Microsoft StudentSim trains AI tutors with simulated student responses

What is changing?

Microsoft Research announced StudentSim on October 12, a research system that imitates how an individual student answers and how that answer changes after tutor guidance. StudentSimEval jointly measures behavioral fidelity and responsiveness to guidance across chess, English writing by second-language learners, and mathematics.

The system first learns shared patterns from records across many students in the same domain, then specializes a LoRA adapter on the limited data of one learner. Microsoft released the code and research paper and notes that the reported results come from its own experiments.

StudentSim is not an education product and has not been deployed in classrooms. It is intended to compare tutoring strategies and training signals before slower, more expensive studies with real learners, not to replace those studies.