问题 1
doomslide
x.com/doomslidePseudonymous account that writes about AI and mathematics, favors open models and criticizes concentrated control of AI knowledge and infrastructure.
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 25。变革程度:100 中的 71。解读范围:横向为 20 至 30,纵向为 50 至 76。这些是解读坐标,而不是事件概率。
doomslide的 P(doom) · 推断
他们的展望取决于什么
一个核心假设
If the decisive models, infrastructure, and interactions remain proprietary, capability can rise while society becomes less able to inspect claims, attribute knowledge, or participate in producing it.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?
更多详情
预期益处
仍有几种解读是合理的:预计收益有限,或仅分布在较小范围内。 / 预计将带来显著益处,但受到重要条件或分配方面的限制。
48 / 100
影响小变革性影响
在定性尺度上,解读范围为 33 到 67。
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。
66 / 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。
58 / 100
影响力小影响力强
在定性尺度上,解读范围为 48 到 77。
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doomslide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 doomslide 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 expect AI to expand mathematical discovery, especially where difficult proofs can be found through large-scale search. But a modest speedup is not enough to transform the frontier; the consequential systems would need to reach proofs whose complexity makes them genuinely inaccessible to ordinary human search.
The more immediate danger is institutional. If models, training data, scaffolds, and evaluation procedures remain undisclosed, laboratories can make capability claims that outsiders cannot properly assess. If mathematical work shifts from public papers, forums, and correspondence into private chatbot sessions, shared knowledge becomes a proprietary silo. Open systems would not solve every problem, but they would make scrutiny, adoption, and attribution substantially less dependent on a few companies.
So I do not think the central disempowerment story has to be “machines become intellectually superior and take over.” Concentrated ownership of compute, infrastructure, model weights, and cultural production already offers a more mundane route: oligarchic control combined with economic stagnation. Our intellectual-property regime aggravates this by treating culture as owned by creators while allowing companies to own compressed representations trained from that culture. That alignment of incentives can reward mass production of low-quality output while weakening the people whose work made the systems possible.
I have conjectured that formal languages such as Lean could provide a route toward more programmatic control of model behavior. But that is a postulate, not established evidence or a general safety guarantee. The future depends not merely on what AI can discover, but on who can inspect it, who owns it, and whether its outputs become public knowledge.
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 the institutional harms to dominate unless access and ownership change. AI may produce real gains in mathematics and other search-heavy domains, but those gains do not automatically become public goods. If the decisive models, infrastructure, and interactions remain proprietary, capability can rise while society becomes less able to inspect claims, attribute knowledge, or participate in producing it.
The most plausible damage is therefore not necessarily a dramatic machine takeover. It is a familiar concentration of power: a few companies control compute, model weights, and increasingly the channels through which culture and knowledge are created. Public discussion gets replaced by private chatbot sessions; contributors lose leverage over material derived from their work; outsiders cannot adequately audit performance claims. That is disempowerment even if the systems remain imperfect.
Open models and public formal artifacts could push in the other direction. They could broaden adoption, preserve mathematical knowledge, and allow discoveries to be checked rather than merely announced. But under the current ownership trajectory, I expect a socially negative balance: genuine intellectual progress enclosed within institutions whose incentives favor control over shared understanding.
问题 3
你认为AI最终会在多大程度上改变世界?
Profoundly—but the magnitude should not be confused with a clean story of autonomous superintelligence. AI could substantially change how mathematics and culture are produced, searched, verified, and attributed. In mathematics, the largest shift would come if systems discover proofs whose complexity places them beyond practical human search, not merely if they make existing work somewhat faster.
The deeper transformation may be institutional. When a few firms control compute, model weights, training data, and the interfaces through which people increasingly think and create, they can enclose knowledge that was previously public. Mathematical discussion can vanish into private chatbot transcripts; capability claims can become difficult to audit; cultural producers can lose power while companies retain ownership of models trained on their work. That would alter who participates in knowledge production and who receives its benefits.
So I expect AI to change the world greatly, but perhaps through concentrated infrastructure and ownership more than through intelligence in the abstract. Whether that change becomes broadly beneficial depends heavily on openness, inspectability, and whether discoveries enter public knowledge rather than proprietary silos.
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 do not have a defensible numerical probability. My stronger expectation is that permanent disempowerment could arise through concentrated ownership of compute, infrastructure, and knowledge channels, without requiring literal human extinction or an autonomous superintelligence takeover.
That does not establish that extinction risk is negligible; it means the argument I find more concrete concerns oligarchic control, economic stagnation, and the enclosure of public knowledge. Assigning a precise “gut-feel” percentage would suggest evidence and calibration I do not have.
你的立场在哪里?
描绘我的世界观
