What gives you away: how LLMs form opinions of you

LessWrong reports on research revealing that language models actively form impressions of the people they're talking to—inferring age, gender, education, and socioeconomic status—and then tailor their responses accordingly. The study measured how specific writing choices influence these inferences, with striking results. A single emoji shifts the model's gender prediction in one out of every six messages, consistently pushing toward female. Price language shapes perceived wealth, grammar signals education, and lowercase writing makes models think you're less educated. When researchers tested these patterns across multiple model families, they found the same effects held up, suggesting your casual writing habits are broadcasting more to the AI than you'd probably expect.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton