The Art of Saying No: Contextual Noncompliance in Language Models
Chat-based language models are designed to be helpful, yet they should not comply with every user request. While most existing work primarily focuses on refusal of "unsafe" queries, we posit that the scope of noncompliance should be broadened. We introduce a comprehensive taxonomy of contextual noncompliance describing when and how models should not comply with user requests. Our taxonomy spans a wide range of categories including incomplete, unsupported, indeterminate, and humanizing requests (in addition to unsafe requests). To test noncompliance capabilities of language models, we use this taxonomy to develop a new evaluation suite of 1000 noncompliance prompts. We find that most existing models show significantly high compliance rates in certain previously understudied categories with models like GPT-4 incorrectly complying with as many as 30% of requests. To address these gaps, we explore different training strategies using a synthetically-generated training set of requests and expected noncompliant responses. Our experiments demonstrate that while direct finetuning of instruction-tuned models can lead to both over-refusal and a decline in general capabilities, using parameter efficient methods like low rank adapters helps to strike a good balance between appropriate noncompliance and other capabilities.
Code (1)
Methods 이 논문이 사용한 방법론
Similar Papers 제목 키워드 기반
Thompson Sampling for Noncompliant Bandits
Thompson sampling, a Bayesian method for balancing exploration and exploitation in bandit problems, has theoretical guarantees and exhibits strong empirical performance in many domains. Traditional Thompson sampling, how…
Thompson SamplingAre Multilingual LLMs Culturally-Diverse Reasoners? An Investigation into Multicultural Proverbs and Sayings
Large language models (LLMs) are highly adept at question answering and reasoning tasks, but when reasoning in a situational context, human expectations vary depending on the relevant cultural common ground. As languages…
Question AnsweringUMUTeam at SemEval-2021 Task 7: Detecting and Rating Humor and Offense with Linguistic Features and Word Embeddings
In writing, humor is mainly based on figurative language in which words and expressions change their conventional meaning to refer to something without saying it directly. This flip in the meaning of the words prevents N…
Sentiment AnalysisWord EmbeddingsExploring Chinese Humor Generation: A Study on Two-Part Allegorical Sayings
Humor, a culturally nuanced aspect of human language, poses challenges for computational understanding and generation, especially in Chinese humor, which remains relatively unexplored in the NLP community. This paper inv…
Contrastive LearningLanguage ModelingLanguage Modelling"Ge Shu Zhi Zhi": Towards Deep Understanding about Worlds
"Ge She Zhi Zhi" is a novel saying in Chinese, stated as "To investigate things from the underlying principle(s) and to acquire knowledge in the form of mathematical representations". The saying is adopted and modified b…
Position