Policy optimisation, reward modelling and RLHF — how models are trained by feedback rather than by labels.
LLM-as-judge is essential for evaluating open-ended text and steering post-training, yet improving the judge itself typically relies on expensive annotations, reward models, or distillation from stron...
Non-verbal vocalizations (NVs), such as laughter, coughs, and sighs, are essential for expressive TTS, but the effectiveness of preference optimization for NV generation remains poorly understood. We...
In reinforcement learning policy evaluation, classic on-policy methods often suffer from high variance when estimating policy performance. To mitigate this issue, behavior policy search has been propo...
arXiv:2608.21863v1 Announce Type: cross Abstract: Tool-Integrated Reasoning (TIR) is a fundamental capability for LLM agents to solve complex tasks by interacting with external tools iteratively. Rei...
arXiv:2608.21500v1 Announce Type: cross Abstract: Prompt injection is listed as the \#1 threat to AI agents. When an agent accesses external data from websites, files, or emails, an attacker may inje...
GRACE is a step‑level benchmark for evaluating the faithfulness of chain‑of‑thought reasoning over context. It provides human annotations for each step in CoT traces from 10 models across 4 datasets, labeling faithfulness, error category, and natural‑language explanations. The benchmark introduces a data‑driven taxonomy that splits errors into GRACE‑Inference (deductive) and GRACE‑Grounding (factual) tracks, each with four categories, and demonstrates that incorporating step‑level faithfulness signals can improve downstream accuracy and reasoning reliability.
The paper proposes shifting AI agent training from isolated task completion to collaborative interaction, defining three key dimensions—Productivity, Proactivity, and Personalization (PPP). It introduces UserVille, an environment with LLM-based user simulators and user-centric feedback, and a multi-objective reinforcement learning framework that optimizes PPP using rewards from task outcomes, question effort, and preference adherence. Experiments on SWE-Bench and BrowseComp-Plus show PPP-trained agents outperform strong LLM baselines, ask more targeted questions, and generalize to unseen preferences and tasks, with a user study underscoring the value of user-centric feedback for effective, supervised collaboration.
arXiv:2507.04356v3 Announce Type: replace-cross Abstract: Research, innovation and practical capital investment have been increasing rapidly toward the realization of autonomous physical agents. This...
GeoRisk-RAG is a hierarchy‑aware framework that improves the reliability of Retrieval‑Augmented Generation (RAG) by incorporating geographic validity. It estimates geographic applicability using a Directed Acyclic Graph (DAG)‑based distance during context retrieval, enabling selective answering. Experiments on a wildfire‑related QA dataset show that GeoRisk‑RAG reduces false confidence rates from ~0.090 to 0.009 and aligns better with human preferences.
arXiv:2608.21925v1 Announce Type: new Abstract: Emotional Support Conversation (ESC) systems aim to provide holistic support by balancing professional therapeutic competence with natural empathy. How...
arXiv:2608.21595v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards (RLVR) improves the reasoning ability of vision-language models (VLMs), and diversifying the rollouts wi...
arXiv:2507.17448v2 Announce Type: replace-cross Abstract: Retrosynthetic planning is a cornerstone of organic synthesis and drug discovery. Yet existing AI methods often rely on pattern matching rath...
arXiv:2606.27504v2 Announce Type: replace Abstract: World Action Models (WAMs) unify future environment prediction with action generation for autonomous driving, yet existing approaches optimize only...
arXiv:2608.22266v1 Announce Type: new Abstract: In the context of information seeking, conversational agents are undergoing an evolution from reactive tools to proactive, personalized assistants. A c...
arXiv:2502.14643v3 Announce Type: replace Abstract: Direct Preference Optimization (DPO) is a widely adopted offline algorithm for preference-based reinforcement learning from human feedback (RLHF),...
arXiv:2501.06926v5 Announce Type: replace Abstract: Double reinforcement learning (DRL) provides efficient off-policy inference for policy values in nonparametric Markov decision processes (MDPs), bu...
arXiv:2608.21871v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards has become the dominant recipe for improving large language model reasoning, yet it presumes large human...
arXiv:2608.23329v1 Announce Type: cross Abstract: Open-world video understanding often requires a model to locate sparse visual evidence and acquire external knowledge that is absent from the video a...
arXiv:2608.21808v1 Announce Type: new Abstract: Multimodal Retrieval-Augmented Generation (RAG) with visual citation is crucial for ensuring the traceability and verifiability of MLLMs. However, curr...
arXiv:2608.23318v1 Announce Type: new Abstract: Hint-based reinforcement learning addresses reward sparsity in long-horizon agentic tasks by retaining a prefix of an expert trajectory before each rol...