Learning from "Silly" Questions Improves Large Language Models, But Only Slightly
This work addresses the challenge of optimizing supervised fine-tuning datasets for LLMs, though it is incremental as it builds on existing methods with new insights.
The paper investigates how fine-tuning large language models with data derived from 'silly' questions can affect performance, finding that specific rules extracted from these questions lead to mixed results, such as a 5% improvement on one task but a 6.14% drop on another.
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.