# ETH Zürich 研究：CS 成绩比写作能力更能预测 vibe coding 水平

- 来源：Rohan Paul (@rohanpaul_ai)
- 发布时间：2026-09-10 09:29
- AIHOT 分数：51
- AIHOT 链接：https://aihot.news/items/cmtuv1m9b0fo4rorp72d4ryrz
- 原文链接：https://x.com/rohanpaul_ai/status/2097859979269345464

## AI 摘要

ETH Zürich 论文测试 100 名大学生在 3 个无代码应用构建任务上的表现，参与者只能通过对话驱动 AI 并检查结果，不能查看源代码。CS 成绩与 vibe coding 表现相关性为 0.39，写作能力为 0.29，两者都显著但 CS 的独立预测价值约为写作的 2 倍；写作能力通过产出更清晰的提示词间接起作用。论文：arxiv.org/abs/2603.14133。

## 正文

If you think vibe coding makes computer science fundamentals irrelevant, this study points the other way.

Vibe coding still rewards computer-science knowledge: it was a stronger predictor than writing skill.

Researchers tested 100 university students on 3 no-code app-building tasks where they could only talk to the AI and inspect the result, never the source code.

Both skills mattered, but CS knowledge was the stronger signal.

CS achievement correlated 0.39 with vibe-coding performance, compared with 0.29 for writing.

When both were considered together, CS added about 2x as much unique predictive value as writing.

Writing still helped because stronger writers tended to produce clearer, better-organized prompts, and better prompts were linked to better apps.

– arxiv. org/abs/2603.14133

Title: "Computer Science Achievement and Writing Skills Predict Vibe Coding Proficiency"