Under persona simulation with hidden ground-truth mastery, Adaptive sessions yield lower final-belief mastery MAE (0.12) than Random-topic (0.16) and Frozen (0.20) controls, and belief updates shrink as estimates converge¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Final-belief mastery MAE is lowest under Adaptive (0.12 vs 0.16/0.20) in the 6-persona Claude Sonnet 4.5-graded run. → Kailai He 2026 q2 i? Mean update magnitude |Δp| shrinks across the six Adaptive persona sessions (0.14→0.10) but stays flat or drifts upward under both controls. → Kailai He 2026 (2)

Evidence¶

Kailai He 2026¶

Kailai He, Zhihao Wu, Linhai Zhang, Runcong Zhao, Yulan He, Jiazheng Li. (2026). CoLearn: An Agentic Tutor that Learns its Learner in a Human–AI Co-Learning Loop. https://arxiv.org/abs/2609.21154 doi:10.48550/arxiv.2609.21154

q2 · i? · design · r2

Persona-simulation run (6 personas, 132 rounds) driving the live grading and BKT services with hidden ground-truth mastery; the article reports "mastery MAE is lowest underAdaptive( 0.12 vs. 0.16/0.20; Table 2)".

"belief mastery MAE is lowest underAdaptive( 0.12 vs. 0.16/0.20; Table 2), and the per-learner belief error|ˆp−p∗| falls in step (Figure 3a)."

Kailai He 2026 (2)¶

Kailai He, Zhihao Wu, Linhai Zhang, Runcong Zhao, Yulan He, Jiazheng Li. (2026). CoLearn: An Agentic Tutor that Learns its Learner in a Human–AI Co-Learning Loop. https://arxiv.org/abs/2609.21154

q2 · i? · design · r2

Secondary aggregate signal from the same 6-persona simulation: the article reports that "mean|∆p| shrinks across the sixAdaptivepersona sessions ( 0.14→0.10 )" while both controls stay flat or drift upward.

"mean|∆p| shrinks across the sixAdaptivepersona sessions ( 0.14→0.10 ) but stays flat or drifts upward under both controls (Figure 3b)."

Discussion¶

Related Claims¶

• In a preliminary 2×3 controlled lesson study across five backbone LLMs, structured student agents produce more differentiated mastery and misconception traces than a baseline simulator — related
• In LLM-simulated subjects with known skill levels, the Executive LLM yields better skill-level recovery (lower MAE) than Independent Agents — related