%0 Generic %1 zheng2023step %A Zheng, Huaixiu Steven %A Mishra, Swaroop %A Chen, Xinyun %A Cheng, Heng-Tze %A Chi, Ed H. %A Le, Quoc V %A Zhou, Denny %D 2023 %K Google deep-mind llms prompt-engineering step-back-prompting %T Take a Step Back: Evoking Reasoning via Abstraction in Large Language Models %U https://arxiv.org/pdf/2310.06117.pdf %X 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 the reasoning steps, LLMs significantly improve their abilities in following a correct reasoning path towards the solution. We conduct experiments of Step-Back Prompting with PaLM-2L models and observe substantial performance gains on a wide range of 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%, TimeQA by 27%, and MuSiQue by 7%.

@misc{zheng2023step, 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 the reasoning steps, LLMs significantly improve their abilities in following a correct reasoning path towards the solution. We conduct experiments of Step-Back Prompting with PaLM-2L models and observe substantial performance gains on a wide range of 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%, TimeQA by 27%, and MuSiQue by 7%.}, added-at = {2023-11-27T23:41:54.000+0100}, archiveprefix = {arXiv}, author = {Zheng, Huaixiu Steven and Mishra, Swaroop and Chen, Xinyun and Cheng, Heng-Tze and Chi, Ed H. and Le, Quoc V and Zhou, Denny}, biburl = {https://www.bibsonomy.org/bibtex/21635492f4be91b1f17e7246c20ebe81b/ghagerer}, eprint = {2310.06117}, interhash = {6967f98b06dc602e7d5ba7e07e131ab4}, intrahash = {1635492f4be91b1f17e7246c20ebe81b}, keywords = {Google deep-mind llms prompt-engineering step-back-prompting}, primaryclass = {cs.LG}, timestamp = {2023-11-27T23:42:53.000+0100}, title = {Take a Step Back: Evoking Reasoning via Abstraction in Large Language Models}, url = {https://arxiv.org/pdf/2310.06117.pdf}, year = 2023 }