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

Citations Over Time

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%.

Related Papers

• → In search of abstraction: The varying abstraction model of categorization(2008)102 cited
• → Draw Me a Flower: Processing and Grounding Abstraction in Natural Language(2022)5 cited
• → Two Parts of Reflective abstraction: For New Problem Solving and Mathematical Concept(2020)4 cited
• The Image of Abstraction(1988)
• → Draw Me a Flower: Processing and Grounding Abstraction in Natural Language(2021)