Binghamton Researchers Halt AI Hallucinations in Science

Researchers at Binghamton University developed a multi-agent protocol to stop AI hallucinations in scientific and medical fields ^[1][3]. The system uses Retrieval-Augmented Generation (RAG) to force models to reference authoritative databases ^[1]. This method ensures large language models remain accurate in high-stakes healthcare environments ^[2].

The new verification method involves a “voting” system using seven different large language models ^[1][2]. Published in the journal STAR Protocols, the system cross-references data to catch and remove fake information ^[1][3]. This approach provides a more reliable framework for using AI in complex research ^[2].