SLMs vs LLMs for Educational Assessment Design
A new arXiv paper (2605.15015) compares Small Language Models (SLMs) and Large Language Models (LLMs) for generating assessment questions aligned with Bloom's taxonomy. The study finds that while LLMs are powerful, they often rely on subjective evaluation and proprietary models. SLMs offer privacy and resource advantages but are underexplored. The research uses reproducible metrics and expert-informed evaluation to systematically compare generation quality across taxonomy levels.
Key facts
- arXiv paper 2605.15015 compares SLMs and LLMs for assessment design
- Focus on Bloom's taxonomy alignment
- SLMs address privacy and resource limitations
- LLMs rely on subjective or limited evaluation methods
- Study uses reproducible, pedagogically grounded metrics
- Expert-informed evaluation is used for comparison
- SLMs remain underexplored for assessment tasks
- Proprietary models are a limitation of current LLM research
Entities
Institutions
- arXiv