We introduce LLM CHESS, an evaluation framework designed to probe the generalization of reasoning and instruction-following abilities in large language models (LLMs) through extended agentic interaction in the domain of chess. We rank over 50 open and closed source models by playing against a random opponent using a range of behavioral metrics, including win and loss rates, move quality, move legality, hallucinated actions, and game duration. For a subset of top reasoning models, we derive an Elo estimate by playing against a chess engine with variably configured skill, which allows for comparisons between models in an easily understandable way. Despite the simplicity of the instruction-following task and the weakness of the opponent, many state-of-the-art models struggle to complete games or achieve consistent wins. Similar to other benchmarks on complex reasoning tasks, our experiments reveal a clear separation between reasoning and non-reasoning models. However, unlike existing static benchmarks, the stochastic and dynamic nature of LLM CHESS uniquely reduces overfitting and memorization while preventing benchmark saturation, proving difficult even for top reasoning models. To support future work on evaluating reasoning and instruction-following in LLMs, we release our experimental framework, a public leaderboard, and a dataset of associated games.

Third, the current game state data--number of players, last played card, hand contents, next player, recent moves, and legal actions. Finally, the LLM was asked to choose the best action according to the specified prompting method. The game state information was extracted from RLCard and reformatted for readability. While RLCard encodes cards using shorthand (e.g., "r-5" for red 5), we expanded these into full descriptions to improve the model's comprehension. An example of the complete prompt format is shown in Figure 3. To drive the model's action selection, we applied two prompting strategies inspired by Moore et al. [17]: cloze prompting and counterfactual prompting. These methods determine how the model interprets the prompt and evaluates its legal actions during gameplay. Cloze Prompting: In this method, legal actions were labeled with sequential letters (A, B, C, etc.), and the LLM was instructed to choose the letter corresponding to the best move. Only one token was allowed in the output, and the highest-probability token from the set of allowable actions was selected as the action.

Researchers worried about catastrophic risks from advanced AI have argued that we should expect sufficiently capable AI agents to pursue power over humanity because power is a convergent instrumental goal, something that is useful for a wide range of final goals. Others have recently expressed skepticism of these claims. This paper aims to formalize the concepts of instrumental convergence and power-seeking in an abstract, decision-theoretic framework, and to assess the claim that power is a convergent instrumental goal. I conclude that this claim contains at least an element of truth, but might turn out to have limited predictive utility, since an agent's options cannot always be ranked in terms of power in the absence of substantive information about the agent's final goals. However, the fact of instrumental convergence is more predictive for agents who have a good shot at attaining absolute or near-absolute power.

Hex and Counter Wargames are adversarial two-player simulations of real military conflicts requiring complex strategic decision-making. Unlike classical board games, these games feature intricate terrain/unit interactions, unit stacking, large maps of varying sizes, and simultaneous move and combat decisions involving hundreds of units. This paper introduces a novel system designed to address the strategic complexity of Hex and Counter Wargames by integrating cutting-edge advancements in Recurrent Neural Networks with AlphaZero, a reliable modern Reinforcement Learning algorithm. The system utilizes a new Neural Network architecture developed from existing research, incorporating innovative state and action representations tailored to these specific game environments. With minimal training, our solution has shown promising results in typical scenarios, demonstrating the ability to generalize across different terrain and tactical situations. Additionally, we explore the system's potential to scale to larger map sizes. The developed system is openly accessible, facilitating continued research and exploration within this challenging domain.

In drug discovery, highly automated high-throughput laboratories are used to screen a large number of compounds in search of effective drugs. These experiments are expensive, so one might hope to reduce their cost by experimenting on a subset of the compounds, and predicting the outcomes of the remaining experiments. In this work, we model this scenario as a sequential subset selection problem: we aim to select the smallest set of candidates in order to achieve some desired level of accuracy for the system as a whole. Our key observation is that, if there is heterogeneity in the difficulty of the prediction problem across the input space, selectively obtaining the labels for the hardest examples in the acquisition pool will leave only the relatively easy examples to remain in the inference set, leading to better overall system performance. We call this mechanism inference set design, and propose the use of an confidence-based active learning solution to prune out these challenging examples. Our algorithm includes an explicit stopping criterion that stops running the experiments when it is sufficiently confident that the system has reached the target performance. Our empirical studies on image and molecular datasets, as well as a real-world large-scale biological assay, show that active learning for inference set design leads to significant reduction in experimental cost while retaining high system performance. Automated high-throughput screening (HTS) laboratories have enabled scientists to screen large compound libraries to find effective therapeutic compounds and screen whole-genome CRISPR knockouts to understand the effects of genes on cell function (Mayr & Bojanic, 2009; Wildey et al., 2017; Blay et al., 2020; Tom et al., 2024; Fay et al., 2023). However, conducting experiments on every compound or gene in these vast design spaces remains very resource-intensive. Reducing experimental costs without compromising the quality of the generated data would allow us to accelerate biology and pharmaceutical research and expand the set of molecules considered for testing. To avoid costs scaling with the number of experiments, we can train a model from a subset of the target library that has been tested in the lab, and then predict experimental outcomes for the remainder of the library using the trained model (Naik et al., 2013; Reker & Schneider, 2015; Dara et al., 2022), thereby building a hybrid screen of the library.

Many enhancements to Monte-Carlo Tree Search (MCTS) have been proposed over almost two decades of general game playing and other artificial intelligence research. However, our ability to characterise and understand which variants work well or poorly in which games is still lacking. This paper describes work on an initial dataset that we have built to make progress towards such an understanding: 268,386 plays among 61 different agents across 1494 distinct games. We describe a preliminary analysis and work on training predictive models on this dataset, as well as lessons learned and future plans for a new and improved version of the dataset.

In strategy games, one of the most important aspects of game design is maintaining a sense of challenge for players. Many mobile titles feature quick gameplay loops that allow players to progress steadily, requiring an abundance of levels and puzzles to prevent them from reaching the end too quickly. As with any content creation, testing and validation are essential to ensure engaging gameplay mechanics, enjoyable game assets, and playable levels. In this paper, we propose an automated approach that can be leveraged for gameplay testing and validation that combines traditional scripted methods with reinforcement learning, reaping the benefits of both approaches while adapting to new situations similarly to how a human player would. We test our solution on a popular tower defense game, Plants vs. Zombies. The results show that combining a learned approach, such as reinforcement learning, with a scripted AI produces a higher-performing and more robust agent than using only heuristic AI, achieving a 57.12% success rate compared to 47.95% in a set of 40 levels. Moreover, the results demonstrate the difficulty of training a general agent for this type of puzzle-like game.

Both entropy-minimizing and entropy-maximizing (curiosity) objectives for unsupervised reinforcement learning (RL) have been shown to be effective in different environments, depending on the environment's level of natural entropy. However, neither method alone results in an agent that will consistently learn intelligent behavior across environments. In an effort to find a single entropy-based method that will encourage emergent behaviors in any environment, we propose an agent that can adapt its objective online, depending on the entropy conditions by framing the choice as a multi-armed bandit problem. We devise a novel intrinsic feedback signal for the bandit, which captures the agent's ability to control the entropy in its environment. We demonstrate that such agents can learn to control entropy and exhibit emergent behaviors in both high- and low-entropy regimes and can learn skillful behaviors in benchmark tasks. Videos of the trained agents and summarized findings can be found on our project page https://sites.google.com/view/surprise-adaptive-agents

This is the essence of a planning problem, and one instance of planning problems of particular import is wargaming, which is a simulated military exercise to test strategies and operational plans in a controlled environment. Within the context of planning problems, it is easy to envision a variety of applications for AI, ranging from decision support systems (DSS), intelligent opponents, scenario generation, and so on. This work will focus primarily on agents usable for the first two applications: DSS and intelligent opponents. As a result, we seek not only agents that can select a good quality action, but also agents that can analyze multi-player interactions, adapt rapidly to changing conditions, and provide interpretable insights. The U.S. Department of Defense introduced Battlespace [5] as a platform for wargaming [6].

Given a network, allocating resources at clusters level, rather than at each node, enhances efficiency in resource allocation and usage. In this paper, we study the problem of finding fully connected disjoint clusters to minimize the intra-cluster distances and maximize the number of nodes assigned to the clusters, while also ensuring that no two nodes within a cluster exceed a threshold distance. While the problem can easily be formulated using a binary linear model, traditional combinatorial optimization solvers struggle when dealing with large-scale instances. We propose an approach to solve this constrained clustering problem via reinforcement learning. Our method involves training an agent to generate both feasible and (near) optimal solutions. The agent learns problem-specific heuristics, tailored to the instances encountered in this task. In the results section, we show that our algorithm finds near optimal solutions, even for large scale instances.