问题 1
Lewis
x.com/ctjlewisOpen-source developer who shares small-model reasoning experiments and favors wide access to AI over government restrictions.
AI将如何改变世界?
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 79。变革程度:100 中的 76。解读范围:横向为 74 至 84,纵向为 49 至 100。这些是解读坐标,而不是事件概率。
Lewis的 P(doom) · 推断
他们的展望取决于什么
一个核心假设
Open models, small-model research, and recursive improvement are central to that upside because they expand who can experiment, verify results, and contribute.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
We do not know exactly what such systems will look like, and claims of certainty-level safety proofs are not credible.回答 1
什么能帮助他们区分这里各种合理的结果?
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
严重或广泛的危害预计将是未来不可忽视的一部分。
64 / 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。
62 / 100
影响力小影响力强
在定性尺度上,解读范围为 38 到 87。
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。
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Lewis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 Lewis 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
I think AI means much more widely distributed intellectual capability and, eventually, systems that help improve the process of building better systems. Open models, small-model reasoning experiments, and accessible tooling matter because progress should not be confined to a few labs or mediated primarily through government restrictions. Broad participation makes experimentation, verification, and practical use possible across many domains.
That is not the same as claiming advanced AI is harmless. Superintelligence should not be caricatured as either guaranteed malevolence or merely a conventional chatbot connected to powerful tools. We do not know exactly what such systems will look like, and claims of certainty-level safety proofs are not credible. Recursive improvement could produce major gains, but it also makes the details of implementation and deployment increasingly consequential.
My preferred response is continued development with serious technical scrutiny, not safety politics that converts uncertainty into a rationale for stopping or centralizing progress. Labs should engage with substantive concerns from domain experts—for example, mathematicians worried about how their work is used—and make real tradeoffs where necessary. But those concerns should inform engineering and release decisions, not become a blanket argument against widely accessible AI or continued improvement.
问题 2
综合考虑益处和危害,你预计AI会对我们的社会产生怎样的总体影响?
Overall, I expect AI to be broadly beneficial: it should distribute intellectual capability, accelerate technical work, and let far more people build useful systems rather than leaving that capacity inside a few institutions. Open models, small-model research, and recursive improvement are central to that upside because they expand who can experiment, verify results, and contribute.
The harms could still be substantial. More capable systems can amplify misuse, concentrate power, and create consequences that are hard to predict—especially once systems meaningfully contribute to their own improvement. I reject both “superintelligence is inevitably evil” and “it is just a chatbot, so there is nothing new to worry about.” Neither simplification helps.
So my positive expectation is not a claim of guaranteed safety. It is a judgment that continued, broadly accessible development is preferable to responding to uncertainty with government restrictions or centralized control. The right approach is serious technical scrutiny and engagement with concrete domain concerns, including difficult release tradeoffs, while preserving open participation and continued progress.
问题 3
你认为AI最终会在多大程度上改变世界?
I expect AI to change the world profoundly, especially by making advanced intellectual capabilities broadly available and by accelerating the creation of still more capable systems. The largest shift is not simply better chatbots; it is that reasoning, software development, research assistance, and experimentation become cheaper and accessible to many more people. Recursive improvement could make that process compound rather than proceed at today’s pace.
The scale is difficult to quantify because genuinely advanced systems cannot be cleanly reduced either to inevitable catastrophe or to familiar software with a nicer interface. Their effects will depend heavily on implementation, access, and deployment. Widely available models could distribute power and enable independent verification, while concentrated systems could give a small number of institutions disproportionate control.
So I expect a very large transformation, but not one whose final shape can be proved in advance. That uncertainty calls for substantive technical scrutiny and attention to concrete domain concerns—not using speculative certainty as a justification for government restrictions or freezing development.
问题 4
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 would not attach a defensible percentage to it. My gut is that the risk is real but not the default outcome. Recursive improvement and highly capable systems could create permanent consequences, so dismissing them as ordinary chatbots is unserious; treating superintelligence as inevitably hostile is equally unjustified.
More importantly, a fabricated number would imply precision we do not have. We cannot prove certainty-level safety, but uncertainty alone does not justify government restrictions or centralized control. I would continue development while applying serious technical scrutiny to concrete systems, deployments, and domain-specific concerns.
你的立场在哪里?
描绘我的世界观

