Dialogue evidence shows coach feedback induces genuine conceptual change, shifting the agent from misconception to conceptually correct explanation rather than superficial answer alignment¶

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? Before instruction the agent misinterpreted exception chaining as error-hiding for security; after coach feedback it selected the correct option and gave a conceptually accurate explanation. → Jiajia Song 2026

Evidence¶

Jiajia Song 2026¶

Jiajia Song, Zhihan Guo, and Jionghao Lin. (2026). Simulating Novice Students Using Machine Unlearning and Relearning in Large Language Models. https://arxiv.org/abs/2603.26142 doi:10.48550/arxiv.2603.26142

q2 · i? · design · r2

Qualitative dialogue comparison in Table 1 (10% unlearning split): the agent moved from Option A with a security misunderstanding of the from keyword to Option B with an explanation that "fromis used to explicitly preserve and link the exception cause".

"This shift reflects a correction of the underlying mental model rather than superficial answer alignment, demonstrating genuine conceptual change induced through interaction."

Discussion¶

Related Claims¶

• Coach-guided interactive teaching yields higher and more progressive recovery than static relearning, particularly under moderate forgetting levels (10%–30%) — related
• Peer Discussion Improves Conceptual Understanding — related
• Conflict-based instruction improves science conceptual learning, though staged contradictions helped only learners who reported being confused, and no study isolates disequilibrium as the mechanism — a broader claim this one bears on
• Sycophantic failure manifests as indiscriminate correct flips rather than explicit sycophantic agreement, and SFT shifts behavior toward pushback and maintain under non-targeted feedback — related
• Step-by-step automated annotations accelerated correction of procedural misconceptions, with intervention students rectifying fundamental errors weeks before control peers — related
• Instruction that confronts science misconceptions (conceptual-conflict interventions, refutation text) produces conceptual change more often than standard exposition — a broader claim this one bears on
• Machine unlearning induces progressively stronger novice-level performance in an LLM teachable agent, with accuracy and F1 declining as the unlearning ratio increases — related
• Feedback probing for deeper conceptual understanding is reported to be particularly effective in classwide peer feedback, though cognitively more demanding — related
• The deterministic reflection gate reliably distinguished low-effort defenses from substantive ones during validation, coaching filler answers rather than unlocking — related