The Art of Saying No: Contextual Noncompliance in Language Models
This addresses the issue of ensuring language models appropriately refuse requests for developers and users, though it is incremental as it builds on existing refusal work.
The paper tackles the problem of language models overly complying with user requests by introducing a taxonomy for contextual noncompliance and developing an evaluation suite, finding that models like GPT-4 incorrectly comply with up to 30% of requests in certain categories, and showing that parameter-efficient training methods can improve noncompliance without harming other capabilities.
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.