SibylSense: Adaptive Rubric Learning via Memory Tuning and Adversarial Probing

TL;DR

SibylSense introduces an adaptive learning method for rubrics in open-ended generation, using memory tuning and adversarial probing to improve reward quality and robustness in reinforcement learning tasks.

Contribution

The paper proposes SibylSense, a novel inference-time approach that dynamically adapts rubric generators through memory updates and adversarial training, enhancing reward discrimination.

Findings

Improves discriminative power of rubrics over static baselines.

Enhances downstream RL performance in open-ended tasks.

Demonstrates robustness and adaptability in rubric construction.

Abstract

Designing aligned and robust rewards for open-ended generation remains a key barrier to RL post-training. Rubrics provide structured, interpretable supervision, but scaling rubric construction is difficult: expert rubrics are costly, prompted rubrics are often superficial or inconsistent, and fixed-pool discriminative rubrics can saturate and drift, enabling reward hacking. We present SibylSense, an inference-time learning approach that adapts a frozen rubric generator through a tunable memory bank of validated rubric items. Memory is updated via verifier-based item rewards measured by reference-candidate answer discriminative gaps from a handful of examples. SibylSense alternates memory tuning with a rubric-adversarial policy update that produces rubric-satisfying candidate answers, shrinking discriminative gaps and driving the rubric generator to capture new quality dimensions. Experiments on two open-ended tasks show that SibylSense yields more discriminative rubrics and improves downstream RL performance over static and non-adaptive baselines.