On its own, an AI plays a student who is too capable and too neat. QLM StandIn makes AI-played students answer like your real students: one student at a time, or a whole class.
Start a free pilotWhat we need ↓Illustration. Each row is a student, each square an answer (teal = right). An AI-played class gets nearly everything right. A real class differs from student to student, answers in streaks, and misses the hard questions more.
Some students are strong, some struggle. Not everyone sits near the top.
A right answer tends to follow a right one, and a wrong one follows a wrong one.
More students miss the hard questions, not just a few.
Students get better as they practice.
Practice with a room that answers like a real one.
Try a lesson before real students see it.
Test a tutor against students who act like yours.
Ask what-if questions without putting real children at risk.
Or write to kumar@quantumlearningmachines.com with "QLM StandIn pilot" in the subject.
One CSV file, one row per answer:
| Column | What it is | Example |
|---|---|---|
student_id | An anonymous code. Never a name. | s0412 |
question_id | Which question or skill it was | frac-07 |
right | 1 if right on the first try without a hint, else 0 | 1 |
order | Which answer this was for that student: 1 for their first, 2 for their second, and so on. Each student starts again at 1. | 3 |
A file for two students looks like this:
student_id,question_id,right,order s0412,frac-07,1,1 s0412,frac-09,0,2 s0412,frac-07,1,3 s0538,frac-02,0,1 s0538,frac-07,1,2
Only have times, not counts? Send the times; we'll turn them into the order.
With 40 or more answers each. More is better.
Tell us what was taught and to whom.
Class, seconds per answer, and hints used, if you have them.
A 2025 study tested eleven AI models on U.S. national test questions for grades 4, 8 and 12. None matched real students: they scored too high and made different mistakes. Microsoft researchers warn that an AI student can sound young and still be too capable, so a tutor looks better than it is.
Sources: Srivatsa, Maurya & Kochmar, 2025 (paper, data and code); Microsoft StudentSim (code).