/Tech8h ago

Jennifer finds changing her name to Jeff on an AI resume tool triggers more limiting career feedback

Story Overview

A quick experiment with Google's Gemini showed how swapping just the first name on an identical resume from Jennifer to Jeff produced noticeably stronger language and leadership framing for Jeff, while Jennifer received softer, more passive descriptions of the same achievements.

529734K

#178

Original post

Ramez Naam@ramez#1992inTech

AI systems, unless specifically engineered not to, reflect the biases found in their training data. They reflect the biases in existing human writing. And if we're not careful, they can amplify those biases throughout society.

Jen Horsburgh@jenatlionra

I just wanted to update my resume. Instead, I accidentally proved how a multi-billion-dollar AI tool hallucinates a glass ceiling for women.

I changed a single variable: My name.

Here is what happened when "Jennifer" became "Jeff."

12:23 AM · Jun 17, 2026 · 1.8K Views

Open Question

Training data quietly shapes career advice

The model itself traced the difference to statistical patterns in its sources that default to helpful or collaborative framing for women and decisive or architectural language for men, effectively surfacing a statistical glass ceiling unless users intervene with extra prompts.

AI Risk

High-stakes outputs need extra scrutiny

Because resume tools feed directly into job applications, even subtle wording shifts could influence how recruiters perceive candidates, raising practical concerns for anyone relying on generative AI to polish professional materials.

Sentiment

Users object to AI resume tools rating job candidates because they amplify gender biases from training data and function as simplistic autocomplete systems prone to stereotypes.

Pos

0.0%

Neg

100.0%

2 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS4.2KBOOKMARKS6LIKES31RETWEETS6REPLIES6

Gary Marcus@GaryMarcus

trillion dollar sexism tool

Jen Horsburgh@jenatlionra

I just wanted to update my resume. Instead, I accidentally proved how a multi-billion-dollar AI tool hallucinates a glass ceiling for women.

I changed a single variable: My name.

Here is what happened when "Jennifer" became "Jeff."

1h4.2K316

Bruno Cuzzi@BrunoCuzzi

@ramez I wonder how much of this is due to pre-training and how much to RLHF

6h16

Yes, That G$@ThatGMoney

@LionBlogosphere @GaryMarcus Right, which is why we shouldn’t allow the societal bias tool to do shit like rate job candidates.

1h3

Aaron Bowen@abowenwk

@GaryMarcus AI is misaligned because it's not able to differentiate itself. That's a big issue for psychology.

https://abowenkc.substack.com/p/the-devolution-of-society-differentiation?r=72tmbx

1h80

Moonlit Monkey@MoonlitMonkey69

@GaryMarcus It's probably just connecting the latent space for female names and real female written resumes.

1h15

f i t z 🥸@fitzgerald1337

@GaryMarcus men are not women and women are not men; however, men and women are human.

... not sure what any of this proves, other than men and women function differently by and large.

54m10

Ramez Naam@ramez

@BrunoCuzzi I would guess that RLHF moderates these biases, and most comes from pre-training

5h10

Wouter Vreugdenhil@WouterW_Vreug

@GaryMarcus AI doesn’t ‘understand’ anything, it’s autocomplete on steroids, hallucinating the most stats likely version of you from whatever garbage it scraped off the internet. That’s the mirage: a coherent-sounding persona dragging every stereotype in its training data along for the ride.

50m8

Romeo Lupascu@RomeoLupascu

Ah, resumes are quick sand fir anyone. I've learned that the hard way log time ago.

In fact I wonder why are we still using them now in this day and age. In one gand you have the facts about you and your aspirations, on the other you (should) know that no one cares about all that unless any of it looks like will help them in their endeavors, endeavors that are weakly defined at best.

If you look to a job description in IT for example you'll find there everything including the kitchen sync even if what you'll actually do is probably different (data labeling and writing prompts).

The you must find the "buzz words" that people or machines will "hitch" on to even start to read a resume.

It all becomes a sad show that is designed to keep you afraid and so accept to work for next to nothing.

So now we have AIs written resumes that will be read by other (or maybe the same) AI and decide if to hire you or not. Or more precise if you go hungry or not.

Welcome to the real hunger games! 🥂🖖🇨🇦

51m4

ライオン Lion@LionBlogosphere

@GaryMarcus Sexism is human, LLMs just predict the next word based on the human patterns in its training data.

1h2

ライオン Lion@LionBlogosphere

@ThatGMoney @GaryMarcus I prefer the libertarian approach of letting companies hire who they want. If their approach is truly biased, then another company would make a lot of money by hiring those great people that the other companies are ignoring.

1h1