Is this a new role, or are you replacing someone?
If it’s a replacement, what changed after the last person left?
你要知道这份工作为何现在出现。新增岗位可能是真实扩张,也可能是尚未定义的愿望;replacement 可能暴露范围、level 或经理预期的问题。
COMPANY · TEAM · MANAGER
Reading ↗A REVERSE INTERVIEW FIELD GUIDE
怎么判断一家公司和一个团队值不值得去。
职位描述写的是承诺,面试官讲的是叙事。你需要的是能互相核对的事实:为什么招人,工作怎样被决定,经理实际上怎样带人,同事每天在做什么,研究有没有资源和验证链,公司还能走多久。
00 / THE PROTOCOL
经理说团队重视研究自主权,就问同事最近一个由研究者提出并推动的项目。公司说模型与实验紧密协作,就请他们讲上一次模型结果如何改变实验,再问实验侧 collaborator。创始人给出 runway,就确认这个数字是否包含正在招聘的人。
答案不一致不一定说明有人撒谎。它说明你终于找到了需要继续问的地方。
01 / RECRUITER
Recruiter 最适合回答岗位来源、汇报线、流程和正式政策。不要让 recruiter 替未来经理解释管理方式。
Is this a new role, or are you replacing someone?
If it’s a replacement, what changed after the last person left?
你要知道这份工作为何现在出现。新增岗位可能是真实扩张,也可能是尚未定义的愿望;replacement 可能暴露范围、level 或经理预期的问题。
Who would I report to, and who would I work with most closely?
这个回答决定你还需要见谁。公司品牌和总监愿景不能替代实际经理与日常 collaborator。
Is there a chance I could talk to the hiring manager and someone on the team?
流程暂时不能安排,不等于团队一定有问题。但在接受 offer 前,实际经理和未来同事始终不可见,就是重要的信息缺口。
02 / HIRING MANAGER
职位是否值得去,首先取决于这个人能否说清问题、资源、决定和已经发生的行为。
What do you want me to have accomplished in my first six months?
What’s the first thing this role is meant to solve?
听具体问题、可交付结果和现实依赖。若回答只是一串 personality traits,岗位可能还没有被定义。再与 JD、recruiter 和团队当前项目核对。
How does work usually end up on this team’s plate?
Can you walk me through the last time priorities changed mid-project?
你要听见客户、实验、产品、研究判断和领导层分别怎样进入决策。一次变化不是红旗;关键是新证据还是随机压力改变了方向,以及谁承担 tradeoff。
What are the last three projects this team shipped?
Who proposed it, and who decided it was ready?
项目名不重要。你在确认团队是在启动、交付、验证,还是长期停留在 roadmap 和 prototype。
Who was the last person you promoted, and what had they done?
When was the last time you went to bat for someone on your team?
新团队没有 promotion 历史不自动构成问题。对于有历史的团队,具体行为比 “we care about growth” 更有用。
Who was the last person to leave the team, and why?
What do new people usually struggle with here?
What advice would you give me to be successful here?
放在对话中后段。经理不应披露私人信息;观察他是否能承担团队的一部分责任,并把隐性技能、onboarding 债务和成功标准说具体。
03 / FUTURE TEAMMATE
公司安排的人仍在职,而且通常由 hiring manager 挑选。正面答案一致不足以定论。优先记录具体事件;如果可以,再见一位跨职能或不在同一直接汇报线上的人。
What have you been working on today?
这是自然入口。继续听哪里最花时间,哪些是计划内工作,哪些来自打断、救火或等待。
What’s the difference between a good day and a bad day for you?
How often do the bad days happen? What usually causes them?
“坏日子是没做成事”没有信息。频率和原因才会暴露会议、救火、等待实验、需求变化或技术债。
What’s it like working with [manager’s name]?
What do you like about it, and what’s the hard part?
Can you give me an example of a time they were really helpful?
适合对话中段。先让对方讲背景,再追优点、困难和具体事件;不要用它当第一句话。
What do you usually talk about in your one-on-ones?
What happens when someone pushes back on a deadline?
如果 1:1 只有 status update,继续了解 feedback 和 career conversation 在哪里发生。deadline 的目标不是永远没有,而是不同意见能否进入决定。
What surprised you after you joined?
向另一位同事问同一句。重复出现的 surprise 值得继续核对,但不要因为几个人答案一致就自动当成结论。
04 / AI FOR SCIENCE
这一节是工业研究场景适配。研究实验室资料直接支持 ownership、funding、mentorship、authorship 和 current/former members;model-to-experiment 问法是谨慎迁移,不冒充已经验证的工业标准问题。
Who decides which research questions the team pursues?
Can you give me a recent example of a researcher changing what the team was working on?
区分提出 idea、批准资源、执行项目和决定停止。它们可能属于不同的人。
Which projects are already funded and staffed for the next year?
有 roadmap 不等于有数据、算力、实验人员或预算。`funded` 和 `staffed` 要分别听。
Can you walk me through the last time a model result changed an experiment?
How long did it take to get the next experimental result?
完整回答应说清模型输出、谁决定实验、实验容量、等待时间和结果如何回到下一轮。不同实验速度天然不同,不要用天数跨公司排名;要找实际 bottleneck。
Could I speak with someone on the experimental side who has worked with this team?
模型团队的 “close collaboration” 应能被实验侧讲成同一个流程。不能安排时,把它记录为未知项,不自行补齐。
Could I speak with someone who recently left the team?
I’m considering an offer to join [team]. Would you be open to a short, off-the-record conversation about your experience there?
第二句用于独立联系,不在正式面试中使用。单个前成员不能代表全组;核对在职时间、汇报线、离开背景和事件日期。
How are publication and authorship decisions made?
What can researchers publish, and what can they take with them when they leave?
口头回答只用于发现政策。真正影响职业记录和后续研究的内容,要回到 written policy、offer、IP agreement 或明确审批流程。
05 / OFFER
这些问题通常放在 mutual interest 已明确或 offer 阶段。你不是要求创始人预测未来,而是要求公司说明它现在知道什么、用什么口径知道。
What is the runway at the current headcount?
What is the runway with the hiring plan you just described?
两个数字可能差很多。记录计算日期,不要只记“18 months”。
When do you expect to raise again or become profitable?
What gives you confidence that plan will work?
How much has the company raised, and when was the last round?
找已经发生的 milestone、signed partnership、revenue、实验结果或明确融资条件。公开融资记录可核实历史,不能替代内部 runway。
What percentage of the company does this grant represent on a fully diluted basis?
What is the current 409A valuation and the strike price?
What happens to the exercise window if I leave?
Do you offer refresh grants or employee liquidity?
Option 数量本身没有可比性。Private equity 仍受稀释、股份类别、liquidation preference、税务和流动性限制。行权可能要求先支付现金并产生税务后果;需要时在签字前咨询合格专业人士。
06 / LISTENING
客户、收入和个人离职原因可能受保密约束。新团队可能没有 promotion 历史。实验速度也取决于实验类型。区分“不方便透露”“这个人不知道”和“组织自己没有答案”。
Who would be the best person to ask about that?
So if I’m hearing you right, [restatement]. Is that accurate?
07 / AFTER THE INTERVIEW
最危险的不是一个你不喜欢的答案,而是一个会改变决定的重要问题,直到接受 offer 时仍没有人能回答。
Question asked:
Answer, as close to their words as possible:
Specific example or number:
What is still unclear:
Who else can verify it:
External or written evidence:
What decision this changes:
Confidence: low / medium / high
EVIDENCE NOTE
这份手册依据真实候选人、招聘经理、研究人员和 startup/equity 指南整理。页面逐题标明来源原问、依据来源改写或场景迁移。核心方法有多来源支持,但没有纵向研究证明某一句英文问题能预测入职后的满意度。