AI 面试
Day 21 - Handling Follow-up Questions about AI Testing
今日目标
能回答关于 AI 测试可靠性、假设、限制、误报漏报和人工把关的面试追问。
30 分钟学习安排
| 时间 | 模块 | 做什么 |
|---|---|---|
| 0-5 分钟 | 核心词汇 | 读词汇、短语和中文含义，重点记能直接在 QA 场景复用的表达。 |
| 5-12 分钟 | 阅读/听力输入 | 阅读当天短文；第二遍当听力材料朗读或用 TTS 播放，只抓问题、证据、动作、结论。 |
| 12-18 分钟 | 句型拆解 | 把长句拆成可替换模板，换成自己的工作内容。 |
| 18-25 分钟 | 口语输出 | 完成 60-120 秒英文表达，必须录音。 |
| 25-30 分钟 | 复盘 | 标记卡住的词、句子和下一次要改进的一点。 |
核心词汇
| English | 中文 | QA 场景用法 |
|---|---|---|
| clarify | 澄清 | Use it when you describe clarify in a defect, test plan, review, meeting, or interview. |
| assumption | 假设 | Use it when you describe assumption in a defect, test plan, review, meeting, or interview. |
| constraint | 约束 | Use it when you describe constraint in a defect, test plan, review, meeting, or interview. |
| alternative | 替代方案 | Use it when you describe alternative in a defect, test plan, review, meeting, or interview. |
| limitation | 限制 | Use it when you describe limitation in a defect, test plan, review, meeting, or interview. |
| bottleneck | 瓶颈 | Use it when you describe bottleneck in a defect, test plan, review, meeting, or interview. |
| edge case | 边界场景 | Use it when you describe edge case in a defect, test plan, review, meeting, or interview. |
| follow-up | 追问 | Use it when you describe follow-up in a defect, test plan, review, meeting, or interview. |
| trade-off | 取舍 | Use it when you describe trade-off in a defect, test plan, review, meeting, or interview. |
| evidence | 证据 | Use it when you describe evidence in a defect, test plan, review, meeting, or interview. |
| not enough context | 上下文不足 | Use it when you describe not enough context in a defect, test plan, review, meeting, or interview. |
| reasonable approach | 合理方案 | Use it when you describe reasonable approach in a defect, test plan, review, meeting, or interview. |
高频短语
- from a QA perspective
- validate the expected behavior
- cover the edge cases
- reduce regression risk
- collect enough evidence
- clarify the acceptance criteria
- prioritize the critical path
- follow up with a verification note
- take ownership of the quality risk
- explain the trade-off clearly
- measure the impact with metrics
阅读 / 听力材料
先慢读一遍，再用正常语速朗读一遍。不要逐字翻译，重点抓 problem -> evidence -> action -> result。
An interviewer asks how you would test a system you have never seen. You clarify users, critical flows, data sensitivity, dependencies, and release risk before answering. The most important point is not just to say that something is broken, but to explain why it matters, who may be affected, and how the team can gain confidence before release. A strong QA explanation is specific, evidence-based, and calm. It connects user impact with technical details, and it gives the team a clear next step.理解检查
- What did you own in this situation?
- What technical details prove your contribution?
- What trade-off did you make?
- What result or metric can you mention?
句型拆解
At a high level, my role was...The challenge was...I approached it by...The result was...What I learned from this was...
可直接替换的输出模板
One relevant example is [project or situation].
My role was to [your responsibility].
The main challenge was [challenge].
I approached it by [actions and technical details].
As a result, [impact or metric].
What I learned was [lesson].示范口语稿
这一段先照读，再改成你的真实项目。
One example I can share is related to quality improvement in a real delivery process. An interviewer asks how you would test a system you have never seen. My role was to identify the quality risk, communicate it clearly, and turn it into concrete testing actions. I used evidence such as failure patterns, test results, logs, or release metrics to support my recommendation. The main trade-off was balancing speed and coverage. As a result, the team had better visibility into release readiness and could make decisions with more confidence. This experience also shaped how I think about QA: quality is not only finding bugs, but reducing uncertainty for the whole team.追问练习
- What was your personal contribution?
- How did you measure the result?
- What would you do differently next time?
- How did you influence developers or product managers?
跟读训练
- My role was to define the test strategy and communicate the release risk.
- I used both manual exploration and automated checks to increase confidence.
- The biggest challenge was balancing speed and coverage.
- The result was fewer regressions and a more predictable release process.
- This experience taught me to connect quality decisions with business impact.
口语任务
录一段 60-120 秒英文。必须包含以下 5 点：
- context
- quality risk
- evidence
- action
- next step or result
QA 角色强化
- 面试里 QA 的竞争力不是“会点点点”，而是能解释风险、策略、自动化、系统理解和协作影响。
- 每个回答至少加入一个技术细节和一个结果指标，例如 defect leakage, execution time, flaky rate, coverage, release confidence。
- 避免把责任推给开发或产品；用 ownership, evidence, follow-up action 来表达成熟度。
AI 时代扩展：Testing Agents / Skills / MCP
新增词汇
| English | 中文 |
|---|---|
| AI-assisted testing | AI 辅助测试 |
| test agent | 测试 agent |
| agent workflow | agent 工作流 |
| human-in-the-loop | 人工把关 |
| LLM evaluation / eval | 大模型评估 |
| tool calling | 工具调用 |
| MCP server | MCP 服务器 |
| skill | 可复用技能/流程能力 |
| prompt injection | 提示注入 |
| hallucination | 幻觉 |
| false confidence | 虚假信心 |
| ground truth | 标准答案/真实依据 |
| tool schema | 工具参数结构 |
| context window | 上下文窗口 |
| permission boundary | 权限边界 |
| idempotency | 幂等性 |
| audit trail | 审计轨迹 |
| benchmark dataset | 基准数据集 |
| rubric | 评分标准 |
| precision and recall | 准确率与召回率 |
| regression eval | 回归评估 |
| failure analysis | 失败分析 |
| tool schema | 工具参数结构 |
| context window | 上下文窗口 |
| permission boundary | 权限边界 |
| idempotency | 幂等性 |
| audit trail | 审计轨迹 |
| tool schema | 工具参数结构 |
| context window | 上下文窗口 |
| permission boundary | 权限边界 |
| idempotency | 幂等性 |
| audit trail | 审计轨迹 |
| benchmark dataset | 基准数据集 |
| rubric | 评分标准 |
| precision and recall | 准确率与召回率 |
| regression eval | 回归评估 |
| failure analysis | 失败分析 |
QA 新场景
In interviews, position yourself as a QA engineer who can combine domain knowledge, automation, AI agents, MCP tools, and careful evaluation.表达重点
面试周要避免只说“我会用 AI”。要说清楚场景、架构、质量指标、风险控制、人工复核和业务影响。
可复用表达
I treat AI agents as accelerators, not decision makers. My job is to design the workflow, validate the evidence, and protect release quality.追问加练
- What is your personal contribution beyond using AI?
- How do you measure agent quality?
- What are the failure modes?
- How do you prevent hallucinated test results?
今日作业
- 录音 1-3 分钟，先看稿读一遍，再只看关键词复述一遍。
- 把今天的模板替换成你真实工作中的项目、缺陷、接口或测试任务。
- 整理 8 个你能在工作会议或面试里复用的表达。
- 写一版 80-150 词英文稿，明天开始前先复述一次。