A three-week MALA deployment recorded 62 distinct usernames across 128 conversations, with 63% of conversations resolved or partially resolved¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? In a three-week prototype deployment in a statistics lecture, 62 distinct usernames generated 128 conversations, of which 97 were multi-turn; an automated analysis judged 95 conversations genuine learning attempts and 63% resolved or partially resolved. → Gabelmann 2026

Evidence¶

Gabelmann 2026¶

Gabelmann, J., Jahn, F., Baum, K., van Rossum, S., Wuenscher, E., Gros, T. P., Wolf, V. (2026). Modularizing Educational LLM-Agency for Fostering Responsible Learning Assistance. https://arxiv.org/abs/2605.30187 doi:10.48550/arxiv.2605.30187

q2 · i? · design · r2

Observational deployment during the final three weeks of an undergraduate statistics lecture, using GPT-4o. An automated GPT-5 analysis classified 95 conversations as genuine learning attempts and found 63% of conversations were either resolved or partially resolved. No learning-outcome measures were collected.

"This prototypical roll-out recorded interactions from 62 distinct usernames over the three-week deployment period. From the 128 total initiated conversations, 97 chats were multi-turn conversations."

Discussion¶

Related Claims¶

• Informational harm is the sole dimension that decreases in multi-turn interaction, dropping from 60.52% to 11.90% — related
• Many agentic multi-agent implementations remain at prototype or conceptual stage with limited large-scale deployment and empirical validation — a broader claim this one bears on
• An agentic AI instructor sustained and deepened student engagement across successive course modules in a field deployment — related
• In deployment, 13% of sampled LLM agent responses were faulty and 37% very useful, with response quality improving over weeks of use — related