Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansion

• URL: http://arxiv.org/abs/2605.11679v2
• Date: Wed, 13 May 2026 09:28:34 GMT
• Title: Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansion
• Abstract summary: In the realm of multi-objective alignment for large language models, balancing disparate human preferences often manifests as a zero-sum conflict.<n>We propose MORA: Multi-Objective Reward Assimilation. Specifically, MORA isolates single-reward prompts through pre-sampling and expands their reward diversity by rewriting the original questions to incorporate multi-dimensional intents.
• Score: 75.37308041820552
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: In the realm of multi-objective alignment for large language models, balancing disparate human preferences often manifests as a zero-sum conflict. Specifically, the intrinsic tension between competing goals dictates that aggressively optimizing for one metric (e.g., helpfulness) frequently incurs a substantial penalty on another (e.g., harmlessness). While prior work mainly focuses on data selection, parameter merging, or algorithmic balancing during training, these approaches merely force compromises between divergent preferences along a fixed Pareto frontier, failing to fundamentally resolve the inherent trade-off. In this work, we approach this problem from a novel perspective of multi-dimensional rewards. By scaling up the model's rollouts and analyzing the outputs across different reward dimensions, we arrive at a critical conclusion: the conflict among multiple objectives stems from the fact that the prompt itself inherently restricts the achievable multi-dimensional rewards. Based on this core observation, we propose MORA: Multi-Objective Reward Assimilation. Specifically, MORA isolates single-reward prompts through pre-sampling and expands their reward diversity by rewriting the original questions to incorporate multi-dimensional intents. Extensive experiments demonstrate that: (1) in sequential alignment, MORA achieves single-preference improvements ranging from 5% to 12.4%, with exceptional gains in harmlessness, after multiple-preference alignment across helpful, harmless, and truthful dimensions. (2) In simultaneous alignment, MORA achieves an average overall reward improvement of 4.6%. Our codes are available at https://github.com/Shiying-Huang/MORA-MPA.

Related papers

• A Better Spur Should Start From Each Objective [12.849739663589316]
  Real-world Multi-Objective Reinforcement Learning (MORL) often suffers from sparse rewards, reward conflicts, and late-stage reward tug-of-war.<n>We propose Multi-Marginal Preference Optimization (MMPO), a fine-grained framework that intervenes at the data, gradient, and constraint levels.<n>Experiments on real-world e-commerce datasets show that MMPO improves training stability and consistently mitigates better performance across conflicting metrics.
  arXiv  Detail & Related papers  (2026-09-08T03:50:21Z)
• Reward-free Alignment for Conflicting Objectives [12.275610380458119]
  We propose a Reward-free Alignment framework for Conflicted Objectives (RACO)<n>RACO directly leverages pairwise preference data and resolves gradient conflicts via a novel clipped variant of conflict-averse gradient descent.<n>We provide convergence guarantees to Pareto-critical points that respect user-specified objective weights, and further show that clipping can strictly improve convergence rate in the two-objective setting.
  arXiv  Detail & Related papers  (2026-02-02T18:59:52Z)
• Multi-dimensional Preference Alignment by Conditioning Reward Itself [32.33870784484853]
  Multi Reward Conditional DPO resolves reward conflicts by introducing a disentangled Bradley-Terry objective.<n>Experiments on Stable 1.5 and SDXL demonstrate that MCDPO achieves superior performance on benchmarks.
  arXiv  Detail & Related papers  (2025-12-11T02:44:31Z)
• OrthAlign: Orthogonal Subspace Decomposition for Non-Interfering Multi-Objective Alignment [61.02595549125661]
  Large language model (LLM) alignment faces a critical dilemma when addressing multiple human preferences.<n>We present OrthAlign, an innovative approach to resolve gradient-level conflicts in preference alignment.<n>We show that OrthAlign achieves maximum single-preference improvements ranging from 34.61% to 50.89% after multiple-objective alignment.
  arXiv  Detail & Related papers  (2025-09-29T11:16:30Z)
• Reinforcing Video Reasoning with Focused Thinking [65.85683941058916]
  We propose TW-GRPO, a novel framework that enhances visual reasoning with focused thinking and dense reward granularity.<n>Specifically, we employ a token weighting mechanism that prioritizes tokens with high informational density.<n>We also reformulate RL training by shifting from single-choice to multi-choice QA tasks.
  arXiv  Detail & Related papers  (2025-05-30T15:42:19Z)
• Controllable Preference Optimization: Toward Controllable Multi-Objective Alignment [103.12563033438715]
  Alignment in artificial intelligence pursues consistency between model responses and human preferences as well as values. Existing alignment techniques are mostly unidirectional, leading to suboptimal trade-offs and poor flexibility over various objectives. We introduce controllable preference optimization (CPO), which explicitly specifies preference scores for different objectives.
  arXiv  Detail & Related papers  (2024-02-29T12:12:30Z)
• Rewarded soups: towards Pareto-optimal alignment by interpolating weights fine-tuned on diverse rewards [101.7246658985579]
  Foundation models are first pre-trained on vast unsupervised datasets and then fine-tuned on labeled data. We propose embracing the heterogeneity of diverse rewards by following a multi-policy strategy. We demonstrate the effectiveness of our approach for text-to-text (summarization, Q&A, helpful assistant, review), text-image (image captioning, text-to-image generation, visual grounding, VQA), and control (locomotion) tasks.
  arXiv  Detail & Related papers  (2023-06-07T14:58:15Z)
• DORB: Dynamically Optimizing Multiple Rewards with Bandits [101.68525259222164]
  Policy-based reinforcement learning has proven to be a promising approach for optimizing non-differentiable evaluation metrics for language generation tasks. We use the Exp3 algorithm for bandits and formulate two approaches for bandit rewards: (1) Single Multi-reward Bandit (SM-Bandit); (2) Hierarchical Multi-reward Bandit (HM-Bandit) We empirically show the effectiveness of our approaches via various automatic metrics and human evaluation on two important NLG tasks.
  arXiv  Detail & Related papers  (2020-11-15T21:57:47Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.