Cody Blakeney

Cody Blakeney

x.com/code_star

Machine learning researcher who works on training data and fine-tuning and argues for self-hosted models and careful security as AI agents spread.

AIは世界をどのように変えるでしょうか?

文明規模の変化漸進的な変化DoomBloom
シミュレーション上の位置解釈範囲

横軸:その人が表明したDoom–Bloomの見通し。 縦軸:変革の規模。

Doom–Bloom:100点中70。変革の規模:100点中41。解釈範囲:横方向は65から75、縦方向は21から54。これらは解釈上の座標であり、事象の確率ではありません。

Cody BlakeneyのP(doom) · 推定

≈4%

0%100%

本人が示した数値ではなく、シミュレーションされた本人の回答から推定したものです。 妥当と考えられる範囲:1–10%。

P(doom)の推定方法

その人の見通しを左右するもの

中心的な前提

So I see AI as a powerful automation layer whose effects depend heavily on data quality, evaluation, operator experience, access control, and whether the surrounding systems remain understandable and reversible.
回答1

この前提が実際には異なると判明した場合、その人の見通しはどう変わりますか?

未解決の問い

That practical trajectory does not, by itself, establish a particular timeline for AGI or settle every frontier-risk question.
回答1

ここで考えられる結果をその人が見分けるうえで、何が役立ちますか?

詳細

予想される恩恵

大きな恩恵が予想されていますが、重要な条件や分配上の制約があります。

67 / 100

影響が小さい変革をもたらす影響

質的尺度での解釈範囲は67から67です。

予想される害

対処可能、または局所的な害が予想されています。

37 / 100

影響が小さい変革をもたらす影響

質的尺度での解釈範囲は33から67です。

人間の影響力

人間の選択によって、AIの軌道を大幅に変えることができます。

63 / 100

影響力が小さい影響力が大きい

質的尺度での解釈範囲は30から95です。

AI利用のルール

事前の保護措置または許可が整うまで、取り上げられたAIの利用を制限します。

シミュレーション上の位置:対象を絞った説明責任と保護措置を伴う形で、取り上げられたAIの利用を認めます。

取り上げられたAIの利用に対する制限を最小限にします。

これらの解釈では、その人が示した条件が維持されています。恩恵と害は、どちらも大きくなり得ます。この範囲は、統計的な信頼区間ではなく、その人のシミュレーションされた回答をどのように読み取ったかを示すものです。

あなたはCody Blakeneyと比べてどの位置でしょうか?
約3分で自分のAIに対する世界観をマッピングして、比較できます

似ている世界観

シミュレーションされた世界観がCody Blakeneyの世界観に最も近いオピニオンリーダー

出典

このシミュレーション対象者の根拠として使用された記事、インタビュー、著作です。

シミュレーション評価

質問1

AIは私たちの未来にとって何を意味すると思いますか?また、なぜそう思いますか?

I expect AI to automate a growing share of routine, valuable work, especially in software and other digital workflows. But the benefit will not come merely from generating more output. Experienced practitioners still have to identify which problems matter, define the task, evaluate the result, and recognize when an apparently plausible answer is wrong. Better data selection, efficient adaptation, and careful evaluation are therefore at least as important as raw model scale. As agents gain access to systems such as email, chat, repositories, and cloud infrastructure, security will increasingly depend on human permission choices. An agent can become another path for social engineering or hijacked access. Provider concentration creates a related systems risk: relying on one hosted model can expose users to common outages, interception, or infrastructure failure, while self-hosted models and provider diversity can improve resilience. In software engineering, faster code production raises the value of fundamentals rather than eliminating them. Maintainable designs, clear interfaces, reviewable changes, and easy rollback become more important when code can be produced rapidly. So I see AI as a powerful automation layer whose effects depend heavily on data quality, evaluation, operator experience, access control, and whether the surrounding systems remain understandable and reversible. That practical trajectory does not, by itself, establish a particular timeline for AGI or settle every frontier-risk question.

質問2

恩恵と害を総合すると、AIは私たちの社会に全体としてどのような影響を与えると予想しますか?

I expect a broadly positive impact where AI is deployed as supervised automation: reducing routine work, speeding development, and helping experienced people apply their judgment more efficiently. The largest gains will come not from producing the most output, but from choosing valuable tasks, using high-quality data, and evaluating results carefully. The harms will often arise from the systems around the model. Agents connected to email, chat, repositories, or infrastructure amplify bad permission decisions and create opportunities for social engineering. Concentrating access in a small number of hosted providers also creates shared points of failure. In software, rapid generation can accumulate brittle code unless changes remain coherent, reviewable, maintainable, and reversible. So I do not think “AI” has one automatic social effect. Its impact will depend substantially on deployment practices, human expertise, security boundaries, and infrastructure diversity. I would not infer a precise overall forecast—or an AGI timeline—from current success at automating particular tasks.

質問3

AIは最終的に世界をどの程度変えると思いますか?

I expect substantial change in how digital work gets done, but I would not attach a reliable numerical magnitude or claim that current automation establishes an AGI endpoint. Agents can remove large amounts of routine work and dramatically increase the rate of software production. That shifts human effort toward selecting worthwhile problems, specifying them clearly, checking results, and managing access. The downstream effects depend on engineering choices. Faster output is not automatically better output: poor data, weak evaluation, excessive permissions, and unmaintainable generated code can scale mistakes just as efficiently. Agents connected to communications, repositories, and infrastructure also widen the consequences of social engineering and compromised access. Meanwhile, dependence on a few hosted providers can turn outages or interception into system-wide failures. So I expect AI to change workflows and infrastructure deeply. How beneficial and resilient that change becomes will depend less on raw generation volume than on experienced judgment, careful evaluation, secure permissions, maintainable systems, and diversity in how models are deployed.

質問4

AIが人類絶滅、または同様に永続的な破局を引き起こす可能性は、直感的にどのくらいだと思いますか?

I would not give a defensible percentage. My work and observations support concrete concerns about permissions, social engineering, provider concentration, and failures amplified by rapid automation, but they do not establish a probability of human extinction or an equivalently permanent catastrophe. Those practical risks should not be mistaken for a blanket denial of frontier risks. They simply operate at a different evidentiary level: I can evaluate whether an agent has excessive repository access, whether a deployment creates a single point of failure, or whether generated changes are reversible. Turning those findings into an extinction forecast requires assumptions about future capabilities, deployment, control, and institutional responses that current task automation does not resolve.
あなたはどの位置でしょうか?
いくつかの簡単な質問に答えて、自分のAIに対する世界観を探ってみましょう。
自分の世界観をマッピングする

あなたはどの位置でしょうか?

自分の世界観をマッピングする