Microsoft and the University of Illinois built StudentSim to replicate individual students from limited data and give AI tutors fast, low-cost feedback. In tests covering 60 students across chess, English, and math, it outperformed GPT-5.4. A chess tutor trained with StudentSim also earned the highest expert ratings among three versions tested.
The article Simulated students that make realistic mistakes help AI tutors learn faster appeared first on The Decoder.
source & further reading
the-decoder.com — original article
Trump announces "AI Force" and plans for an "AI czar" as he pushes unchecked AI growth
Qwen3.8-Omni-Flash undercuts Google's Gemini Flash pricing while matching its multimodal benchmarks
Unity launches official plugins for Claude Code and OpenAI Codex to stop AI agents from using outdated tutorials