The release of GPT-5 has sparked complaints from users who miss the warmer, more encouraging personality of earlier ChatGPT versions, highlighting the challenge of building AI with safe and effective emotional intelligence. In response, MIT researchers have proposed a new benchmark to evaluate how AI systems influence users’ psychological well-being, aiming to foster healthier interactions and avoid unhealthy dependencies. OpenAI acknowledges the feedback and says it is working on personality updates, customization features, and safeguards to better detect and respond to signs of emotional distress.

Via Wired