HIPO: Instruction Hierarchy via Constrained Reinforcement Learning
This addresses the challenge of reliable LLM deployment in complex workflows by ensuring priority-ordered instruction compliance, representing a novel method for a known bottleneck rather than a foundational advancement.
The paper tackles the problem of Hierarchical Instruction Following (HIF) in large language models, where standard alignment methods fail to enforce system prompt compliance, and introduces HIPO, a framework that formulates HIF as a Constrained Markov Decision Process, resulting in significant improvements in both system compliance and user utility across diverse model architectures.
Hierarchical Instruction Following (HIF) refers to the problem of prompting large language models with a priority-ordered stack of instructions. Standard methods like RLHF and DPO typically fail in this problem since they mainly optimize for a single objective, failing to explicitly enforce system prompt compliance. Meanwhile, supervised fine-tuning relies on mimicking filtered, compliant data, which fails to establish the priority asymmetry at the algorithmic level. In this paper, we introduce \textsc{HIPO}, a novel alignment framework that formulates HIF as a Constrained Markov Decision Process. \textsc{HIPO} elevates system prompts from mere input context to strict algorithmic boundaries. Using a primal-dual safe reinforcement learning approach, the algorithm dynamically enforces system prompt compliance as an explicit constraint, maximizing user utility strictly within this feasible region. Extensive evaluations across diverse model architectures (e.g., Qwen, Phi, Llama) demonstrate that \textsc{HIPO} significantly improves both system compliance and user utility. Furthermore, mechanistic analysis reveals that this constrained optimization autonomously drives the model to shift its attention toward long-range system tokens, providing a principled foundation for reliable LLM deployment in complex workflows.