A new study published on arXiv analyzed the pedagogical quality of large language models (LLMs) in math tutoring compared to human experts. The research found that while larger LLMs approach expert-level performance on average, they exhibit distinct instructional and linguistic patterns. LLMs tend to use less of the specific discursive strategies employed by expert human tutors, such as restating and revoicing student errors, while producing responses that are longer, more lexically diverse, and more polite. The study suggests that focusing on these specific instructional strategies and linguistic features is crucial for evaluating tutoring systems. AI
IMPACT LLMs are nearing expert-level pedagogical quality in math tutoring, but their distinct communication styles may require further refinement for optimal student engagement.
RANK_REASON The cluster contains an academic paper detailing research findings on LLM performance in a specific domain. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
