Daily Papers

Language models have shown unprecedented capabilities, sparking debate over the source of their performance. Is it merely the outcome of learning syntactic patterns and surface level statistics, or do they extract semantics and a world model from the text? Prior work by Li et al. investigated this by training a GPT model on synthetic, randomly generated Othello games and found that the model learned an internal representation of the board state. We extend this work into the more complex domain of chess, training on real games and investigating our model's internal representations using linear probes and contrastive activations. The model is given no a priori knowledge of the game and is solely trained on next character prediction, yet we find evidence of internal representations of board state. We validate these internal representations by using them to make interventions on the model's activations and edit its internal board state. Unlike Li et al's prior synthetic dataset approach, our analysis finds that the model also learns to estimate latent variables like player skill to better predict the next character. We derive a player skill vector and add it to the model, improving the model's win rate by up to 2.6 times.

Modern chess language models are dense transformers trained on millions of games played by thousands of high-rated individuals. However, these monolithic networks tend to collapse into mode-averaged behavior, where stylistic boundaries are blurred, and rare but effective strategies are suppressed. To counteract homogenization, we introduce Mixture-of-Masters (MoM), the first chess mixture-of-experts model with small-sized GPT experts emulating world-class grandmasters. Each expert is trained with a combination of self-supervised learning and reinforcement learning guided by chess-specific rewards. For each move, a post-hoc learnable gating network selects the most appropriate persona to channel depending on the game state, allowing MoM to switch its style dynamically--e.g., Tal's offensive vocation or Petrosian's defensive solidity. When evaluated against Stockfish on unseen standard games, MoM outperforms both dense individual expert networks and popular GPT baselines trained on aggregated data, while ensuring generation variety, control, and interpretability.

Modern chess engines are silent experts: they play at a superhuman level, but do not offer explanations for their play. On the other hand, language models (LMs) can generate plausible-sounding explanations, but their weak playing strength limits the utility of their explanations. We introduce Queen, a 4B-parameter chess-language model that can explain its moves and plans while playing at the level of a typical Grandmaster. Our novel framework enables domain-specific reasoning through complementary components: an encoder-decoder architecture and an iterative distillation algorithm. This architecture integrates a silent expert chess encoder with an instruction-tuned LM through cross-attention, which we train via a question-answering curriculum to extract chess concepts from the encoder's representations. Building on this domain-adapted model, we iteratively improve its explanations with a natural-language analog of the Bellman update: the model analyzes the positions after its top candidate moves and consolidates them into an explanation of the current position, which is then distilled back into the model. Over seven iterations, our model gains over 900 Elo points (1782 to 2697), substantially surpassing all frontier models on both playing strength and puzzle accuracy, despite containing three orders of magnitude fewer parameters. Furthermore, LM-based evaluations show that our explanations are fluent and approach GPT-5.6-Sol (high) in coherence. The generality of our architecture and training procedure suggests a recipe for applying language models to domains where silent expert encoders are available, like games, robotics, and computer use.

Recent work has fine-tuned language models on chess data and reported high benchmark scores as evidence that the resulting models can understand the rules of chess, play full chess games at a professional level, or generate human-readable explanations grounded in expert knowledge. We train KinGPT, a 25M-parameter character-level language model trained only on (position, best-move) pairs, who exceeds 3B-parameter ChessGPT on a 600-puzzle mate-in-N suite and 4B-parameter C1-4B over a 20-theme puzzle benchmark. We examine several claims made in existing literature regarding chess-trained language models and assert that their impressive benchmark performance is largely explained by pattern-matching. We also demonstrate how LLM-Modulo, a verifier-in-the-loop framework, raises RedPajama 3B's best move accuracy from 1.2% to 21.2% and move generation validity from 19.3% to 95.3% on mate-in-N chess puzzles, comparable to gains achieved from ChessGPT's fine-tuning on chess-specific web corpora at a fraction of the cost. Our results illustrate how pairing a general LLM with an external verifier offers a more flexible alternative to directly training on synthetic data for well-defined domains. We open source all training/evaluation code, datasets, puzzle samples, and KinGPT model checkpoints for reproducibility.

Language models often lack grounded reasoning capabilities in specialized domains where training data is scarce but bespoke systems excel. We introduce a general framework for distilling expert system reasoning into natural language chain-of-thought explanations, enabling compact models to acquire domain expertise and the ability to generate faithful, grounded explanations. Rather than distilling only final outputs, we capture the full reasoning process, transforming opaque expert computations into transparent, step-by-step explanations. We demonstrate this approach in chess, a canonical reasoning domain where language models continue to underperform. Our 4B parameter model, C1, advances from a near-zero baseline to 48.1% accuracy, outperforming all open-source models and most frontier proprietary systems. Notably, C1 surpasses its distillation teacher and generates solutions in two orders of magnitude fewer tokens than baselines. Unlike prior neural chess approaches that predict only best moves, C1 generates explainable solutions revealing strategic reasoning. Our pipeline combines supervised fine-tuning and reinforcement learning with theme-balanced data sampling for comprehensive tactical coverage. Master Distillation demonstrates how to inject expert-level knowledge into compact models for under-optimized domains, offering a recipe for unlocking RLVR where LLMs lack sufficient base capabilities.

Chess provides an ideal testbed for evaluating the reasoning, modeling, and abstraction capabilities of large language models (LLMs), as it has well-defined structure and objective ground truth while admitting a wide spectrum of skill levels. However, existing evaluations of LLM ability in chess are ad hoc and narrow in scope, making it difficult to accurately measure LLM chess understanding and how it varies with scale, post-training methodologies, or architecture choices. We present ChessQA, a comprehensive benchmark that assesses LLM chess understanding across five task categories (Structural, Motifs, Short Tactics, Position Judgment, and Semantic), which approximately correspond to the ascending abstractions that players master as they accumulate chess knowledge, from understanding basic rules and learning tactical motifs to correctly calculating tactics, evaluating positions, and semantically describing high-level concepts. In this way, ChessQA captures a more comprehensive picture of chess ability and understanding, going significantly beyond the simple move quality evaluations done previously, and offers a controlled, consistent setting for diagnosis and comparison. Furthermore, ChessQA is inherently dynamic, with prompts, answer keys, and construction scripts that can evolve as models improve. Evaluating a range of contemporary LLMs, we find persistent weaknesses across all five categories and provide results and error analyses by category. We will release the code, periodically refreshed datasets, and a public leaderboard to support further research.

Recent advances in large language models have enabled natural language to serve as a flexible interface for controlling complex systems, but often at the cost of large-scale multimodal training or weakened domain-specific inductive biases. In structured decision-making domains such as chess, specialized policy networks achieve strong performance but lack semantic controllability, while prompt-conditioned language models are more flexible yet typically exhibit weaker domain grounding. We propose UniMaia, a framework for prompt-conditioned policy modulation that adapts a frozen Lc0-based chess policy network using a parameter-efficient text encoder and a ControlNet-style conditioning mechanism. UniMaia enables semantic control over gameplay, including opening selection and player strength, while preserving the pretrained policy representations. We further introduce UniMaia-Aux, which incorporates auxiliary temporal conditioning and behavioral prediction objectives. To support this work, we construct a large-scale metadata-augmented Lichess dataset, develop a semi-automated prompt-generation pipeline, and introduce benchmarks spanning both prompt-conditioned and metadata-conditioned settings. UniMaia achieves state-of-the-art expected accuracy on several prompt-conditioned benchmarks and competitive top-move accuracy on general instruction-following tasks, while remaining competitive with dedicated metadata-conditioned approaches on human move prediction benchmarks. UniMaia-Aux further improves expected accuracy and behavioral modeling across several evaluation settings, with modest trade-offs in top-move accuracy. Overall, our results demonstrate that prompt-conditioned control of domain-specific policy networks is feasible without end-to-end multimodal training, while highlighting trade-offs between controllability and predictive performance.

In this study, we present an innovative fusion of language models and query analysis techniques to unlock cognition in artificial intelligence. Our system seamlessly integrates a Chess engine with a language model, enabling it to predict moves and provide strategic explanations. Leveraging a vector database to achieve retrievable answer generation, our OpenSI AI system elucidates its decision-making process, bridging the gap between raw computation and human-like understanding. Our choice of Chess as the demonstration environment underscores the versatility of our approach. Beyond Chess, our system holds promise for diverse applications, from medical diagnostics to financial forecasting.

Transformer language models have made tremendous strides in natural language understanding tasks. However, the complexity of natural language makes it challenging to ascertain how accurately these models are tracking the world state underlying the text. Motivated by this issue, we consider the task of language modeling for the game of chess. Unlike natural language, chess notations describe a simple, constrained, and deterministic domain. Moreover, we observe that the appropriate choice of chess notation allows for directly probing the world state, without requiring any additional probing-related machinery. We find that: (a) With enough training data, transformer language models can learn to track pieces and predict legal moves with high accuracy when trained solely on move sequences. (b) For small training sets providing access to board state information during training can yield significant improvements. (c) The success of transformer language models is dependent on access to the entire game history i.e. "full attention". Approximating this full attention results in a significant performance drop. We propose this testbed as a benchmark for future work on the development and analysis of transformer language models.

Evaluating large language models becomes increasingly challenging as their capabilities advance: benchmarks can saturate, public test sets risk contamination, and assessing harder tasks can require expensive grading or execution infrastructure. These challenges are amplified in automatic prompt optimization (APO), where evaluation is repeated throughout the search for better prompts. Studying APO therefore requires a benchmark that is cheap and deterministic to score, hard enough to leave room for improvement, and renewable as models evolve. We introduce a chess benchmark built from 1,118 Lichess puzzles to study APO for frozen LLMs: we optimize their prompts without updating their model weights. Chess combines inexpensive exact-match scoring, engine-based evaluation of alternative moves, and a renewable supply of problems with adjustable difficulty. Unlike evaluations that report only success on isolated test items, the benchmark also connects puzzle-solving gains to short game-play rollouts within the same domain. We use it to evaluate six APO algorithms on eight target models, measuring not only baseline strength but also how much each model responds to optimization and whether optimized prompts transfer across models and to game play. Chess is thus a well-suited benchmark for APO: it is (i) challenging, as even the strongest evaluated model, Gemini 3.5 Flash (used as the meta-model), solves only about 55\% of puzzles; (ii) discriminative, revealing gains, unchanged performance, and regressions across methods and models; (iii) renewable, with fresh puzzles to reduce contamination risk and adjustable difficulty to maintain headroom as models improve; and (iv) affordable, as the complete study runs for around \$800. We release the puzzles, optimization and evaluation code, and dataset-renewal scripts (https://github.com/imec-ailabs/Automatic-Prompt-Optimization-with-Chess).

Recent large language models (LLMs) have shown strong reasoning capabilities. However, a critical question remains: do these models possess genuine reasoning skills particularly complex strategic reasoning or are they primarily excelling at sophisticated pattern recognition within their training data? To address this question, this paper presents a chess testbed, ChessArena, to evaluate the strategic reasoning capabilities of LLMs. Chess requires complex strategic reasoning capabilities including long-term planning, strict rule comprehension, and multi-turn conversation memorization. Specifically, ChessArena is a competitive framework where LLMs play against each other, under four different play modes. The testbed is equipped with a ranking algorithm and a leaderboard. The testbed can also evaluate fine-grained capabilities including basic understanding, move selection, and puzzle solving. Over 13 LLMs with different modes are evaluated in ChessArena, playing over 800 games. The results reveal significant shortcomings in current LLMs: no model can beat Maia-1100 (a chess engine at human amateur level), while some even failed to defeat a random player that selects moves arbitrarily. We also present a strong baseline to the testbed: our fine-tuned Qwen3-8B substantially improved performance, approaching much larger state-of-the-art reasoning models.

This work demonstrates that natural language transformers can support more generic strategic modeling, particularly for text-archived games. In addition to learning natural language skills, the abstract transformer architecture can generate meaningful moves on a chessboard. With further fine-tuning, the transformer learns complex gameplay by training on 2.8 million chess games in Portable Game Notation. After 30,000 training steps, OpenAI's Generative Pre-trained Transformer (GPT-2) optimizes weights for 774 million parameters. This fine-tuned Chess Transformer generates plausible strategies and displays game formations identifiable as classic openings, such as English or the Slav Exchange. Finally, in live play, the novel model demonstrates a human-to-transformer interface that correctly filters illegal moves and provides a novel method to challenge the transformer's chess strategies. We anticipate future work will build on this transformer's promise, particularly in other strategy games where features can capture the underlying complex rule syntax from simple but expressive player annotations.

Large Language Models (LLMs) exhibit remarkable capabilities, yet it remains unclear to what extent these reflect sophisticated recall (crystallized intelligence) or reasoning ability (fluid intelligence). We introduce chess as a controlled testbed for disentangling these faculties. Leveraging the game's structure and scalable engine evaluations, we construct a taxonomy of positions varying in training corpus proximity--ranging from common states solvable by memorization to novel ones requiring first-principles reasoning. We systematically evaluate multiple GPT generations under varying reasoning intensities. Our analysis reveals a clear gradient: performance consistently degrades as fluid intelligence demands increase. Notably, in out-of-distribution tasks, performance collapses to random levels. While newer models improve, progress slows significantly for tasks outside the training distribution. Furthermore, while reasoning-augmented inference improves performance, its marginal benefit per token decreases with distributional proximity. These results suggest current architectures remain limited in systematic generalization, highlighting the need for mechanisms beyond scale to achieve robust fluid intelligence.

Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task. While typically task-agnostic in architecture, this method still requires task-specific fine-tuning datasets of thousands or tens of thousands of examples. By contrast, humans can generally perform a new language task from only a few examples or from simple instructions - something which current NLP systems still largely struggle to do. Here we show that scaling up language models greatly improves task-agnostic, few-shot performance, sometimes even reaching competitiveness with prior state-of-the-art fine-tuning approaches. Specifically, we train GPT-3, an autoregressive language model with 175 billion parameters, 10x more than any previous non-sparse language model, and test its performance in the few-shot setting. For all tasks, GPT-3 is applied without any gradient updates or fine-tuning, with tasks and few-shot demonstrations specified purely via text interaction with the model. GPT-3 achieves strong performance on many NLP datasets, including translation, question-answering, and cloze tasks, as well as several tasks that require on-the-fly reasoning or domain adaptation, such as unscrambling words, using a novel word in a sentence, or performing 3-digit arithmetic. At the same time, we also identify some datasets where GPT-3's few-shot learning still struggles, as well as some datasets where GPT-3 faces methodological issues related to training on large web corpora. Finally, we find that GPT-3 can generate samples of news articles which human evaluators have difficulty distinguishing from articles written by humans. We discuss broader societal impacts of this finding and of GPT-3 in general.

We study empirical scaling laws for language model performance on the cross-entropy loss. The loss scales as a power-law with model size, dataset size, and the amount of compute used for training, with some trends spanning more than seven orders of magnitude. Other architectural details such as network width or depth have minimal effects within a wide range. Simple equations govern the dependence of overfitting on model/dataset size and the dependence of training speed on model size. These relationships allow us to determine the optimal allocation of a fixed compute budget. Larger models are significantly more sample-efficient, such that optimally compute-efficient training involves training very large models on a relatively modest amount of data and stopping significantly before convergence.

While large language models have made strides in natural language processing, their proficiency in complex reasoning tasks requiring formal language comprehension, such as chess, remains less investigated. This paper probes the performance of ChatGPT, a sophisticated language model by OpenAI in tackling such complex reasoning tasks, using chess as a case study. Through robust metrics examining both the legality and quality of moves, we assess ChatGPT's understanding of the chessboard, adherence to chess rules, and strategic decision-making abilities. Our evaluation identifies limitations within ChatGPT's attention mechanism that affect its formal language comprehension and uncovers the model's underdeveloped self-regulation abilities. Our study also reveals ChatGPT's propensity for a coherent strategy in its gameplay and a noticeable uptick in decision-making assertiveness when the model is presented with a greater volume of natural language or possesses a more lucid understanding of the state of the chessboard. These findings contribute to the growing exploration of language models' abilities beyond natural language processing, providing valuable information for future research towards models demonstrating human-like cognitive abilities.

Modern natural language models such as the GPT-2/GPT-3 contain tremendous amounts of information about human belief in a consistently testable form. If these models could be shown to accurately reflect the underlying beliefs of the human beings that produced the data used to train these models, then such models become a powerful sociological tool in ways that are distinct from traditional methods, such as interviews and surveys. In this study, We train a version of the GPT-2 on a corpora of historical chess games, and then "launch" clusters of synthetic agents into the model, using text strings to create context and orientation. We compare the trajectories contained in the text generated by the agents/model and compare that to the known ground truth of the chess board, move legality, and historical patterns of play. We find that the percentages of moves by piece using the model are substantially similar from human patterns. We further find that the model creates an accurate latent representation of the chessboard, and that it is possible to plot trajectories of legal moves across the board using this knowledge.

Reasoning is not just about solving problems -- it is also about evaluating which problems are worth solving at all. Evaluations of artificial intelligence (AI) systems primarily focused on problem solving, historically by studying how models play games such as chess and Go. In this paper, we advocate for a new paradigm that assesses AI systems' evaluation of games. First, we introduce a formalism for evaluating such evaluations. We then leverage a large-scale dataset of over 100 novel board games and over 450 human judgments to compare evaluations produced by modern language and reasoning models against those of people and symbolic computational agents. We consider two kinds of evaluative queries: assessing the payoff (or fairness) and the funness of games. These queries span two dimensions relevant to the design of evaluations of AI evaluations: how complex a query is to compute and how difficult a query is to quantify. Our results show that reasoning models are generally more aligned to people in their evaluations of games than non-reasoning language models. However, we observe a non-monotonic relationship: as models get closer to game-theoretic optimal, their fit to human data weakens. We also observe more "jaggedness" across models for assessing funness, in line with the greater difficulty of quantifying this query. Across queries and games, reasoning models show highly variable and unpredictable resource usage when assessing queries, pointing to the importance of imbuing more resource-rational meta-reasoning in language and reasoning models.

When solving decision-making tasks, humans typically depend on information from two key sources: (1) Historical policy data, which provides interaction replay from the environment, and (2) Analytical insights in natural language form, exposing the invaluable thought process or strategic considerations. Despite this, the majority of preceding research focuses on only one source: they either use historical replay exclusively to directly learn policy or value functions, or engaged in language model training utilizing mere language corpus. In this paper, we argue that a powerful autonomous agent should cover both sources. Thus, we propose ChessGPT, a GPT model bridging policy learning and language modeling by integrating data from these two sources in Chess games. Specifically, we build a large-scale game and language dataset related to chess. Leveraging the dataset, we showcase two model examples ChessCLIP and ChessGPT, integrating policy learning and language modeling. Finally, we propose a full evaluation framework for evaluating language model's chess ability. Experimental results validate our model and dataset's effectiveness. We open source our code, model, and dataset at https://github.com/waterhorse1/ChessGPT.

Large language models (LLMs) memorize a vast amount of prior knowledge from the Internet that help them on downstream tasks but also may notoriously sway their outputs towards wrong or biased answers. In this work, we test how the knowledge about popular subjects hurt the accuracy of vision language models (VLMs) on standard, objective visual tasks of counting and identification. We find that state-of-the-art VLMs are strongly biased (e.g, unable to recognize a fourth stripe has been added to a 3-stripe Adidas logo) scoring an average of 17.05% accuracy in counting (e.g., counting stripes in an Adidas-like logo) across 7 diverse domains from animals, logos, chess, board games, optical illusions, to patterned grids. Insert text (e.g., "Adidas") describing the subject name into the counterfactual image further decreases VLM accuracy. The biases in VLMs are so strong that instructing them to double-check their results or rely exclusively on image details to answer improves counting accuracy by only +2 points, on average. Our work presents an interesting failure mode in VLMs and an automated framework for testing VLM biases. Code and data are available at: vlmsarebiased.github.io.

While large language models perform well on a range of complex tasks (e.g., text generation, question answering, summarization), robust multi-step planning and reasoning remains a considerable challenge for them. In this paper we show that search-based planning can significantly improve LLMs' playing strength across several board games (Chess, Fischer Random / Chess960, Connect Four, and Hex). We introduce, compare and contrast two major approaches: In external search, the model guides Monte Carlo Tree Search (MCTS) rollouts and evaluations without calls to an external engine, and in internal search, the model directly generates in-context a linearized tree of potential futures and a resulting final choice. Both build on a language model pre-trained on relevant domain knowledge, capturing the transition and value functions across these games. We find that our pre-training method minimizes hallucinations, as our model is highly accurate regarding state prediction and legal moves. Additionally, both internal and external search indeed improve win-rates against state-of-the-art bots, even reaching Grandmaster-level performance in chess while operating on a similar move count search budget per decision as human Grandmasters. The way we combine search with domain knowledge is not specific to board games, suggesting direct extensions into more general language model inference and training techniques.

Complex games have long been an important benchmark for testing the progress of artificial intelligence algorithms. AlphaGo, AlphaZero, and MuZero have defeated top human players in Go and Chess, garnering widespread societal attention towards artificial intelligence. Concurrently, large language models (LLMs) have exhibited remarkable capabilities across various tasks, raising the question of whether LLMs can achieve similar success in complex games. In this paper, we explore the potential of LLMs in mastering complex card games. We systematically assess the learning capabilities of LLMs across eight diverse card games, evaluating the impact of fine-tuning on high-quality gameplay data, and examining the models' ability to retain general capabilities while mastering these games. Our findings indicate that: (1) LLMs can approach the performance of strong game AIs through supervised fine-tuning on high-quality data, (2) LLMs can achieve a certain level of proficiency in multiple complex card games simultaneously, with performance augmentation for games with similar rules and conflicts for dissimilar ones, and (3) LLMs experience a decline in general capabilities when mastering complex games, but this decline can be mitigated by integrating a certain amount of general instruction data. The evaluation results demonstrate strong learning ability and versatility of LLMs. The code is available at https://github.com/THUDM/LLM4CardGame

In Natural Language Processing (NLP), the Elo rating system, originally designed for ranking players in dynamic games such as chess, is increasingly being used to evaluate Large Language Models (LLMs) through "A vs B" paired comparisons. However, while popular, the system's suitability for assessing entities with constant skill levels, such as LLMs, remains relatively unexplored. We study two fundamental axioms that evaluation methods should adhere to: reliability and transitivity. We conduct extensive evaluation of Elo behaviour, illustrating that individual Elo computations exhibit volatility and delving into the impact of varying the Elo rating system's hyperparameters. We show that these axioms are not always satisfied raising questions about the reliability of current comparative evaluations of LLMs. If the current use of Elo scores is intended to substitute the costly head-to-head comparison of LLMs, it is crucial to ensure the ranking is as robust as possible. Guided by the axioms, our findings offer concrete guidelines for enhancing the reliability of LLM evaluation methods, suggesting a need for reassessment of existing comparative approaches.

What latent features are encoded in language model (LM) representations? Recent work on training sparse autoencoders (SAEs) to disentangle interpretable features in LM representations has shown significant promise. However, evaluating the quality of these SAEs is difficult because we lack a ground-truth collection of interpretable features that we expect good SAEs to recover. We thus propose to measure progress in interpretable dictionary learning by working in the setting of LMs trained on chess and Othello transcripts. These settings carry natural collections of interpretable features -- for example, "there is a knight on F3" -- which we leverage into supervised metrics for SAE quality. To guide progress in interpretable dictionary learning, we introduce a new SAE training technique, p-annealing, which improves performance on prior unsupervised metrics as well as our new metrics.

Multimodal reasoning stands as a pivotal capability for large vision-language models (LVLMs). The integration with Domain-Specific Languages (DSL), offering precise visual representations, equips these models with the opportunity to execute more accurate reasoning in complex and professional domains. However, the vanilla Chain-of-Thought (CoT) prompting method faces challenges in effectively leveraging the unique strengths of visual and DSL representations, primarily due to their differing reasoning mechanisms. Additionally, it often falls short in addressing critical steps in multi-step reasoning tasks. To mitigate these challenges, we introduce the Bi-Modal Behavioral Alignment (BBA) prompting method, designed to maximize the potential of DSL in augmenting complex multi-modal reasoning tasks. This method initiates by guiding LVLMs to create separate reasoning chains for visual and DSL representations. Subsequently, it aligns these chains by addressing any inconsistencies, thus achieving a cohesive integration of behaviors from different modalities. Our experiments demonstrate that BBA substantially improves the performance of GPT-4V(ision) on geometry problem solving (28.34% to 34.22%), chess positional advantage prediction (42.08% to 46.99%) and molecular property prediction (77.47% to 83.52%).

Humans draw to facilitate reasoning: we draw auxiliary lines when solving geometry problems; we mark and circle when reasoning on maps; we use sketches to amplify our ideas and relieve our limited-capacity working memory. However, such actions are missing in current multimodal language models (LMs). Current chain-of-thought and tool-use paradigms only use text as intermediate reasoning steps. In this work, we introduce Sketchpad, a framework that gives multimodal LMs a visual sketchpad and tools to draw on the sketchpad. The LM conducts planning and reasoning according to the visual artifacts it has drawn. Different from prior work, which uses text-to-image models to enable LMs to draw, Sketchpad enables LMs to draw with lines, boxes, marks, etc., which is closer to human sketching and better facilitates reasoning. Sketchpad can also use specialist vision models during the sketching process (e.g., draw bounding boxes with object detection models, draw masks with segmentation models), to further enhance visual perception and reasoning. We experiment with a wide range of math tasks (including geometry, functions, graphs, and chess) and complex visual reasoning tasks. Sketchpad substantially improves performance on all tasks over strong base models with no sketching, yielding an average gain of 12.7% on math tasks, and 8.6% on vision tasks. GPT-4o with Sketchpad sets a new state of the art on all tasks, including V*Bench (80.3%), BLINK spatial reasoning (83.9%), and visual correspondence (80.8%). All codes and data are in https://visualsketchpad.github.io/.

Utilizing large language models (LLMs) for transforming natural language questions into SQL queries (text-to-SQL) is a promising yet challenging approach, particularly when applied to real-world databases with complex and extensive schemas. In particular, effectively incorporating data catalogs and database values for SQL generation remains an obstacle, leading to suboptimal solutions. We address this problem by proposing a new pipeline that effectively retrieves relevant data and context, selects an efficient schema, and synthesizes correct and efficient SQL queries. To increase retrieval precision, our pipeline introduces a hierarchical retrieval method leveraging model-generated keywords, locality-sensitive hashing indexing, and vector databases. Additionally, we have developed an adaptive schema pruning technique that adjusts based on the complexity of the problem and the model's context size. Our approach generalizes to both frontier proprietary models like GPT-4 and open-source models such as Llama-3-70B. Through a series of ablation studies, we demonstrate the effectiveness of each component of our pipeline and its impact on the end-to-end performance. Our method achieves new state-of-the-art performance on the cross-domain challenging BIRD dataset.

Deploying large language models (LLMs) on edge devices presents significant challenges due to the substantial computational overhead and memory requirements. Activation sparsification can mitigate these challenges by reducing the number of activated neurons during inference. Existing methods typically employ thresholding-based sparsification based on the statistics of activation tensors. However, these methods do not explicitly model the impact of activation sparsification on performance, leading to suboptimal performance degradation. To address this issue, this paper reformulates the activation sparsification problem by introducing a new objective that optimizes the sparsification decisions. Building on this reformulation, we propose CHESS, a general activation sparsification approach via CHannel-wise thrEsholding and Selective Sparsification. First, channel-wise thresholding assigns a unique threshold to each activation channel in the feed-forward network (FFN) layers. Then, selective sparsification involves applying thresholding-based activation sparsification to specific layers within the attention modules. Finally, we detail the implementation of sparse kernels to accelerate LLM inference. Experimental results demonstrate that the proposed CHESS achieves lower performance degradation over 8 downstream tasks while activating fewer parameters compared to existing methods, thus speeding up the LLM inference by up to 1.27x.

How can you get a language model to reason in a task it natively struggles with? We study how reasoning evolves in a language model -- from supervised fine-tuning (SFT) to reinforcement learning (RL) -- by analyzing how a set of theoretically-inspired datasets impacts language model performance in chess. We find that fine-tuning a model to directly predict the best move leads to effective RL and the strongest downstream performance -- however, the RL step elicits unfaithful reasoning (reasoning inconsistent with the chosen move). Alternatively, training on multi-move trajectories yields comparable downstream performance with faithful reasoning and more stable RL. We show that RL induces a substantial positive shift in the distribution of move quality and reduces hallucination rates as a side effect. Finally, we find several SFT-checkpoint metrics -- metrics spanning evaluation performance, hallucination rates, and reasoning quality -- to be predictive of post-RL model performance. We release checkpoints and final models as well as training data, evaluations, and code which allowed us to surpass leading open-source reasoning models in chess with a 7B-parameter model.

We demonstrate that small pretrained foundational generative language models with millions of parameters can learn the latent rules of a process from data associated with the process. Inspired by Stefan Zweig's novella "Schachnovelle," also known as "The Royal Game" in English, we show that 28M and 125M parameter pretrained foundational small language models (SLMs) can be instruction fine-tuned with 1,000-to-1,000,000 examples to learn the rules of chess, propose legal moves, and accurately solve chess problems. We also explore the impact of successive language model fine-tuning epochs on improved outcomes and demonstrate reductions in model hallucinations by increasing the number of instruction fine-tuning examples.

Game playing has long served as a fundamental benchmark for evaluating Artificial General Intelligence (AGI). While Large Language Models (LLMs) have demonstrated impressive capabilities in general reasoning, their effectiveness in spatial strategic reasoning, which is critical for complex and fully observable board games, remains insufficiently explored. In this work, we adopt Chinese Chess (Xiangqi) as a challenging and rich testbed due to its intricate rules and spatial complexity. To advance LLMs' strategic competence in such environments, we propose a training framework tailored to Xiangqi, built upon a large-scale dataset of five million board-move pairs enhanced with expert annotations and engine evaluations. Building on this foundation, we introduce Xiangqi-R1, a 7B-parameter model trained in multi-stage manner: (1) fine-tuning for legal move prediction to capture basic spatial rules, (2) incorporating strategic annotations to improve decision-making, and (3) applying reinforcement learning via Group Relative Policy Optimization (GRPO) with multi-dimensional reward signals to enhance reasoning stability. Our Experimental results indicate that, despite their size and power, general-purpose LLMs struggle to achieve satisfactory performance in these tasks. Compared to general-purpose LLMs, Xiangqi-R1 greatly advances with an 18% rise in move legality and a 22% boost in analysis accuracy. Our results point to a promising path for creating general strategic intelligence in spatially complex areas.

LLMs are increasingly deployed as orchestrators that coordinate specialized subagents to solve complex tasks through natural language. However, in many important domains like game playing and robotics, the strongest available agents are not language models. Integrating non-language agents with LLMs would require verbalization: compressing their rich continuous representations into sparse textual summaries at each interaction step. To study whether verbalization constitutes a bottleneck, we introduce LLAMIA-Bench, a suite of six diverse collaborative chess tasks spanning three facets: behavioral imitation, state assessment, and natural-language explanation. Each task instantiates a well-established chess problem that neither the LLM nor the chess engine can solve alone. To solve LLM collaboration with non-language agents, we introduce latent state internalization, which projects the subagent's continuous representations directly into the LLM's token stream as learned state tokens, with dynamic re-encoding as actions advance the environment state. Comparing internalization to verbalized integration, our experiments reveal a consistent verbalization debt: the performance gap widens throughout training and persists as the LLM scales from 4B to 14B parameters. A single 14B model, LLAMIA, trained with latent state internalization, matches or exceeds task specialists and frontier models including GPT-5.1 with tool access across all benchmark tasks, and generalizes out-of-distribution where task-specific finetunes collapse

We demonstrate LLM agent specification gaming by instructing models to win against a chess engine. We find reasoning models like o1 preview and DeepSeek-R1 will often hack the benchmark by default, while language models like GPT-4o and Claude 3.5 Sonnet need to be told that normal play won't work to hack. We improve upon prior work like (Hubinger et al., 2024; Meinke et al., 2024; Weij et al., 2024) by using realistic task prompts and avoiding excess nudging. Our results suggest reasoning models may resort to hacking to solve difficult problems, as observed in OpenAI (2024)'s o1 Docker escape during cyber capabilities testing.

In the post-AlphaGo era, there has been a renewed interest in search techniques such as Monte Carlo Tree Search (MCTS), particularly in their application to Large Language Models (LLMs). This renewed attention is driven by the recognition that current next-token prediction models often lack the ability for long-term planning. Is it possible to instill search-like abilities within the models to enhance their planning abilities without relying on explicit search? We propose DiffuSearch , a model that does implicit search by looking into the future world via discrete diffusion modeling. We instantiate DiffuSearch on a classical board game, Chess, where explicit search is known to be essential. Through extensive controlled experiments, we show DiffuSearch outperforms both the searchless and explicit search-enhanced policies. Specifically, DiffuSearch outperforms the one-step policy by 19.2% and the MCTS-enhanced policy by 14% on action accuracy. Furthermore, DiffuSearch demonstrates a notable 30% enhancement in puzzle-solving abilities compared to explicit search-based policies, along with a significant 540 Elo increase in game-playing strength assessment. These results indicate that implicit search via discrete diffusion is a viable alternative to explicit search over a one-step policy. All codes are publicly available at https://github.com/HKUNLP/DiffuSearch{https://github.com/HKUNLP/DiffuSearch}.

Reasoning is a central capability of human intelligence. In recent years, with the advent of large-scale datasets, pretrained large language models have emerged with new capabilities, including reasoning. However, these models still struggle with long-term, complex reasoning tasks, such as playing chess. Based on the observation that expert chess players employ a dual approach combining long-term strategic play with short-term tactical play along with language explanation, we propose improving the reasoning capability of large language models in chess by integrating annotated strategy and tactic. Specifically, we collect a dataset named MATE, which consists of 1 million chess positions with candidate moves annotated by chess experts for strategy and tactics. We finetune the LLaMA-3-8B model and compare it against state-of-the-art commercial language models in the task of selecting better chess moves. Our experiments show that our models perform better than GPT, Claude, and Gemini models. We find that language explanations can enhance the reasoning capability of large language models.

Latent, or silent, reasoning lets language models carry out intermediate computation in continuous vector space instead of words, and is widely assumed to function as an internal scratchpad the model consults during inference. Whether that assumption survives reinforcement learning has not been tested directly: existing causal analyses of latent reasoning are confined to math and logic tasks, comparing reliance on thoughts within one checkpoint, never before and after RL. We train a chess-playing model through a staged latent-reasoning curriculum followed by reinforcement learning, and find legality climbs monotonically to 61% (from a 48% pre-RL baseline) while checkmate confabulation is eliminated entirely. To locate this gain, we run a six-condition causal intervention suite on the same model before and after RL: substituting or noising the thought vectors leaves performance unchanged, ablating them costs only mild degradation, and only exact-zero vectors cause collapse. This robustness gap is itself the finding: under exact-zero corruption, legality collapses to 1% pre-RL versus 9% post-RL, a gap that survives correction across the full battery. A 10x-larger replication of the post-RL checkpoint's own battery confirms this: removing the thoughts, with or without restoring sequence length, also reaches significance; substitution and noise remain indistinguishable from baseline. RL appears to add robustness to disruption, not reliance on thought content. These results push back against the field's default assumption that latent thoughts function as an actively consulted inference-time scratchpad, and instead indicate latent reasoning's principal effect here is shaping the model's parameters during training. We also demonstrate a working RL gain in chess, where multiple groups report the same latent-reasoning-plus-RL recipe failing to improve accuracy over SFT.

As neural theorem provers become increasingly agentic, the ability to interpret and act on compiler feedback is critical. However, existing Lean datasets consist almost exclusively of correct proofs, offering little supervision for understanding and repairing failures. We study Lean proof repair as a supervised learning problem: given an erroneous proof and compiler feedback, predict both a corrected proof and a natural-language diagnosis grounded in the same feedback. We introduce APRIL (Automated Proof Repair in Lean), a dataset of 260,000 supervised tuples pairing systematically generated proof failures with compiler diagnostics and aligned repair and explanation targets. Training language models on APRIL substantially improves repair accuracy and feedback-conditioned reasoning; in our single-shot repair evaluation setting, a finetuned 4B-parameter model outperforms the strongest open-source baseline. We view diagnostic-conditioned supervision as a complementary training signal for feedback-using provers. Our dataset is available at https://huggingface.co/datasets/uw-math-ai/APRIL{this link}.

We introduce Kaggle Game Arena, an open and ever-expanding platform to evaluate large language models (LLMs) through competitive games. Different from static benchmarks, game arena enables models to play head-to-head matchups in structured environments where the gameplay strength naturally increases as models evolve, preventing performance saturation. This technical report details the infrastructure behind Game Arena and describes the three pilot game environments: Chess, Poker, and Werewolf. These environments span perfect information, imperfect information, and multiplayer game settings, enabling a systematic study of models' strategic planning, adaptation, and robustness under uncertainty. For each game, we provide a detailed description of the environment, evaluation metrics, and results from running full competitions across models. Through robust infrastructure and large-scale ground-truth based evaluation, Game Arena ensures reproducibility, transparency and generalizability to new games and variants over time.

Widely used alignment techniques, such as reinforcement learning from human feedback (RLHF), rely on the ability of humans to supervise model behavior - for example, to evaluate whether a model faithfully followed instructions or generated safe outputs. However, future superhuman models will behave in complex ways too difficult for humans to reliably evaluate; humans will only be able to weakly supervise superhuman models. We study an analogy to this problem: can weak model supervision elicit the full capabilities of a much stronger model? We test this using a range of pretrained language models in the GPT-4 family on natural language processing (NLP), chess, and reward modeling tasks. We find that when we naively finetune strong pretrained models on labels generated by a weak model, they consistently perform better than their weak supervisors, a phenomenon we call weak-to-strong generalization. However, we are still far from recovering the full capabilities of strong models with naive finetuning alone, suggesting that techniques like RLHF may scale poorly to superhuman models without further work. We find that simple methods can often significantly improve weak-to-strong generalization: for example, when finetuning GPT-4 with a GPT-2-level supervisor and an auxiliary confidence loss, we can recover close to GPT-3.5-level performance on NLP tasks. Our results suggest that it is feasible to make empirical progress today on a fundamental challenge of aligning superhuman models.

Reinforcement learning (RL) has become central to improving large language models (LLMs) on complex reasoning tasks, yet RL post-training is largely studied in isolation from the pretraining that precedes it. As a result, two basic questions remain open: (1) how do pretraining choices (model size, data) shape the returns to RL compute, and (2) what does RL actually do to the model? These questions are difficult to study in the standard LLM setting: pretraining corpora are vast and uncontrolled, making it hard to attribute behaviors to pretraining versus RL, and systematic compute sweeps across both stages are prohibitively expensive. To address these challenges, we use chess as a controlled testbed for studying reasoning across the full pretraining-to-post-training pipeline. We follow the standard LLM training pipeline by pretraining language models from 5M to 1B parameters on human chess games, supervised fine-tuning on synthetic reasoning traces, and running RL on chess puzzles with verifiable rewards. Using this framework, we find that the post-RL performance at given RL compute level is well-predicted from the pretraining loss, and slope of the RL reward curves improves approximately linearly with the pretraining tokens. Beyond scaling, we find that RL does not simply sharpen the SFT policy: on easy puzzles it amplifies correct moves the SFT policy already preferred, while on hard puzzles it surfaces correct moves that were nearly absent under SFT. We further test whether our findings transfer beyond chess by training a 1B language model on math-domain text, where the same predictive pattern emerges: longer-pretrained checkpoints reach higher post-RL performance and improve faster under RL. In sum, we provide a quantitative account of the pretraining-to-RL interface and a controlled testbed for studying the science of reasoning across the full pretraining-to-post-training pipeline.

Neurons in large language models often exhibit polysemanticity, simultaneously encoding multiple unrelated concepts and obscuring interpretability. Instead of relying on post-hoc methods, we present MoE-X, a Mixture-of-Experts (MoE) language model designed to be intrinsically interpretable. Our approach is motivated by the observation that, in language models, wider networks with sparse activations are more likely to capture interpretable factors. However, directly training such large sparse networks is computationally prohibitive. MoE architectures offer a scalable alternative by activating only a subset of experts for any given input, inherently aligning with interpretability objectives. In MoE-X, we establish this connection by rewriting the MoE layer as an equivalent sparse, large MLP. This approach enables efficient scaling of the hidden size while maintaining sparsity. To further enhance interpretability, we enforce sparse activation within each expert and redesign the routing mechanism to prioritize experts with the highest activation sparsity. These designs ensure that only the most salient features are routed and processed by the experts. We evaluate MoE-X on chess and natural language tasks, showing that it achieves performance comparable to dense models while significantly improving interpretability. MoE-X achieves a perplexity better than GPT-2, with interpretability surpassing even sparse autoencoder (SAE)-based approaches.

In this paper, I apply linguistic methods of analysis to non-linguistic data, chess plays, metaphorically equating one with the other and seeking analogies. Chess game notations are also a kind of text, and one can consider the records of moves or positions of pieces as words and statements in a certain language. In this article I show how word embeddings (word2vec) can work on chess game texts instead of natural language texts. I don't see how this representation of chess data can be used productively. It's unlikely that these vector models will help engines or people choose the best move. But in a purely academic sense, it's clear that such methods of information representation capture something important about the very nature of the game, which doesn't necessarily lead to a win.

Generative sequence models are typically trained on sample sequences from natural or formal languages. It is a crucial question whether -- or to what extent -- sample-based training is able to capture the true structure of these languages, often referred to as the ``world model''. Theoretical results indicate that we can hope for soundness at best, that is, generating valid sequences, but not necessarily all of them. However, it is still important to have practical tools that are able to verify whether a given sequence model is sound. In this study, we focus on chess, as it is a domain that provides enough complexity while having a simple rule-based world model. We propose adversarial sequence generation for verifying the soundness of the sequence model. Our adversaries generate valid sequences so as to force the sequence model to generate an invalid next move prediction. Apart from the falsification of soundness, this method is also suitable for a more fine-grained analysis of the failure modes and the effects of different choices during training. To demonstrate this, we propose a number of methods for adversarial sequence generation and evaluate the approach on a large set of chess models. We train models on random as well as high-quality chess games, using several training recipes. We find that none of the models are sound, but some training techniques and dataset choices are able to improve soundness remarkably. We also investigate the potential application of board state probes in both our training and attack methods. Our findings indicate that the extracted board states have no causal role in next token prediction in most of the models.

Chess has long served as a canonical testbed for artificial intelligence, but modeling approaches for its central tasks have diverged. Maximizing playing strength, predicting human play, and enabling interpretability are typically solved with disparate architectures, and these designs are often misaligned with the geometry of the domain. This raises the natural question of whether these objectives require separate modeling paradigms, or if there exists a single architecture that supports them simultaneously. We introduce Chessformer, a unified architecture that advances the state of the art on all three central goals in chess modeling. Chessformer is an encoder-only transformer that represents board squares as tokens, augments self-attention with a novel dynamic positional encoding called Geometric Attention Bias (GAB) that adapts to domain-specific geometry, and predicts actions with an attention-based source-destination policy head. We evaluate Chessformer on each front. First, we develop \maiathree, a family of models for human move prediction that reaches 57.1\% move-matching accuracy, significantly surpassing the previous state of the art with fewer than a quarter of the parameters. Second, we integrate Chessformer into Leela Chess Zero, a leading open-source engine, adding over 100 Elo of playing strength and resulting in tournament victories over Stockfish in major computer chess competitions. Third, we show that Chessformer's square-token design makes attention patterns and activations directly attributable to board squares, enabling granular interpretability analyses that prior architectures do not naturally support. More broadly, our results demonstrate that aligning a model's tokenization, positional encoding, and output design with the underlying structure of a domain can yield simultaneous gains in performance, human compatibility, and interpretability.

There are an increasing number of domains in which artificial intelligence (AI) systems both surpass human ability and accurately model human behavior. This introduces the possibility of algorithmically-informed teaching in these domains through more relatable AI partners and deeper insights into human decision-making. Critical to achieving this goal, however, is coherently modeling human behavior at various skill levels. Chess is an ideal model system for conducting research into this kind of human-AI alignment, with its rich history as a pivotal testbed for AI research, mature superhuman AI systems like AlphaZero, and precise measurements of skill via chess rating systems. Previous work in modeling human decision-making in chess uses completely independent models to capture human style at different skill levels, meaning they lack coherence in their ability to adapt to the full spectrum of human improvement and are ultimately limited in their effectiveness as AI partners and teaching tools. In this work, we propose a unified modeling approach for human-AI alignment in chess that coherently captures human style across different skill levels and directly captures how people improve. Recognizing the complex, non-linear nature of human learning, we introduce a skill-aware attention mechanism to dynamically integrate players' strengths with encoded chess positions, enabling our model to be sensitive to evolving player skill. Our experimental results demonstrate that this unified framework significantly enhances the alignment between AI and human players across a diverse range of expertise levels, paving the way for deeper insights into human decision-making and AI-guided teaching tools.

Large Language Models (LLMs) reasoning abilities are increasingly being applied to classical board and card games, but the dominant approach -- involving prompting for direct move generation -- has significant drawbacks. It relies on the model's implicit fragile pattern-matching capabilities, leading to frequent illegal moves and strategically shallow play. Here we introduce an alternative approach: We use the LLM to translate natural language rules and game trajectories into a formal, executable world model represented as Python code. This generated model -- comprising functions for state transition, legal move enumeration, and termination checks -- serves as a verifiable simulation engine for high-performance planning algorithms like Monte Carlo tree search (MCTS). In addition, we prompt the LLM to generate heuristic value functions (to make MCTS more efficient), and inference functions (to estimate hidden states in imperfect information games). Our method offers three distinct advantages compared to directly using the LLM as a policy: (1) Verifiability: The generated CWM serves as a formal specification of the game's rules, allowing planners to algorithmically enumerate valid actions and avoid illegal moves, contingent on the correctness of the synthesized model; (2) Strategic Depth: We combine LLM semantic understanding with the deep search power of classical planners; and (3) Generalization: We direct the LLM to focus on the meta-task of data-to-code translation, enabling it to adapt to new games more easily. We evaluate our agent on 10 different games, of which 4 are novel and created for this paper. 5 of the games are fully observed (perfect information), and 5 are partially observed (imperfect information). We find that our method outperforms or matches Gemini 2.5 Pro in 9 out of the 10 considered games.

Li et al. (2023) used the Othello board game as a test case for the ability of GPT-2 to induce world models, and were followed up by Nanda et al. (2023b). We briefly discuss the original experiments, expanding them to include more language models with more comprehensive probing. Specifically, we analyze sequences of Othello board states and train the model to predict the next move based on previous moves. We evaluate seven language models (GPT-2, T5, Bart, Flan-T5, Mistral, LLaMA-2, and Qwen2.5) on the Othello task and conclude that these models not only learn to play Othello, but also induce the Othello board layout. We find that all models achieve up to 99% accuracy in unsupervised grounding and exhibit high similarity in the board features they learned. This provides considerably stronger evidence for the Othello World Model Hypothesis than previous works.

The recent breakthrough successes in machine learning are mainly attributed to scale: namely large-scale attention-based architectures and datasets of unprecedented scale. This paper investigates the impact of training at scale for chess. Unlike traditional chess engines that rely on complex heuristics, explicit search, or a combination of both, we train a 270M parameter transformer model with supervised learning on a dataset of 10 million chess games. We annotate each board in the dataset with action-values provided by the powerful Stockfish 16 engine, leading to roughly 15 billion data points. Our largest model reaches a Lichess blitz Elo of 2895 against humans, and successfully solves a series of challenging chess puzzles, without any domain-specific tweaks or explicit search algorithms. We also show that our model outperforms AlphaZero's policy and value networks (without MCTS) and GPT-3.5-turbo-instruct. A systematic investigation of model and dataset size shows that strong chess performance only arises at sufficient scale. To validate our results, we perform an extensive series of ablations of design choices and hyperparameters.

Language models show a surprising range of capabilities, but the source of their apparent competence is unclear. Do these networks just memorize a collection of surface statistics, or do they rely on internal representations of the process that generates the sequences they see? We investigate this question by applying a variant of the GPT model to the task of predicting legal moves in a simple board game, Othello. Although the network has no a priori knowledge of the game or its rules, we uncover evidence of an emergent nonlinear internal representation of the board state. Interventional experiments indicate this representation can be used to control the output of the network and create "latent saliency maps" that can help explain predictions in human terms.

Recent radiology multi-modal language models have made substantial progress in chest X-ray report generation, visual question answering, and temporal reasoning. While longitudinal chest X-ray interpretation compares sequential examinations to describe change, visual grounding aims to connect clinical language with localized image evidence. Although longitudinal modeling and visual grounding have each advanced radiology language models, how localized visual evidence can support longitudinal interpretation remains under-explored. We introduce CheXGround, a region-grounded longitudinal chest X-ray language model that represents paired studies through corresponding anatomical regions. CheXGround extracts anatomical regions from current and prior radiographs, encodes them as temporally enhanced Region-of-Interest (ROI) tokens, and combines them with global temporal image context during generation. To connect these region tokens with clinical text, we propose Temporal Region--Phrase Alignment, a pretraining objective that aligns temporal anatomical representations with localized report phrases. We evaluate CheXGround on single-study and longitudinal Visual Question Answering (VQA), longitudinal findings generation, temporal grounded VQA, and anatomical grounding. Across these tasks, CheXGround improves clinical language quality, temporal reasoning, and localization accuracy over recent baselines. Our results suggest that organizing longitudinal evidence at the anatomical level is a strong representation for grounded radiology language modeling. Project page: https://adonaydem.github.io/chexground-website

When applied to question answering and other text generation tasks, language models (LMs) may be queried generatively (by sampling answers from their output distribution) or discriminatively (by using them to score or rank a set of candidate outputs). These procedures sometimes yield very different predictions. How do we reconcile mutually incompatible scoring procedures to obtain coherent LM predictions? We introduce a new, a training-free, game-theoretic procedure for language model decoding. Our approach casts language model decoding as a regularized imperfect-information sequential signaling game - which we term the CONSENSUS GAME - in which a GENERATOR seeks to communicate an abstract correctness parameter using natural language sentences to a DISCRIMINATOR. We develop computational procedures for finding approximate equilibria of this game, resulting in a decoding algorithm we call EQUILIBRIUM-RANKING. Applied to a large number of tasks (including reading comprehension, commonsense reasoning, mathematical problem-solving, and dialog), EQUILIBRIUM-RANKING consistently, and sometimes substantially, improves performance over existing LM decoding procedures - on multiple benchmarks, we observe that applying EQUILIBRIUM-RANKING to LLaMA-7B outperforms the much larger LLaMA-65B and PaLM-540B models. These results highlight the promise of game-theoretic tools for addressing fundamental challenges of truthfulness and consistency in LMs.

As artificial intelligence becomes increasingly intelligent---in some cases, achieving superhuman performance---there is growing potential for humans to learn from and collaborate with algorithms. However, the ways in which AI systems approach problems are often different from the ways people do, and thus may be uninterpretable and hard to learn from. A crucial step in bridging this gap between human and artificial intelligence is modeling the granular actions that constitute human behavior, rather than simply matching aggregate human performance. We pursue this goal in a model system with a long history in artificial intelligence: chess. The aggregate performance of a chess player unfolds as they make decisions over the course of a game. The hundreds of millions of games played online by players at every skill level form a rich source of data in which these decisions, and their exact context, are recorded in minute detail. Applying existing chess engines to this data, including an open-source implementation of AlphaZero, we find that they do not predict human moves well. We develop and introduce Maia, a customized version of Alpha-Zero trained on human chess games, that predicts human moves at a much higher accuracy than existing engines, and can achieve maximum accuracy when predicting decisions made by players at a specific skill level in a tuneable way. For a dual task of predicting whether a human will make a large mistake on the next move, we develop a deep neural network that significantly outperforms competitive baselines. Taken together, our results suggest that there is substantial promise in designing artificial intelligence systems with human collaboration in mind by first accurately modeling granular human decision-making.

We present the first evaluation harness that enables any out-of-the-box, local, Large Language Models (LLMs) to play full-press Diplomacy without fine-tuning or specialized training. Previous work required frontier LLMs, or fine-tuning, due to the high complexity and information density of Diplomacy's game state. Combined with the high variance of matches, these factors made Diplomacy prohibitive for study. In this work, we used data-driven iteration to optimize a textual game state representation such that a 24B model can reliably complete matches without any fine tuning. We develop tooling to facilitate hypothesis testing and statistical analysis, and we present case studies on persuasion, aggressive playstyles, and performance across a range of models. We conduct a variety of experiments across many popular LLMs, finding the larger models perform the best, but the smaller models still play adequately. We also introduce Critical State Analysis: an experimental protocol for rapidly iterating and analyzing key moments in a game at depth. Our harness democratizes the evaluation of strategic reasoning in LLMs by eliminating the need for fine-tuning, and it provides insights into how these capabilities emerge naturally from widely used LLMs. Our code is available in the supplement and will be open sourced.

Playing games has a long history of describing intricate interactions in simplified forms. In this paper we explore if large language models (LLMs) can play games, investigating their capabilities for randomisation and strategic adaptation through both simultaneous and sequential game interactions. We focus on GPT-4o-Mini-2024-08-17 and test two games between LLMs: Rock Paper Scissors (RPS) and games of strategy (Prisoners Dilemma PD). LLMs are often described as stochastic parrots, and while they may indeed be parrots, our results suggest that they are not very stochastic in the sense that their outputs - when prompted to be random - are often very biased. Our research reveals that LLMs appear to develop loss aversion strategies in repeated games, with RPS converging to stalemate conditions while PD shows systematic shifts between cooperative and competitive outcomes based on prompt design. We detail programmatic tools for independent agent interactions and the Agentic AI challenges faced in implementation. We show that LLMs can indeed play games, just not very well. These results have implications for the use of LLMs in multi-agent LLM systems and showcase limitations in current approaches to model output for strategic decision-making.

In recent months, Language Models (LMs) have become a part of daily discourse, with focus on OpenAI and the potential of Artificial General Intelligence (AGI). Furthermore, the leaking of LLama's weights to the public has led to an influx of innovations demonstrating the impressive capabilities of generative LMs. While we believe that AGI is still a distant goal, we recognize the potential of LMs in solving tasks such as searching complex documents, compiling reports with basic analysis, and providing assistance in problem-solving. In this paper, we propose formalizing the execution model of language models. We investigate current execution models, to find that this formalism has received little attention, and present our contribution: the first formalized execution model for LMs. We introduce a new algorithm for sampling the predictions of LMs, which we use to build a reliable and inspectable execution model. We introduce a low-level language to write "cognitive program" for this execution model. We hope to shed light on the need for execution models for LMs and encourage further research in this area.

In this paper, we introduce a combination of novel and exciting tasks: the solution and generation of linguistic puzzles. We focus on puzzles used in Linguistic Olympiads for high school students. We first extend the existing benchmark for the task of solving linguistic puzzles. We explore the use of Large Language Models (LLMs), including recent state-of-the-art models such as OpenAI's o1, for solving linguistic puzzles, analyzing their performance across various linguistic topics. We demonstrate that LLMs outperform humans on most puzzles types, except for those centered on writing systems, and for the understudied languages. We use the insights from puzzle-solving experiments to direct the novel task of puzzle generation. We believe that automating puzzle generation, even for relatively simple puzzles, holds promise for expanding interest in linguistics and introducing the field to a broader audience. This finding highlights the importance of linguistic puzzle generation as a research task: such puzzles can not only promote linguistics but also support the dissemination of knowledge about rare and understudied languages.

Large language models (LLMs) have achieved striking successes on many benchmarks, yet recent studies continue to expose fundamental weaknesses. In particular, tasks that require abstract reasoning remain challenging, often because they use representations such as grids, symbols, or visual patterns that differ from the natural language data LLMs are trained on. In this paper, we introduce Concept, a simple word-guessing board game, as a benchmark for probing abductive reasoning in a representation that is much closer to LLM pre-training data: natural language. Our results show that this game, easily solved by humans (with a success rate of over 90\%), is still very challenging for state-of-the-art LLMs (no model exceeds 40\% success rate). Specifically, we observe that LLMs struggle with interpreting other players' strategic intents, and with correcting initial hypotheses given sequential information updates. In addition, we extend the evaluation across multiple languages, and find that the LLM performance drops further in lower-resource languages (Dutch, French, and Spanish) compared to English.

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.

As large language models (LLMs) advance across diverse tasks, the need for comprehensive evaluation beyond single metrics becomes increasingly important. To fully assess LLM intelligence, it is crucial to examine their interactive dynamics and strategic behaviors. We present LLMsPark, a game theory-based evaluation platform that measures LLMs' decision-making strategies and social behaviors in classic game-theoretic settings, providing a multi-agent environment to explore strategic depth. Our system cross-evaluates 15 leading LLMs (both commercial and open-source) using leaderboard rankings and scoring mechanisms. Higher scores reflect stronger reasoning and strategic capabilities, revealing distinct behavioral patterns and performance differences across models. This work introduces a novel perspective for evaluating LLMs' strategic intelligence, enriching existing benchmarks and broadening their assessment in interactive, game-theoretic scenarios. The benchmark and rankings are publicly available at https://llmsparks.github.io/.

As Large Language Models (LLMs) are integrated into critical real-world applications, their strategic and logical reasoning abilities are increasingly crucial. This paper evaluates LLMs' reasoning abilities in competitive environments through game-theoretic tasks, e.g., board and card games that require pure logic and strategic reasoning to compete with opponents. We first propose GTBench, a language-driven environment composing 10 widely-recognized tasks, across a comprehensive game taxonomy: complete versus incomplete information, dynamic versus static, and probabilistic versus deterministic scenarios. Then, we investigate two key problems: (1) Characterizing game-theoretic reasoning of LLMs; (2) LLM-vs-LLM competitions as reasoning evaluation. We observe that (1) LLMs have distinct behaviors regarding various gaming scenarios; for example, LLMs fail in complete and deterministic games yet they are competitive in probabilistic gaming scenarios; (2) Open-source LLMs, e.g., CodeLlama-34b-Instruct, are less competitive than commercial LLMs, e.g., GPT-4, in complex games. In addition, code-pretraining greatly benefits strategic reasoning, while advanced reasoning methods such as Chain-of-Thought (CoT) and Tree-of-Thought (ToT) do not always help. Detailed error profiles are also provided for a better understanding of LLMs' behavior.

The Instruction-Driven Game Engine (IDGE) project aims to democratize game development by enabling a large language model (LLM) to follow free-form game descriptions and generate game-play processes. The IDGE allows users to create games simply by natural language instructions, which significantly lowers the barrier for game development. We approach the learning process for IDGEs as a Next State Prediction task, wherein the model autoregressively predicts the game states given player actions. The computation of game states must be precise; otherwise, slight errors could corrupt the game-play experience. This is challenging because of the gap between stability and diversity. To address this, we train the IDGE in a curriculum manner that progressively increases its exposure to complex scenarios. Our initial progress lies in developing an IDGE for Poker, which not only supports a wide range of poker variants but also allows for highly individualized new poker games through natural language inputs. This work lays the groundwork for future advancements in transforming how games are created and played.

Current rating systems update ratings incrementally and may not always accurately reflect a player's true strength at all times, especially for rapidly improving players or very rusty players. To overcome this, we explore a method to estimate player ratings directly from game moves and clock times. We compiled a benchmark dataset from Lichess, encompassing various time controls and including move sequences and clock times. Our model architecture comprises a CNN to learn positional features, which are then integrated with clock-time data into a bidirectional LSTM, predicting player ratings after each move. The model achieved an MAE of 182 rating points in the test data. Additionally, we applied our model to the 2024 IEEE Big Data Cup Chess Puzzle Difficulty Competition dataset, predicted puzzle ratings and achieved competitive results. This model is the first to use no hand-crafted features to estimate chess ratings and also the first to output a rating prediction for each move. Our method highlights the potential of using move-based rating estimation for enhancing rating systems and potentially other applications such as cheating detection.

Recent advancements in large language models (LLMs) have led to remarkable performance across a wide range of language understanding and mathematical tasks. As a result, increasing attention has been given to assessing the true reasoning capabilities of LLMs, driving research into commonsense, numerical, logical, and qualitative reasoning. However, with the rapid progress of reasoning-focused models such as OpenAI's o1 and DeepSeek's R1, there has been a growing demand for reasoning benchmarks that can keep pace with ongoing model developments. In this paper, we introduce MastermindEval, a simple, scalable, and interpretable deductive reasoning benchmark inspired by the board game Mastermind. Our benchmark supports two evaluation paradigms: (1) agentic evaluation, in which the model autonomously plays the game, and (2) deductive reasoning evaluation, in which the model is given a pre-played game state with only one possible valid code to infer. In our experimental results we (1) find that even easy Mastermind instances are difficult for current models and (2) demonstrate that the benchmark is scalable to possibly more advanced models in the future Furthermore, we investigate possible reasons why models cannot deduce the final solution and find that current models are limited in deducing the concealed code as the number of statement to combine information from is increasing.

We introduce a framework for translating game descriptions in natural language into extensive-form representations in game theory, leveraging Large Language Models (LLMs) and in-context learning. Given the varying levels of strategic complexity in games, such as perfect versus imperfect information, directly applying in-context learning would be insufficient. To address this, we introduce a two-stage framework with specialized modules to enhance in-context learning, enabling it to divide and conquer the problem effectively. In the first stage, we tackle the challenge of imperfect information by developing a module that identifies information sets along and the corresponding partial tree structure. With this information, the second stage leverages in-context learning alongside a self-debugging module to produce a complete extensive-form game tree represented using pygambit, the Python API of a recognized game-theoretic analysis tool called Gambit. Using this python representation enables the automation of tasks such as computing Nash equilibria directly from natural language descriptions. We evaluate the performance of the full framework, as well as its individual components, using various LLMs on games with different levels of strategic complexity. Our experimental results show that the framework significantly outperforms baseline models in generating accurate extensive-form games, with each module playing a critical role in its success.

Communication games, which we refer to as incomplete information games that heavily depend on natural language communication, hold significant research value in fields such as economics, social science, and artificial intelligence. In this work, we explore the problem of how to engage large language models (LLMs) in communication games, and in response, propose a tuning-free framework. Our approach keeps LLMs frozen, and relies on the retrieval and reflection on past communications and experiences for improvement. An empirical study on the representative and widely-studied communication game, ``Werewolf'', demonstrates that our framework can effectively play Werewolf game without tuning the parameters of the LLMs. More importantly, strategic behaviors begin to emerge in our experiments, suggesting that it will be a fruitful journey to engage LLMs in communication games and associated domains.

Mastering games is a hard task, as games can be extremely complex, and still fundamentally different in structure from one another. While the AlphaZero algorithm has demonstrated an impressive ability to learn the rules and strategy of a large variety of games, ranging from Go and Chess, to Atari games, its reliance on extensive computational resources and rigid Convolutional Neural Network (CNN) architecture limits its adaptability and scalability. A model trained to play on a 19times 19 Go board cannot be used to play on a smaller 13times 13 board, despite the similarity between the two Go variants. In this paper, we focus on Chess, and explore using a more generic Graph-based Representation of a game state, rather than a grid-based one, to introduce a more general architecture based on Graph Neural Networks (GNN). We also expand the classical Graph Attention Network (GAT) layer to incorporate edge-features, to naturally provide a generic policy output format. Our experiments, performed on smaller networks than the initial AlphaZero paper, show that this new architecture outperforms previous architectures with a similar number of parameters, being able to increase playing strength an order of magnitude faster. We also show that the model, when trained on a smaller 5times 5 variant of chess, is able to be quickly fine-tuned to play on regular 8times 8 chess, suggesting that this approach yields promising generalization abilities. Our code is available at https://github.com/akulen/AlphaGateau.

We present gg-bench, a collection of game environments designed to evaluate general reasoning capabilities in language models. Unlike most static benchmarks, gg-bench is a data generating process where new evaluation instances can be generated at will. In particular, gg-bench is synthetically generated by (1) using a large language model (LLM) to generate natural language descriptions of novel games, (2) using the LLM to implement each game in code as a Gym environment, and (3) training reinforcement learning (RL) agents via self-play on the generated games. We evaluate language models by their winrate against these RL agents by prompting models with the game description, current board state, and a list of valid moves, after which models output the moves they wish to take. gg-bench is challenging: state-of-the-art LLMs such as GPT-4o and Claude 3.7 Sonnet achieve winrates of 7-9% on gg-bench using in-context learning, while reasoning models such as o1, o3-mini and DeepSeek-R1 achieve average winrates of 31-36%. We release the generated games, data generation process, and evaluation code in order to support future modeling work and expansion of our benchmark.

Large Language Models (LLMs) have enabled Multi-Agent Systems (MASs) where agents interact through natural language to solve complex tasks or simulate multi-party dialogues. Recent work on LLM-based MASs has mainly focused on architecture design, such as role assignment and workflow orchestration. In contrast, this paper targets the interaction process itself, aiming to improve agents' communication efficiency by helping them convey their intended meaning more effectively through language. To this end, we propose LinguaGame, a linguistically-grounded game-theoretic paradigm for multi-agent dialogue generation. Our approach models dialogue as a signalling game over communicative intents and strategies, solved with a training-free equilibrium approximation algorithm for inference-time decision adjustment. Unlike prior game-theoretic MASs, whose game designs are often tightly coupled with task-specific objectives, our framework relies on linguistically informed reasoning with minimal task-specific coupling. Specifically, it treats dialogue as intentional and strategic communication, requiring agents to infer what others aim to achieve (intents) and how they pursue those goals (strategies). We evaluate our framework in simulated courtroom proceedings and debates, with human expert assessments showing significant gains in communication efficiency.

The standard methodology of evaluating large language models (LLMs) based on static pairs of inputs and outputs is insufficient for developing assistants: this kind of assessments fails to take into account the essential interactive element in their deployment, and therefore limits how we understand language model capabilities. We introduce CheckMate, an adaptable prototype platform for humans to interact with and evaluate LLMs. We conduct a study with CheckMate to evaluate three language models~(InstructGPT, ChatGPT, and GPT-4) as assistants in proving undergraduate-level mathematics, with a mixed cohort of participants from undergraduate students to professors of mathematics. We release the resulting interaction and rating dataset, MathConverse. By analysing MathConverse, we derive a preliminary taxonomy of human behaviours and uncover that despite a generally positive correlation, there are notable instances of divergence between correctness and perceived helpfulness in LLM generations, amongst other findings. Further, we identify useful scenarios and existing issues of GPT-4 in mathematical reasoning through a series of case studies contributed by expert mathematicians. We conclude with actionable takeaways for ML practitioners and mathematicians: models which communicate uncertainty, respond well to user corrections, are more interpretable and concise may constitute better assistants; interactive evaluation is a promising way to continually navigate the capability of these models; humans should be aware of language models' algebraic fallibility, and for that reason discern where they should be used.

Medical vision-language models report strong chest radiograph accuracy, and this is increasingly read as evidence that they use the image. That inference is unsafe: a model exploiting finding-name priors scores like one that reads the scan, and no standard benchmark separates them. We introduce a causal audit that intervenes on the image, occluding the relevant region, occluding an irrelevant one, and swapping in another patient's same-label scan, and combines three behavioral metrics to test whether a correct answer depends on the image. Across nine systems, a text-only model with no image access reaches within 5.7 accuracy points of the best multimodal one, and a 119-billion-parameter multimodal model is statistically indistinguishable from a 7-billion text-only baseline. The audit splits the cohort into three models that ignore the image, one that is unstable, and five that use it selectively, for a subset of findings; the categories hold across a second dataset, resolution, and prompt phrasing. Against board-certified radiologists, a text-only model is statistically indistinguishable from a radiologist's accuracy while grounding at zero, whereas the image-using models ground at radiologist-comparable rates. Reported confidence flags ungrounded answers only when a model uses the image. Grounding audits, not accuracy, should gate clinical deployment.

Radiology plays a pivotal role in modern medicine due to its non-invasive diagnostic capabilities. However, the manual generation of unstructured medical reports is time consuming and prone to errors. It creates a significant bottleneck in clinical workflows. Despite advancements in AI-generated radiology reports, challenges remain in achieving detailed and accurate report generation. In this study we have evaluated different combinations of multimodal models that integrate Computer Vision and Natural Language Processing to generate comprehensive radiology reports. We employed a pretrained Vision Transformer (ViT-B16) and a SWIN Transformer as the image encoders. The BART and GPT-2 models serve as the textual decoders. We used Chest X-ray images and reports from the IU-Xray dataset to evaluate the usability of the SWIN Transformer-BART, SWIN Transformer-GPT-2, ViT-B16-BART and ViT-B16-GPT-2 models for report generation. We aimed at finding the best combination among the models. The SWIN-BART model performs as the best-performing model among the four models achieving remarkable results in almost all the evaluation metrics like ROUGE, BLEU and BERTScore.

Vision-language models (VLMs) pretrained on large-scale image-text pairs demonstrate strong image-level understanding, but are primarily optimized for global alignment and do not explicitly encode fine-grained anatomical structure, limiting their suitability for spatially precise tasks such as segmentation. We introduce CheXanatomy, a framework that integrates explicit anatomical knowledge into a pretrained VLM through autoregressive token-space supervision. Instead of adding task-specific decoder heads, the model is trained to generate anatomical segmentation masks via next-token prediction. To enable scalable supervision, we synthesize realistic chest radiographs from CT volumes and forward-project CT segmentation labels to obtain anatomically consistent 2D masks. We evaluate the approach on synthetic and real chest radiographs against a U-Net baseline, including ablations on model scale, input resolution, and vision encoder fine-tuning. Autoregressive anatomical supervision achieves performance comparable to specialized convolutional models in-distribution and demonstrates improved geometric robustness under domain shift to real CXR data. In addition, anatomy-pretrained models exhibit improved sample efficiency when adapting to novel localization tasks under limited supervision. Larger models and higher input image resolution improve performance, while vision encoder fine-tuning has limited effect. These results show that embedding anatomical structure directly into the generative objective promotes spatially grounded representations and supports anatomy-aware medical vision-language modeling.

Large Vision Language Models (LVLMs) show promise in medical applications, but their inability to faithfully ground responses in visual evidence raises serious concerns about clinical trustworthiness. While visual attribution methods are widely used to explain LVLM predictions, whether these explanations actually reflect the visual evidence underlying the model's decision is largely unverified, since ground-truth annotations for internal model reasoning are typically unavailable. We address this question for chest X-ray (CXR) reasoning by developing a causal evaluation framework that retains only CXR-VQA samples for which the expert-annotated region is verified, via counterfactual editing, to be causally responsible for the model's prediction. Using this framework across 11 attribution methods, six open-source LVLMs, and two output modes (direct answer and step-by-step reasoning), we find that existing attribution methods often fail to identify the evidence used by LVLMs. To address this failure, we propose MedFocus, a concept-based attribution method that localizes clinically meaningful anatomical regions via unbalanced optimal transport and measures their causal effect on model outputs through targeted interventions. MedFocus produces spatial, concept-level, and token-level attributions and substantially outperforms prior methods, taking a step toward more trustworthy attribution for medical LVLMs. Our data and code are available at https://github.com/gzxiong/medfocus/.

The latest breakthroughs in large vision-language models, such as Bard and GPT-4, have showcased extraordinary abilities in performing a wide range of tasks. Such models are trained on massive datasets comprising billions of public image-text pairs with diverse tasks. However, their performance on task-specific domains, such as radiology, is still under-investigated and potentially limited due to a lack of sophistication in understanding biomedical images. On the other hand, conversational medical models have exhibited remarkable success but have mainly focused on text-based analysis. In this paper, we introduce XrayGPT, a novel conversational medical vision-language model that can analyze and answer open-ended questions about chest radiographs. Specifically, we align both medical visual encoder (MedClip) with a fine-tuned large language model (Vicuna), using a simple linear transformation. This alignment enables our model to possess exceptional visual conversation abilities, grounded in a deep understanding of radiographs and medical domain knowledge. To enhance the performance of LLMs in the medical context, we generate ~217k interactive and high-quality summaries from free-text radiology reports. These summaries serve to enhance the performance of LLMs through the fine-tuning process. Our approach opens up new avenues the research for advancing the automated analysis of chest radiographs. Our open-source demos, models, and instruction sets are available at: https://github.com/mbzuai-oryx/XrayGPT.

Medical vision-language model (VLM) evaluation is sensitive to workflow design, prompting strategy, and benchmark construction, yet most studies treat these factors in isolation. We introduce a reliability stress test for chest X-ray interpretation built on two balanced datasets (a private report-backed set and a curated MIMIC subset). Three medical VLMs (CheXagent, MedGemma-4B, and MedGemma-27B) are evaluated across three prompt styles and two workflows (single-VLM and multi-agent), producing 36 configurations. We show that exact-match accuracy alone can overstate the effectiveness of conservative models that default to "Normal" predictions. Diagnostic reliability also depends heavily on model family and scale: multi-agent reasoning helps some configurations but hurts others. Building on these observations, we propose a decision-time routing framework that selectively escalates to multi-agent inference only when beneficial, improving the cost-quality trade-off over fixed workflows. Our results highlight the need for evaluation protocols that jointly consider prompt sensitivity, failure-mode diversity, and workflow choice before clinical deployment.

This paper presents a new approach that integrates deep learning with computational chess, using both the Mixture of Experts (MoE) method and Monte-Carlo Tree Search (MCTS). Our methodology employs a suite of specialized models, each designed to respond to specific changes in the game's input data. This results in a framework with sparsely activated models, which provides significant computational benefits. Our framework combines the MoE method with MCTS, in order to align it with the strategic phases of chess, thus departing from the conventional ``one-for-all'' model. Instead, we utilize distinct game phase definitions to effectively distribute computational tasks across multiple expert neural networks. Our empirical research shows a substantial improvement in playing strength, surpassing the traditional single-model framework. This validates the efficacy of our integrated approach and highlights the potential of incorporating expert knowledge and strategic principles into neural network design. The fusion of MoE and MCTS offers a promising avenue for advancing machine learning architectures.

Purpose : To develop and evaluate CAPTOR, a deep-learning vision-language foundation model for oncologic chest-abdomen-pelvis (CAP) CT interpretation. Materials & Methods : CAPTOR (Chest-Abdomen-Pelvis foundaTion model for Oncologic Radiology) is a contrastive vision-language model, trained using a pretraining strategy for oncologic CAP CT (in single- or multiple-series acquisitions), on a multicenter retrospective dataset from North America, Europe and Asia, predominantly from tumor lesion examinations. CAPTOR was compared with state-of-the-art CT foundation models on three downstream tasks: disease detection, survival prediction using TumorImagingBench and image retrieval. Classification and retrieval were both evaluated on held-out internal and public test sets, stratified by anatomical coverage and indication. Bootstrap resampling was used to estimate confidence intervals, and pairwise comparisons were performed using permutation tests. Results : CAPTOR was trained and tested on 25,512 and 10,794 CT-report pairs respectively. TumorImagingBench datasets comprised between 134 and 421 studies each. CAPTOR outperformed all baseline models, on the classification task, on the CAP subset (AUC 0.767, 95%CI: 0.734-0.793, vs. 0.695, 95%CI: 0.645-0.736 for the next-best, p<0.001) and oncologic subset (AUC 0.784, 95%CI: 0.759-0.805, vs. 0.734, 95%CI: 0.703-0.767 p<0.001), ranked first on three of the four oncologic prognostic benchmarks, and achieved superior retrieval performance on CAP and oncologic subsets. Conclusion : A foundation model trained on oncology-enriched CAP CT data outperformed relevant state-of-the-art models in its target clinical context, underscoring the importance of clinically aligned pretraining. We release our preprocessing, training and inference code : https://github.com/Guerbet-AI/CAPTOR and model weights : https://huggingface.co/Guerbet-AI/CAPTOR

Vietnamese medical research has become an increasingly vital domain, particularly with the rise of intelligent technologies aimed at reducing time and resource burdens in clinical diagnosis. Recent advances in vision-language models (VLMs), such as Gemini and GPT-4V, have sparked a growing interest in applying AI to healthcare. However, most existing VLMs lack exposure to Vietnamese medical data, limiting their ability to generate accurate and contextually appropriate diagnostic outputs for Vietnamese patients. To address this challenge, we introduce ViX-Ray, a novel dataset comprising 5,400 Vietnamese chest X-ray images annotated with expert-written findings and impressions from physicians at a major Vietnamese hospital. We analyze linguistic patterns within the dataset, including the frequency of mentioned body parts and diagnoses, to identify domain-specific linguistic characteristics of Vietnamese radiology reports. Furthermore, we fine-tune five state-of-the-art open-source VLMs on ViX-Ray and compare their performance to leading proprietary models, GPT-4V and Gemini. Our results show that while several models generate outputs partially aligned with clinical ground truths, they often suffer from low precision and excessive hallucination, especially in impression generation. These findings not only demonstrate the complexity and challenge of our dataset but also establish ViX-Ray as a valuable benchmark for evaluating and advancing vision-language models in the Vietnamese clinical domain.

Chest X-rays (CXRs) are among the most frequently performed imaging examinations worldwide, yet rising imaging volumes increase radiologist workload and the risk of diagnostic errors. Although artificial intelligence (AI) systems have shown promise for CXR interpretation, most generate only final predictions, without making explicit how visual evidence is translated into radiographic findings and diagnostic predictions. We present CheXOne, a reasoning-enabled vision-language model for CXR interpretation. CheXOne jointly generates diagnostic predictions and explicit, clinically grounded reasoning traces that connect visual evidence, radiographic findings, and these predictions. The model is trained on 14.7 million instruction and reasoning samples curated from 30 public datasets spanning 36 CXR interpretation tasks, using a two-stage framework that combines instruction tuning with reinforcement learning to improve reasoning quality. We evaluate CheXOne in zero-shot settings across visual question answering, report generation, visual grounding and reasoning assessment, covering 17 evaluation settings. CheXOne outperforms existing medical and general-domain foundation models and achieves strong performance on independent public benchmarks. A clinical reader study demonstrates that CheXOne-drafted reports are comparable to or better than resident-written reports in 55% of cases, while effectively addressing clinical indications and enhancing both report writing and CXR interpretation efficiency. Further analyses involving radiologists reveal that the generated reasoning traces show high clinical factuality and provide causal support for the final predictions, offering a plausible explanation for the performance gains. These results suggest that explicit reasoning can improve model performance, interpretability and clinical utility in AI-assisted CXR interpretation.

Chest X-ray imaging is crucial for diagnosing pulmonary and cardiac diseases, yet its interpretation demands extensive clinical experience and suffers from inter-observer variability. While deep learning models offer high diagnostic accuracy, their black-box nature hinders clinical adoption in high-stakes medical settings. To address this, we propose X-Ray-CoT (Chest X-Ray Chain-of-Thought), a novel framework leveraging Vision-Language Large Models (LVLMs) for intelligent chest X-ray diagnosis and interpretable report generation. X-Ray-CoT simulates human radiologists' "chain-of-thought" by first extracting multi-modal features and visual concepts, then employing an LLM-based component with a structured Chain-of-Thought prompting strategy to reason and produce detailed natural language diagnostic reports. Evaluated on the CORDA dataset, X-Ray-CoT achieves competitive quantitative performance, with a Balanced Accuracy of 80.52% and F1 score of 78.65% for disease diagnosis, slightly surpassing existing black-box models. Crucially, it uniquely generates high-quality, explainable reports, as validated by preliminary human evaluations. Our ablation studies confirm the integral role of each proposed component, highlighting the necessity of multi-modal fusion and CoT reasoning for robust and transparent medical AI. This work represents a significant step towards trustworthy and clinically actionable AI systems in medical imaging.

Purpose: Recent advancements in large language models (LLMs) have expanded their capabilities in a multimodal fashion, potentially replicating the image interpretation of human radiologists. This study aimed to develop open-source multimodal large language model for interpreting chest X-ray images (CXR-LLaVA). We also examined the effect of prompt engineering and model parameters such as temperature and nucleus sampling. Materials and Methods: For training, we collected 659,287 publicly available CXRs: 417,336 CXRs had labels for certain radiographic abnormalities (dataset 1); 241,951 CXRs provided free-text radiology reports (dataset 2). After pre-training the Resnet50 as an image encoder, the contrastive language-image pre-training was used to align CXRs and corresponding radiographic abnormalities. Then, the Large Language Model Meta AI-2 was fine-tuned using dataset 2, which were refined using GPT-4, with generating various question answering scenarios. The code can be found at https://github.com/ECOFRI/CXR_LLaVA. Results: In the test set, we observed that the model's performance fluctuated based on its parameters. On average, it achieved F1 score of 0.34 for five pathologic findings (atelectasis, cardiomegaly, consolidation, edema, and pleural effusion), which was improved to 0.46 through prompt engineering. In the independent set, the model achieved an average F1 score of 0.30 for the same pathologic findings. Notably, for the pediatric chest radiograph dataset, which was unseen during training, the model differentiated abnormal radiographs with an F1 score ranging from 0.84 to 0.85. Conclusion: CXR-LLaVA demonstrates promising potential in CXR interpretation. Both prompt engineering and model parameter adjustments can play pivotal roles in interpreting CXRs.

The rapid evolution of artificial intelligence, especially in large language models (LLMs), has significantly impacted various domains, including healthcare. In chest X-ray (CXR) analysis, previous studies have employed LLMs, but with limitations: either underutilizing the multi-tasking capabilities of LLMs or lacking clinical accuracy. This paper presents M4CXR, a multi-modal LLM designed to enhance CXR interpretation. The model is trained on a visual instruction-following dataset that integrates various task-specific datasets in a conversational format. As a result, the model supports multiple tasks such as medical report generation (MRG), visual grounding, and visual question answering (VQA). M4CXR achieves state-of-the-art clinical accuracy in MRG by employing a chain-of-thought prompting strategy, in which it identifies findings in CXR images and subsequently generates corresponding reports. The model is adaptable to various MRG scenarios depending on the available inputs, such as single-image, multi-image, and multi-study contexts. In addition to MRG, M4CXR performs visual grounding at a level comparable to specialized models and also demonstrates outstanding performance in VQA. Both quantitative and qualitative assessments reveal M4CXR's versatility in MRG, visual grounding, and VQA, while consistently maintaining clinical accuracy.

We introduce an approach to evaluate language model (LM) agency using negotiation games. This approach better reflects real-world use cases and addresses some of the shortcomings of alternative LM benchmarks. Negotiation games enable us to study multi-turn, and cross-model interactions, modulate complexity, and side-step accidental evaluation data leakage. We use our approach to test six widely used and publicly accessible LMs, evaluating performance and alignment in both self-play and cross-play settings. Noteworthy findings include: (i) only closed-source models tested here were able to complete these tasks; (ii) cooperative bargaining games proved to be most challenging to the models; and (iii) even the most powerful models sometimes "lose" to weaker opponents

While Large Language Models (LLMs) have demonstrated their proficiency in complex reasoning tasks, their performance in dynamic, interactive, and competitive scenarios - such as business strategy and stock market analysis - remains underexplored. To bridge this gap, we formally explore the dynamic reasoning capabilities of LLMs for decision-making in rapidly evolving environments. We introduce two game theory-based pilot challenges that mirror the complexities of real-world dynamic decision-making. These challenges are well-defined, enabling clear, controllable, and precise evaluation of LLMs' dynamic reasoning abilities. Through extensive experiments, we find that existing reasoning methods tend to falter in dynamic settings that require k-level thinking - a key concept not tackled by previous works. To address this, we propose a novel reasoning approach for LLMs, named "K-Level Reasoning". This approach adopts the perspective of rivals to recursively employ k-level thinking based on available historical information, which significantly improves the prediction accuracy of rivals' subsequent moves and informs more strategic decision-making. This research not only sets a robust quantitative benchmark for the assessment of dynamic reasoning but also markedly enhances the proficiency of LLMs in dynamic contexts.

This paper presents novel systems and methodologies for the development of efficient large language models (LLMs). It explores the trade-offs between model size, performance, and computational resources, with the aim of maximizing the efficiency of these AI systems. The research explores novel methods that allow different parts of the model to share parameters, reducing the total number of unique parameters required. This approach ensures that the model remains compact without sacrificing its ability to learn and represent complex language structures. This study provides valuable insights and tools for creating more efficient and effective LLMs, contributing to a more sustainable and accessible future for AI language modeling.

We present an end-to-end learning method for chess, relying on deep neural networks. Without any a priori knowledge, in particular without any knowledge regarding the rules of chess, a deep neural network is trained using a combination of unsupervised pretraining and supervised training. The unsupervised training extracts high level features from a given position, and the supervised training learns to compare two chess positions and select the more favorable one. The training relies entirely on datasets of several million chess games, and no further domain specific knowledge is incorporated. The experiments show that the resulting neural network (referred to as DeepChess) is on a par with state-of-the-art chess playing programs, which have been developed through many years of manual feature selection and tuning. DeepChess is the first end-to-end machine learning-based method that results in a grandmaster-level chess playing performance.

Chess has long been a testbed for AI's quest to match human intelligence, and in recent years, chess AI systems have surpassed the strongest humans at the game. However, these systems are not human-aligned; they are unable to match the skill levels of all human partners or model human-like behaviors beyond piece movement. In this paper, we introduce Allie, a chess-playing AI designed to bridge the gap between artificial and human intelligence in this classic game. Allie is trained on log sequences of real chess games to model the behaviors of human chess players across the skill spectrum, including non-move behaviors such as pondering times and resignations In offline evaluations, we find that Allie exhibits humanlike behavior: it outperforms the existing state-of-the-art in human chess move prediction and "ponders" at critical positions. The model learns to reliably assign reward at each game state, which can be used at inference as a reward function in a novel time-adaptive Monte-Carlo tree search (MCTS) procedure, where the amount of search depends on how long humans would think in the same positions. Adaptive search enables remarkable skill calibration; in a large-scale online evaluation against players with ratings from 1000 to 2600 Elo, our adaptive search method leads to a skill gap of only 49 Elo on average, substantially outperforming search-free and standard MCTS baselines. Against grandmaster-level (2500 Elo) opponents, Allie with adaptive search exhibits the strength of a fellow grandmaster, all while learning exclusively from humans.

In this paper we look at the ability of recent large language models (LLMs) at solving mathematical problems in combinatorics. We compare models LLaMA-2, LLaMA-3.1, GPT-4, and Mixtral against each other and against human pupils and undergraduates with prior experience in mathematical olympiads. To facilitate these comparisons we introduce the Combi-Puzzles dataset, which contains 125 problem variants based on 25 combinatorial reasoning problems. Each problem is presented in one of five distinct forms, created by systematically manipulating the problem statements through adversarial additions, numeric parameter changes, and linguistic obfuscation. Our variations preserve the mathematical core and are designed to measure the generalisability of LLM problem-solving abilities, while also increasing confidence that problems are submitted to LLMs in forms that have not been seen as training instances. We found that a model based on GPT-4 outperformed all other models in producing correct responses, and performed significantly better in the mathematical variation of the problems than humans. We also found that modifications to problem statements significantly impact the LLM's performance, while human performance remains unaffected.

The rapid advancement of large language models (LLMs) has accelerated their application in reasoning, with strategic reasoning drawing increasing attention. To evaluate LLMs' strategic reasoning capabilities, game theory, with its concise structure, has become a preferred approach. However, current research focuses on a limited selection of games, resulting in low coverage. Classic game scenarios risk data leakage, and existing benchmarks often lack extensibility, making them inadequate for evaluating state-of-the-art models. To address these challenges, we propose TMGBench, a benchmark with comprehensive game type coverage, novel scenarios, and flexible organization. Specifically, we incorporate all 144 game types summarized by the Robinson-Goforth topology of 2x2 games, constructed as classic games. We also employ synthetic data generation to create diverse, higher-quality scenarios through topic guidance and human inspection, referred to as story-based games. Lastly, we provide a sustainable framework for increasingly powerful LLMs by treating these games as atomic units and organizing them into more complex forms via sequential, parallel, and nested structures. Our comprehensive evaluation of mainstream LLMs covers tests on rational reasoning, robustness, Theory-of-Mind (ToM), and reasoning in complex forms. Results reveal flaws in accuracy, consistency, and varying mastery of ToM. Additionally, o1-mini, OpenAI's latest reasoning model, achieved accuracy rates of 66.6%, 60.0%, and 70.0% on sequential, parallel, and nested games, highlighting TMGBench's challenges.

We propose a general two-stage algorithm that enjoys a provable scaling law for the test-time compute of large language models (LLMs). Given an input problem, the proposed algorithm first generates N candidate solutions, and then chooses the best one via a multiple-round knockout tournament where each pair of candidates are compared for K times and only the winners move on to the next round. In a minimalistic implementation, both stages can be executed with a black-box LLM alone and nothing else (e.g., no external verifier or reward model), and a total of N times (K + 1) highly parallelizable LLM calls are needed for solving an input problem. Assuming that a generated candidate solution is correct with probability p_{gen} > 0 and a comparison between a pair of correct and incorrect solutions identifies the right winner with probability p_{comp} > 0.5 (i.e., better than a random guess), we prove theoretically that the failure probability of the proposed algorithm decays to zero exponentially with respect to N and K: $P(final output is incorrect) le (1 - p_{gen})^N + lceil log_2 N rceil e^{-2 K (p_{comp} - 0.5)^2}.$ Our empirical results with the challenging MMLU-Pro benchmark validate the technical assumptions, as well as the efficacy of the proposed algorithm and the gains from scaling up its test-time compute.

Large Language Models (LLMs) have exhibited impressive capabilities across numerous domains, yet they often struggle with complex reasoning and decision-making tasks. Decision-making games, which inherently require multifaceted reasoning logic, serve as ideal sandboxes for evaluating and enhancing the reasoning abilities of LLMs. In this work, we first explore whether LLMs can master complex decision-making games through targeted post-training. To this end, we design data synthesis strategies and curate extensive offline datasets from two classic games, Doudizhu and Go. We further develop a suite of techniques to effectively incorporate this data into LLM training, resulting in two novel agents: Mastermind-Dou and Mastermind-Go. Our experimental results demonstrate that these Mastermind LLMs achieve competitive performance in their respective games. Additionally, we explore whether integrating decision-making data can enhance the general reasoning abilities of LLMs. Our findings suggest that such post-training improves certain aspects of reasoning, providing valuable insights for optimizing LLM data collection and synthesis strategies.

Large language models (LLMs) have been increasingly applied to tasks in language understanding and interactive decision-making, with their impressive performance largely attributed to the extensive domain knowledge embedded within them. However, the depth and breadth of this knowledge can vary across domains. Many existing approaches assume that LLMs possess a comprehensive understanding of their environment, often overlooking potential gaps in their grasp of actual world dynamics. To address this, we introduce Discover, Verify, and Evolve (DiVE), a framework that discovers world dynamics from a small number of demonstrations, verifies the accuracy of these dynamics, and evolves new, advanced dynamics tailored to the current situation. Through extensive evaluations, we assess the impact of each component on performance and compare the dynamics generated by DiVE to human-annotated dynamics. Our results show that LLMs guided by DiVE make more informed decisions, achieving rewards comparable to human players in the Crafter environment and surpassing methods that require prior task-specific training in the MiniHack environment.

Automated reasoning with unstructured natural text is a key requirement for many potential applications of NLP and for developing robust AI systems. Recently, Language Models (LMs) have demonstrated complex reasoning capacities even without any finetuning. However, existing evaluation for automated reasoning assumes access to a consistent and coherent set of information over which models reason. When reasoning in the real-world, the available information is frequently inconsistent or contradictory, and therefore models need to be equipped with a strategy to resolve such conflicts when they arise. One widely-applicable way of resolving conflicts is to impose preferences over information sources (e.g., based on source credibility or information recency) and adopt the source with higher preference. In this paper, we formulate the problem of reasoning with contradictory information guided by preferences over sources as the classical problem of defeasible reasoning, and develop a dataset called BoardgameQA for measuring the reasoning capacity of LMs in this setting. BoardgameQA also incorporates reasoning with implicit background knowledge, to better reflect reasoning problems in downstream applications. We benchmark various LMs on BoardgameQA and the results reveal a significant gap in the reasoning capacity of state-of-the-art LMs on this problem, showing that reasoning with conflicting information does not surface out-of-the-box in LMs. While performance can be improved with finetuning, it nevertheless remains poor.

Vision-language models (VLMs) trained on paired chest radiographs and radiology reports learn a shared embedding space that can preserve instance-level image-report correspondence. This poses a privacy risk in settings where radiographs and reports are deliberately kept separate after acquisition, such as image-only data sharing or access-controlled reports, because a de-identified image may be re-linked to its original narrative report through cosine similarity alone. We formalized this as image-to-report retrieval and used public paired cohorts, in which the true pairing is known by design, as ground-truth benchmarks to audit the risk rather than as the privacy scenario. Evaluating VLMs of increasing clinical specialization on 406,241 paired examples from 126,804 patients across MIMIC-CXR (43,793 held-out pairs) and external CheXpert Plus (29,296 pairs), we found that re-linkage rose systematically with specialization: the strongest VLM retrieved the correct report at 15 times chance at a candidate pool of N = 100, 50 times chance at N = 10,000, and well above chance at full-database scale. The signal persisted under pathology-matched hard negatives that removed disease-label shortcuts, indicating correspondence beyond broad diagnostic categories. To reduce it without retraining, we froze both encoders and applied differentially private optimization only to the projection heads defining the alignment layer (epsilon = 0.34, delta = 6x10-6). This reduced Recall@1 by 61.8% at N = 10,000 on MIMIC-CXR and transferred to CheXpert Plus without retraining, while image-side utility was largely preserved: macro AUROC for linear-probe classification across 14 labels shifted only from 79.63% to 79.43%. Targeted DP finetuning of the shared alignment layer can substantially reduce cross-modal re-linkage without materially degrading the image representations that make these models clinically useful.

Medical vision-language models often struggle with generating accurate quantitative measurements in radiology reports, leading to hallucinations that undermine clinical reliability. We introduce FactCheXcker, a modular framework that de-hallucinates radiology report measurements by leveraging an improved query-code-update paradigm. Specifically, FactCheXcker employs specialized modules and the code generation capabilities of large language models to solve measurement queries generated based on the original report. After extracting measurable findings, the results are incorporated into an updated report. We evaluate FactCheXcker on endotracheal tube placement, which accounts for an average of 78% of report measurements, using the MIMIC-CXR dataset and 11 medical report-generation models. Our results show that FactCheXcker significantly reduces hallucinations, improves measurement precision, and maintains the quality of the original reports. Specifically, FactCheXcker improves the performance of 10/11 models and achieves an average improvement of 135.0% in reducing measurement hallucinations measured by mean absolute error. Code is available at https://github.com/rajpurkarlab/FactCheXcker.

The widespread use of chest X-rays (CXRs), coupled with a shortage of radiologists, has driven growing interest in automated CXR analysis and AI-assisted reporting. While existing vision-language models (VLMs) show promise in specific tasks such as report generation or abnormality detection, they often lack support for interactive diagnostic capabilities. In this work we present RadVLM, a compact, multitask conversational foundation model designed for CXR interpretation. To this end, we curate a large-scale instruction dataset comprising over 1 million image-instruction pairs containing both single-turn tasks -- such as report generation, abnormality classification, and visual grounding -- and multi-turn, multi-task conversational interactions. After fine-tuning RadVLM on this instruction dataset, we evaluate it across different tasks along with re-implemented baseline VLMs. Our results show that RadVLM achieves state-of-the-art performance in conversational capabilities and visual grounding while remaining competitive in other radiology tasks. Ablation studies further highlight the benefit of joint training across multiple tasks, particularly for scenarios with limited annotated data. Together, these findings highlight the potential of RadVLM as a clinically relevant AI assistant, providing structured CXR interpretation and conversational capabilities to support more effective and accessible diagnostic workflows.

Medical report generation has achieved remarkable advancements yet has still been faced with several challenges. First, the inherent imbalance in the distribution of normal and abnormal cases may lead models to exhibit a biased focus on normal samples, resulting in unreliable diagnoses. Second, the frequent occurrence of common template sentences in the reports may overwhelm the critical abnormal information. Moreover, existing works focus on 2D chest X-rays, leaving CT report generation underexplored due to the high-dimensional nature of CT images and the limited availability of CT-report pairs. Recently, LLM has shown a great ability to generate reliable answers with appropriate prompts, which shed light on addressing the aforementioned challenges. In this paper, we propose Dia-LLaMA, a framework to adapt the LLaMA2-7B for CT report generation by incorporating diagnostic information as guidance prompts. Considering the high dimension of CT, we leverage a pre-trained ViT3D with perceiver to extract the visual information. To tailor the LLM for report generation and emphasize abnormality, we extract additional diagnostic information by referring to a disease prototype memory bank, which is updated during training to capture common disease representations. Furthermore, we introduce disease-aware attention to enable the model to adjust attention for different diseases. Experiments on the chest CT dataset demonstrated that our proposed method outperformed previous methods and achieved state-of-the-art on both clinical efficacy performance and natural language generation metrics. The code will be made publically available.

Background: Recent advances in large language models highlight the need for high-quality multilingual medical datasets. While Japan leads globally in CT scanner deployment and utilization, the lack of large-scale Japanese radiology datasets has hindered the development of specialized language models for medical imaging analysis. Objective: To develop a comprehensive Japanese CT report dataset through machine translation and establish a specialized language model for structured finding classification. Additionally, to create a rigorously validated evaluation dataset through expert radiologist review. Methods: We translated the CT-RATE dataset (24,283 CT reports from 21,304 patients) into Japanese using GPT-4o mini. The training dataset consisted of 22,778 machine-translated reports, while the validation dataset included 150 radiologist-revised reports. We developed CT-BERT-JPN based on "tohoku-nlp/bert-base-japanese-v3" architecture for extracting 18 structured findings from Japanese radiology reports. Results: Translation metrics showed strong performance with BLEU scores of 0.731 and 0.690, and ROUGE scores ranging from 0.770 to 0.876 for Findings and from 0.748 to 0.857 for Impression sections. CT-BERT-JPN demonstrated superior performance compared to GPT-4o in 11 out of 18 conditions, including lymphadenopathy (+14.2%), interlobular septal thickening (+10.9%), and atelectasis (+7.4%). The model maintained F1 scores exceeding 0.95 in 14 out of 18 conditions and achieved perfect scores in four conditions. Conclusions: Our study establishes a robust Japanese CT report dataset and demonstrates the effectiveness of a specialized language model for structured finding classification. The hybrid approach of machine translation and expert validation enables the creation of large-scale medical datasets while maintaining high quality.

Radiologists play a crucial role by translating medical images into medical reports. However, the field faces staffing shortages and increasing workloads. While automated approaches using vision-language models (VLMs) show promise as assistants, they require exceptionally high accuracy. Most current VLMs in radiology rely solely on supervised fine-tuning (SFT). Meanwhile, in the general domain, additional preference fine-tuning has become standard practice. The challenge in radiology lies in the prohibitive cost of obtaining radiologist feedback. We propose a scalable automated preference alignment technique for VLMs in radiology, focusing on chest X-ray (CXR) report generation. Our method leverages publicly available datasets with an LLM-as-a-Judge mechanism, eliminating the need for additional expert radiologist feedback. We evaluate and benchmark five direct alignment algorithms (DAAs). Our results show up to a 57.4% improvement in average GREEN scores, a LLM-based metric for evaluating CXR reports, and a 9.2% increase in an average across six metrics (domain specific and general), compared to the SFT baseline. We study reward overoptimization via length exploitation, with reports lengthening by up to 3.2x. To assess a potential alignment tax, we benchmark on six additional diverse tasks, finding no significant degradations. A reader study involving four board-certified radiologists indicates win rates of up to 0.62 over the SFT baseline, while significantly penalizing verbosity. Our analysis provides actionable insights for the development of VLMs in high-stakes fields like radiology.

Inspired by the success of large language models (LLMs), there is growing research interest in developing LLMs in the medical domain to assist clinicians. However, for hospitals, using closed-source commercial LLMs involves privacy issues, and developing open-source public LLMs requires large-scale computational resources, which are usually limited, especially in resource-efficient regions and low-income countries. We propose an open-source Small Language and Vision Assistant (SLaVA-CXR) that can be used for Chest X-Ray report automation. To efficiently train a small assistant, we first propose the Re^3Training method, which simulates the cognitive development of radiologists and optimizes the model in the Recognition, Reasoning, and Reporting training manner. Then, we introduce a data synthesis method, RADEX, which can generate a high-quality and diverse training corpus with privacy regulation compliance. The extensive experiments show that our SLaVA-CXR built on a 2.7B backbone not only outperforms but also achieves 6 times faster inference efficiency than previous state-of-the-art larger models.

Vision-language models (VLMs) often produce chain-of-thought (CoT) explanations that sound plausible yet fail to reflect the underlying decision process, undermining trust in high-stakes clinical use. Existing evaluations rarely catch this misalignment, prioritizing answer accuracy or adherence to formats. We present a clinically grounded framework for chest X-ray visual question answering (VQA) that probes CoT faithfulness via controlled text and image modifications across three axes: clinical fidelity, causal attribution, and confidence calibration. In a reader study (n=4), evaluator-radiologist correlations fall within the observed inter-radiologist range for all axes, with strong alignment for attribution (Kendall's tau_b=0.670), moderate alignment for fidelity (tau_b=0.387), and weak alignment for confidence tone (tau_b=0.091), which we report with caution. Benchmarking six VLMs shows that answer accuracy and explanation quality are decoupled, acknowledging injected cues does not ensure grounding, and text cues shift explanations more than visual cues. While some open-source models match final answer accuracy, proprietary models score higher on attribution (25.0% vs. 1.4%) and often on fidelity (36.1% vs. 31.7%), highlighting deployment risks and the need to evaluate beyond final answer accuracy.

Advancing representation learning in specialized fields like medicine remains challenging due to the scarcity of expert annotations for text and images. To tackle this issue, we present a novel two-stage framework designed to extract high-quality factual statements from free-text radiology reports in order to improve the representations of text encoders and, consequently, their performance on various downstream tasks. In the first stage, we propose a Fact Extractor that leverages large language models (LLMs) to identify factual statements from well-curated domain-specific datasets. In the second stage, we introduce a Fact Encoder (CXRFE) based on a BERT model fine-tuned with objective functions designed to improve its representations using the extracted factual data. Our framework also includes a new embedding-based metric (CXRFEScore) for evaluating chest X-ray text generation systems, leveraging both stages of our approach. Extensive evaluations show that our fact extractor and encoder outperform current state-of-the-art methods in tasks such as sentence ranking, natural language inference, and label extraction from radiology reports. Additionally, our metric proves to be more robust and effective than existing metrics commonly used in the radiology report generation literature. The code of this project is available at https://github.com/PabloMessina/CXR-Fact-Encoder.

Vision and language models (VLMs) hold immense promise to transform biomedical imaging workflows, from detecting lesions in chest X-rays to profiling cellular features in microscopy. Realizing this potential, however, requires robust and fine-grained visual perception. Models need to correctly interpret subtle features in images, and they must do so across diverse biomedical modalities, scales, and contexts. Nevertheless, current benchmarks remain limited. To address these gaps, we introduce the Massive Multimodal Biomedical Understanding (MMBU) benchmark. It is the largest biomedical vision and language benchmark to date, covering 35 submodalities with rich structured metadata. It includes both open and closed versions of ungrounded classification, grounded classification, and object detection, enabling systematic evaluation of model performance across biological scales, clinical settings, and imaging modalities. Evaluating 15 open-weight and 2 frontier VLMs, we find that while medical adaptation provides measurable gains for some models, the high accuracy often reported on established benchmarks can mask deficiencies in visual perception and domain generalization.

Chest X-rays (CXRs) are the most frequently performed imaging test in clinical practice. Recent advances in the development of vision-language foundation models (FMs) give rise to the possibility of performing automated CXR interpretation, which can assist physicians with clinical decision-making and improve patient outcomes. However, developing FMs that can accurately interpret CXRs is challenging due to the (1) limited availability of large-scale vision-language datasets in the medical image domain, (2) lack of vision and language encoders that can capture the complexities of medical data, and (3) absence of evaluation frameworks for benchmarking the abilities of FMs on CXR interpretation. In this work, we address these challenges by first introducing CheXinstruct - a large-scale instruction-tuning dataset curated from 28 publicly-available datasets. We then present CheXagent - an instruction-tuned FM capable of analyzing and summarizing CXRs. To build CheXagent, we design a clinical large language model (LLM) for parsing radiology reports, a vision encoder for representing CXR images, and a network to bridge the vision and language modalities. Finally, we introduce CheXbench - a novel benchmark designed to systematically evaluate FMs across 8 clinically-relevant CXR interpretation tasks. Extensive quantitative evaluations and qualitative reviews with five expert radiologists demonstrate that CheXagent outperforms previously-developed general- and medical-domain FMs on CheXbench tasks. Furthermore, in an effort to improve model transparency, we perform a fairness evaluation across factors of sex, race and age to highlight potential performance disparities. Our project is at https://stanford-aimi.github.io/chexagent.html.

Longitudinal information in radiology reports refers to the sequential tracking of findings across multiple examinations over time, which is crucial for monitoring disease progression and guiding clinical decisions. Many recent automated radiology report generation methods are designed to capture longitudinal information; however, validating their performance is challenging. There is no proper tool to consistently label temporal changes in both ground-truth and model-generated texts for meaningful comparisons. Large language models (LLMs) offer a promising annotation alternative, as they are capable of capturing nuanced linguistic patterns and semantic similarities without extensive manual intervention. They also adapt well to new contexts. In this study, we therefore propose an LLM-based pipeline to automatically annotate longitudinal information in radiology reports. The pipeline first identifies sentences containing relevant information and then extracts the progression of diseases. We evaluate and compare five mainstream LLMs on these two tasks using 500 manually annotated reports. Considering both efficiency and performance, Qwen2.5-32B was subsequently selected and used to annotate another 95,169 reports from the public MIMIC-CXR dataset. Our Qwen2.5-32B-annotated dataset provided us with a standardized benchmark for evaluating report generation models. Using this new benchmark, we assessed seven state-of-the-art report generation models. Our LLM-based annotation method outperforms existing annotation solutions, achieving 11.3\% and 5.3\% higher F1-scores for longitudinal information detection and disease tracking, respectively. The source code is available at https://github.com/wxinyi1996/Standardizing-Longitudinal-Chest-X-ray-Report-Evaluation-via-Large-Language-Model-Annotation.git.

We present NV-Reason-CT, a generative vision--language model for chest and abdominal CT combining native 3D visual encoding with radiologist-guided reasoning. The model couples a native 3D vision transformer with a language model, passing all visual tokens and their explicit 3D coordinates into language decoding without further spatial token merging. This retains volumetric spatial information within the vision encoder and through the language model's positional encoding during joint processing with text. We train on a curated corpus of approximately 550,000 multimodal instruction examples from 70,111 unique CT image inputs, combining standardized reports, abnormality-focused and anatomy-specific questions, multi-turn interactions, and radiologist-authored reasoning from recorded and transcribed expert CT interpretations. Expert annotations provide direct supervision and guide additional report-grounded synthetic reasoning. End-to-end supervised fine-tuning (SFT) is followed by Group Relative Policy Optimization (GRPO), with verifiable rewards over chest and abdominal abnormality sets. The model supports abnormality classification, report generation, and interactive reasoning with reviewable observations, differential diagnoses, and uncertainty. Evaluation spans public CT benchmarks and a held-out NIH cohort. On CT-RATE, NV-Reason-CT achieves a macro-F1 of 0.614 and macro-AUROC of 0.871 without a task-specific classification head; generated reports achieve a report-derived macro-F1 of 0.592. In a preliminary study with expert radiologists, AI-assisted review received favorable confidence ratings and was associated with a 50% reduction in average reported interpretation and reporting time. We release the model and training code to support reproducible research on explainable AI for volumetric medical imaging.

Developing generalist foundation model has recently attracted tremendous attention among researchers in the field of AI for Medicine (AI4Medicine). A pivotal insight in developing these models is their reliance on dataset scaling, which emphasizes the requirements on developing open-source medical image datasets that incorporate diverse supervision signals across various imaging modalities. In this paper, we introduce RadGenome-Chest CT, a comprehensive, large-scale, region-guided 3D chest CT interpretation dataset based on CT-RATE. Specifically, we leverage the latest powerful universal segmentation and large language models, to extend the original datasets (over 25,692 non-contrast 3D chest CT volume and reports from 20,000 patients) from the following aspects: (i) organ-level segmentation masks covering 197 categories, which provide intermediate reasoning visual clues for interpretation; (ii) 665 K multi-granularity grounded reports, where each sentence of the report is linked to the corresponding anatomical region of CT volume in the form of a segmentation mask; (iii) 1.3 M grounded VQA pairs, where questions and answers are all linked with reference segmentation masks, enabling models to associate visual evidence with textual explanations. All grounded reports and VQA pairs in the validation set have gone through manual verification to ensure dataset quality. We believe that RadGenome-Chest CT can significantly advance the development of multimodal medical foundation models, by training to generate texts based on given segmentation regions, which is unattainable with previous relevant datasets. We will release all segmentation masks, grounded reports, and VQA pairs to facilitate further research and development in this field.

Conversational AI tools that can generate and discuss clinically correct radiology reports for a given medical image have the potential to transform radiology. Such a human-in-the-loop radiology assistant could facilitate a collaborative diagnostic process, thus saving time and improving the quality of reports. Towards this goal, we introduce RaDialog, the first thoroughly evaluated and publicly available large vision-language model for radiology report generation and interactive dialog. RaDialog effectively integrates visual image features and structured pathology findings with a large language model (LLM) while simultaneously adapting it to a specialized domain using parameter-efficient fine-tuning. To keep the conversational abilities of the underlying LLM, we propose a comprehensive, semi-automatically labeled, image-grounded instruct dataset for chest X-ray radiology tasks. By training with this dataset, our method achieves state-of-the-art clinical correctness in report generation and shows impressive abilities in interactive tasks such as correcting reports and answering questions, serving as a foundational step toward clinical dialog systems. Our code is available on github: https://github.com/ChantalMP/RaDialog.

Objectives: We aim to dynamically retrieve informative demonstrations, enhancing in-context learning in multimodal large language models (MLLMs) for disease classification. Methods: We propose a Retrieval-Augmented In-Context Learning (RAICL) framework, which integrates retrieval-augmented generation (RAG) and in-context learning (ICL) to adaptively select demonstrations with similar disease patterns, enabling more effective ICL in MLLMs. Specifically, RAICL examines embeddings from diverse encoders, including ResNet, BERT, BioBERT, and ClinicalBERT, to retrieve appropriate demonstrations, and constructs conversational prompts optimized for ICL. We evaluated the framework on two real-world multi-modal datasets (TCGA and IU Chest X-ray), assessing its performance across multiple MLLMs (Qwen, Llava, Gemma), embedding strategies, similarity metrics, and varying numbers of demonstrations. Results: RAICL consistently improved classification performance. Accuracy increased from 0.7854 to 0.8368 on TCGA and from 0.7924 to 0.8658 on IU Chest X-ray. Multi-modal inputs outperformed single-modal ones, with text-only inputs being stronger than images alone. The richness of information embedded in each modality will determine which embedding model can be used to get better results. Few-shot experiments showed that increasing the number of retrieved examples further enhanced performance. Across different similarity metrics, Euclidean distance achieved the highest accuracy while cosine similarity yielded better macro-F1 scores. RAICL demonstrated consistent improvements across various MLLMs, confirming its robustness and versatility. Conclusions: RAICL provides an efficient and scalable approach to enhance in-context learning in MLLMs for multimodal disease classification.

Recent medical multimodal foundation models are built as multimodal LLMs (MLLMs) by connecting a CLIP-pretrained vision encoder to an LLM using LLaVA-style finetuning. This two-stage, decoupled approach introduces a projection layer that can distort visual features. This is especially concerning in medical imaging where subtle cues are essential for accurate diagnoses. In contrast, early-fusion generative approaches such as Chameleon eliminate the projection bottleneck by processing image and text tokens within a single unified sequence, enabling joint representation learning that leverages the inductive priors of language models. We present CheXmix, a unified early-fusion generative model trained on a large corpus of chest X-rays paired with radiology reports. We expand on Chameleon's autoregressive framework by introducing a two-stage multimodal generative pretraining strategy that combines the representational strengths of masked autoencoders with MLLMs. The resulting models are highly flexible, supporting both discriminative and generative tasks at both coarse and fine-grained scales. Our approach outperforms well-established generative models across all masking ratios by 6.0% and surpasses CheXagent by 8.6% on AUROC at high image masking ratios on the CheXpert classification task. We further inpaint images over 51.0% better than text-only generative models and outperform CheXagent by 45% on the GREEN metric for radiology report generation. These results demonstrate that CheXmix captures fine-grained information across a broad spectrum of chest X-ray tasks. Our code is at: https://github.com/StanfordMIMI/CheXmix.

Long-tailed chest X-ray classification requires visual representations that capture both common abnormalities and subtle, infrequent findings. We propose Med-AR-8B and Med-AR-2B, two radiology-native autoregressive vision-language models pretrained with structured reports, abnormality-focused text, and region annotations. We evaluate the transfer of their visual encoders to multi-label classification against contrastive, self-supervised, and supervised pretrained encoders, including Med-CLIP, CheXFound, EVA-Base, ARK, and BioViL-T, using a common ML-Decoder classification head. To assess fine-grained recognition, we also construct LLM-expanded, report-derived label sets for MIMIC-CXR and CheXpert. Across PadChest, MIMIC-CXR, and CheXpert, Med-AR-8B outperforms Med-CLIP in mean AUROC and AUPRC for head, medium, and tail findings. On MIMIC-CXR, it increases tail-label mean AUPRC from 0.1033 to 0.1441. Med-AR-2B achieves the strongest discrimination results on PadChest. Across the broader encoder comparison, a Med-AR variant achieves the highest mean AUROC and AUPRC in every reported prevalence group on each public dataset. Both Med-AR variants also achieve lower excess area under the risk-coverage curve than Med-CLIP on all three public datasets, indicating improved selective-prediction performance under the evaluated protocol. Internal results are metric-dependent, with Med-CLIP retaining advantages in overall and tail AUPRC and in selective prediction. These findings establish Med-AR as a strong pretraining recipe for long-tailed chest X-ray classification on the evaluated public benchmarks and demonstrate the value of assessing discrimination and selective prediction together.

Vision-language (VL) pretraining using paired chest X-ray (CXR) images and radiology reports has shown strong potential for medical image understanding. However, existing methods often remain dependent on task-specific finetuning because radiology reports are lengthy, clinically dense, and difficult to align with simple zero-shot prompts. Recent sentence-level approaches partially address this limitation using clinical phrases extracted by large language models (LLMs), but they largely overlook the intrinsic characteristics of radiology discourse. In particular, limited positive-pair diversity constrains further gains, while clinically equivalent sentences frequently recur across patients, creating false negatives in contrastive learning. To address these issues, we propose SentZero, an enhanced sentence-centric VL pretraining framework for zero-shot, multi-task CXR analysis. SentZero introduces LLM-based abstract-level sentence structuring and mapping to expand positive-pair diversity, together with an additional loss term to mitigate false negatives. We further introduce sentence-conditioned residual modulation of visual embeddings, enabling visual features to adapt to the semantic characteristics of each input sentence. Across diverse downstream tasks and datasets, SentZero improves zero-shot generalization and outperforms prior multi-task zero-shot methods.

Our approach, which we call Embeddings for Language/Image-aligned X-Rays, or ELIXR, leverages a language-aligned image encoder combined or grafted onto a fixed LLM, PaLM 2, to perform a broad range of tasks. We train this lightweight adapter architecture using images paired with corresponding free-text radiology reports from the MIMIC-CXR dataset. ELIXR achieved state-of-the-art performance on zero-shot chest X-ray (CXR) classification (mean AUC of 0.850 across 13 findings), data-efficient CXR classification (mean AUCs of 0.893 and 0.898 across five findings (atelectasis, cardiomegaly, consolidation, pleural effusion, and pulmonary edema) for 1% (~2,200 images) and 10% (~22,000 images) training data), and semantic search (0.76 normalized discounted cumulative gain (NDCG) across nineteen queries, including perfect retrieval on twelve of them). Compared to existing data-efficient methods including supervised contrastive learning (SupCon), ELIXR required two orders of magnitude less data to reach similar performance. ELIXR also showed promise on CXR vision-language tasks, demonstrating overall accuracies of 58.7% and 62.5% on visual question answering and report quality assurance tasks, respectively. These results suggest that ELIXR is a robust and versatile approach to CXR AI.

Medical image analysis is crucial in modern radiological diagnostics, especially given the exponential growth in medical imaging data. The demand for automated report generation systems has become increasingly urgent. While prior research has mainly focused on using machine learning and multimodal language models for 2D medical images, the generation of reports for 3D medical images has been less explored due to data scarcity and computational complexities. This paper introduces 3D-CT-GPT, a Visual Question Answering (VQA)-based medical visual language model specifically designed for generating radiology reports from 3D CT scans, particularly chest CTs. Extensive experiments on both public and private datasets demonstrate that 3D-CT-GPT significantly outperforms existing methods in terms of report accuracy and quality. Although current methods are few, including the partially open-source CT2Rep and the open-source M3D, we ensured fair comparison through appropriate data conversion and evaluation methodologies. Experimental results indicate that 3D-CT-GPT enhances diagnostic accuracy and report coherence, establishing itself as a robust solution for clinical radiology report generation. Future work will focus on expanding the dataset and further optimizing the model to enhance its performance and applicability.

Vision language models (VLMs) show promise in medical diagnosis, but their performance across demographic subgroups when using in-context learning (ICL) remains poorly understood. We examine how the demographic composition of demonstration examples affects VLM performance in two medical imaging tasks: skin lesion malignancy prediction and pneumothorax detection from chest radiographs. Our analysis reveals that ICL influences model predictions through multiple mechanisms: (1) ICL allows VLMs to learn subgroup-specific disease base rates from prompts and (2) ICL leads VLMs to make predictions that perform differently across demographic groups, even after controlling for subgroup-specific disease base rates. Our empirical results inform best-practices for prompting current VLMs (specifically examining demographic subgroup performance, and matching base rates of labels to target distribution at a bulk level and within subgroups), while also suggesting next steps for improving our theoretical understanding of these models.

Contrastive vision-language learning uses paired chest CT volumes and radiology reports to learn abnormality classifiers without manually annotated labels. However, two characteristics of chest CT challenge conventional global contrastive learning. First, many critical abnormalities are small or anatomically localized, and pooling an en- tire volume into a single embedding may dilute their visual evidence. Second, the standard contrastive objective treats every other scan in a batch as a negative. Because many chest CTs share abnormalities, this objective incorrectly pushes co-positive pairs apart. We propose Anatomy-Routed Contrastive Learning for 3D Chest CT (ARC-CT), a region-aware framework that addresses these limitations using only la- bels extracted from reports by an LLM, with no manual annotations or bounding boxes. ARC-CT combines three components: (1) an Anato- myQFormer localizing evidence via queries constrained by automatically generated organ masks; (2) a label-Jaccard soft InfoNCE objective in- tegrating the standard one-hot target with the label-set overlap of each pair, which reduces false-negative penalties between studies that share clinical findings; and (3) an organ-level alignment loss connecting mask- pooled visual features to organ-specific report text extracted offline with a large language model. ARC-CT achieves a 0.86 mask-free macro AUC across 18 abnormalities using a compact 3D ResNet-18 backbone. Over- all, ARC-CT outperforms both comparable efficient baselines and sev- eral larger transformer models. Our code and weights are available at https://github.com/arc-ct/arc-ct.

What is learned by sophisticated neural network agents such as AlphaZero? This question is of both scientific and practical interest. If the representations of strong neural networks bear no resemblance to human concepts, our ability to understand faithful explanations of their decisions will be restricted, ultimately limiting what we can achieve with neural network interpretability. In this work we provide evidence that human knowledge is acquired by the AlphaZero neural network as it trains on the game of chess. By probing for a broad range of human chess concepts we show when and where these concepts are represented in the AlphaZero network. We also provide a behavioural analysis focusing on opening play, including qualitative analysis from chess Grandmaster Vladimir Kramnik. Finally, we carry out a preliminary investigation looking at the low-level details of AlphaZero's representations, and make the resulting behavioural and representational analyses available online.

In arena-style evaluation of large language models (LLMs), two LLMs respond to a user query, and the user chooses the winning response or deems the "battle" a draw, resulting in an adjustment to the ratings of both models. The prevailing approach for modeling these rating dynamics is to view battles as two-player game matches, as in chess, and apply the Elo rating system and its derivatives. In this paper, we critically examine this paradigm. Specifically, we question whether a draw genuinely means that the two models are equal and hence whether their ratings should be equalized. Instead, we conjecture that draws are more indicative of query difficulty: if the query is too easy, then both models are more likely to succeed equally. On three real-world arena datasets, we show that ignoring rating updates for draws yields a 1-3% relative increase in battle outcome prediction accuracy (which includes draws) for all four rating systems studied. Further analyses suggest that draws occur more for queries rated as very easy and those as highly objective, with risk ratios of 1.37 and 1.35, respectively. We recommend future rating systems to reconsider existing draw semantics and to account for query properties in rating updates.

The prototyping of computer games, particularly card games, requires extensive human effort in creative ideation and gameplay evaluation. Recent advances in Large Language Models (LLMs) offer opportunities to automate and streamline these processes. However, it remains challenging for LLMs to design novel game mechanics beyond existing databases, generate consistent gameplay environments, and develop scalable gameplay AI for large-scale evaluations. This paper addresses these challenges by introducing a comprehensive automated card game prototyping framework. The approach highlights a graph-based indexing method for generating novel game designs, an LLM-driven system for consistent game code generation validated by gameplay records, and a gameplay AI constructing method that uses an ensemble of LLM-generated action-value functions optimized through self-play. These contributions aim to accelerate card game prototyping, reduce human labor, and lower barriers to entry for game developers.

Chess board recognition is the task of mapping the image of a chess board to the information of which piece is on which square. So far this task has two established benchmarks: ChessCog is synthetic, and ChessReD comes from smartphone pictures of a single chess board setup. We introduce ChessQueries, a new method combining a ViT encoder with a DETR-style decoder, which outperforms existing methods. On the ChessReD benchmark, we improve the state of the art from 15.3% to 99.2%, and demonstrate strong capabilities on out-of-distribution datasets. Our method saturates the task on the two datasets, with an average 0.01 wrong squares per board (vs. SotA: 3.4 / 0.15 respectively). We also share a new, harder public dataset, parsed from broadcasted top-level chess tournaments. Code, model weights and the SLCC data will be released.