Abstract:

The rapid adoption of Large Language Models (LLMs) such as ChatGPT and Claude is transforming learning practices in higher education, particularly in computer science programs where students increasingly rely on AI-assisted tools during problem-solving activities. Understanding how students interact with these systems and how they perceive their educational value is becoming an important challenge for educators within the current European higher education context, where generative AI is rapidly reshaping digital learning environments.

This study explores the perceived usefulness and interaction strategies associated with LLM tools among fourth-year Computer Science Engineering students at a European university. Senior students were selected because they typically demonstrate higher levels of academic maturity and self-regulated learning skills than early-year cohorts.

Data were collected through an anonymous online questionnaire distributed via the institutional learning platform after completion of first-semester courses. The survey, implemented using Microsoft Forms, included 15 items combining closed questions and optional open-ended responses. The instrument examined several aspects of LLM usage, including frequency of use, prompting strategies depending on the type of academic task (e.g., programming exercises or exam preparation), validation behaviors such as fact-checking or logical verification of generated responses, and students’ perceived impact on their learning outcomes as well as their attitudes toward the reliability and responsible use of AI-generated content.

The present paper reports the results of an initial pilot dataset consisting of 17 valid responses. Preliminary findings indicate widespread adoption of LLM tools among students, with ChatGPT being the most frequently used system, often complemented by other platforms such as Gemini, Copilot, or Perplexity. Student perceptions reveal heterogeneous views of these tools: while some students mainly perceive LLMs as mechanisms for increasing productivity and saving time, others report improvements in conceptual understanding and learning efficiency. Qualitative responses also suggest the emergence of reflective interaction strategies, including prompt refinement, verification of generated content, and critical evaluation of AI outputs.

This ongoing study aims to further expand the dataset and deepen the analysis of student interaction patterns with LLM tools. The results are expected to provide insights that may support instructors in developing pedagogical guidelines for the responsible and effective integration of generative AI in computer science education, as well as practical recommendations for adapting assessment strategies and fostering critical thinking in AI-supported learning environments.

Keywords:

Large Language Models (LLMs), AI-assisted learning, Computer Science education, Student interaction strategies, Perceived usefulness of AI tools.