Control Barrier Function for Aligning Large Language Models
cs.CL, cs.AI, cs.SY, eess.SY
Submitted: 2025-11-05
Updated: 2025-11-06
Comments: This work is an extenede version of arXiv:2408.15625 and has been submitted to the IEEE for possible publication
Journal ref: IEEE Transactions on Control Systems Technology 2026
DOI: 10.1109/TCST.2026.3675329
License: http://creativecommons.org/licenses/by-nc-sa/4.0/
The gist: This paper proposes a control-based framework for aligning large language models (LLMs) by leveraging a control barrier function (CBF) to ensure user-desirable text generation.
Terminology
Abstract
This paper proposes a control-based framework for aligning large language models (LLMs) by leveraging a control barrier function (CBF) to ensure user-desirable text generation. The presented framework applies the CBF safety filter to the predicted token generated from the baseline LLM, to intervene in the generated text. The safety filter includes two significant advantages: this safety filter is an add-on type, allowing it to be used for alignment purposes without fine-tuning the baseline LLM, and if there is an evaluation model regarding the desired alignment, it can be directly applied to the filter design. The overall text-generation system is implemented with open-source language models, aiming to generate positive text.
Sources
- Large Language Model Alignment: A Survey
- Large Language Models: A Survey
- Aligning Large Language Models with Human: A Survey
- Aligning Language Models with Preferences through f-divergence Minimization
- Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
- Constitutional AI: Harmlessness from AI Feedback
- A Survey on In-context Learning
- Is In-Context Learning Sufficient for Instruction Following in LLMs?
- What's the Magic Word? A Control Theory of LLM Prompting
- Taming AI Bots: Controllability of Neural States in Large Language Models
- The Llama 3 Herd of Models
- Autoregressive Entity Retrieval
- CTRL: A Conditional Transformer Language Model for Controllable Generation
- Controlled Decoding from Language Models
- DeAL: Decoding-time Alignment for Large Language Models
- Inference-Time Intervention: Eliciting Truthful Answers from a Language Model
- Unlearning Traces the Influential Training Data of Language Models
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering