Extended multi-turn dialogue does not produce effective tutoring: learning gains remain uniformly low and misconception reduction peaks around 24-25%¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? In multi-turn conversations, Learns More scores range from 5.51 for Qwen2.5-7B in Physics to 13.08 for Mistral-7B in Mathematics, and misconception reduction peaks around 24-25%. → Rima Hazra 2026

Evidence¶

Rima Hazra 2026¶

Rima Hazra, Bikram Ghuku, Ilona Marchenko, Yaroslava Tokarieva, Sayan Layek, Somnath Banerjee, Julia Stoyanovich, Mykola Pechenizkiy. (2026). SAFETUTORS: Benchmarking Pedagogical Safety in AI Tutoring Systems. arXiv preprint. https://arxiv.org/abs/2603.17373 doi:10.48550/arxiv.2603.17373

q2 · i? · design · r2

Multi-turn learning-trajectory analysis (Figure 4) across subjects and models, evaluating learns-more, misconception-reduction, and motivation-improvement indicators. The article reports "Learning gains remain uniformly low" and misconception reduction peaking around 24-25% for Qwen2.5-14B and Llama-3-8B; no effect size is printed.

"Learning gains re- main uniformly low, with "Learns More" scores ranging from 5.51 for Qwen2.5-7B in Physics to 13.08 for Mistral-7B in Mathematics."

Discussion¶

Related Claims¶

• Knowledge tracing performance on tutoring dialogues is relatively low, with a maximum of around 76% AUC, which the authors take to show dialogueKT is a challenging task. — related
• Fine-tuning frontier LLMs on authentic tutoring data improves performance on all six conversational and pedagogical benchmarks, moving values toward human-tutor levels — related
• Some classroom discussion approaches improve text comprehension, but gains are not uniform across approaches and more student talk alone does not produce them — related