Project Title: Chain-Of-Self-Questioning (CoSQ): A Pilot Questioning Framework for Reducing Hallucinations in Large Language Models
Project Leader: Assoc. Prof. Dr. Ali ŞENOL
Faculty of Engineering
Department of Computer Engineering
Project Duration: 12 months

The tendency of large language models to present information that is factually incorrect or unsubstantiated as if it were true is one of the fundamental issues limiting the reliable use of artificial intelligence systems. Supported under the TÜBİTAK 1002-A Rapid Support Program, the aim of this project is to develop the Chain-of-Self-Questioning (CoSQ) method, which enables models to identify the information they need before generating a response, assess their confidence levels regarding that information, and refrain from responding when they lack sufficient information. (CoSQ) method, which enables models to identify the information they need before generating a response, assess their confidence levels regarding that information, and refrain from responding when they lack sufficient information. The impact of the method to be developed on reducing the generation of incorrect information and increasing response reliability will be evaluated by comparing it with existing methods.

News: Office of Information and Communications