Dear Sir
We read with great interest the study by Nikolovski et al. comparing the performance of two large language models (LLMs) when tested on a pediatric cardiology board-style examination.[1] The authors show that LLMs perform adequately and with high accuracy, approaching passing thresholds and therefore acting as educational aids for trainees undergoing the board-examination journey.
While the study appropriately frames LLMs as “test-takers,” these findings raise a potentially complementary role that may aid the trainees even further, as “generators” of board examination-like questions. Given that LLMs can engage with board-style examination questions at almost passing levels, this fact may be used to produce, using the LLMs, clinically relevant and vignette-based multiple-choice questions (MCQs) that align with established teaching curricula. The capability would even expand to provide “question bank-style” questions generated using LLMs, which would support trainee preparation immensely.
LLMs may even provide explanations alongside answers and therefore act as “personal tutors,” reinforcing concept retention and higher-order clinical understanding. If guidelines are added into the LLM chatbots, these systems may be able to generate MCQs that are guideline-aligned and therefore highly relevant for trainees.
However, there are important limitations that must be acknowledged. Artificial intelligence (AI)-generated content remains susceptible to inaccuracies, variability, and potential misalignment with evolving clinical standards. Accordingly, expert oversight and validation are essential to ensure educational reliability and prevent propagation of errors.
Integrating AI-assisted question generation and explanation within existing educational resources may represent a pragmatic step toward more coordinated and accessible examination preparation. Such an approach would not replace established learning tools, but rather complement them by enhancing scalability, adaptability, and alignment with defined educational standards, ultimately supporting greater consistency across diverse training environments.
Conflicts of interest
There are no conflicts of interest.
Funding Statement
Nil.
REFERENCE
- 1.Nikolovski N, Morgan CT, Gritti MN. Comparing closed and open large language models on pediatric cardiology board exam performance. Ann Pediatr Cardiol. 2025;18:590–3. doi: 10.4103/apc.apc_301_25. [DOI] [PMC free article] [PubMed] [Google Scholar]