EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·6d ago·5 min read
AI Tutoring with Khanmigo in a Two-Year School Experiment
TL;DR
A two-year study in Tennessee middle schools found that using the AI tutor Khanmigo during math sessions improved student achievement by 1.3 national percentile ranks per term. However, low engagement levels limited its effectiveness.
✦ Why It Matters
Focus on increasing student engagement with AI tools to maximize their educational impact.
Key Takeaways
- Implement strategies to boost student interaction with AI tutors.
- Monitor engagement levels to assess AI tool effectiveness.
- Encourage meaningful dialogue between students and AI.
- Consider combining AI tutoring with traditional methods for better outcomes.
Full Summary
Generative AI has been heralded as a potential game-changer in education, particularly through personalized tutoring. In a two-year cluster randomized trial involving 18 middle schools in Tennessee, students used Khan Academy's AI tutor, Khanmigo, designed to coach rather than provide direct answers during remedial math sessions.
The study revealed that students' math achievement increased by 1.3 national percentile ranks per term, translating to an effect size of approximately 0.14 standard deviations after a full year of participation. Despite these gains, the effectiveness of Khanmigo was hampered by low engagement; while 96 percent of students tried the AI tutor, the median student interacted with it only a third of the days they practiced.
Most interactions consisted of simple responses rather than meaningful dialogue. This suggests that simply providing access to AI tutoring is insufficient; fostering student engagement is crucial for realizing its full potential.