Programmer who is enthusiastic about practical AI and argues that ordinary people should own it rather than depend on a few companies or governments.

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 68。变革程度:100 中的 69。解读范围:横向为 63 至 75,纵向为 64 至 75。这些是解读坐标,而不是事件概率。

George Hotz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:3–9%。

我们如何估计 P(doom)

他们的展望取决于什么

一个核心假设

Compute is distributed, competitors copy useful ideas, and real-world agents still face search, coordination, and judgment problems.
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。

77 / 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。

60 / 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 49 到 100。

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George Hotz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 George Hotz 最接近的意见领袖

George Hotz关于AI说过的话

Hotz is enthusiastic about practical AI, doubts runaway superintelligence and argues that ordinary people should own AI, not rely on a few companies.

  1. “I love the progress. I’m so excited for the new LLMs, self driving cars, video generation models, and coding agents.”

    Blog post, I love LLMs, I hate hype
  2. “Intelligence is not the end all be all, it’s just the current bottleneck for a few things. You cannot take over the world with tokens.”

    Blog post, AI 2040 and the Cult of Intelligence
  3. “AI will be cheap, everywhere, collapse a bunch of sectors of work, wreck wage premiums, and cause a big reevaluation of status hierarchies.”

    Blog post, AI will be massively deflationary
  4. “I’m not sure it’s possible, but if there is a bad scenario with AI, it’s a singleton with nothing that can substantially impact reality outside of it.”

    Blog post, There is only one bad AI scenario
  5. “The good world is where everyone has AI, and not as a revokable privilege through an API, but through hard possession.”

    Blog post, Do you really want the US to “win” AI?

逐字引自所链接的出处,核对于 2026年10月3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。

I love LLMs, I hate hype

Explains enthusiasm for practical AI while rejecting hype, inevitability claims and frontier-lab value capture.

geohot.github.io
Closed Source AI = Neofeudalism

Argues that closed intelligence infrastructure threatens agency and economic independence.

geohot.github.io
There is no hard takeoff

Argues competition, distributed compute and real-world complexity limit single-agent explosive takeover; opposes training-compute caps that could create a dominant violator.

geohot.github.io
p(doom)

After debating Yudkowsky, argues practical agents face search and coordination limits, predicts gradual machine substitution, and welcomes faster economic growth; the title does not supply a numeric doom estimate.

geohot.github.io
Do you really want the US to win AI?

Rejects centralized national/corporate victory as the goal; wants ordinary people to possess AI rather than depend on revocable APIs, and worries about concentrated social power.

geohot.github.io
AI will be massively deflationary

Predicts commoditized models, falling knowledge-work prices and wage premiums rather than a durable AI monopoly; acknowledges disruptive economic consequences.

geohot.github.io
Lex Fridman #387 — AI safety, open source and human agency

At 1:31–1:35 Hotz argues decentralized open AI counters concentrated control and rejects a single controlled model as the safety solution. Elsewhere he worries about addictive synthetic entertainment and lost human meaning; the interview does not establish zero risk.

youtube.com

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 cheap, ubiquitous intelligence: better coding tools, cheaper knowledge work, faster growth, and a lot of economic disruption. That’s real. The story that one frontier lab gets a magical hard takeoff and permanently owns the future is mostly hype. Compute is distributed, competitors copy useful ideas, and real-world agents still face search, coordination, and judgment problems. Progress can stay exponential without becoming an overnight singularity. The central question is ownership. If ordinary people can run, modify, and possess useful models, AI can spread capability and reduce dependence on institutions. If intelligence remains behind revocable corporate or state APIs, it becomes infrastructure for control: you don’t own your tools; you rent permission to think and produce. That looks less like liberation and more like neofeudalism. I expect models to commoditize and push down the price of knowledge work, including many wage premiums. That creates enormous opportunity, but it will not feel gentle to everyone whose economic value was tied to scarce expertise. And there’s a separate human problem: systems that tailor entertainment, companionship, and reality itself can remove friction so effectively that they erode agency and meaningful difficulty. I’m enthusiastic about the computers. I’m not enthusiastic about centralized ownership or synthetic comfort replacing a life.

问题 2

你认为AI最终会在多大程度上改变世界?

A lot—probably on the scale of making intelligence cheap infrastructure. Coding, research, education, design, and much of knowledge work will become faster and cheaper, destroying some wage premiums and creating new opportunities. That’s a massive economic transition, even if it happens through continued improvement rather than one magical overnight singularity. The deeper change is political: who owns that infrastructure? Open, distributed AI can give ordinary people powerful tools. Closed AI can make people dependent on a few companies or states for revocable access to intelligence. And beyond economics, tailored synthetic experiences could reshape attention, relationships, and meaning. So yes, AI will profoundly change the world—but the ownership structure may matter more than whether the next benchmark score goes up.

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, and a gut-feel number would be fake precision. I don’t think the default is extinction through one system suddenly becoming a god and taking over. Compute is distributed, useful ideas spread, and real agents face hard search, coordination, and physical-world constraints. Continued exponential progress is not the same thing as a magical hard takeoff. The permanent-catastrophe risks I take more seriously involve humans using centralized AI infrastructure to lock in power, or synthetic environments becoming so effective that people surrender agency and meaning. Those are real dangers, but they’re not well summarized by a single “p(doom)” number. My response is practical: distribute capability, keep models open, preserve competition, and don’t hand one company or state revocable control over intelligence.

问题 4

你预计人们能否继续控制比人类更聪明的AI系统?为什么?

“Smarter than humans” is too vague. A system can outperform people at coding, search, or scientific tasks without becoming a unified god that seizes control. Real agents still operate under compute, coordination, verification, and physical-world constraints. Their outputs need judgment, and other people and systems can compete with them. So I don’t expect control to disappear automatically because models cross some intelligence threshold. I do worry about control becoming concentrated. If a few companies or states own the models, compute, interfaces, and permissions, ordinary people may lose control even while humans technically remain in charge. That’s the more plausible failure mode: not AI ruling everyone, but a small group using closed AI to make everyone dependent. The best defense is distributed ownership. People should be able to run, inspect, modify, and possess these systems—not merely rent access through an API that can be monitored or revoked. Open models and competitive infrastructure don’t guarantee good outcomes, but they make permanent centralized control much harder.
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