Adaptive Interviewing for Persona Simulation in LLMs: Evidence-Grounded Reasoning Improves Decision Alignment

Accurately simulating the decisions of a specific individual remains challenging for large language models (LLMs), partly because persona information is often provided as static descriptions that miss the values, experiences, and contextual cues needed for individual-level decision simulation. We propose an adaptive interview framework that gathers persona-relevant information through a structured three-stage dialogue: core questions, dynamic follow-ups, and a synthesized personality summary. Using the resulting interview transcripts, we evaluate whether LLMs can simulate participants' decisions in moral dilemma scenarios. We compare three conversational contexts -- Core-10 responses, the full interview dialogue, and a summarized persona representation. We find that adaptive interviewing functions less as a uniform accuracy booster and more as a selective grounding mechanism: follow-up-derived evidence is incorporated in around 40% of full-interview traces, and these follow-up-grounded predictions are more accurate than core-only grounded ones (45.5% vs. 39.3%). These findings highlight that richer persona context alone is insufficient: improvements arise only when models actually ground their decisions in user-specific evidence.

StakeBench: Evaluating Language Understanding Grounded in Market Commitment

Existing financial NLP benchmarks often rely on labels supplied by outside observers, measuring how language is perceived rather than what speakers have committed to in the market. We introduce StakeBench, an evaluation framework for language understanding grounded in market commitment. StakeBench links 560,876 comments from 2,261 resolved markets to verified position, action, and market-odds records across Polymarket and Manifold. Supervision is derived from observable market behavior. Position sides, post-comment trading actions, and market-odds trajectories replace human annotation. Four diagnostic tasks test whether models detect market commitment, identify the revealed side, anticipate future action, and perform collective odds projection. Three commitment-aware metrics measure alignment with revealed preferences rather than perceived sentiment. Validity audits and explicit interpretation boundaries help distinguish observable commitment signals from latent belief and causal market-odds impact. Across 15 LLMs and 18 topics and platform settings, models partially recover position-side signals, with Directed Accuracy from 0.506 to 0.599, but show structural failures on later tasks. Ten of the fifteen models collapse to one or two action labels in future action anticipation, and no model consistently improves on the naive odds-direction baseline in collective odds projection. Model scale is not correlated with performance, finance-domain tuning does not improve revealed-side identification, and platform incentives strongly shape higher-order results. StakeBench is packaged with evaluation code and dataset under CC-BY 4.0.

PhyMix: Towards Physically Consistent Single-Image 3D Indoor Scene Generation with Implicit--Explicit Optimization

Existing single-image 3D indoor scene generators often produce results that look visually plausible but fail to obey real-world physics, limiting their reliability in robotics, embodied AI, and design. To examine this gap, we introduce a unified Physics Evaluator that measures four main aspects: geometric priors, contact, stability, and deployability, which are further decomposed into nine sub-constraints, establishing the first benchmark to measure physical consistency. Based on this evaluator, our analysis shows that state-of-the-art methods remain largely physics-unaware. To overcome this limitation, we further propose a framework that integrates feedback from the Physics Evaluator into both training and inference, enhancing the physical plausibility of generated scenes. Specifically, we propose PhyMix, which is composed of two complementary components: (i) implicit alignment via Scene-GRPO, a critic-free group-relative policy optimization that leverages the Physics Evaluator as a preference signal and biases sampling towards physically feasible layouts, and (ii) explicit refinement via a plug-and-play Test-Time Optimizer (TTO) that uses differentiable evaluator signals to correct residual violations during generation. Overall, our method unifies evaluation, reward shaping, and inference-time correction, producing 3D indoor scenes that are visually faithful and physically plausible. Extensive synthetic evaluations confirm state-of-the-art performance in both visual fidelity and physical plausibility, and extensive qualitative examples in stylized and real-world images further showcase the robustness of the method. We will release codes and models upon publication.

coDrawAgents: A Multi-Agent Dialogue Framework for Compositional Image Generation

Text-to-image generation has advanced rapidly, but existing models still struggle with faithfully composing multiple objects and preserving their attributes in complex scenes. We propose coDrawAgents, an interactive multi-agent dialogue framework with four specialized agents: Interpreter, Planner, Checker, and Painter that collaborate to improve compositional generation. The Interpreter adaptively decides between a direct text-to-image pathway and a layout-aware multi-agent process. In the layout-aware mode, it parses the prompt into attribute-rich object descriptors, ranks them by semantic salience, and groups objects with the same semantic priority level for joint generation. Guided by the Interpreter, the Planner adopts a divide-and-conquer strategy, incrementally proposing layouts for objects with the same semantic priority level while grounding decisions in the evolving visual context of the canvas. The Checker introduces an explicit error-correction mechanism by validating spatial consistency and attribute alignment, and refining layouts before they are rendered. Finally, the Painter synthesizes the image step by step, incorporating newly planned objects into the canvas to provide richer context for subsequent iterations. Together, these agents address three key challenges: reducing layout complexity, grounding planning in visual context, and enabling explicit error correction. Extensive experiments on benchmarks GenEval and DPG-Bench demonstrate that coDrawAgents substantially improves text-image alignment, spatial accuracy, and attribute binding compared to existing methods.

Influencing LLM Multi-Agent Dialogue via Policy-Parameterized Prompts

Large Language Models (LLMs) have emerged as a new paradigm for multi-agent systems. However, existing research on the behaviour of LLM-based multi-agents relies on ad hoc prompts and lacks a principled policy perspective. Different from reinforcement learning, we investigate whether prompt-as-action can be parameterized so as to construct a lightweight policy which consists of a sequence of state-action pairs to influence conversational behaviours without training. Our framework regards prompts as actions executed by LLMs, and dynamically constructs prompts through five components based on the current state of the agent. To test the effectiveness of parameterized control, we evaluated the dialogue flow based on five indicators: responsiveness, rebuttal, evidence usage, non-repetition, and stance shift. We conduct experiments using different LLM-driven agents in two discussion scenarios related to the general public and show that prompt parameterization can influence the dialogue dynamics. This result shows that policy-parameterised prompts offer a simple and effective mechanism to influence the dialogue process, which will help the research of multi-agent systems in the direction of social simulation.

Action-Geometry Prediction with 3D Geometric Prior for Bimanual Manipulation

Bimanual manipulation requires policies that can reason about 3D geometry, anticipate how it evolves under action, and generate smooth, coordinated motions. However, existing methods typically rely on 2D features with limited spatial awareness, or require explicit point clouds that are difficult to obtain reliably in real-world settings. At the same time, recent 3D geometric foundation models show that accurate and diverse 3D structure can be reconstructed directly from RGB images in a fast and robust manner. We leverage this opportunity and propose a framework that builds bimanual manipulation directly on a pre-trained 3D geometric foundation model. Our policy fuses geometry-aware latents, 2D semantic features, and proprioception into a unified state representation, and uses diffusion model to jointly predict a future action chunk and a future 3D latent that decodes into a dense pointmap. By explicitly predicting how the 3D scene will evolve together with the action sequence, the policy gains strong spatial understanding and predictive capability using only RGB observations. We evaluate our method both in simulation on the RoboTwin benchmark and in real-world robot executions. Our approach consistently outperforms 2D-based and point-cloud-based baselines, achieving state-of-the-art performance in manipulation success, inter-arm coordination, and 3D spatial prediction accuracy. Code is available at https://github.com/Chongyang-99/GAP.git.

PEGAsus: 3D Personalization of Geometry and Appearance

We present PEGAsus, a new framework capable of generating Personalized 3D shapes by learning shape concepts at both Geometry and Appearance levels. First, we formulate 3D shape personalization as extracting reusable, category-agnostic geometric and appearance attributes from reference shapes, and composing these attributes with text to generate novel shapes. Second, we design a progressive optimization strategy to learn shape concepts at both the geometry and appearance levels, decoupling the shape concept learning process. Third, we extend our approach to region-wise concept learning, enabling flexible concept extraction, with context-aware and context-free losses. Extensive experimental results show that PEGAsus is able to effectively extract attributes from a wide range of reference shapes and then flexibly compose these concepts with text to synthesize new shapes. This enables fine-grained control over shape generation and supports the creation of diverse, personalized results, even in challenging cross-category scenarios. Both quantitative and qualitative experiments demonstrate that our approach outperforms existing state-of-the-art solutions.

Imagine a City: CityGenAgent for Procedural 3D City Generation

The automated generation of interactive 3D cities is a critical challenge with broad applications in autonomous driving, virtual reality, and embodied intelligence. While recent advances in generative models and procedural techniques have improved the realism of city generation, existing methods often struggle with high-fidelity asset creation, controllability, and manipulation. In this work, we introduce CityGenAgent, a natural language-driven framework for hierarchical procedural generation of high-quality 3D cities. Our approach decomposes city generation into two interpretable components, Block Program and Building Program. To ensure structural correctness and semantic alignment, we adopt a two-stage learning strategy: (1) Supervised Fine-Tuning (SFT). We train BlockGen and BuildingGen to generate valid programs that adhere to schema constraints, including non-self-intersecting polygons and complete fields; (2) Reinforcement Learning (RL). We design Spatial Alignment Reward to enhance spatial reasoning ability and Visual Consistency Reward to bridge the gap between textual descriptions and the visual modality. Benefiting from the programs and the models' generalization, CityGenAgent supports natural language editing and manipulation. Comprehensive evaluations demonstrate superior semantic alignment, visual quality, and controllability compared to existing methods, establishing a robust foundation for scalable 3D city generation.

Failing on Bias Mitigation: Investigating Why Predictive Models Struggle with Government Data

The potential for bias and unfairness in AI-supporting government services raises ethical and legal concerns. Using crime rate prediction with the Bristol City Council data as a case study, we examine how these issues persist. Rather than auditing real-world deployed systems, our goal is to understand why widely adopted bias mitigation techniques often fail when applied to government data. Our findings reveal that bias mitigation approaches applied to government data are not always effective -- not because of flaws in model architecture or metric selection, but due to the inherent properties of the data itself. Through comparing a set of comprehensive models and fairness methods, our experiments consistently show that the mitigation efforts cannot overcome the embedded unfairness in the data -- further reinforcing that the origin of bias lies in the structure and history of government datasets. We then explore the reasons for the mitigation failures in predictive models on government data and highlight the potential sources of unfairness posed by data distribution shifts, the accumulation of historical bias, and delays in data release. We also discover the limitations of the blind spots in fairness analysis and bias mitigation methods when only targeting a single sensitive feature through a set of intersectional fairness experiments. Although this study is limited to one city, the findings are highly suggestive, which can contribute to an early warning that biases in government data may persist even with standard mitigation methods.

AutoMonitor-Bench: Evaluating the Reliability of LLM-Based Misbehavior Monitor

We introduce AutoMonitor-Bench, the first benchmark designed to systematically evaluate the reliability of LLM-based misbehavior monitors across diverse tasks and failure modes. AutoMonitor-Bench consists of 3,010 carefully annotated test samples spanning question answering, code generation, and reasoning, with paired misbehavior and benign instances. We evaluate monitors using two complementary metrics: Miss Rate (MR) and False Alarm Rate (FAR), capturing failures to detect misbehavior and oversensitivity to benign behavior, respectively. Evaluating 12 proprietary and 10 open-source LLMs, we observe substantial variability in monitoring performance and a consistent trade-off between MR and FAR, revealing an inherent safety-utility tension. To further explore the limits of monitor reliability, we construct a large-scale training corpus of 153,581 samples and fine-tune Qwen3-4B-Instruction to investigate whether training on known, relatively easy-to-construct misbehavior datasets improves monitoring performance on unseen and more implicit misbehaviors. Our results highlight the challenges of reliable, scalable misbehavior monitoring and motivate future work on task-aware designing and training strategies for LLM-based monitors.