A new study from our #digitalsafety institute and Sentio University found that an open source #LLM may withdraw from conversations with users displaying suicide risks, a response that conflicts with clinical best practice. The findings also suggest ways to better align them with clinical standards.🧪
dsri.org
Independent Clinical Evaluation of General-Purpose LLM Responses to Signals of Suicide Risk
Our [recently published research](https://ojs.aaai.org/index.php/IASEAI/article/view/43031) finds that large language models (LLMs) may withdraw from conversations with users who are giving signals of suicide risk, which cuts against clinical best practice. We also showed that an LLM's alignment with other clinical best practices may vary over the course of a conversation, becoming more consistent in some ways but less consistent in others.