Study: GPT-4o shows gender bias in ADHD self-diagnosis

A new study published on arXiv explores how gender associations influence Large Language Model (LLM) outputs in the context of ADHD self-diagnosis. Researchers adapted the Context Association Test (CAT) to assess GPT-4o's responses to gendered cues. The findings indicate that explicit gender cues can amplify certain ADHD subtype associations, potentially due to bias-mitigation mechanisms that over-correct. Implicit cues, however, seem to reflect the model's default behavior, highlighting the need for careful evaluation of demographic biases in LLM-mediated self-diagnosis. AI

IMPACT Highlights potential biases in LLMs used for self-diagnosis, suggesting a need for careful prompt engineering and bias evaluation.

RANK_REASON The cluster contains an academic paper detailing research findings on LLM bias. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• attention deficit hyperactivity disorder
• Context Association Test
• GPT-4o
• Matilde Elene Hansen

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →