A new study reveals that large language models like GPT-5, Gemini 2.5 Pro, and Claude Opus 4 often misinterpret psychological distress in online posts, particularly those from identity-based communities. Researchers found that open-weight models and even frontier models tend to overestimate distress, especially in cases of none-to-mild distress, by producing a high number of false positives. This contrasts with human judgments, where out-group assessments were more balanced, suggesting the models possess a AI
RANK_REASON [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
