The paper introduces Test‑Time Policy Optimization (TTPO), an approach that enables large language models to improve mathematical reasoning without relying on ground‑truth labels. TTPO uses majority‑vote pseudo‑labels and an asymmetric objective: it distills rollouts that agree with the pseudo‑label via On‑Policy Self‑Distillation and penalizes disagreeing rollouts with Grouped Reinforcement Learning. Token‑level selection further refines the process, down‑weighting already‑converged positions during distillation and penalizing only confident errors during RL. Experiments show that TTPO matches label‑supervised OPSD on five competition‑level benchmarks, boosts Qwen3‑1.7B from 38.0 % to 45.2 % in test‑time training, and achieves significant gains without explicit reasoning steps, while also generalizing well across tasks.
By Aozhe Wang, Zhengxi Lu, Jianze Wang, Shangke Lv, Ying Liu, Weiming Lu, Jun Xiao, Yueting Zhuang, Hua Yang, Qianglong Chen, Yongliang Shen
The paper introduces Test‑Time Policy Optimization (TTPO), a method that enables large language models to improve mathematical reasoning during inference without relying on ground‑truth labels. TTPO uses majority‑vote pseudo‑labels to guide an asymmetric objective: agreeing rollouts are distilled via On‑Policy Self‑Distillation, while disagreeing rollouts are penalized with Grouped Reinforcement Learning, with token‑level selection refining both branches. Experiments show that TTPO matches label‑supervised OPSD on five benchmarks, boosts Qwen3‑1.7B from 38.0 % to 45.2 % in test‑time training, and achieves significant gains without explicit reasoning steps, demonstrating strong cross‑task generalization.
arXiv:2604.24957v3 Announce Type: replace-cross
Abstract: Scaling test-time compute has emerged as a powerful mechanism for enhancing Large Language Model (LLM) performance. However, standard post-tr...
By Adam Ousherovitch, Ambuj Tewari
The paper introduces Discriminative World Models for Web Agents, proposing a predicted-state matching objective that trains world models to produce representations that can distinguish the true resulting state from those of alternative actions. Using a branching dataset from WebArena Go-Browse, the authors demonstrate that this approach outperforms traditional supervised next-state prediction on a held‑out benchmark and improves action ranking on WebPRMBench. Additionally, employing the discriminative world model for test‑time action selection boosts end‑to‑end task success on WebArena‑Lite.
By Kelvin Li, Dhruv Pendharkar, Anish Pahilajani, Chuyi Shang, Leon Oks, Leonid Karlinsky, Rogerio Feris, Trevor Darrell, Roei Herzig
arXiv:2609.37119v1 Announce Type: cross
Abstract: Recent approaches to reinforcement learning (RL) post-training for large language models increasingly remove the critic to reduce training instabilit...
By Hongyang Li, Xiao Li, Caesar Wu, Said Mammar, Gr\'egoire Danoy, Pascal Bouvry
The paper introduces Jev, a reinforcement‑learning‑trained model that provides calibrated probability answers to typed questions about a single input in one call. Jev is evaluated on RLCDAlignBench, a benchmark covering ten alignment failures across 44 tests and five target models, achieving a median AUROC of 0.886 zero‑shot and outperforming supervised baselines on most tasks. The study shows that question wording has little impact, while contextual fields that encode labels are more influential, and that Jev matches human‑label agreement while being 63× cheaper than LLM‑judge scorers.
By Ruoqi Guo, Yi Liu, Gelei Deng, Yuekang Li, Lida Zhao, Yutao Wu, Simin Chen, Ying Zhang, Leo Yu Zhang
The paper introduces POISE, a reinforcement learning algorithm that uses a model’s internal states as a value estimator to reduce variance in reinforcement learning with verifiable rewards (RLVR). By employing a lightweight probe that reads internal signals during the forward pass, POISE predicts baselines online and uses a cross‑rollout construction to keep gradients unbiased. Experiments on Qwen3‑4B and OLMo3‑7B‑Instruct‑DPO across six domains show POISE outperforms existing RLVR baselines, offering more stable training and a value model that generalizes across tasks and scales with the policy.
By Yunho Choi, Jongwon Lim, Woojin Ahn, Minjae Oh, Jeonghoon Shim, Yohan Jo
The paper introduces Best‑Practice Critic Optimization (BPCO), a stable and efficient recipe for training a critic in reinforcement learning for large language models. BPCO combines DPPO, bounded value predictions, Monte Carlo targets, unnormalized policy advantages, and length‑adaptive advantage estimation, allowing the critic to be conditioned on hidden reward information. Experiments on mathematical reasoning tasks with models from 1.5B to 30B parameters show that BPCO consistently outperforms a strong critic‑based baseline and matches or exceeds group‑based methods while sampling only one response per prompt.
By Penghui Qi, Xiangxin Zhou, Wee Sun Lee
Discriminative World Models for Web Agents proposes a new training objective called predicted‑state matching, which forces a world model to produce representations that can distinguish the true resulting web state from those produced by alternative actions. The authors train these models on a branching dataset from WebArena Go‑Browse, where each decision point includes multiple actions and their outcomes. Experiments show that models trained with predicted‑state matching outperform those trained with standard supervised next‑state prediction on a held‑out benchmark, improve PRM‑style action ranking on WebPRMBench, and enhance end‑to‑end task success on WebArena‑Lite when used for test‑time action selection.
arXiv:2606.16011v2 Announce Type: replace
Abstract: Standard accuracy benchmarks evaluate whether large language models (LLMs) reach correct answers. However, they do not test whether models maintain...
By Nafiseh Nikeghbal, Amir Hossein Kargaran, Shaghayegh Kolli, Jana Diesner
The paper introduces Alignment Forecasting, a method for predicting whether fine‑tuning a language model on a given dataset will increase specific alignment failures such as deception or sycophancy. It presents ALIGNMENTFORECASTBENCH, a benchmark of over 5,000 forecasting questions across many models, datasets, and failure modes, and shows that a simple forecasting scaffold using an LLM’s assessment of dataset bias can outperform baseline forecasters. The authors demonstrate that filtering out high‑risk training examples identified by the forecaster can improve alignment in multiple‑choice evaluations, though benefits in open‑ended conversations remain uncertain.
By Chen Yueh-Han, Bruce W. Lee, Ilia Sucholutsky, Tomek Korbak
arXiv:2608.23566v2 Announce Type: replace-cross
Abstract: Group-based reinforcement learning methods such as GRPO for large language models avoid training a critic by sampling multiple responses for...
By Penghui Qi, Xiangxin Zhou, Wee Sun Lee