Off-the-shelf LLM tutors are more verbose than human tutors, leaving students only 5-15% of talk time versus close to 30% in human tutoring¶

Claim · All claims
Evidence · 1 study · design r2 · q2
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¶

q2 i? In benchmark simulations, off-the-shelf models leave students 5-15% of talk time while human tutors leave students close to 30%. → Janos Perczel 2025

Evidence¶

Janos Perczel 2025¶

Janos Perczel, Jin Chow, and Dorottya Demszky. (2025). TeachLM: Post-Training LLMs for Education Using Authentic Learning Data. https://arxiv.org/abs/2510.05087 doi:10.48550/arxiv.2510.05087

q2 · i? · design · r2

Benchmarking study (Fig. 6) comparing human tutors from 80,000 hours of Polygence data with state-of-the-art LLMs from OpenAI, Google, Anthropic, Meta, and XAI using the fine-tuned student model protocol. Human tutors also speak only 72 words per turn versus 150-300 for most models.

"In Fig. 6(a), we find that in human-to-human tutor-led dialogues students speak close to 30% of the time. Off-the-shelf models are typically more verbose than humans, leaving only 5-15% of talk time to students."

Discussion¶

Related Claims¶

• Fine-tuning frontier LLMs on authentic tutoring data improves performance on all six conversational and pedagogical benchmarks, moving values toward human-tutor levels — related