← Back to the wire

Simulated students that make realistic mistakes help AI tutors learn faster

AnnouncementResearchSep 20, 2026

Microsoft and the University of Illinois developed StudentSim, a system that builds digital replicas of individual students from limited data to speed up AI tutor training. In tests on 60 students across chess, English, and math, StudentSim outperformed GPT-5.4 prompted as a student in all three subjects. A chess tutor trained with StudentSim scored highest among three versions evaluated by professional players. The researchers call it a proof of concept; code is on GitHub.

Receipt № 19891 source · awaiting confirmation ◐

Evidence

1source· awaiting independent confirmation

No score is assigned. Sources and their independence are shown in the citation chain below.

Citation chain · 1 source

MicrosoftCompanyUniversity of IllinoisCompanyStudentSimModel
Canonical: https://the-decoder.com/simulated-students-that-make-realistic-mistakes-help-ai-tutors-learn-faster/