arXiv Machine Learning By Zhecheng Sheng, Yongsen Tan, Xiruo Ding, Trevor Cohen, Serguei Pakhomov

When Prompts Interact: Assessing Prompt Arithmetic for Deconfounding under Distribution Shift

Read the original on arXiv Machine Learning →

The paper investigates how combining soft prompts via task arithmetic can reduce reliance on confounding variables in classification models. It introduces Hybrid Prompt Arithmetic (HyPA), which merges task prompts with linearized confounder prompts to counteract spurious correlations. Experiments across multiple benchmarks show that HyPA consistently improves the robustness‑performance trade‑off under distribution shift, and analysis of hidden representations suggests it mitigates confounding by diminishing the influence of confounder signals.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

arXiv AI
Aug 24

Beyond Prompt Engineering: A Systematic Analysis of Prompt Lexical Sensitivity and Its Impacts on Quality

The paper investigates how small lexical changes in prompts can cause large performance swings in large language models. Using a dataset of 132,000 prompt variants, the authors uncover a scaling law linking higher average task performance to lower variance and greater robustness. They identify domain-specific terminology and explicit action directives as key linguistic factors that stabilize prompts, and propose an automated Prompt-Refining Agent that reduces performance variance by 40.7% in code generation while maintaining or improving mean performance.

By Qipeng Xie, Zi Liang, Jiafei Wu, Yufei Chen, Weizheng Wang, Wenao Ma, Zhong Ming, Haiqin Yang, Kaishun Wu
arXiv Machine Learning
Aug 28

$p1$: Better Prompt Optimization with Fewer Prompts

The paper investigates why prompt optimization works better for some tasks than others by decomposing reward variance into response variance and system‑prompt variance. It finds that optimization succeeds when system‑prompt variance dominates, and that adding more user prompts can actually reduce this variance, especially on heterogeneous datasets. To address this, the authors propose $p1$, a filtering method that selects a small set of high‑variance user prompts, which improves optimization on reasoning benchmarks and even allows a system prompt trained on just two AIME 24 prompts to generalize well.

By Zhaolin Gao (Sid), Yu (Sid), Wang, Bo Liu, Thorsten Joachims, Kiant\'e Brantley, Wen Sun
arXiv AI
Aug 20

Evaluating and Explaining Prompt Sensitivity of LLMs Using Interactions

The paper introduces a fine-grained method called interactions to analyze prompt sensitivity in large language models (LLMs). By decomposing output scores into nonlinear interactions, the authors show that subtle prompt changes can destabilize these interactions even when overall outputs stay unchanged. They propose an Interaction-based Prompt Sensitivity (IPS) metric and use it to evaluate 50 open-source LLMs, finding that supervised fine‑tuning, larger model scales, dense architectures, and few‑shot learning all reduce prompt sensitivity, primarily by stabilizing low‑order interactions.

By Ruiyang Qin, Qingzhuo Wang, Tian Wang, Zhihua Wei, Wen Shen