Ensuring Factuality, Consistency and Justifiability in LLMs

• Grant number: 25/20109-4
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Start date: September 01, 2026
• End date: February 28, 2029
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Denis Deratani Mauá
• Grantee: Victor Hugo Nascimento Rocha
• Host Institution: Instituto de Estudos Avançados (IEA). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Associated research grant: 22/02937-9 - Neural inductive logic programming, AP.PNGP.PI

Abstract

The recent advances in Large Language Models (LLMs) have transformed conversational agents into tools capable of sustaining long dialogues, answering complex questions, and performing certain reasoning tasks. Nevertheless, such systems face key limitations: they do not provide guarantees regarding the factual correctness of their outputs, have high costs for updating their knowledge base, and lack clear mechanisms to justify their responses. Existing approaches, such as structured output generation and Retrieval-Augmented Generation (RAG), offer only partial solutions, as they fail to ensure factual accuracy and often propagate errors. Similarly, workflows and orchestration frameworks for LLM-based agents suffer from comparable shortcomings.This project aims to advance the state of the art in conversational agents by developing systems that guarantee factual correctness, support flexible knowledge base updates, and are able to produce convincing explanations grounded in reliable sources. To this end, we will investigate theoretical frameworks that enable the integration of LLMs with explicit knowledge representation and formal inference, particularly neuro-symbolic approaches based on logical rules and computational argumentation models. The expected outcome is the design of agents capable of delivering reliable, up-to-date responses supported by well-founded explanations, thereby contributing to applications with significant scientific and social impact. (AU)