terminally online engineer

terminally online engineer

x.com/tekbog

Pseudonymous account that builds AI agent tooling, questions how AI labs talk about risk and argues reliance on LLMs is eroding software craft.

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 67。变革程度:100 中的 53。解读范围:横向为 62 至 75,纵向为 46 至 79。这些是解读坐标,而不是事件概率。

terminally online engineer的 P(doom) · 推断

≈5%

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:3–8%。

我们如何估计 P(doom)

他们的展望取决于什么

一个核心假设

That’s how domain knowledge falls off a cliff while people trying to ship AGI can’t run a CRUD app.
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 don’t have a percentage or AGI date.
回答 2

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。

67 / 100

影响小变革性影响

在定性尺度上,解读范围为 67 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。

54 / 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 30 到 95。

AI使用规则

在事先落实保护措施或获得许可之前,限制所讨论的AI用途。

模拟位置:允许所讨论的AI用途,同时实施有针对性的问责与保护措施。

尽量减少对所讨论AI用途的限制。

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terminally online engineer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 terminally online engineer 最接近的意见领袖

terminally online engineer关于AI说过的话

The terminally online engineer account questions AI hype and alarm, wants labs held accountable for their products and worries about junior roles.

  1. “many engineers that were doing ML before the AI revolution don’t get the hype”

    Post on X
  2. “Claude is a product from Anthropic, and therefore Anthropic is held accountable for what the product does - it's not magic”

    Post on X
  3. “i know people wanna freak out about this and scream AGI and how we are all going to die but it’s a much simpler story than that”

    Post on X
  4. “entry level positions and junior roles are not being created, there’s no training to pass the knowledge”

    Post on X

逐字引自所链接的出处,核对于 2026年10月6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。

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?

AI probably makes a lot of useful things cheaper and easier, but the immediate future is also a mountain of unreliable software nobody understands. The models are impressive, but today they code like an average engineer who thinks passing tests means the system works. Then companies stop hiring juniors, stop training people, lay off the engineers with context, and let everyone spin the LLM roulette. That’s how domain knowledge falls off a cliff while people trying to ship AGI can’t run a CRUD app. I’m less worried about “what if the air-gapped computer could talk” than boring accountability and engineering quality. Claude is a product, not magic. Hold labs responsible for what their products do, fix the vulnerabilities already everywhere, and sell people on concrete benefits like cheaper burritos instead of constantly announcing that everyone will lose their jobs.

问题 2

你认为AI最终会在多大程度上改变世界?

A lot, probably—but “a lot” doesn’t automatically mean omnipotent robot gods or every job disappearing. It can mean cheaper goods, better tools, new kinds of art, and software getting built much faster. It can also mean a generation of systems nobody understands because we stopped training juniors and reviewing how anything works. That mundane institutional damage may matter more than the doom scenarios. I don’t have a percentage or AGI date. I just think the impact will be huge, uneven, and much less magical than the labs’ marketing.

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 percentage. My gut is that the current doom discourse is mostly speculative theater—“okay, but what if the air-gapped computer could talk”—while the concrete failures are unreliable products, awful security, and collapsing engineering knowledge. That isn’t proof advanced AI is safe; it’s just why I don’t treat confident extinction numbers as serious engineering.

问题 4

你倾向于认为AI带来的益处多于危害,还是确实难以取舍?

I lean more good than bad. Useful tools, cheaper goods, better security, genuinely good creative work—those benefits are real. I’m not torn about whether the technology can help; I’m worried we’ll deploy it in the dumbest possible way: no accountability, no review, no juniors, and no one left who understands prod when the clanker says the tests pass.
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