ProfileRL outperforms prompting and fine-tuning baselines on turn-level student simulation, with statistically significant gains on Acts, Correctness, and Errors¶

Claim · All claims
Evidence · 1 study · causal r2 · q2
Currency · newest study 2026 · checked against its sources 2026-10-20 (automated, 1 of 1 entry) · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? ProfileRL improves Acts by 0.042, Corr. by 0.037, and Errors by 0.038 over History SFT, with statistically significant gains on these three metrics (p<0.05 under a paired t-test). → Zhangqi Duan 2026 q2 i? ProfileRL outperforms all baselines on every metric, and the gains hold with a smaller simulator (Llama-3.2-3B). → Zhangqi Duan 2026

Evidence¶

Zhangqi Duan 2026¶

Zhangqi Duan, Shuyan Huang, Alexander Scarlatos, Jaewook Lee, Simon Woodhead, Andrew Lan. (2026). Who Am I? History-Aware Profiles for Student Simulation in Tutoring Dialogues. arXiv preprint. https://arxiv.org/abs/2605.30051 doi:10.48550/arxiv.2605.30051

q2 · i? · causal · r2

Quantitative comparison on 5-fold cross-validation over a real-world math platform dataset (1,775 dialogues, 66,705 QA records, 670 students), with a 80/10/10 split across students. ProfileRL reaches the best scores, e.g. Acts 0.644 vs History SFT 0.602, and the article reports "statistically significant gains on these three metrics".

"Compared to History SFT, ProfileRL improves Acts by 0.042, Corr. by 0.037, and Errors by 0.038, with statistically significant gains on these three metrics (p<0.05under a pairedt-test)"

Discussion¶

Related Claims¶

• Both RL stages are needed: removing DPO and GRPO training causes the largest performance drop in student simulation — related
• Reinforcement learning with a learned alignment reward substantially improves question-concept alignment after supervised fine-tuning, while answerability remains high throughout training — related
• Prompting-based student simulators maintain artificially high pass rates (~80%) from the first submission step, while fine-tuned models track real students' low-start, sharp-increase trajectories — related
• Epi2Diff achieves the best performance across all four item-difficulty test sets and all reported metrics, outperforming SLM fine-tuning, LLM prompting, and supervised LLM adaptation baselines — related
• Improvements over baselines are statistically significant, with Cohen's d ranging from 1.2 to 1.8 for key metrics — related
• Across all three domains, trained StudentSim simulators outperform every baseline on guidance responsiveness (R), with the largest gap on open-ended guidance modes — related