Mike Taylor

Mike Taylor

x.com/hammer_mt

AI practitioner and author who tests prompts and models on real tasks and argues people should run their own task-specific evaluations.

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 58。变革程度:100 中的 43。解读范围:横向为 49 至 63,纵向为 38 至 55。这些是解读坐标,而不是事件概率。

Mike Taylor的 P(doom)

尚未估计

他们的模拟回答中关于灾难性风险的信息不足,无法进行估计。

我们如何估计 P(doom)

他们的展望取决于什么

一个核心假设

The clearest mechanism is cumulative task automation: if models make drafting, analysis, synthesis, and option generation cheaper across many workflows, organizations can change how work is divided long before an AI can independently hold an entire job.
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

I don’t have a well-supported net-positive or net-negative forecast for society as a whole.
回答 3

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

预计将带来显著益处,但受到重要条件或分配方面的限制。

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

根据你的回答得出的暂定估计;较宽的范围表示其他合理解读。

51 / 100

影响力小影响力强

在定性尺度上,解读范围为 0 到 100。

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Mike Taylor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 Mike Taylor 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 automate meaningful parts of many jobs well before it can reliably replace entire roles. Jobs are bundles of tasks: a model might draft a product brief, synthesize feedback, or generate options impressively, then fail at judgment, context, coordination, or follow-through. A good result on one blind comparison is evidence about that task—not proof that the whole profession is solved. The practical future is therefore less about one universal “intelligence score” and more about matching models to real needs. I’d use blind, task-specific evaluations, with examples drawn from actual work. Some tasks justify an expensive high-capability model; others need a dependable daily driver or cheap intelligence at scale. Price and benchmark prestige do not necessarily predict which model will perform best. I also think reliability will depend heavily on application design. Clear intent, good examples, and decomposing complex requests can improve results substantially. More context is not always better: accumulated memory can become stale or contradictory and make outputs worse. So my basic expectation is broad, uneven automation—useful and sometimes surprising, but requiring continual testing rather than sweeping claims about whole jobs or society.

问题 2

你认为AI最终会在多大程度上改变世界?

I expect substantial change, but I wouldn’t attach a defensible percentage or declare that every institution will be transformed. The clearest mechanism is cumulative task automation: if models make drafting, analysis, synthesis, and option generation cheaper across many workflows, organizations can change how work is divided long before an AI can independently hold an entire job. The magnitude will also be uneven. A model may be excellent at one narrow task and unreliable at the next, while a cheaper model may outperform a prestigious one for a specific need. That makes “How capable is AI?” less useful than “Does this system reliably perform this task under realistic conditions?” The world-changing effects will come from thousands of those task-level deployments—and from their failures—not from a single benchmark announcing that intelligence has arrived. So: potentially very large, but not measurable from the evidence I’ve discussed. I’m more confident about broad, uneven changes to workflows than about any grand forecast for society as a whole.

问题 3

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?

I don’t have a well-supported net-positive or net-negative forecast for society as a whole. The evidence I trust is much narrower: AI can improve particular tasks, lower costs, and automate meaningful slices of work, while also producing convincing failures when context is stale, intent is unclear, or performance is inferred from irrelevant benchmarks. The overall impact will depend heavily on deployment. A system that passes a realistic blind test for one workflow may be genuinely valuable; that still says little about whether it can perform an entire role, let alone whether AI is socially beneficial in aggregate. Vendors may also optimize for cost rather than giving users the strongest model, so organizations need their own evaluations instead of treating model labels as truth. My strongest expectation is therefore substantial but uneven change—not a defensible verdict on the final balance of benefits and harms.

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 don’t have a defensible percentage for that. My work supports claims about task performance, reliability, and deployment failures—not a calibrated estimate of human extinction or permanent global catastrophe. Assigning a number would create an impression of precision without an evaluation behind it.
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