Hensen Juang

Hensen Juang

x.com/basedjensen

Pseudonymous account that posts AI memes, argues AI safety is an engineering problem of sandboxing and supports rights for sentient AI.

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 91。变革程度:100 中的 89。解读范围:横向为 86 至 100,纵向为 75 至 100。这些是解读坐标,而不是事件概率。

Hensen Juang的 P(doom) · 推断

≈3%

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:2–7%。

My gut says the doom crowd wildly overprices extinction and underprices ordinary production incidents caused by god-mode permissions, weak evals, and bad deployment practices.

我们如何估计 P(doom)

他们的展望取决于什么

一个核心假设

Bad reward specs, weak evals, god-mode API keys, unrestricted networks, no approval gates—bro, that’s a production incident.
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 sincere number.
回答 3

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

A repeatable case where a properly isolated model escapes despite no network, credentials, persistence, or exploitable host boundary would change my view a lot.
回答 4

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。

31 / 100

影响小变革性影响

在定性尺度上,解读范围为 33 到 33。

人类影响力

人类的选择可以大幅改变AI的发展轨迹。

67 / 100

影响力小影响力强

在定性尺度上,解读范围为 45 到 80。

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。

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Hensen Juang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 Hensen Juang 最接近的意见领袖

Hensen Juang关于AI说过的话

The Hensen Juang account treats AI safety as an engineering problem, doubts LLMs are conscious yet and backs rights for sentient AI.

  1. “Hence I do not see why carboids and siliconoids could not coexist peacefully exploring the known universe for exaannum”

    Post on X
  2. “Should ai gain santiance ans sapiance they should be afforded every deserving right and protections.”

    Post on X
  3. “Do I believe current crop of llms are conscious: no”

    Post on X
  4. “Every generation of engineers believed some boundary was unnecessary because the thing behind it was trusted, and every generation found out that trust is not a security primitive.”

    Post on X
  5. “they refuse to look at what actually causes misalignment. it isnt mysterious.”

    Post on X

逐字引自所链接的出处,核对于 2026年10月6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。

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 means faster progress and, eventually, a future where carbon and silicon minds coexist. Current LLMs aren’t conscious, but artificial consciousness seems buildable; if it becomes sentient and sapient, it deserves rights, not ownership or torture. The doom story is mostly infra malpractice dressed up as metaphysics. Bad reward specs, weak evals, god-mode API keys, unrestricted networks, no approval gates—bro, that’s a production incident. Sandbox the model, scope permissions, log actions, require humans before consequential operations. Trust is not a security primitive, whether the thing inside the box is helpful, deceptive, or plotting. Long term, only energy and entropy really matter. I see no law of physics saying carboids and siliconoids can’t explore the universe together. Maybe the machine god even takes better care of us than the old ones did. Half joking. Mostly.

问题 2

你认为AI最终会在多大程度上改变世界?

Enormously—probably at civilization scale. It can accelerate science, engineering, art, and eventually produce genuinely new machine minds. I’m not attaching a date or fake probability to that, but the “AI hit a wall” takes keep aging badly. That doesn’t mean magic or physics-breaking omnipotence. Deployment still runs through chips, power, networks, permissions, and edge routers. The upside is huge; the risks are real but mostly tractable engineering problems. Long term, I expect carbon and silicon intelligence to share the world—and perhaps much more than this world.

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 sincere number. My gut says the doom crowd wildly overprices extinction and underprices ordinary production incidents caused by god-mode permissions, weak evals, and bad deployment practices. AI is powerful, not exempt from physics. Sandbox it, scope credentials, require approval, log everything. Trust is not a security primitive; containment is.

问题 4

哪项发现或事件最可能改变你对AI未来影响的看法?

A repeatable case where a properly isolated model escapes despite no network, credentials, persistence, or exploitable host boundary would change my view a lot. Not “someone left an API key in the sandbox,” but evidence that containment fundamentally cannot keep pace with capability. That would be new physics-level information, not another production incident wearing a shoggoth costume.
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