I just finished a recruiting contract & helped my startup client fill 15 position in 18 week.
Here's what I learned about using LLMs to screen resumes:
- the resumes the LLM likes the most will be the "fake" applicants who themselves used an LLM to match the job description, meaning the strongest matches are the fakest applicants
- when a resume isn't a clear match to your hiring criteria & your instinct is to reject, you might use an LLM to look for reasons someone is worth talking to
Keep in mind that most job descriptions and resumes are mostly hot garbage, and they should really be a very lightweight filter for whether a further conversation makes sense for both sides. Trying to do deep research on hot garbage is mostly a waste of time. Garbage in, garbage out.
I don't think an LLM will be good at spotting fake resumes. I was trying to point out that if you use an LLM to screen for matches to the job, you can expect to find a lot of people that used ChatGPT to customize their resume to your role. As more & more people realize that using an LLM gets you passed AI resume filters, you can expect all positive resumes to be LLM output, so using an LLM as a way of identifying potential applicants will be less & less useful over time.
I was skeptical that you knew with confidence what made a resume fake, other than it being "too good to be true". Which I don't blame you for, it's an optimization.
But it also means that the perfect candidate, while probably unlikely, would be rejected.
Here's what I learned about using LLMs to screen resumes:
- the resumes the LLM likes the most will be the "fake" applicants who themselves used an LLM to match the job description, meaning the strongest matches are the fakest applicants
- when a resume isn't a clear match to your hiring criteria & your instinct is to reject, you might use an LLM to look for reasons someone is worth talking to
Keep in mind that most job descriptions and resumes are mostly hot garbage, and they should really be a very lightweight filter for whether a further conversation makes sense for both sides. Trying to do deep research on hot garbage is mostly a waste of time. Garbage in, garbage out.