Comparing Large Language Models: Analysis of Functionality and Limitations

In recent years, the field of artificial intelligence and machine learning has evolved at an unprecedented rate, with large language models (LLM) being a key element of this development. These models have become fundamental tools in many applications from natural language processing to content generation and solving complex problems. This bachelor's thesis focuses on comparing the performance of selected LLM with an emphasis on the Czech language to explore their ability to understand and generate content in this specific linguistic context. The work evaluates and compares the capabilities of models such as GPT-3.5, Claude 3 Sonnet, and others in areas that include text comprehension, grammar, mathematics, and biology. Quantitative data were obtained through standardized tests, including high school didactic tests and SCIO tests, and supplemented with qualitative analysis of the models' responses. Special attention is given to identifying specific errors and limitations of these models in the context of the Czech language, revealing key areas for further development and improvement. The results suggest that despite an elevated level of knowledge in domains, such as biology, LLM face significant challenges in areas requiring deeper linguistic understanding and complex logical reasoning, especially in mathematics and advanced Czech grammar. This work contributes to a better understanding of the possibilities and limitations of LLM in the Czech linguistic environment and offers valuable insights for research communities and developers of artificial intelligence. Discussions of the results and recommendations for future research emphasize the need for expanded data sets and algorithm improvements to enhance the universality and accuracy of LLM.