Title: Large Language Models, Reasoning, and the Issue of Plagiarism

Abstract:

Recent assertions suggest that Large Language Models (LLMs) possess the capacity for human-level general intelligence and can generate novel scientific insights. However, these claims are complicated by two conflicting interpretations of LLM functionality: one portrays them as engines of synthesis capable of genuine reasoning to uncover new knowledge, while the other views them as systems that simply retrieve and republish existing work without proper credit. Within the context of scientific inquiry, this dichotomy is best framed as a conflict between authentic reasoning and plagiarism. Determining the actual nature of LLM capabilities is difficult, as critical elements such as training data and interaction logs remain inaccessible. Consequently, assertions regarding LLM reasoning fail to meet Popper’s criteria for refutability. To address this, we outline recommendations for transparency and reproducibility, enabling the scientific community to evaluate reasoning claims through rigorous methodology. Furthermore, we argue that the prevailing emphasis on the reasoning narrative is inadvertently fostering plagiarism within scientific publications, and we examine potential solutions to this growing concern.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC