Microsoft's StudentSim uses error-prone simulated students to train better AI tutors
The Decoder · rss · 2026-09-20
Microsoft and the University of Illinois built StudentSim, which recreates individual students from limited data and makes them make realistic mistakes, giving AI tutors fast, low-cost feedback. Across tests with 60 students in chess, English, and math, tutors trained with StudentSim outperformed GPT-5.4, and a chess tutor trained this way earned the highest expert ratings among three versions tested.
More from Research
- Sentence Transformers models quietly dominate Hugging Face's most-downloaded list — tomaarsen · 2026-09-20
- "Recipe for intelligence" paper published in Neuron — summerfieldlab · 2026-09-20
- Elegant Derivation Shows Policy Gradient Score Centering Is a Straight-Through Estimator — YouJiacheng · 2026-09-20
- WarpState: a 2B-param architecture with dual-timescale associative memory replacing ever-growing KV caches — zemondza · 2026-09-20
- Reproducing Jev's ideas from scratch: a parallel decision system with shared KV caches on Qwen3 1.5B — Sam_Witteveen · 2026-09-20
- LoRA, QLoRA, VeRA and more: what each of 6 fine-tuning variants actually optimizes — techNmak · 2026-09-20