A recent study based on a questionnaire of 100 higher-education students, predominantly from engineering-related fields, examines how students use and perceive Large Language Models (LLMs) in engineering education. The study, conducted by researcher Kudina and published on arXiv, reveals a balancing act between LLM benefits and serious concerns about accuracy, bias, and academic integrity.
How Students Use Large Language Models
According to the study, students primarily value LLMs for four key tasks:
- Writing support – generating drafts, improving grammar, and structuring papers.
- Conceptual clarification – explaining difficult topics in simpler terms.
- Coding assistance – debugging code, generating snippets, and learning syntax.
- Brainstorming – generating ideas for projects or assignments.
These uses reflect the efficiency and personalized learning that LLMs promise. However, the study notes that students simultaneously expressed concerns about inaccuracies, bias, overreliance, academic integrity, and the burden of verification.
Concerns Over Inaccuracy and Bias
Students reported that LLM outputs can contain errors or subtle biases that are hard to detect, especially for learners who lack deep domain expertise. The burden of verification – requiring students to check each output – was cited as a significant drawback. Overreliance on LLMs could undermine the development of critical thinking and problem-solving skills that engineering education aims to foster.
The Oracle vs Tutor Metaphor
Kudina analyzes two dominant metaphors in student discourse: LLMs as an 'oracle' and as a 'tutor'.
| Metaphor | Student Expectation | Risk |
|---|---|---|
| Oracle | LLM provides authoritative, expert answers. | Students may accept outputs uncritically, assuming correctness. |
| Tutor | LLM offers personalized, adaptive instruction. | May give false sense of learning; actual skill development requires struggle and reflection. |
These metaphors, the study argues, cultivate expectations of authority, expertise, and personalized learning that often exceed LLMs' actual capabilities.
The Concept of 'Cruel Optimism'
The paper introduces the concept of 'cruel optimism' to describe students' attachment to LLM promises of efficiency and personalized support. As Kudina writes, the perceived benefits depend on the very skills, vigilance, and expertise that students are still developing. The more students rely on LLMs for tasks they cannot yet perform independently, the more they risk stunting their own growth – yet they feel compelled to use these tools to keep up with peers and institutional pressures.
Recommendations for Purpose-Driven Integration
The study argues for a purpose-driven and context-sensitive approach to AI integration in engineering education. Key recommendations include:
- Developing critical AI literacy so students can evaluate LLM outputs and limitations.
- Reflective assessment design that rewards process and reasoning, not just final products.
- Pedagogical caution – using LLMs as supplementary tools, not replacements for foundational learning.
- Considering broader ethical and environmental impacts, such as energy consumption of large models.
While this research focuses on academia, its findings are directly relevant for enterprise technology leaders who oversee AI training programs or digital learning platforms. The same tensions between efficiency and skill erosion appear in corporate upskilling efforts. For CTOs and Chief Digital Officers, the study underscores that deploying AI tools without careful attention to user expertise and critical thinking can undermine long-term competence. The balancing act Kudina describes is not limited to the classroom – it applies to any organization integrating AI into knowledge work.
