Classroom-scale data (5-80 samples per misconception) is insufficient for reliable tutor-model misconception acquisition¶

Claim · All claims
Evidence · 1 study · causal r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? In the small-sample regime, only one of ten misconceptions reaches MA above 90% at 80 samples per misconception. → Naiming Liu 2026

Evidence¶

Naiming Liu 2026¶

Naiming Liu, Xinghe Chen, Richard Baraniuk, Mrinmaya Sachan, and Shashank Sonkar. (2026). Misconception Acquisition Dynamics in Large Language Models. https://arxiv.org/abs/2604.00818 doi:10.48550/arxiv.2604.00818

q2 · i? · causal · r2

Small-sample regime experiments (nm in {5,10,20,40,80}) training tutor models on ten misconceptions simultaneously; OCA remained stable but average MA failed to reach the 90% threshold for most misconceptions.

"With only 80 samples per misconception, only one out of ten misconceptions achieves MA above 90%."

Discussion¶

Related Claims¶

• Without step-level supervision, neither the student model nor the tutor model can acquire misconceptions regardless of training data size — related
• Tutor models trained on ten misconceptions simultaneously learn each misconception while overall correct accuracy remains stable or improves (93% to 98%) — related
• Current GenAI tools are not designed to support active learning and their apparent domain expertise is brittle — related