Search and Filter

Classroom Ai: Large Language Models As Grade-Specific Teachers

Date

AI Overview

Why it matters

Large language models struggle to provide grade-appropriate explanations, systematically exceeding young students' comprehension levels even when explicitly prompted. Adapting explanations to match students' grade levels is essential for addressing teacher shortages and supporting learning.

Study at a glance
  • Question: Can finetuning large language models on readability-curated synthetic data produce grade-appropriate educational explanations while maintaining factual accuracy?
  • Setting and data: Benchmark and synthetic datasets, 208 human survey participants (108 from Korea, the US, and England; 100 on MTurk), and over 1.28 million generated question-answer pairs.
  • Method: Supervised finetuning of language models using synthetic data categorized by readability metrics, followed by automated benchmark evaluations and human validation surveys.
  • AI tools: GPT4o-mini, GPT4o, LLaMA3.1:70B, LLaMA3.1:8B, Qwen 2.5:72B, Mixtral 8x7B
Key findings
  • Finetuned models increased grade-level alignment by an average of 35.64 percentage points over prompt-based approaches while maintaining factual accuracy.
  • Across 108 human participants ranking responses, perceived difficulty strongly agreed with the models' intended grade levels, achieving a Kendall's rank correlation of 0.76.
  • Lower-grade models used simpler words, shorter sentences, and more direct explanations, whereas higher-grade models internally utilized more complex vocabulary and longer explanations.

Summary drafted with AI from the full paper and may contain errors. See the full paper for details.

Large Language Models (LLMs) offer a promising solution to complement traditional teaching and address global teacher shortages that affect hundreds of millions of children, but they fail to provide grade-appropriate responses for students at different educational levels. We introduce a framework for finetuning LLMs to generate age-appropriate educational content across six grade levels, from lower elementary to adult education. Our framework successfully adapts explanations to match students' comprehension capacities without sacrificing factual correctness. This approach integrates seven established readability metrics through a clustering method and builds a comprehensive dataset for grade-specific content generation. Evaluations across multiple datasets with 208 human participants demonstrate substantial improvements in grade-level alignment, achieving a 35.64 percentage point increase compared to prompt-based methods while maintaining response accuracy. AI-assisted learning tailored to different grade levels has the potential to advance educational engagement and equity.