Positive and negative impacts of large language models

This bachelor's thesis analyzes the positive and negative impacts of large language models (LLMs) on society. The aim is to evaluate the benefits and risks across various areas of application. The thesis also focuses on the architecture of LLMs, their development, and their primary applications. The theoretical part is based on a literature review, which serves to define key concepts, explain the principles of generative artificial intelligence (GenAI), and identify major advantages and challenges associated with the use of these technologies. The practical part is built on a case study and qualitative content analysis of outputs from four selected models (ChatGPT-4o, Claude Sonnet 4, Gemini 2.5 Pro, and LLaMa 4 Maverick), which were tested on analytical, creative, and informational tasks. The results show that the models differ in linguistic quality, factual accuracy, and stylistic expression, with none of them being universally superior. The thesis also emphasizes the importance of well-crafted prompt formulation and a critical approach to interpreting model outputs. Based on the findings, recommendations are proposed for the practical and responsible use of LLMs, taking into account their specific strengths and limitations.