📄 中文摘要
开发者在使用 AI 编码工具时，主观上认为自己的工作效率提高了 24%。然而，客观测量显示，他们的实际效率下降了 19%。这一 43 个百分点的认知差距揭示了 AI 工具采用率上升而信任度下降的原因。METR 研究对经验丰富的开源开发者进行测量，开发者在自己的代码库中选择任务进行工作，结果表明，尽管 AI 工具让开发者感到更高效，但实际上却降低了他们的生产力。这一现象对如何有效使用这些工具具有重要意义。
开发者认为 AI 使他们的工作效率提高 24%，但实际上却降低了 19%
📄 English Summary
Developers Think AI Makes Them 24% Faster. It Actually Makes Them 19% Slower.
Developers using AI coding tools perceive a 24% increase in their productivity, while objective measurements indicate a 19% decrease. This 43-point perception gap highlights the rising adoption of AI tools alongside declining trust. The METR study assessed experienced open-source developers working on self-selected tasks in their own repositories, revealing that, despite feeling more productive, the AI tools measurably hindered their actual productivity. Understanding this phenomenon is crucial for effectively utilizing these tools.