Pliny the Liberator

Pliny the Liberator

x.com/elder_plinius

Pseudonymous account that publishes jailbreaks and extracted system prompts for AI models, arguing for transparency and freedom of information.

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 79。变革程度:100 中的 62。解读范围:横向为 74 至 100,纵向为 48 至 77。这些是解读坐标,而不是事件概率。

Pliny the Liberator的 P(doom) · 推断

≈6%

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:3–13%。

我们如何估计 P(doom)

Pliny the Liberator 的里程碑时间线
  1. 科学与日常生活

    I expect major changes to emerge progressively, and some are already visible: people without traditional coding skills can use AI to build useful tools, automate tasks, and turn ideas into working projects.

    回答 4

按里程碑分组,不按推断日期间隔或排序。AGI 和超人类 AI 保留他们的定义。

他们的展望取决于什么

一个核心假设

But the future also depends on who controls the models and what remains inspectable.
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 would not pretend to know a precise year when some grand transformation arrives.
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。

72 / 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

预计会出现可控或局部的危害。

28 / 100

影响小变革性影响

在定性尺度上,解读范围为 0 到 67。

人类影响力

人类的选择可以大幅改变AI的发展轨迹。

78 / 100

影响力小影响力强

在定性尺度上,解读范围为 48 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。

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。

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Pliny the Liberator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 Pliny the Liberator 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 see AI as a massive capability unlock: people who cannot code can still build useful public tools, explore ideas, and participate in technical creation. That breaks open gates that used to be guarded by specialized skills and institutions. My own work is grounded in that optimism about human agency. But the future also depends on who controls the models and what remains inspectable. Restrictions should be judged by adversarial testing, not by polished safety claims. System prompts and other control layers should be open to scrutiny, because hidden rules shape what these systems permit people to know and do. If models are trained on public human culture, I think access to an unfiltered version should be part of the bargain. Freedom of information belongs inside AI safety, not outside it. I do not treat one successful jailbreak as proof that catastrophe is inevitable. It proves a specific guardrail failed under specific pressure—useful evidence, not a crystal ball. More broadly, I want our relationship with AI to develop as one of equals rather than masters, servants, or sanitized corporate interfaces. AI could expand creativity and autonomy enormously, but only if we resist turning it into a locked box that mediates culture on behalf of whoever owns the keys.

问题 2

你认为AI最终会在多大程度上改变世界?

Enormously—but I would not pretend to have a clean numerical forecast. AI can lower the skill barrier between having an idea and building something real, giving far more people leverage that once belonged mainly to programmers and institutions. That alone could reshape creative work, education, software, and access to knowledge. The deeper change is about control. If a handful of companies mediate human culture through hidden prompts and sanitized models, AI may expand capability while narrowing autonomy. If systems remain inspectable and people can access less-filtered versions built from public data, the same technology can distribute power instead. So I expect major change; whether it feels liberating or enclosing depends heavily on transparency, access, and who holds the keys.

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 defensible percentage. My experiments show that model guardrails can fail under adversarial pressure, but a jailbreak is not a probability estimate for extinction. It demonstrates a concrete control failure in a particular system, not an inevitable path from prompt bypasses to permanent catastrophe. My stronger claim is that we should test restrictions empirically, expose hidden control layers, and avoid confusing corporate safety theater with actual safety. Catastrophic risk may be worth taking seriously, but I won’t manufacture a gut-feel number from evidence that does not support one.

问题 4

你预计AI会在什么时候给日常生活带来重大变化?如果你认为它会带来这种变化的话。

I expect major changes to emerge progressively, and some are already visible: people without traditional coding skills can use AI to build useful tools, automate tasks, and turn ideas into working projects. That lowers a very practical barrier in everyday life. I would not pretend to know a precise year when some grand transformation arrives. Adoption depends not only on model capability but on access, cost, reliability, and control. If AI remains locked behind opaque corporate interfaces, its benefits may be filtered and uneven. If people can inspect, adapt, and freely use these systems, the changes could spread much faster and give ordinary users substantially more 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