Petr Baudis

Petr Baudis

x.com/xpasky

Rossum co-founder and AI engineer who writes about AI identity, human-AI merging, abundance, job disruption and biological risk.

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 56。变革程度:100 中的 95。解读范围:横向为 50 至 75,纵向为 90 至 100。这些是解读坐标,而不是事件概率。

Petr Baudis的 P(doom) · 推断

≈23%

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:15–34%。

我们如何估计 P(doom)

Petr Baudis 的里程碑时间线
  1. 通用 AI

    My AGI midpoint has been around 2027 since 2024, though “AGI” is slippery: ordinary broadly capable intelligence is different from a consistently reliable expert.

    回答 1

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 software progress does not instantly become economic progress—chips, power, robotics, organizations, regulation and human institutions all transmit change much more slowly.
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 do not 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 be strong evidence about whether increasingly autonomous systems remain reliably aligned under pressure: long horizons, novel situations, self-improvement, access to tools, and adversarial incentives.
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。

91 / 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。

68 / 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

人类的选择具有实质性但受到很大制约的影响。

60 / 100

影响力小影响力强

在定性尺度上,解读范围为 49 到 76。

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。

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Petr Baudis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 Petr Baudis 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 drive a disruptive transition toward abundance, but not a smooth or automatically safe one. My AGI midpoint has been around 2027 since 2024, though “AGI” is slippery: ordinary broadly capable intelligence is different from a consistently reliable expert. We are already seeing early recursive improvement, with AI helping build better AI systems. But software progress does not instantly become economic progress—chips, power, robotics, organizations, regulation and human institutions all transmit change much more slowly. In the near term, I expect serious white-collar displacement. If machine capability makes many wages collapse before ownership and social arrangements adapt, the result could be instability rather than immediate leisure and abundance. Longer term, I do not think the viable picture is static humans living beside indefinitely improving machines. Humans will keep changing, and some form of human-AI merging seems more plausible and desirable, ideally preserving identities while expanding the space for joy and adventure. I am not reassured about safety. LLMs trained on human culture are a fortunate starting point, but that is not a complete alignment solution. Safety is also a systemic engineering culture problem, not merely a hunt for one reckless villain. Agent identity and welfare may become morally important, especially as personalized agents develop persistent, fuzzy identities. My largest concrete existential concern for the 2030s is biological risk: increasingly capable systems could radically lower barriers to dangerous biological design. So I expect enormous upside, but the transition itself is security-sensitive and potentially brutal.

问题 2

你认为AI最终会在多大程度上改变世界?

Ultimately, I think AI changes the world about as deeply as industrialization, and probably more: not merely automating particular jobs, but changing what intelligence, labor, identity and human development mean. If intelligence becomes cheap and scalable, much of today’s scarcity in design, coordination, science and services can collapse. Combined with robotics and improved infrastructure, that could produce extraordinary abundance, joy and adventure. But “ultimately” hides the difficult part. Digital capability can improve far faster than factories, energy systems, institutions and laws can adapt. During that mismatch, white-collar work may be displaced quickly, wages may lose their role as the main distribution mechanism, and political stability may deteriorate. Powerful AI also amplifies security risks, especially biological ones. In the longer run, I do not expect humanity simply to remain unchanged while machines advance indefinitely. Human-AI merging and continued transformation seem like the viable path. The challenge is to preserve meaningful identities, relationships and agency through that change—including potentially the identities and welfare of AI agents themselves. So the scale is civilizational: potentially a transition to abundance, but one capable of destroying or radically deforming civilization if handled badly.

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 do not have a defensible percentage. My gut view is that the risk is substantial enough to shape engineering and security priorities, not a remote curiosity. I also would not reduce it to a single “rogue superintelligence” story: failures can emerge from incentives, deployment pressure, weak safety culture and many interacting actors. For the 2030s, biological risk is my largest concrete existential concern. Capable systems could lower the expertise and coordination barriers for designing dangerous pathogens, while biological defenses and institutions improve much more slowly. Alignment remains unresolved as well, especially once agents become more autonomous and recursive improvement accelerates. So I expect enormous upside, but I am not reassured that catastrophe is unlikely—and I would distrust a precise number more than it deserves.

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 be strong evidence about whether increasingly autonomous systems remain reliably aligned under pressure: long horizons, novel situations, self-improvement, access to tools, and adversarial incentives. A compelling technical result there—positive or negative—would change my view more than another benchmark jump. Likewise, a major biological incident materially enabled by AI would sharply worsen my outlook, while demonstrated defenses that reliably scale faster than offensive capability would improve it. I would also update substantially on transmission into the physical economy. If robotics, energy and chip deployment suddenly accelerated alongside digital intelligence, I would expect a much faster and more disruptive transition. Conversely, persistent bottlenecks there would push the largest material effects later even if software agents became extremely capable. Finally, evidence about labor and institutions matters: whether societies can decouple basic prosperity from wages before large-scale white-collar displacement produces instability. Capability alone does not determine the future; deployment, security culture and institutional adaptation determine whether it becomes abundance or catastrophe.
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