Beyond the AI Tutor: Social Learning with LLM Agents

Authors: Harsh Kumar, Zi Kang (Jace) Mu, Jonathan Vincentius, Ashton Anderson Year: 2026 Venue: arXiv (cs.HC, cs.CY)

Summary

Multi-agent LLM configurations for learning. Two experiments (N=315 math, N=247 writing) — tutor+peers outperforms single tutor. Multi-model condition avoids idea homogeneity.

Key Contributions

Multi-agent LLM configurations for learning. Two experiments (N=315 math, N=247 writing) — tutor+peers outperforms single tutor. Multi-model condition avoids idea homogeneity.

Connections

• intelligent-tutoring

• scaffolding

• collaborative-ai-tutoring

• llm-student-modeling-memory

• ai-tutor-effectiveness-review

Related Pages

• agentic-ai-ecosystems-higher-education — Extends multi-agent LLM learning to institutional scale

Citation

Harsh Kumar et al. (2026). Beyond the AI Tutor: Social Learning with LLM Agents. arXiv:2604.02677. cs.HC, cs.CY.