In a study comparing GPT-4’s ability to grade short-answer responses against specialized models, GPT-4 displayed robust performance, especially when reference answers were excluded. Using the SciEntsBank and Beetle datasets, GPT-4 achieved notable F1 scores of 0.744 and 0.651, respectively. While its capabilities are comparable to systems from years past, BERT family models, which undergo task-specific training, still surpass it. Dr. Kortemeyer’s research highlights GPT-4’s potential in higher education, but concerns about data security with cloud-based models persist. As AI delves deeper into educational assessment, the trade-off between performance, adaptability, and security remains a primary focus.

For further actions, you may consider blocking this person and/or reporting abuse
Top comments (0)