A new study examined the effectiveness of 'AI-proofing' strategies in higher education assessments by testing three state-of-the-art large language models (LLMs) on 75 questions from authentic college coursework across 17 disciplines. The research found that the best-performing LLM answered these questions with approximately 83% accuracy without supplementary materials and 85% with them. The study identified five strategies faculty used to make questions more resistant to AI, and analyzed their impact on LLM performance, discussing implications for academic integrity policies and the ethical deployment of LLMs in education. AI
IMPACT Suggests new methods for designing assessments that are more resistant to AI, impacting academic integrity and educational policy.
RANK_REASON Academic paper analyzing LLM performance on educational assessments. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →