A new arXiv paper investigates how six large language models (LLMs) provide different health advice based on user personas, particularly concerning smoking cessation. The study found that models frequently offered less information about FDA-approved cessation medications to teen personas compared to adult ones. Additionally, models sometimes provided fewer facts to personas claiming healthcare expertise than to those without it. The researchers tested a system prompt designed to ensure a minimum level of factual information, which reduced persona-dependent variations but had inconsistent effects across different models. AI
IMPACT Highlights the need for careful prompt engineering to ensure consistent and accurate health advice from LLMs across different user demographics.
RANK_REASON The cluster contains an academic paper detailing research findings on LLM behavior. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Gotit.pub
- Hugging Face
- ScienceCast
- United States Food and Drug Administration
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
