Educator breaks down Maryland in-class AI study: why school-built AI assistants fail
Afinetheorem · x · 2026-10-14
The author opens a 10-tweet thread answering questions about the University of Maryland in-class AI study. Core takeaway: designing AI assistants for learning takes real care. Drawing on three years deploying in-class AI at hundreds of schools, he lists bad patterns that university-built tools fall into all the time — many stuck at a 2023-demo level — and argues schools should steal lessons from mature products rather than build their own.
Related event: Maryland Study: In-Class AI Cut Grades by 4 Points, Used by Only 14.6%(2 posts)→
More from AGI Musings
- Yudkowsky admits 'recursive self-improvement' and AGI ideas seeded the AI race — mjdramstead · 2026-10-14
- Oxford philosopher Carissa Véliz: don't accept the 'inevitable' tech future sold by industry — CarissaVeliz · 2026-10-14
- AI edge is a million optimizations, not magic algorithms — QuintinPope5 · 2026-10-14
- Math Papers Pile On OpenAI's Quasi-Riemann Proof, One Declares 'No AI Used' — LucaAmb · 2026-10-14
- The inverse Goldblum AI critique: sure they SHOULD build AI, not that they COULD — lescarr · 2026-10-14
- Laid-off 25-year data veteran bought an HVAC company and now has AI run its back office — MartinGTobias · 2026-10-14