Lune

AAAI2026顶会

Reducing the Scope of Language Models

David Yunis, Siyu Huo, Chulaka Gunasekara, Danish Contractor

摘要

Large language models (LLMs) are deployed in a wide variety of user-facing applications. Typically, these deployments have some specific purpose, like answering questions grounded on documentation or acting as coding assistants, but they require general language understanding. In such deployments, LLMs should respond only to queries that align with the intended purpose and reject all other requests, such as generating poetry or answering questions about physics, a task we refer to as 'scoping'. We conduct a comprehensive empirical evaluation of various methods, ranging from prompting, fine-tuning to preference learning and the recently proposed general alignment technique known as Circuit Breakers (CB). Across three families of language models and a broad variety of tasks, we show that it is possible to scope language models. We examine scoping for multiple topics, and fine-grained topics. We ablate diversity of irrelevant queries, layer different techniques, conduct adversarial evaluations and more. Among other results, we find that when diverse examples of irrelevant queries are available, simple supervised fine-tuning produces the best results, but when such diversity is low, Circuit Breakers perform quite well. One can often get the benefits of both methods by layering them in succession. We intend our study to serve as a practitioner's guide to scoping LLMs.

问 Lune

几分钟读透这篇 Paper,不用再花一下午。

Lune 能读完全文,也能读引用它的顶会 Paper,讲清它的结论、证据和局限,每一点都注明出自哪篇 Paper。不用硬啃,也不靠猜。

可以从这些问题问起

Lune 可以查阅

全文 · 相关的 Paper

获取答案

免费开始,无需绑卡

脉络

  • Direct Preference Optimization: Your Language Model is Secretly a Reward Model 引用了 Training language models to follow instructions with human feedback
  • Finetuned Language Models are Zero-Shot Learners 引用了 Training language models to follow instructions with human feedback
  • Training language models to follow instructions with human feedback 引用了 Finetuned Language Models are Zero-Shot Learners
  • Jailbroken: How Does LLM Safety Training Fail? 引用了 Training language models to follow instructions with human feedback
它引用的引用它的从左到右按年份排列,引用越多,位置越高。

它引用的 Paper18

  1. Training language models to follow instructions with human feedbackLong Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida 等NeurIPS 2022 · 24,707 次引用
  2. LoRA: Low-Rank Adaptation of Large Language ModelsEdward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu 等ICLR 2022 · 18,833 次引用
  3. Direct Preference Optimization: Your Language Model is Secretly a Reward ModelRafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning 等NeurIPS 2023 · 10,924 次引用
  4. Finetuned Language Models are Zero-Shot LearnersJason Wei, Maarten Bosma, Vincent Y. Zhao, Kelvin Guu 等ICLR 2022 · 4,966 次引用
  5. Jailbroken: How Does LLM Safety Training Fail?Alexander Wei, Nika Haghtalab, Jacob SteinhardtNeurIPS 2023 · 2,230 次引用

相关 Paper

  1. Steering Out-of-Distribution Generalization with Concept Ablation Fine-TuningHelena Casademunt, Caden Juang, Adam Karvonen, Samuel Marks 等ICML 2026 · 32 次引用
  2. Knowledgeable Preference Alignment for LLMs in Domain-specific Question AnsweringYichi Zhang, Zhuo Chen, Yin Fang, Yanxi Lu 等ACL Findings 2024 · 24 次引用
  3. REINFORCE Adversarial Attacks on Large Language Models: An Adaptive, Distributional, and Semantic ObjectiveSimon Geisler, Tom Wollschläger, M. H. I. Abdalla, Vincent Cohen-Addad 等ICML 2025
  4. SHARP: Self-adaptive Harmful Category-aware Prompt Generation for Black-box JailbreakingYingjie Xue, Xingyou Xia, Jun Zhang, Yunbo Cao 等ACL 2026
  5. DeAL: Decoding-time Alignment for Large Language ModelsJames Y. Huang, Sailik Sengupta, Daniele Bonadiman, Yi-An Lai 等ACL 2025
黄昏的海面,两侧是细线勾勒的悬崖