Design and Implementation of a Large Language Model(LLM) Based Science Rationality Assistant to Promote Scientific Temper

Abstract

Scientific temper-the capacity for evidence-based reasoning, skepticism, and critical inquiry-is a fundamental attribute of modern citizenship and a constitutional duty in India. However, its development is often hindered by widespread pseudoscience, misinformation, and lack of access to scientific education. This paper presents the design and implementation of a Science Rationality Assistant (GEMMASRA) powered by Gemma Large Language Models (LLM) to address this gap. The assistant is designed to engage users in interactive dialogue, explain scientific concepts in accessible language, and critically evaluate unverified claims. It leverages Gemma 3n LLM capabilities in natural language understanding, multilingual generation, and context-sensitive reasoning to simulate scientific thinking processes. A custom prompt-engineering strategy and curated dataset of verified scientific content were used to align responses with rational inquiry principles. The assistant was evaluated on three dimensions: accuracy, engagement, and promotion of critical thinking, using a cohort of students and general users. Results indicate that the system significantly improved users' ability to differentiate between evidence-based and pseudoscientific claims. The paper concludes with a discussion on ethical safeguards, limitations of current LLMs, and the broader implications for deploying AI tools to foster scientific temper in educational and public discourse settings.

Ähnliche Arbeiten