AI ethics evidence index Search
Source-linked daily record
Evidence record 4419 · automatically gathered

Revisiting Reinforcement Learning with Verifiable Rewards from a Contrastive Perspective

Group Relative Policy Optimization (GRPO) is one of the most widely adopted RLVR algorithms for post-training large language models on reasoning tasks. We first show that GRPO admits an equivalent discriminative reformulation, in which policy optimization maximizes the expected score gap between verified positive and negative rollouts. This reformulation reveals two objective-level limitations: likelihood-misaligned surrogate scores, in which clipped ratio-based scores are optimized rather than

Record details

Published: 13 May 2026
Source: arXiv
Category: Research
Topics: Bias & fairness · Regulation · Safety & alignment
Retrieved: 14 July 2026

source-onlyevidence status

These records share source-supplied organisations, an exact publisher byline, automatic topics or regions. The reason is shown on every link; related does not mean supporting, agreeing with or verifying this record.

Shared topic: Bias & fairness · Shared topic: Regulation

LC-ERD: Mining Latent Logic for Self-Evolving Reasoning via Consistency-Regulated Reward Decomposition

arXiv · 19 May 2026

Shared topic: Bias & fairness · Shared topic: Regulation

OceanPile: A Large-Scale Multimodal Ocean Corpus for Foundation Models

arXiv · 25 April 2026

Shared topic: Bias & fairness · Shared topic: Regulation

Silent Failures in Federated Personalization of Foundation Models

arXiv · 31 May 2026

Shared topic: Bias & fairness · Shared topic: Regulation

BiasGRPO: Stabilizing Bias Mitigation in High-Variance Reward Landscapes via Group-Relative Policy Optimization

arXiv · 3 June 2026

Shared topic: Bias & fairness · Shared topic: Regulation

Prompt Programming for Cultural Bias and Alignment of Large Language Models

arXiv · 17 March 2026

Shared topic: Bias & fairness · Shared topic: Regulation

LLM Constitutional Multi-Agent Governance

arXiv · 13 March 2026

How to cite this record

ethics.ai (13 May 2026), “Revisiting Reinforcement Learning with Verifiable Rewards from a Contrastive Perspective,” evidence record 4419, https://ethics.ai/record/4419 (originally published by arXiv).

JSON

Use and limitations

This page is a stable index and citation surface for a source record. ethics.ai did not author the underlying report and has not independently verified every claim. Automatic topics may be imperfect. For consequential use, quote and cite the original publisher.