DiPT: Enhancing LLM reasoning through diversified perspective-taking

TL;DR

DiPT introduces diversified perspective-taking to enhance language model reasoning, improving accuracy, stability, and safety by integrating multiple viewpoints and enriching training data.

Contribution

This paper presents DiPT, a novel method that incorporates diversified viewpoints into LLM reasoning and offers a data augmentation recipe to improve fine-tuning outcomes.

Findings

Enhanced reasoning performance and stability with DiPT.

Improved context understanding and safety against jailbreak prompts.

Data enrichment with perspectives boosts fine-tuned model capabilities.

Abstract

Existing work on improving language model reasoning typically explores a single solution path, which can be prone to errors. Inspired by perspective-taking in social studies, this paper introduces DiPT, a novel approach that complements current reasoning methods by explicitly incorporating diversified viewpoints. This approach allows the model to gain a deeper understanding of the problem's context and identify the most effective solution path during the inference stage. Additionally, it provides a general data-centric AI recipe for augmenting existing data to improve their quality for fine-tuning. Our empirical results demonstrate that DiPT can be flexibly integrated into existing methods that focus on a single reasoning approach, enhancing their reasoning performance and stability when presented with paraphrased problems. Furthermore, we illustrate improved context understanding by maintaining the model's safe outputs against "jailbreaking" prompts intentionally designed to bypass safeguards built into deployed models. Lastly, we show that fine-tuning with data enriched with diverse perspectives can boost the reasoning capabilities of the model compared to fine-tuning with raw data alone.