Do Large Language Models Mentalize When They Teach?

TL;DR

This study investigates whether large language models teach by mentalizing or using simple heuristics, finding that they often resemble human-like Bayesian inference but do not consistently improve with scaffolding prompts.

Contribution

It applies cognitive models to analyze LLM teaching strategies, revealing that most LLMs behave similarly to Bayesian teachers and that scaffolding does not reliably enhance their teaching performance.

Findings

Most LLMs perform well and resemble human teaching strategies.

Bayes-Optimal models best explain LLM teaching choices.

Scaffolding prompts do not reliably improve LLM teaching performance.

Abstract

How do LLMs decide what to teach next: by reasoning about a learner's knowledge, or by using simpler rules of thumb? We test this in a controlled task previously used to study human teaching strategies. On each trial, a teacher LLM sees a hypothetical learner's trajectory through a reward-annotated directed graph and must reveal a single edge so the learner would choose a better path if they replanned. We run a range of LLMs as simulated teachers and fit their trial-by-trial choices with the same cognitive models used for humans: a Bayes-Optimal teacher that infers which transitions the learner is missing (inverse planning), weaker Bayesian variants, heuristic baselines (e.g., reward based), and non-mentalizing utility models. In a baseline experiment matched to the stimuli presented to human subjects, most LLMs perform well, show little change in strategy over trials, and their graph-by-graph performance is similar to that of humans. Model comparison (BIC) shows that Bayes-Optimal teaching best explains most models' choices. When given a scaffolding intervention, models follow auxiliary inference- or reward-focused prompts, but these scaffolds do not reliably improve later teaching on heuristic-incongruent test graphs and can sometimes reduce performance. Overall, cognitive model fits provide insight into LLM tutoring policies and show that prompt compliance does not guarantee better teaching decisions.