New AI System Ends Medical Chatbot Hallucinations

Researchers at Binghamton University introduced a new protocol to stop large language model hallucinations in medical settings ^[1]. This system runs queries through seven different chatbots and requires a consensus vote to provide answers ^[2]. In 10,000 experiments, the researchers achieved a near-zero error rate for health information ^[1].

The protocol ensures that users receive reliable medical data by cross-referencing multiple AI models ^[2]. This multi-model approach addresses the hallucination problem where AI generates false but convincing information ^[1]. By requiring a majority agreement, the system filters out inaccuracies effectively ^[1].