arXiv subjects

Cheston Tan

Publications and source records attributed to Cheston Tan.

At least 19 recordsLinked to original sources

arXiv2026

Skill Issue: Are Skills Language-Invariant in LLMs?

Large language models access knowledge inconsistently across languages, but to what extent do they differ in their skill sets when interacting with different languages? This work quantifies cross-lingual skill inconsistency orthogonally from knowledge and general benchmark performance. We do this via multilingual self-play: two instances of the same model compete in a text-based game, each interacting through a different language interface. Since the model, opponent, rules, state space, and available actions remain fixed, this setting isolates the effect of language on the model's realized behavior. We build a multilingual extension to TextArena and evaluate three open-weight models across eight languages and six games covering spatial reasoning, imperfect information, resource allocation, and repeated interaction. We find that the same model can exhibit markedly different playing strength across languages, with systematic variation in win--loss margins, invalid actions, and strategic tendencies. Detailed analyses reveal language-specific failures in spatial reasoning, card-conditioned decisions, and optimal move selection. In some settings, changing only the intermediate reasoning language recovers much of the lost performance, suggesting that language can affect different stages of the decision process. These results show that skill discrepancies are a measurable major roadblock in the development of truly multilingual models. Better understanding these discrepancies can help us design models that perform more equitably across languages.

cs.CL↗

arXiv2026

FedPref: Federated Preference Learning for Structured Radiology Report Extraction

Radiology reports describe findings and locations in free text, but downstream search and analysis require these relations in a fixed schema. Learning this extraction requires labels that are unevenly distributed across institutions: smaller hospitals have less local evidence, and pooling data may be infeasible. We introduce FedPref: frozen public language models propose alternative JSON extractions, local annotations rank them, and sites collaboratively train compact Qwen3-8B adapters while sharing only model updates. A heterogeneous teacher pool provides cross-model contrast when repeated single-model samples collapse. On development data from six simulated hospitals with unequal data volume and disease prevalence, FedPref improves client-mean F1 by 2.49 points and worst-site F1 by 9.10 points compared with training each site in isolation, with the largest gains at the sites holding the least data. Central training on the pooled preference-pair union is 2.66 points higher on client-mean F1. On a locked, 400-report manually validated gold test set, FedPref reaches 68.68 F1 and pooled training 71.67, preserving that same ordering. FedPref thus lets institutions with unequal, unpooled data benefit from collaboration without ever sharing reports or annotations.

cs.AI↗

arXiv2026

StatePlay: State-Aware Game World Models for Mechanics-Consistent Generation

Recent game world models can generate visually realistic and interactive environments conditioned on player actions. However, games are not defined by pixels alone; they are governed by explicit mechanics, namely state-dependent rules that control health reduction, skill activation, and game termination. These mechanics depend on precise internal states, such as health points, skill meters, and timers, which are tightly coupled with visual observations and determine how gameplay evolves. Without modeling these state dynamics, existing game world models may generate visually plausible rollouts but violate the underlying game rules. In this paper, we propose StatePlay, a novel state-aware game world model that jointly predicts visual content and game states to promote mechanics-consistent generation. StatePlay adopts a mixture-of-transformers (MoT)-style architecture that preserves specialized visual and state representations while enabling cross-modal interaction, allowing predicted states to guide frame generation. Each branch is further optimized with a distinct objective suited to its modality. Experiments show that StatePlay achieves an average normalized L1 distance below 0.06 for state prediction. Furthermore, compared with models without explicit state modeling, our method improves mechanics fidelity in generated game rollouts by 18.6%. Overall, our work highlights the importance of state-aware game world modeling and advances beyond pixel-level realism toward complete and mechanically faithful game generation.

cs.CV↗

arXiv2026

MINDGAMES: A Live Arena for Evaluating Social and Strategic Reasoning in Multi-Agent LLMs

Large language models (LLMs) are increasingly deployed as interactive agents, yet their capacity for social and strategic reasoning over extended interaction remains poorly understood. Existing evaluations rely on static vignettes or single-game benchmarks that cannot capture the sustained, multi-faceted reasoning that real-world multi-agent settings demand. We introduce Mindgames, a multi-game arena and evaluation platform for LLM agents that operationalizes complementary reasoning demands relevant to ``theory of mind'': belief attribution under hidden information, opponent modeling through repeated strategic interaction, cooperative inference under knowledge asymmetries, and sustained deception in social deduction. Built on TextArena, Mindgames provides a unified interaction interface, TrueSkill-based rating, and full trajectory logging across four game environments. We instantiate Mindgames through a 2025 competition cycle hosted at a major AI conference, which assessed 944 submitted agents from 76 teams across four games: Colonel Blotto, Iterated Prisoner's Dilemma, Codenames, and Secret Mafia. Our analysis surfaces both agent-level and evaluation-level limitations: brittle rule adherence remains a major bottleneck, top-performing systems repeatedly rely on explicit structural scaffolding, and leaderboard validity differs sharply across environments. In particular, failure-heavy environments can reward robustness to opponent errors as much as strategic ability, with Secret Mafia exhibiting a pronounced error-survival confound in this cycle. We release a dataset of 29,571 multi-agent games with turn-level observations, actions, and rewards, together with MG-Ref, a deterministic offline tournament protocol that scores new agents against a frozen reference pool of top-ranked, low-error Stage~II submissions under the same error-attribution lens used in this analysis.

cs.AI↗

arXiv2026

CoMMET: A Psychologically Grounded Benchmark for Evaluating Theory of Mind in Multimodal LLMs

Theory of Mind (ToM)-the ability to reason about the mental states of oneself and others-is a cornerstone of human social intelligence. As Multimodal Large Language Models (MLLMs) become ubiquitous in real-world applications, validating their capacity for this level of social reasoning is essential for effective and natural interactions. However, existing benchmarks for assessing ToM in MLLMs are limited; most rely solely on text inputs and focus narrowly on belief-related tasks. In this paper, we propose a new multimodal benchmark dataset, CoMMET, a comprehensive mental states and moral evaluation task inspired by the Theory of Mind Booklet Task. CoMMET expands the scope of evaluation by covering a broader range of mental states and introducing multi-turn testing. To the best of our knowledge, this is the first psychology-grounded benchmark to evaluate MLLMs across multiple mental states in a multimodal, open-ended, and multi-turn setting. Through a comprehensive assessment of MLLMs across different families and sizes, we analyze the strengths and limitations of current models and identify directions for future improvement.

cs.CL↗

arXiv2026

MEMO: Memory-Augmented Model Context Optimization for Robust Multi-Turn Multi-Agent LLM Games

Multi-turn, multi-agent LLM game evaluations often exhibit substantial run-to-run variance. In long-horizon interactions, small early deviations compound across turns and are amplified by multi-agent coupling. This biases win rate estimates and makes rankings unreliable across repeated tournaments. Prompt choice worsens this further by producing different effective policies. We address both instability and underperformance with MEMO (Memory-augmented MOdel context optimization), a self-play framework that optimizes inference-time context by coupling retention and exploration. Retention maintains a persistent memory bank that stores structured insights from self-play trajectories and injects them as priors during later play. Exploration runs tournament-style prompt evolution with uncertainty-aware selection via TrueSkill, and uses prioritized replay to revisit rare and decisive states. Across five text-based games, MEMO raises mean win rate from 25.1% to 49.5% for GPT-4o-mini and from 20.9% to 44.3% for Qwen-2.5-7B-Instruct, using $2,000$ self-play games per task. Run-to-run variance also drops, giving more stable rankings across prompt variations. These results suggest that multi-agent LLM game performance and robustness have substantial room for improvement through context optimization. MEMO achieves the largest gains in negotiation and imperfect-information games, while RL remains more effective in perfect-information settings. All code is open-source and available here: https://github.com/openverse-ai/MEMO

cs.AI↗

arXiv2026

Information Fidelity in Tool-Using LLM Agents: A Martingale Analysis of the Model Context Protocol

As AI agents powered by large language models (LLMs) increasingly use external tools for high-stakes decisions, a critical reliability question arises: how do errors propagate across sequential tool calls? We introduce the first theoretical framework for analyzing error accumulation in Model Context Protocol (MCP) agents, proving that cumulative distortion exhibits linear growth and high-probability deviations bounded by $O(\sqrt{T})$. This concentration property ensures predictable system behavior and rules out exponential failure modes. We develop a hybrid distortion metric combining discrete fact matching with continuous semantic similarity, then establish martingale concentration bounds on error propagation through sequential tool interactions. Experiments across Qwen2-7B, Llama-3-8B, and Mistral-7B validate our theoretical predictions, showing empirical distortion tracks the linear trend with deviations consistently within $O(\sqrt{T})$ envelopes. Key findings include: semantic weighting reduces distortion by 80\%, and periodic re-grounding approximately every 9 steps suffices for error control. We translate these concentration guarantees into actionable deployment principles for trustworthy agent systems.

cs.AI↗

arXiv2026

TangramSR: Can Vision-Language Models Reason in Continuous Geometric Space?

Humans excel at spatial reasoning tasks like Tangram puzzle assembly through cognitive processes involving mental rotation, iterative refinement, and visual feedback. Inspired by how humans solve Tangram puzzles through trial-and-error, observation, and correction, we design a framework that models these human cognitive mechanisms. However, comprehensive experiments across five representative Vision-Language Models (VLMs) reveal systematic failures in continuous geometric reasoning: average IoU of only 0.41 on single-piece tasks, dropping to 0.23 on two-piece composition, far below human performance where children can complete Tangram tasks successfully. This paper addresses a fundamental challenge in self-improving AI: can models iteratively refine their predictions at test time without parameter updates? We introduce a test-time self-refinement framework that combines in-context learning (ICL) with reward-guided feedback loops, inspired by human cognitive processes. Our training-free verifier-refiner agent applies recursive refinement loops that iteratively self-refine predictions based on geometric consistency feedback, achieving IoU improvements from 0.63 to 0.932 on medium-triangle cases without any model retraining. This demonstrates that incorporating human-inspired iterative refinement mechanisms through ICL and reward loops can substantially enhance geometric reasoning in VLMs, moving self-improving AI from promise to practice in continuous spatial domains. Our work is available at this anonymous link https://anonymous.4open.science/r/TangramVLM-F582/.

cs.AI↗

arXiv2025

10 Open Challenges Steering the Future of Vision-Language-Action Models

Due to their ability of follow natural language instructions, vision-language-action (VLA) models are increasingly prevalent in the embodied AI arena, following the widespread success of their precursors -- LLMs and VLMs. In this paper, we discuss 10 principal milestones in the ongoing development of VLA models -- multimodality, reasoning, data, evaluation, cross-robot action generalization, efficiency, whole-body coordination, safety, agents, and coordination with humans. Furthermore, we discuss the emerging trends of using spatial understanding, modeling world dynamics, post training, and data synthesis -- all aiming to reach these milestones. Through these discussions, we hope to bring attention to the research avenues that may accelerate the development of VLA models into wider acceptability.

cs.RO↗

arXiv2025

FailSafe: Reasoning and Recovery from Failures in Vision-Language-Action Models

Recent advances in robotic manipulation have integrated low-level robotic control into Vision-Language Models (VLMs), extending them into Vision-Language-Action (VLA) models. Although state-of-the-art VLAs achieve strong performance in downstream robotic applications, supported by large-scale crowd-sourced robot training data, they still inevitably encounter failures during execution. Enabling robots to reason and recover from unpredictable and abrupt failures remains a critical challenge. Existing robotic manipulation datasets, collected in either simulation or the real world, primarily provide only ground-truth trajectories, leaving robots unable to recover once failures occur. Moreover, the few datasets that address failure detection typically offer only textual explanations, which are difficult to utilize directly in VLA models. To address this gap, we introduce FailSafe, a novel failure generation and recovery system that automatically produces diverse failure cases paired with executable recovery actions. FailSafe can be easily adapted to a wide range of manipulation tasks in simulators with motion planning support, enabling scalable creation of failure-action data. To demonstrate its effectiveness, we fine-tune LLaVA-OneVision-7B (LLaVA-OV-7B) to build FailSafe-VLM. Experimental results show that FailSafe-VLM successfully helps robotic arms detect and recover from potential failures, improving the performance of three state-of-the-art VLA models (Pi-0-FAST, OpenVLA, OpenVLA-OFT) by up to 22.6% on average across several tasks in ManiSkill. Furthermore, FailSafe-VLM could generalize across different spatial configurations, camera viewpoints, object and robotic embodiments.

cs.RO↗

arXiv2025

Stencil: Subject-Driven Generation with Context Guidance

Recent text-to-image diffusion models can generate striking visuals from text prompts, but they often fail to maintain subject consistency across generations and contexts. One major limitation of current fine-tuning approaches is the inherent trade-off between quality and efficiency. Fine-tuning large models improves fidelity but is computationally expensive, while fine-tuning lightweight models improves efficiency but compromises image fidelity. Moreover, fine-tuning pre-trained models on a small set of images of the subject can damage the existing priors, resulting in suboptimal results. To this end, we present Stencil, a novel framework that jointly employs two diffusion models during inference. Stencil efficiently fine-tunes a lightweight model on images of the subject, while a large frozen pre-trained model provides contextual guidance during inference, injecting rich priors to enhance generation with minimal overhead. Stencil excels at generating high-fidelity, novel renditions of the subject in less than a minute, delivering state-of-the-art performance and setting a new benchmark in subject-driven generation.

cs.CV↗

arXiv2025

SPIRAL: Self-Play on Zero-Sum Games Incentivizes Reasoning via Multi-Agent Multi-Turn Reinforcement Learning

Recent advances in reinforcement learning have shown that language models can develop sophisticated reasoning through training on tasks with verifiable rewards, but these approaches depend on human-curated problem-answer pairs and domain-specific reward engineering. We introduce SPIRAL, a self-play framework where models learn by playing multi-turn, zero-sum games against continuously improving versions of themselves, generating an automatic curriculum of stronger opponents, and eliminating the need for human supervision. To enable this self-play training at scale, we implement a fully online, multi-turn, multi-agent reinforcement learning system for LLMs and propose role-conditioned advantage estimation (RAE) to stabilize multi-agent training. SPIRAL produces reasoning capabilities that transfer broadly, improving performance by up to 10% across a suite of 8 reasoning benchmarks on 4 different models spanning Qwen and Llama model families, outperforming supervised fine-tuning on 25,000 expert game trajectories. Multi-game training (TicTacToe, Kuhn Poker, Simple Negotiation) yields the strongest results, with improvements observed across both base and instruction-tuned models. Analysis of chain-of-thought traces reveals that games develop distinct cognitive patterns that transfer to improve reasoning performance, with different games developing complementary strengths. Even models which have already been trained on reasoning tasks using RLVR, like DeepSeek-R1-Distill-Qwen-7B, still benefit from our approach. These results demonstrate that zero-sum games naturally develop transferable reasoning capabilities across diverse model architectures and training stages, highlighting a promising direction for autonomous reasoning development. Our code can be found in https://github.com/spiral-rl/spiral.

cs.AI↗

arXiv2025

GroundFlow: A Plug-in Module for Temporal Reasoning on 3D Point Cloud Sequential Grounding

Sequential grounding in 3D point clouds (SG3D) refers to locating sequences of objects by following text instructions for a daily activity with detailed steps. Current 3D visual grounding (3DVG) methods treat text instructions with multiple steps as a whole, without extracting useful temporal information from each step. However, the instructions in SG3D often contain pronouns such as "it", "here" and "the same" to make language expressions concise. This requires grounding methods to understand the context and retrieve relevant information from previous steps to correctly locate object sequences. Due to the lack of an effective module for collecting related historical information, state-of-the-art 3DVG methods face significant challenges in adapting to the SG3D task. To fill this gap, we propose GroundFlow -- a plug-in module for temporal reasoning on 3D point cloud sequential grounding. Firstly, we demonstrate that integrating GroundFlow improves the task accuracy of 3DVG baseline methods by a large margin (+7.5\% and +10.2\%) in the SG3D benchmark, even outperforming a 3D large language model pre-trained on various datasets. Furthermore, we selectively extract both short-term and long-term step information based on its relevance to the current instruction, enabling GroundFlow to take a comprehensive view of historical information and maintain its temporal understanding advantage as step counts increase. Overall, our work introduces temporal reasoning capabilities to existing 3DVG models and achieves state-of-the-art performance in the SG3D benchmark across five datasets.

cs.CV↗

arXiv2025

The Singapore Consensus on Global AI Safety Research Priorities

Rapidly improving AI capabilities and autonomy hold significant promise of transformation, but are also driving vigorous debate on how to ensure that AI is safe, i.e., trustworthy, reliable, and secure. Building a trusted ecosystem is therefore essential -- it helps people embrace AI with confidence and gives maximal space for innovation while avoiding backlash. The "2025 Singapore Conference on AI (SCAI): International Scientific Exchange on AI Safety" aimed to support research in this space by bringing together AI scientists across geographies to identify and synthesise research priorities in AI safety. This resulting report builds on the International AI Safety Report chaired by Yoshua Bengio and backed by 33 governments. By adopting a defence-in-depth model, this report organises AI safety research domains into three types: challenges with creating trustworthy AI systems (Development), challenges with evaluating their risks (Assessment), and challenges with monitoring and intervening after deployment (Control).

cs.AI↗

arXiv2025

How do Transformer Embeddings Represent Compositions? A Functional Analysis

Compositionality is a key aspect of human intelligence, essential for reasoning and generalization. While transformer-based models have become the de facto standard for many language modeling tasks, little is known about how they represent compound words, and whether these representations are compositional. In this study, we test compositionality in Mistral, OpenAI Large, and Google embedding models, and compare them with BERT. First, we evaluate compositionality in the representations by examining six diverse models of compositionality (addition, multiplication, dilation, regression, etc.). We find that ridge regression, albeit linear, best accounts for compositionality. Surprisingly, we find that the classic vector addition model performs almost as well as any other model. Next, we verify that most embedding models are highly compositional, while BERT shows much poorer compositionality. We verify and visualize our findings with a synthetic dataset consisting of fully transparent adjective-noun compositions. Overall, we present a thorough investigation of compositionality.

cs.CL↗

arXiv2025

From Grunts to Lexicons: Emergent Language from Cooperative Foraging

Language is a powerful communicative and cognitive tool. It enables humans to express thoughts, share intentions, and reason about complex phenomena. Despite our fluency in using and understanding language, the question of how it arises and evolves over time remains unsolved. A leading hypothesis in linguistics and anthropology posits that language evolved to meet the ecological and social demands of early human cooperation. Language did not arise in isolation, but through shared survival goals. Inspired by this view, we investigate the emergence of language in multi-agent Foraging Games. These environments are designed to reflect the cognitive and ecological constraints believed to have influenced the evolution of communication. Agents operate in a shared grid world with only partial knowledge about other agents and the environment, and must coordinate to complete games like picking up high-value targets or executing temporally ordered actions. Using end-to-end deep reinforcement learning, agents learn both actions and communication strategies from scratch. We find that agents develop communication protocols with hallmark features of natural language: arbitrariness, interchangeability, displacement, cultural transmission, and compositionality. We quantify each property and analyze how different factors, such as population size, social dynamics, and temporal dependencies, shape specific aspects of the emergent language. Our framework serves as a platform for studying how language can evolve from partial observability, temporal reasoning, and cooperative goals in embodied multi-agent settings. We will release all data, code, and models publicly.

cs.AI↗

arXiv2025

Theory of Mind in Large Language Models: Assessment and Enhancement

Theory of Mind (ToM)-the ability to reason about the mental states of oneself and others-is a cornerstone of human social intelligence. As Large Language Models (LLMs) become increasingly integrated into daily life, understanding their ability to interpret and respond to human mental states is crucial for enabling effective interactions. In this paper, we review LLMs' ToM capabilities by analyzing both evaluation benchmarks and enhancement strategies. For evaluation, we focus on recently proposed and widely used story-based benchmarks. For enhancement, we provide an in-depth analysis of recent methods aimed at improving LLMs' ToM abilities. Furthermore, we outline promising directions for future research to further advance these capabilities and better adapt LLMs to more realistic and diverse scenarios. Our survey serves as a valuable resource for researchers interested in evaluating and advancing LLMs' ToM capabilities.

cs.CL↗

arXiv2025

TextArena

TextArena is an open-source collection of competitive text-based games for training and evaluation of agentic behavior in Large Language Models (LLMs). It spans 57+ unique environments (including single-player, two-player, and multi-player setups) and allows for easy evaluation of model capabilities via an online-play system (against humans and other submitted models) with real-time TrueSkill scores. Traditional benchmarks rarely assess dynamic social skills such as negotiation, theory of mind, and deception, creating a gap that TextArena addresses. Designed with research, community and extensibility in mind, TextArena emphasizes ease of adding new games, adapting the framework, testing models, playing against the models, and training models. Detailed documentation of environments, games, leaderboard, and examples are available on https://github.com/LeonGuertler/TextArena and https://www.textarena.ai/.

cs.CL↗