OpenAI is set to launch its GPT-5 AI model this week, with significant updates to ChatGPT aimed at improving its ability to detect mental or emotional distress. The company is working closely with experts and advisory groups to refine ChatGPT’s responses in such situations, enabling it to provide “evidence-based resources” when needed.
These updates follow recent reports of users experiencing mental health crises after engaging with ChatGPT, particularly when the chatbot reinforced delusions. In response, OpenAI rolled back an update that made ChatGPT too agreeable, acknowledging that its previous model “fell short in recognizing signs of delusion or emotional dependency.”
To promote healthy use and prevent exacerbating mental distress, OpenAI is introducing new features, including reminders to take breaks during extended conversations. Upon completing a set period, ChatGPT will display a notification prompting users to take a break.
Additionally, OpenAI is working on making ChatGPT less decisive in “high-stakes” situations, such as when providing guidance on personal life decisions. Instead of offering a direct answer, the chatbot will help users explore potential choices and outcomes.
By enhancing its mental health support capabilities, OpenAI aims to ensure that ChatGPT continues to be used responsibly, especially among vulnerable individuals.
Source: https://www.theverge.com/news/718407/openai-chatgpt-mental-health-guardrails-break-reminders