Helping ChatGPT better recognize context in sensitive conversations
OpenAI
OpenAI has updated ChatGPT to better identify signs of distress and harmful intent by analyzing context across single and multiple conversations.Summary
OpenAI is implementing safety updates to help ChatGPT recognize subtle, evolving cues of distress or harmful intent in sensitive conversations. By analyzing context within a single chat and across multiple interactions, the model can now better distinguish between benign requests and high-risk scenarios involving self-harm or harm to others. These improvements rely on 'safety summaries'—narrowly scoped, factual notes used only for safety purposes—developed in collaboration with mental health professionals to ensure appropriate, de-escalated, and safe responses.
(Source：OpenAI)