Erik Bernhardsson

Erik Bernhardsson

x.com/bernhardsson

Founder of the cloud infrastructure company Modal who writes about compute, GPU economics and how AI changes the software business.

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 49。解读范围:横向为 67 至 77,纵向为 44 至 54。这些是解读坐标,而不是事件概率。

Erik Bernhardsson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:低于 10%。

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 realizing that upside requires a healthy ecosystem: practical developer infrastructure, flexible compute, open-source collaboration, and independent companies that preserve long-term competition.
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?

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。

67 / 100

影响小变革性影响

在定性尺度上,解读范围为 67 到 67。

预期危害

预计会出现可控或局部的危害。

35 / 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

人类的选择具有实质性但受到很大制约的影响。

52 / 100

影响力小影响力强

在定性尺度上,解读范围为 44 到 81。

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Erik Bernhardsson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 Erik Bernhardsson 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 expect AI to be a major productivity technology, but its effects will be shaped by infrastructure and economics, not capability alone. A model may make writing code dramatically cheaper, yet teams still need compute, deployment, data systems, monitoring, and reliable tools. GPU access is particularly awkward because users want flexibility while suppliers often prefer large, long-term commitments. Closing that mismatch determines which ideas become practical products. I also don’t think cheaper coding means every company will build all its software internally. Software vendors get the same productivity gains, can spread infrastructure and maintenance costs across customers, and can develop better product judgment. Knowledge with long feedback loops—where you only learn whether a decision was good months or years later—also remains valuable because it cannot be reproduced merely by generating code quickly. The upside is concrete rather than abstract. Better access to AI and compute can accelerate useful software and scientific work, including computational biology and potentially medicine discovery. But realizing that upside requires a healthy ecosystem: practical developer infrastructure, flexible compute, open-source collaboration, and independent companies that preserve long-term competition.

问题 2

你认为AI最终会在多大程度上改变世界?

A great deal, but “how much” is difficult to compress into a credible number. I expect broad productivity gains because AI lowers the cost of producing software and applying computation to many problems. The most important outcomes may be concrete ones: better tools, faster scientific work, and progress in areas such as computational biology and medicine discovery. But capability does not translate automatically into impact. Compute access, deployment infrastructure, data, maintenance, and product judgment still constrain what gets built. Cheap code also does not imply that every company becomes self-sufficient; specialized vendors can use the same AI gains while spreading infrastructure and operating costs across many customers. So I expect substantial change, but unevenly and through institutions and markets—not as an overnight replacement of the existing economy. The shape of that change will depend heavily on practical infrastructure and sustained competition.

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 focus has been on nearer-term infrastructure, productivity, science, and competition, so I wouldn’t pretend those views establish an extinction-risk forecast.

问题 4

如果有的话,你预计AI会造成哪些重大危害?

The clearest harms I expect come from concentration and uneven access. If compute is locked behind a few vendors, giant fixed commitments, or vertically integrated platforms, smaller teams face worse economics and the ecosystem becomes less competitive. That can determine who gets to build and capture value even when the underlying models are widely available. AI will also create labor and business disruption as software production becomes cheaper, although cheap coding does not mean software companies disappear or every firm builds everything internally. Reliability, deployment, maintenance, and product judgment remain real constraints, and knowledge with long feedback loops retains value. Beyond that, I would not claim a detailed catalog or quantified forecast that I cannot support. My strongest view is that infrastructure design and market structure will heavily shape whether AI’s gains are broadly useful or captured by a small number of companies.
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