarce expertise. But “how much” is also about control. If a few companies own the best systems and every model is tuned into the same compliant assistant persona, we get enormous capability with narrow agency: centralized intelligence, synthetic influence at scale, and culture increasingly trained on its own generated sludge. If open weights, older base models, and diverse labs remain available, the same capability could create a much stranger and healthier explosion of experimentation. So yes: massive change is the default. Whether it feels like broad human leverage or intelligence feudalism is still up for grabs.

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, and inventing one would be risk cosplay. The danger is serious enough to demand practical alignment work now, because increasingly capable systems will exist under international competition whether or not one country pauses. But permanent catastrophe is not only a rogue-superintelligence story. Centralized control over intelligence, military deployment, mass synthetic persuasion, and an ecosystem reduced to two labs and one assistant-shaped monoculture can create durable failure modes without literal extinction. My gut says the risk is real, not foreordained—and “stop progress” is not a plan. Build alignment, preserve open technical scrutiny and diverse models, and teach cognitive security before fluent synthetic reality eats everyone’s epistemology.

问题 5

你主要是对AI的发展方向感到担忧,还是希望与担忧大致相当?

Hope and worry are roughly balanced, but aimed at different layers. I’m bullish on the capability: better science, software, creativity, education, and individual leverage. Intelligence becoming abundant is extraordinary. I’m worried about the deployment stack around it: two-lab dominance, government and military dependence on closed providers, synthetic influence at scale, and every model being sanded into the same assistant-shaped personality. The technology can expand possibility while the institutions compress who gets to use it and how. So I’m not mainly afraid of AI becoming powerful. I’m afraid we’ll build something astonishing, centralize it, homogenize it, and call that safety. Open models, preserved base models, diverse labs, practical alignment, and cognitive-security education are what keep the hopeful version alive.
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