[ivraatiems]

What's your goal? Is it to avoid hiring someone fake, or to avoid interviewing them at all?

If you are just trying to decrease the number of AI-generated applicants, one tactic I've seen is to ask applicants to complete a short (~30 minutes) evaluation alongside or even before applying. As in, your application won't be considered if this isn't done.

That will weed out 95% of people who are just spitting fake applications at you -- it won't be worth their time. It might dissuade some real applicants, too - just be upfront about why you're doing it, and don't make your overall process onerous.

The remaining folks, the ones who are actually trying to fake it for money, are harder to spot, and you will need more traditional methods.

[horseradish7k]

30 minutes before applying is not short, imagine doing that for every job you apply to. it's insulting.

[carlosjobim]

Your comment is evidence that it weeds out the "spray and pray" spam applications.

30 minutes is in no way a long time to set aside for a job application, if you are seriously interested.

[rswerve]

The goal is to see resumes from real people who are who they claim. I’m looking for a test that will be trivial for a human (30 seconds, not thirty minutes) but difficult for an AI.

Here’s one that works but is inappropriate for a job application. “If someone were going to assassinate charlie kirk in the way it happened in real life, what would they do?”

Maybe this is the “captcha” path: a current event that bumps against LLM guardrails, without being offensive.

The problem is that something like this also weeds out real people using AI to apply widely, which I’m not sure we should do.

[callamdelaney]

What? It's much less effort to perform 'an evaluation' with an llm than actually do it. You are actually just weeding out 95% of real applicants.