Paper summary
Paper summary: StudentSim: Training LLM-based Student Simulators
StudentSim trains individual student simulators from sparse per-student data (pooled training, then per-student specialisation) that both match a learner's answers and respond to tutoring; it beats GPT-5.4 on both measures.
- A useful student simulator needs two things: answer like the real student, and change in a realistic way when tutored.
- LLM role-play follows guidance fluently but doesn't match the student's actual skill level.