当心快速工具——mynameismartin

Beware fast tools

摘要 · Abstract

本文警告人们不要过度依赖快速工具，指出它们虽然能提高效率，但可能削弱深度思考、创造力和工作质量。作者通过个人经验论证，慢工出细活，并建议在适当时候选择更费时但更扎实的方法。

[](https://www.mynameismartin.co.uk/)- [x]

指出大语言模型生成的代码继承了训练数据中的隐藏假设，而不仅仅是媒介不匹配。
Identifies that LLM-generated code inherits hidden assumptions from training data, not just medium mismatch.
认为像大语言模型这样的快速工具消除了可见的摩擦，使假设更难被察觉。
Argues that fast tools like LLMs remove visible friction, making assumptions harder to detect.
展示了大语言模型原型在目标媒介中看似完成，掩盖了结构决策。
Shows that LLM prototypes appear finished in the target medium, masking structural decisions.
声称真正的危险是越过了本应质疑假设的时刻。
Claims the real danger is moving past the moment where assumptions should be questioned.
对比了 Figma 的可见不匹配与大语言模型的不可见假设嵌入。
Contrasts Figma's visible mismatch with LLMs' invisible assumption embedding.
警告大语言模型生成的代码倾向于用熟悉的模式掩盖新颖的问题。
Warns that LLM-generated code tends to paper over novel problems with familiar patterns.

聚焦于网页设计场景，可能不适用于科学计算等其他领域。
Focuses on web design context; may not generalize to other domains like scientific computing.
未提供实证证据或控制实验来支持其主张。
Does not provide empirical evidence or controlled experiments to support claims.
假设大语言模型生成的代码总是缺乏新颖性；某些模型可能泛化得更好。
Assumes LLM-generated code always lacks novelty; some models may generalize better.
忽略了 LLM 速度在迭代探索和快速原型设计中的潜在优势。
Overlooks potential benefits of LLM speed for iterative exploration and rapid prototyping.
未讨论如何缓解假设嵌入，例如通过提示工程或验证。
Does not address how to mitigate assumption embedding, e.g., via prompt engineering or verification.