Take a Step Back: Evoking Reasoning via Abstraction in Large Language Models

TL;DR

This paper introduces Step-Back Prompting, a technique that helps large language models abstract high-level concepts from details, significantly enhancing their reasoning abilities across various challenging tasks.

Contribution

The paper proposes Step-Back Prompting, a novel method enabling LLMs to perform abstractions and improve reasoning accuracy on complex tasks.

Findings

Performance improvements on MMLU Physics and Chemistry by 7% and 11%.

27% improvement on TimeQA.

7% improvement on MuSiQue.

Abstract

We present Step-Back Prompting, a simple prompting technique that enables LLMs to do abstractions to derive high-level concepts and first principles from instances containing specific details. Using the concepts and principles to guide reasoning, LLMs significantly improve their abilities in following a correct reasoning path towards the solution. We conduct experiments of Step-Back Prompting with PaLM-2L, GPT-4 and Llama2-70B models, and observe substantial performance gains on various challenging reasoning-intensive tasks including STEM, Knowledge QA, and Multi-Hop Reasoning. For instance, Step-Back Prompting improves PaLM-2L performance on MMLU (Physics and Chemistry) by 7% and 11% respectively, TimeQA by 27%, and MuSiQue by 7%.