Physix Frontier badgePhysix Frontier
PublishChinese | EN

Physix Frontier · News Briefing Card (Hacker News · Oct 6, 2026)

Tennessee Two-Year Trial: Khanmigo Math Gains Modest

KEY FACTS

  • A two-year randomized controlled trial ran across 18 Tennessee middle schools, with students using Khanmigo for math tutoring.
  • Khanmigo was configured for guided tutoring that does not give direct answers, embedded in daily math remediation classes.
  • The trial showed math gains of about 1.3 national percentiles per semester.
  • The extrapolated effect for full-year active participation was 0.14 standard deviations, similar to non-AI Khan Academy practice.
  • 96% of students tried Khanmigo at least once, but only 17% of error-practice sessions included a message to the AI.

KEY DATA

18Trial schools
two yearsTrial duration
1.3 national percentilesPer-semester score gain
0.14 standard deviationsExtrapolated full-year effect

PHYSIX OBSERVATION

The bottleneck for AI tutoring is not the technology, but students' reluctance to use it. Khanmigo's effect is no different from ordinary practice, showing that providing a tool does not equal changing behavior. If an education product cannot design strong interactive motivation, even the best model is just decoration.

Source: Hacker News report