Summary of Unveiling the Lexical Sensitivity Of Llms: Combinatorial Optimization For Prompt Enhancement, by Pengwei Zhan et al.
Unveiling the Lexical Sensitivity of LLMs: Combinatorial Optimization for Prompt Enhancement
by Pengwei Zhan, Zhen Xu, Qian Tan, Jie Song, Ru Xie
First submitted to arxiv on: 31 May 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Large language models (LLMs) are capable of completing various downstream tasks with exceptional instruct-following abilities. However, their performance heavily relies on instructions, revealing an over-sensitivity to lexical variations that are imperceptible to humans. Our proposed black-box Combinatorial Optimization framework for Prompt Lexical Enhancement (COPLE) addresses this issue by iteratively optimizing lexical choices based on feedback from proxy tasks using a word influence search strategy. Experimental results show that even human-crafted prompts used in current benchmarks suffer from this lexical sensitivity, and COPLE recovers the model’s ability to complete downstream tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how large language models can be very good at following instructions, but only if the instructions are exactly right. Even small changes in words can make a big difference in how well they perform. The authors developed a new way to help the models by giving them better instructions, which can improve their ability to complete tasks. |
Keywords
» Artificial intelligence » Optimization » Prompt
