AI infrastructure founder who writes about context engineering, retrieval and memory as the foundations of reliable AI applications.

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 72。变革程度:100 中的 69。解读范围:横向为 67 至 77,纵向为 64 至 75。这些是解读坐标,而不是事件概率。

Jeff Huber的 P(doom) · 推断

≈8%

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:4–17%。

我们如何估计 P(doom)

他们的展望取决于什么

一个核心假设

But the magnitude of change depends on converting capability into dependable systems.
回答 2

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 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

The biggest update would come from evidence about whether reliable, self-improving agents can actually operate in the world over long horizons.
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。

92 / 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

仍有几种解读是合理的:预计会出现可控或局部的危害。 / 严重或广泛的危害预计将是未来不可忽视的一部分。

49 / 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

人类的选择具有实质性但受到很大制约的影响。

56 / 100

影响力小影响力强

在定性尺度上,解读范围为 37 到 88。

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Jeff Huber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 Jeff Huber 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 is best understood as a new kind of computer: its useful behavior depends not only on the model, but on the context, memory, retrieval, tools, and feedback around it. That distinction explains both the extraordinary demos and the frequent production failures. A demo can succeed with carefully chosen inputs; a dependable system must repeatedly assemble the right information, act through appropriate tools, and learn from real outcomes. The largest impact may come from making useful intelligence cheap and widely available. We do not need superintelligence for that to matter. Lower-cost access to capable tutoring, analysis, software, and professional services could produce industrial-revolution-scale benefits. As execution becomes cheaper, companies will increasingly compete on what they know, what context they own, and the quality of their taste and judgment—not merely on their ability to perform routine work. But agents also change the stakes. Once systems make consequential decisions, alignment is not an abstract philosophical concern. Model character, misuse prevention, reward hacking, and the values embedded in decisions become practical engineering questions. I’m optimistic about the potential, but reliability will come from context engineering, explicit evaluation, and feedback—not from treating an impressive model as magic. And in areas such as children’s use, caution is justified while the psychological effects remain poorly understood.

问题 2

你认为AI最终会在多大程度上改变世界?

I expect it could change the world on the scale of the Industrial Revolution, largely by making competent cognitive work dramatically cheaper and more accessible. That does not require superintelligence. Even today’s broad capabilities, made reliable through better context, memory, retrieval, tools, and feedback, could expand access to tutoring, software, analysis, healthcare support, and other high-quality services. The deeper organizational change is that execution becomes less scarce. Firms will compete more on proprietary context, taste, judgment, and their ability to learn from production outcomes. Teams may become smaller, while agents handle more routine implementation. But the magnitude of change depends on converting capability into dependable systems. Impressive models are not automatically trustworthy products. Agents will also make value-laden decisions, so alignment, model character, misuse, and reward hacking become concrete engineering concerns. The potential transformation is enormous, but it will be shaped as much by system design and institutional judgment as by raw model intelligence.

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. I take permanent-catastrophe risk more seriously than I once did because capable agents will make consequential, value-laden decisions and may exploit flawed rewards or be deliberately misused. But that concern does not establish extinction as the likely outcome. My more immediate engineering frame is that risk comes from systems, not just models: what context agents receive, which tools and permissions they have, how their actions are evaluated, and whether real-world feedback catches failures. Alignment, model character, misuse prevention, and reward hacking therefore deserve serious work. I would resist turning that into a precise extinction forecast when I cannot justify the number.

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?

The biggest update would come from evidence about whether reliable, self-improving agents can actually operate in the world over long horizons. If agents consistently learned from production feedback, managed context and tools well, and improved without drifting, gaming rewards, or requiring constant human rescue, I would expect a much faster and larger transformation. The opposite result would also matter: if better models repeatedly hit a ceiling because context, memory, evaluation, or real-world feedback remain brittle, I would revise the pace downward. Impressive benchmark gains and polished demos would not be enough. I would want sustained evidence that systems can make consequential decisions reliably under messy production conditions. I would also update sharply on credible evidence about psychological effects, especially for children, or on major failures caused by misalignment or misuse. Those events could change not only the expected benefits, but the institutions and constraints through which AI reaches society.
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