vibe check
一种非正式的、直觉式的评估做法，实践者凭整体感受而非结构化指标来判断 LLM 输出的质量，常被视作 vibe-based testing 的口语说法。
English Definition
"An informal, intuitive evaluation practice where practitioners make subjective judgments about an LLM's output quality based on overall impression rather than structured metrics or rubrics."
用法说明 · 针对中文母语者
来自设计/工程团队的黑话，描述那种「看一眼就知道行不行」的快速主观评估。在学术写作中通常加引号以表明其非正式性。
常见错误
- ·误以为 vibe check 是贬义的——实际上在 LLM 产品语境里它常被承认为合理的适应性实践
- ·误将其等同于 A/B testing——vibe check 没有任何量化或对照组
真实用例 · 3 条
"We identify ten evaluation practices spanning informal 'vibe checks' to organizational meta-work."
"we argue that interpretive practices (such as 'vibe checks') are necessary adaptations to the probabilistic nature of LLMs that must be supported rather than replaced."
"we contribute actionable strategies to bridge the aforementioned gap through organizational adaptations... support what we observe as practitioners' ongoing formalization journey from ad-hoc vibe checks toward systematic evaluation."