[GoLUG] The Strange Behavior of LLMs in Hiring Decisions: Systemic Gender and Positional Biases in Candidate Selection

[Steve Litt]

On Fri, 23 May 2025 08:05:01 -0700
Kyle Terrien <kyle at terren.us> wrote:

> Someone did a study into whether LLMs are biased when making
> “decisions” as to who to hire.
> 
> https://davidrozado.substack.com/p/the-strange-behavior-of-llms-in-hiring
> 
> The conclusion:
> 
> + It pays to be a woman.
> + It pays to be the first candidate evaluated.
> 
> > In this context, LLMs do not appear to act rationally. Instead, they
> > generate articulate responses that may superficially seem logically
> > sound but ultimately lack grounding in principled reasoning.  
> 
> > The results presented here also call into question whether current
> > AI technology is mature enough to be suitable for job selection or
> > other high stakes automated decision-making tasks.  
> 
> Ouch.  That’s a harsh condemnation of AI.

Overly harsh in my opinion, and to explain why I think that let's
discuss self-driving cars. It's going to be a long, long time before
self driving cars are safer than smart, skilled and attentive human
drivers. One software bug and the self-driving car plunges into the
lake.

But right now they can probably be made safer than 95% of human
drivers, meaning driverless cars will decrease accidents and other
mishaps.

All an AI candidate selector needs to do is be better than 95, or heck,
even 60 percent of current ways it's done. Except...

Except that the way candidates have been chosen, at least from the
1980's, is horribly flawed. HR pens a job ad that screens out 90% of
the people who could actually do the job, so liars flood HR with
resumes, and all too often a fast-talking liar gets the job and then
fails miserably. From the 1980's on, the trick was to do an end run
around HR. Use various human networking techniques to find your new
boss, talk with him or her, get all that stuff out of the way, then
have him or her shepherd your resume and job application through HR.
Same with AI.

In the spring of 1980 I went on a 7 day whirlwind tour of Los Angeles
and San Francisco looking for a job (I was living in Chicago at the
time). In each place, I asked to talk to somebody in charge. Some
places just through a job application in my face: I thanked them and
walked out. Nothing like having only 7 days to teach you how to
prioritize job hunting time.

In the 1990's I was all at every computer fair, every software expo,
everything. I talked with everyone I met, and everyone knew what I did
and how good I was at it. I had my elevator speech down pat, and I had
rehearsed enough material that if asked questions, I could go into
details. Most of the work I ever got was because of who I knew, not
what I knew.

My opinion: If you're sending in resumes and applications, you're dead
meat. You spend hours looking on job sites. You spend what, a half hour
tailoring your resume and cover letter for each job. Twenty, thirty a
day. The day goes by and you've spoken to nobody. And of course you get
ghosted by all the companies you applied with: 5000 other people sent
in resumes and cover letters. You need to make an identity for
yourself, then start going to meetings and expos to talk with others,
and make sure you know how to put yourself in the best light, because
you're darn good. Then put your money where your mouth is, and start
giving presentations so you're THE GOTO guy for skill X. Bonus points
if you start a business in skill X to make yourself more credible,
bypass HR and AI, and maybe get some money coming in the door. Start
the business cheaply so you can get out of it if the right job comes
along.

1980. 2025. HR. AI. You must get face time with the guy with hiring
authority, or at least hiring recommendation.

SteveT

Steve Litt 
Spring 2023 featured book: Troubleshooting Techniques of the Successful
Technologist http://www.troubleshooters.com/techniques