Towards Interpretable Soft Prompts
Oam Patel, Jason Wang, Nikhil Shivakumar Nayak, Suraj Srinivas, Himabindu Lakkaraju
TL;DR
Soft prompts boost task performance but remain opaque to humans. The authors propose a formal interpretability framework with Faithfulness and Scrutability and test two tuners, PEZ and RLPrompt, using a perplexity-based scrutability proxy. Results reveal a fundamental trade-off: increasing interpretability often reduces task performance, and perplexity is an imperfect proxy for scrutability; PEZ may fail to converge while RLPrompt yields only partial, inconsistent interpretability gains. The work clarifies the limits of current interpretability approaches for soft prompts and points to future directions in alternative scrutability metrics and cross-model generalization.
Abstract
Soft prompts have been popularized as a cheap and easy way to improve task-specific LLM performance beyond few-shot prompts. Despite their origin as an automated prompting method, however, soft prompts and other trainable prompts remain a black-box method with no immediately interpretable connections to prompting. We create a novel theoretical framework for evaluating the interpretability of trainable prompts based on two desiderata: faithfulness and scrutability. We find that existing methods do not naturally satisfy our proposed interpretability criterion. Instead, our framework inspires a new direction of trainable prompting methods that explicitly optimizes for interpretability. To this end, we formulate and test new interpretability-oriented objective functions for two state-of-the-art prompt tuners: Hard Prompts Made Easy (PEZ) and RLPrompt. Our experiments with GPT-2 demonstrate a fundamental trade-off between interpretability and the task-performance of the trainable prompt, explicating the hardness of the soft prompt interpretability problem and revealing odd behavior that arises when one optimizes for an interpretability proxy.