Every Question Has Its Own Value: Reinforcement Learning with Explicit Human Values

TL;DR

This paper introduces RLEV, a reinforcement learning method that incorporates explicit human value signals into LLM training, improving alignment with human priorities and demonstrating robustness to noisy signals.

Contribution

RLEV extends existing RL frameworks by integrating human-defined value signals directly into the reward function for better alignment.

Findings

RLEV outperforms correctness-only baselines across multiple RL algorithms and model scales.

RLEV policies learn value-sensitive termination, being concise for low-value prompts and thorough for high-value ones.

The method remains robust under noisy value signals, such as difficulty-based labels.

Abstract

We propose Reinforcement Learning with Explicit Human Values (RLEV), a method that aligns Large Language Model (LLM) optimization directly with quantifiable human value signals. While Reinforcement Learning with Verifiable Rewards (RLVR) effectively trains models in objective domains using binary correctness rewards, it overlooks that not all tasks are equally significant. RLEV extends this framework by incorporating human-defined value signals directly into the reward function. Using exam-style data with explicit ground-truth value labels, RLEV consistently outperforms correctness-only baselines across multiple RL algorithms and model scales. Crucially, RLEV policies not only improve value-weighted accuracy but also learn a value-sensitive termination policy: concise for low-value prompts, thorough for high-value ones. We demonstrate this behavior stems from value-weighted gradient amplification on end-of-sequence tokens. Ablation studies confirm the gain is causally linked to value alignment. RLEV remains robust under noisy value signals, such as difficulty-based labels, demonstrating that optimizing for an explicit utility function offers a practical path to aligning LLMs with human priorities.