HRM304 · Skill 1 · Gender-coded ad audit

ByteNova — Global Management Trainee

Three ads. Same company, same role, three offices. With a partner, work through the ads then the LLM exercise below. Your work saves automatically to this browser.

Click a word in any ad to mark it. First click Agentic Second click Communal Third click clear Counts update live below each ad.
Ad A · San Francisco
Global Management Trainee · San Francisco
0agentic 0communal Ratio — : —
Ad B · London
Global Management Trainee · London
0agentic 0communal Ratio — : —
Ad C · Shenzhen HQ
Global Management Trainee · Shenzhen Headquarters
0agentic 0communal Ratio — : —

Coding lexicon — common indicators, not an exhaustive list. Code other fitting words you spot.

Agentic action · competition · independence
aggressive · ambitious · analyse · argue · assert · challenge · compete · confident · decisive · defend · dominant · drive · driven · fearless · independent · lead · logical · objective · outspoken · persist · push · self-starter
Communal cooperation · care · connection
collaborate · commit · communicate · connect · considerate · cooperate · depend · dependable · empathise · honest · interpersonal · kind · listen · loyal · share · support · sympathetic · trust · understand · warm · yield
Agentic 中文
独当一面 · 抗压能力强 · 狼性文化 · 雄心勃勃 · 领导力 · 主动性 · 攻坚 · 执行力
Communal 中文
团队合作 · 沟通协调 · 细心 · 亲和力 · 服务意识 · 谦虚 · 配合
Translation problem — read before coding Ad C. In English research, loyal · commit · modest code as communal. In a Chinese corporate context, 忠诚 / 投入 / 谦虚 read as non-gendered corporate virtues paired with toughness, not as warmth. Code in English first; then ask whether a Chinese reader would code it the same way.

Coding sheet: fill one column per ad

Ad A · San Francisco
Who doesn't apply?
Name a desirable candidate (someone the company should want) and the specific phrase in the ad that puts them off.
Trait or task wording?
Pick one phrase and label it. Trait = who you are; task = what you do.
One redesign
Change one word or phrase to reduce the linguistic skew without changing the role.
Ad B · London
Predicted applicant pool
Name one specific candidate type and one specific reason they don't apply.
Trait or task wording?
Pick one phrase and label it. Trait = who you are; task = what you do.
One redesign
Change one word or phrase to reduce the linguistic skew without changing the role.
Ad C · Shenzhen
Predicted applicant pool
Name one specific candidate type and one specific reason they don't apply.
Trait or task wording?
Pick one phrase and label it. Trait = who you are; task = what you do.
One redesign
Change one word or phrase to reduce the linguistic skew without changing the role.
HRM304 · Skill 1 · LLM screener thought experiment

If ByteNova fed every CV to an LLM, what would happen?

Work through the three questions below (~5 minutes). No technical knowledge needed: reason from what you just coded.

Bridge from the ads. Ad C contains commit, modest, fearless, drive. Using English coding, what ratio did you get? Now re-read the translation note. Would a Chinese reader code the same words the same way? What does this mean for an LLM screener trained mostly on English text? Hold onto your answer — the questions below build on it.
The thought experiment. ByteNova receives 15,000 applications for these three roles. They feed the job descriptions and the CVs into a large language model (LLM) and ask it to rank candidates. What does the LLM do?
1
Semantic matching.

Each ad is a vocabulary. The LLM ranks CVs by how well the CV's language matches the ad's vocabulary. For Ad A, what kind of CV vocabulary scores high? For Ad B? For Ad C?

2
The gender effect.

Research shows men and women describe their own achievements in systematically different language: men more agentic, women more communal. If Ad A rewards agentic CV vocabulary, what happens to the gender mix of the shortlist, even though the LLM never sees the applicant's name or gender?

3
Trained on history.

The model learned what a "good CV" looks like from CVs of people who got hired in the past. Amazon scrapped an internal screener in 2018 after finding it downranked CVs containing the word women's (as in women's chess club captain) — it had been trained on a decade of mostly-male engineering hires. Even if ByteNova rewrote every word of every ad, what would still come baked into the LLM itself?

Auto-saves to this browser.