As AI assistants serve millions of users daily, evaluating user experience (UX) beyond general model capability has become increasingly important. We present UXBench, the first user-centric benchmark grounded in real user feedback signals for evaluating preference alignment and dialogue generation. The benchmark consists of three interconnected tasks, UX Judge, UX Eval, and UX Recovery, with 7,400 test instances extracted from over 70K interaction logs of a mainstream Chinese AI assistant. The dataset closely reflects real user distributions, covering 8 scenarios, 83 domains, and diverse failure patterns that pose severe challenges. Extensive experiments on 26 frontier language models provide novel insights into how well models perceive user experience and how improvements in model capability contribute to better dialogue engagement. Through analyses of model behavior and performance gaps, we document six important findings, demonstrating that user feedback prediction is a learnable capability and revealing different aspects that influence user experience. UXBench establishes a new evaluation landscape and calls for greater attention to tailored UX optimization, contributing toward a user-centric scaling law for the development of successful AI assistants. The full project is released at https://github.com/mengze-hong/UXBench.

Cultural reasoning lies at the core of many emerging AI applications, yet existing benchmarks provide limited insight into how well models understand and contextualize cultural artifacts. We introduce ExhibitionBench, the first benchmark that frames exhibition curation as a high-level evaluation testbed, comprising 300 institution-derived exhibition and thematic collection records, 23,658 artifacts, and three interconnected tasks that evaluate complementary aspects of cultural understanding and reasoning, translating them into user-perceivable utilities. Extensive experiments across 29 systems reveal a significant gap between frontier LLMs and expert-level competence, particularly on tasks that require collection-level coherence and thematic understanding. Performance gains from scaling, prompting, and multimodal inputs are limited and highly model-dependent, whereas tailored model optimizations can equip open-weight models with better capabilities that support on-premise deployment. These findings highlight the difficulty of reasoning over cultural collections and provide a practical benchmark for developing more culturally grounded and reliable AI systems. The full project is released at https://github.com/mengze-hong/ExhibitionBench.

We deploy large language models (LLMs) as business development (BD) agents for persuasive price negotiation in online travel agencies (OTAs). The agent must follow a multi-stage Standard Operating Procedure (SOP) and strict guardrails (no over-promising and no hallucinations), while remaining human-like and effective over long, multi-turn dialogues.We propose Reward-Enhanced Policy Optimization (REPO), a reinforcement learning post-training method that combines heterogeneous rewards: a preference-trained reward model (RM), an LLM-as-a-judge (RJ) for nuanced behaviors (e.g., emotional value and SOP compliance), and rule-based reward functions (RF) (mainly regex-based) for deterministic checks on numerics, formatting, and guardrails. In expert consensus evaluation (three human experts; 30 online conversations and 45 curated bad cases), REPO improves average dialogue rating to 4.63 (+0.33 over GRPO) and raises the share of conversations with at least one excellent response to 66.67% (+23.34 pp over GRPO), while achieving a 93.33% bad-case fix rate with 75.56% clean fixes.In a production A/B test on 9,653 real customer conversations (vs. an intent-driven dialogue system), REPO improves response rate by +12.14 pp and task success rate by +5.94 pp (p<0.001).

Claim verification is an essential step in the automated fact-checking pipeline which assesses the veracity of a claim against a piece of evidence. In this work, we explore the potential of few-shot claim verification, where only very limited data is available for supervision. We propose MAPLE (Micro Analysis of Pairwise Language Evolution), a pioneering approach that explores the alignment between a claim and its evidence with a small seq2seq model and a novel semantic measure. Its innovative utilization of micro language evolution path leverages unlabelled pairwise data to facilitate claim verification while imposing low demand on data annotations and computing resources. MAPLE demonstrates significant performance improvements over SOTA baselines SEED, PET and LLaMA 2 across three fact-checking datasets: FEVER, Climate FEVER, and SciFact. Data and code are available.

To mitigate the impact of the scarcity of labelled data on fact-checking systems, we focus on few-shot claim verification. Despite recent work on few-shot classification by proposing advanced language models, there is a dearth of research in data annotation prioritisation that improves the selection of the few shots to be labelled for optimal model performance. We propose Active PETs, a novel weighted approach that utilises an ensemble of Pattern Exploiting Training (PET) models based on various language models, to actively select unlabelled data as candidates for annotation. Using Active PETs for few-shot data selection shows consistent improvement over the baseline methods, on two technical fact-checking datasets and using six different pretrained language models. We show further improvement with Active PETs-o, which further integrates an oversampling strategy. Our approach enables effective selection of instances to be labelled where unlabelled data is abundant but resources for labelling are limited, leading to consistently improved few-shot claim verification performance. Our code is available.

Scientific claim verification is a unique challenge that is attracting increasing interest. The SCIVER shared task offers a benchmark scenario to test and compare claim verification approaches by participating teams and consists in three steps: relevant abstract selection, rationale selection and label prediction. In this paper, we present team QMUL-SDS’s participation in the shared task. We propose an approach that performs scientific claim verification by doing binary classifications step-by-step. We trained a BioBERT-large classifier to select abstracts based on pairwise relevance assessments for each <claim, title of the abstract> and continued to train it to select rationales out of each retrieved abstract based on <claim, sentence>. We then propose a two-step setting for label prediction, i.e. first predicting “NOT_ENOUGH_INFO” or “ENOUGH_INFO”, then label those marked as “ENOUGH_INFO” as either “SUPPORT” or “CONTRADICT”. Compared to the baseline system, we achieve substantial improvements on the dev set. As a result, our team is the No. 4 team on the leaderboard.