Our abductive team researches methods of increasing explainability and trustworthiness of Large Language Models (LLM) by reducing the chances of hallucinations. We use a neuro-symbolic approach, where the question and context given to an LLM gets converted to a logic programming language (Prolog) and the answer is deduced by the Prolog interpreter. Currently we are developing a fine tuning method which uses abductive learning to create an LLM that specializes in translating from natural language to Prolog
