The paper presents a scalable method for learning dataweights that optimizes training-example importance using metagradients, guided by a small high‑quality target dataset. It introduces parametric weighting rules that generalize to new and streaming data and can transfer from smaller to larger models. Applied to reward modeling, the learned weights improve held‑out performance, adapt models to specialized domains, and enable human‑synthetic supervision mixtures to outperform either source alone.
arXiv:2605.07872v2 Announce Type: replace-cross Abstract: Multimodal reward models have advanced substantially in text and image domains, yet progress in video understanding reward modeling remains s...
arXiv:2503. 00539v2 Announce Type: replace-cross Abstract: Reinforcement learning from human feedback (RLHF) has evolved to be one of the main methods for fine-tuning large language models (LLMs).
arXiv:2605.25629v3 Announce Type: replace Abstract: Weak-to-strong (W2S) generalization is a promising framework for scalable oversight, yet existing evaluations often test students under matched tra...
The paper presents a post‑training approach for text‑to‑image models that combines a preference reward, trained on large human preference data, with rubric‑based rewards that assess prompt faithfulness and other desirable traits. The authors show that a simple reward composition strategy outperforms a naive weighted average, leading to significant Elo gains on the Arena leaderboard for models like Flux2dev and Ideogram‑4. They also release Arena‑T2I‑Training, a 1K subset of data to aid reproducible research in post‑training.
arXiv:2608. 11746v1 Announce Type: new Abstract: Modern systems are increasingly expected to transfer across tasks not specified during training.
The paper introduces TESS, a scalable data‑selection framework that replaces per‑sample weights with a selection network to improve transferability across datasets and model sizes. It identifies instability in existing meta‑learning for training‑data selection (MTS) due to weight suppression and overreliance on easy features, and proposes a Pointwise Value Matching objective to address these issues. Experiments on large language model safety and instruction tuning show strong transfer from subsets to full corpora and from smaller to larger models.
arXiv:2602. 17658v3 Announce Type: replace-cross Abstract: Reward modeling is central to alignment pipelines such as RLHF, RLAIF, and PPO-based policy optimization, yet its reliability is constrained by limited and heterogeneous human preference data that are expensive to collect at scale.
arXiv:2609.00213v1 Announce Type: new Abstract: Reinforcement learning fine-tuning of large language models increasingly adopts multiple reward dimensions, including verifiable rules, task-specific e...
Reinforcement learning (RL) is vital for optimizing video generation models, with a robust reward model (RM) serving as the cornerstone. However, existing video reward models often produce unstable sc...
arXiv:2609.22947v1 Announce Type: new Abstract: Reinforcement learning (RL) is vital for optimizing video generation models, with a robust reward model (RM) serving as the cornerstone. However, exist...
The paper presents a new scaling law for reward optimization in AI alignment, showing that performance scales as Θ(√min{log(M), K}), where M is the number of preference comparisons used to train a proxy reward model and K is the KL‑divergence budget relative to a reference policy. The authors derive this law using an information‑theoretic model, prove its tightness, and validate it with extensive experiments involving a 70B gold reward model and smaller proxy models (0.6B–4B). The empirical results demonstrate a strong fit (R² 97–99 %) across different model sizes, noise levels, and optimization methods, suggesting that reward optimization behaves like a simple selection task over IID Gaussian variables with noisy feedback.
arXiv:2507. 05019v2 Announce Type: replace-cross Abstract: In-context learning enables transformer models to generalize to new tasks based solely on input prompts, without any need for weight updates.