跳到正文
原文
Hacker News· speckx·· 1 天前精选AI 评分27

开发者吐槽 LLM 生成的招聘邮件:'个性化'开头与岗位描述脱节

The Deeply Impersonal Personalized Recruiter Mail

AI 导读

speckx 在个人博客中批评 LLM 生成的招聘邮件,指出开头会大段复述收件人履历,而真正附上的岗位描述却与经历毫无关联,例如把基础设施工程师推给 Django 全栈岗。作者认为与旧式模板邮件相比,这类'个性化'与岗位错位的组合更冒犯收件人的判断力;LLM 本可根据简历判断候选人是否匹配,却未被真正使用。

推荐理由

作者以亲身经历对比了 LLM 生成的招聘外联与岗位描述之间的脱节,为读者提供了判断 AI 外联质量的参照。

正文 · 原文

08 Oct, 2026

Lately I’ve encountered a new species of software engineer recruiter mail. It goes something like this:

Your work on infrastructure at Railway, along with your earlier years building observability software at City Storage Systems and storage systems on AWS Glacier, is exactly the kind of platform background this role calls for.

So the first paragraph is spent explaining my career to me (thanks I forgot where I worked). The last sentence is coincidentally “that’s exactly what we’re looking for”! Then follows the actual job description for building CI/CD pipelines, which has almost nothing to do with the job experience just mentioned. Clearly the first paragraph is generated by an LLM, but the job description too?!

A few gems I've gotten:

That's the gap. You'd be the first. The work is concrete: multi-tenant isolation for AI model serving, golden-path service templates in Terraform, IAM and secrets management wired into CI/CD, and policy-as-code that makes secure the default rather than the exception.

Observability here is genuinely unsolved. The failure modes are probabilistic rather than deterministic, so standard instrumentation tells you less than you need.

Every document is a referral, a prior auth, or an intake form, which means extraction accuracy isn't a dashboard metric, it's whether a patient gets seen or gets denied. Messy inputs, no tolerance for silent failure, growing fast.

Over the years I’ve gotten lots of emails along the lines of “Hey it’s really neat that you work at X company” immediately followed by the job description, so what’s the difference? So what if it’s generated by an LLM?

To me the difference is twofold:

  1. “Personalized” with zero connection to the job description
  2. Being very obvious LLM text

With the former, I understand recruiters send a lot of emails to hit the numbers they need. The email template with [insert name and previous company] is a shotgun blast, they probably haven't spent more than 10 seconds looking at my profile to see if it’s a fit. So it’s unsurprising to me that it’s for a Django fullstack role when my history shows I’m an infra engineer with no professional Python experience. However when the same job description is preceded by paragraph saying how my previous roles makes me “perfect for this job”, it actively insults my intelligence.

What makes it extra maddening is an LLM could look at my resume to see that I'm a match and write about how my experience directly ties in to the role. Or conversely it could determine “hey why the hell are we even talking to this person, they don't know Verilog”. I suspect the latter is why they don’t actually do this.

All of that coupled with even the job description being full of extremely common LLMisms pushes these emails below the plain old templated emails in my mind. A simple edit pass through another LLM could remove these. Or, you know, doing a quick proofread of the thing being sent out. The overt fakeness of the effort and personalization makes it feel so unapologetically impersonal.

#llms #recruiting #software

来源:Hacker News · blog.pentlander.com