# Large Language Models as Students Who Think Aloud: Overly Coherent, Verbose, and Confident

arXiv: https://arxiv.org/abs/2602.01015 · HTML: https://mlsanity.com/paper/2602.01015/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 8 of its 26 references that are in the map.
- First posted on arXiv 1 Feb 2026.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 8 of 8)

- [Sparks of Artificial General Intelligence: Early experiments with GPT-4](https://mlsanity.com/paper/2303.12712/) (2023)
- [Ruffle&Riley: Insights from Designing and Evaluating a Large Language Model-Based Conversational Tutoring System](https://mlsanity.com/paper/2404.17460/) (2024)
- [LearnLM: Improving Gemini for Learning](https://mlsanity.com/paper/2412.16429/) (2024)
- [Can Large Language Models Match Tutoring System Adaptivity? A Benchmarking Study](https://mlsanity.com/paper/2504.05570/) (2025)
- [Combining Large Language Models with Tutoring System Intelligence: A Case Study in Caregiver Homework Support](https://mlsanity.com/paper/2412.11995/) (2024)
- [3DG: A Framework for Using Generative AI for Handling Sparse Learner Performance Data From Intelligent Tutoring Systems](https://mlsanity.com/paper/2402.01746/) (2024)
- [Do Tutors Learn from Equity Training and Can Generative AI Assess It?](https://mlsanity.com/paper/2412.11255/) (2024)
- [Using Think-Aloud Data to Understand Relations between Self-Regulation Cycle Characteristics and Student Performance in Intelligent Tutoring Systems](https://mlsanity.com/paper/2312.05675/) (2023)

## Built on by (most cited first, 2 of 2)

- [Metacognition in LLMs: Foundations, Progress, and Opportunities](https://mlsanity.com/paper/2607.11881/) (2026)
- [Representation Learning to Study Temporal Dynamics in Tutorial Scaffolding](https://mlsanity.com/paper/2603.24535/) (2026)

## Abstract (arXiv)

Large language models (LLMs) are increasingly embedded in AI-based tutoring systems. Can they faithfully model novice reasoning and metacognitive judgments? Existing evaluations emphasize problem-solving accuracy, overlooking the fragmented and imperfect reasoning that characterizes human learning. We evaluate LLMs as novices using 630 think-aloud utterances from multi-step chemistry tutoring problems with problem-solving logs of student hint use, attempts, and problem context. We compare LLM-generated reasoning to human learner utterances under minimal and extended contextual prompting, and assess the models' ability to predict step-level learner success. Although GPT-4.1 generates fluent and contextually appropriate continuations, its reasoning is systematically over-coherent, verbose, and less variable than human think-alouds. These effects intensify with a richer problem-solving context during prompting. Learner performance was consistently overestimated. These findings highlight epistemic limitations of simulating learning with LLMs. We attribute these limitations to LLM training data, including expert-like solutions devoid of expressions of affect and working memory constraints during problem solving. Our evaluation framework can guide future design of adaptive systems that more faithfully support novice learning and self-regulation using generative artificial intelligence.