Current GenAI tools are not designed to support active learning and their apparent domain expertise is brittle¶

Claim · All claims
Evidence · 1 study · theoretical r3 · q1
Currency · newest study 2025 · not yet checked against its sources · not reviewed by a person · recheck by 2028: tool and model accuracy changes quickly

Subclaims¶

q1 i? Current training regimes of GenAI yield neither robust domain expertise nor pedagogy-aligned tutoring capabilities, so the models cannot engage learners in genuine active learning processes in STEM. → Peter Wulff and Marcus Kubsch 2025

Evidence¶

Peter Wulff and Marcus Kubsch 2025¶

Peter Wulff and Marcus Kubsch. (2025). Learning against the machine: the double edged sword of (Gen)AI in STEM education. International Journal of STEM Education 12:66. https://doi.org/10.1186/s40594-025-00588-6

q1 · i? · theoretical · r3

The commentary reports, citing Gupta et al. (2025), that GenAI training regimes yield neither robust domain expertise nor pedagogy-aligned tutoring capabilities; the reviewed study's design is not described in the article. The authors conclude current GenAI models cannot provide the affordances of active learning.

"Empirical evidence indicated that current training regimes of GenAI neither yield robust domain expertise nor pedagogy-aligned capabilities as expert tutors (Gupta et al., 2025)."

Discussion¶

Related Claims¶

• Both pedagogy-expertise-weighted and unanimous-vote ensembles frequently worsen LLM alignment with student learning — related
• Small input perturbations such as minor spelling errors can reduce state-of-the-art dialogue and assessment model accuracy by over 30% — related
• LLM accuracy on the CDPK pedagogy benchmark ranges from 28% to 89%, with closed-weight reasoning models dominating the top of the leaderboard — related
• Classroom-scale data (5-80 samples per misconception) is insufficient for reliable tutor-model misconception acquisition — related