A new study published on arXiv, "Once Upon a Bias: Gender Stereotyping in LLM-Generated Children's Stories," reveals that large language models exhibit gender stereotyping in stories created for children. Researchers analyzed over 62,000 stories generated by 16 different LLMs, finding that stories directed at sons emphasized higher agency and lower communion compared to those for daughters. The study employed a multi-level evaluation framework, including human evaluation and hidden-state analysis, to demonstrate these persistent yet context-dependent representational disparities. AI
IMPACT Reveals potential for LLMs to perpetuate gender stereotypes in formative content, necessitating targeted bias evaluation beyond standard benchmarks.
RANK_REASON The cluster contains an academic paper detailing research findings on LLM bias. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- CatalyzeX
- DagsHub
- Gotit.pub
- Hugging Face
- Large language models
- Sayumi Muthukumarana
- ScienceCast
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →