Learning from "Silly" Questions Improves Large Language Models, But Only Slightly

TL;DR

This study investigates how incorporating rules derived from analyzing 'silly' questions on Ruozhiba can slightly enhance large language model fine-tuning, revealing mixed effects depending on task and rule applied.

Contribution

The paper introduces a method to analyze and apply educational and cognitive science rules to improve SFT datasets, with large-scale evaluation on MMLU tasks.

Findings

Rules can improve performance on some tasks by around 5%.

Certain rules cause performance drops up to 6.14%.

Effectiveness of rules varies across different tasks.

Abstract

Constructing high-quality Supervised Fine-Tuning (SFT) datasets is critical for the training of large language models (LLMs). Recent studies have shown that using data from a specific source, Ruozhiba, a Chinese website where users ask "silly" questions to better understand certain topics, can lead to better fine-tuning performance. This paper aims to explore some hidden factors: the potential interpretations of its success and a large-scale evaluation of the performance. First, we leverage GPT-4 to analyze the successful cases of Ruozhiba questions from the perspective of education, psychology, and cognitive science, deriving a set of explanatory rules. Then, we construct fine-tuning datasets by applying these rules to the MMLU training set. Surprisingly, our results indicate that rules can significantly improve model performance in certain tasks, while potentially diminishing performance on others. For example, SFT data generated following the "Counterintuitive Thinking" rule can achieve approximately a 5% improvement on the "Global Facts" task, whereas the "Blurring the Conceptual Boundaries" rule leads to a performance drop of 6.14% on the "Econometrics" task. In addition, for specific tasks, different rules tend to have a consistent impact on model performance. This suggests that the differences between the extracted rules are not as significant, and the effectiveness of the rules is relatively consistent across tasks. Our research highlights the importance of considering task diversity and rule applicability when constructing SFT datasets to achieve more comprehensive performance improvements.