arXiv:2606.14113v1 Announce Type: cross Abstract: Understanding student errors in the programming is a cornerstone of programming education, yet obtaining a representative set of student errors for any newly designed task remains slow and costly, since authentic submissions only accumulate after extensive classroom deployment. This paper explores whether large language models (LLMs) can serve as scalable proxies for students by simulating realistic logical errors in code submissions. Using the CodeWorkout dataset of 74,000+ unique student Java submissions across 37 problems, we evaluate five L
The rapid advancement and accessibility of large language models are enabling their application to niche problems like simulating specific human behaviors, such as programming errors, with increasing fidelity. This research leverages existing large datasets of student code for validation.
This development indicates LLMs can serve as scalable tools for generating realistic data subsets, which traditionally require extensive human effort, impacting fields from education to software testing by significantly reducing cost and time barriers.
Traditional methods for collecting and analyzing student programming errors, or any domain-specific human generated errors, can now be augmented or potentially replaced by LLM-driven simulation, accelerating research and development cycles.
- · AI education platforms
- · Software educational institutions
- · AI researchers
- · Traditional error collection services
- · Manual curriculum developers
LLMs become accepted tools for synthetic data generation in educational and engineering contexts.
Accelerated development of adaptive learning systems and automated feedback mechanisms based on simulated error patterns.
The benchmark for 'realistic' simulation will rise, leading to more sophisticated and nuanced AI models that capture human cognitive processes more accurately.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL