AI Agents

Recent embodied agents are primarily built based on reinforcement learning (RL) or large language models (LLMs). Among them, RL agents are efficient for deployment but only perform very few tasks. By contrast, giant LLM agents (often more than 1000B parameters) present strong generalization while demanding enormous computing resources. In this work, we combine their advantages while avoiding the drawbacks by conducting the proposed referee RL on our developed large auto-regressive model (LARM). Specifically, LARM is built upon a lightweight LLM (fewer than 5B parameters) and directly outputs the next action to execute rather than text. We mathematically reveal that classic RL feedbacks vanish in long-horizon embodied exploration and introduce a giant LLM based referee to handle this reward vanishment during training LARM. In this way, LARM learns to complete diverse open-world tasks without human intervention. Especially, LARM successfully harvests enchanted diamond equipment in Minecraft, which demands significantly longer decision-making chains than the highest achievements of prior best methods.

Large language models (LLMs) have revolutionized various AI applications. However, their billions of parameters pose significant challenges for practical deployment. Structured pruning is a hardware-friendly compression technique and receives widespread attention. Nonetheless, existing literature typically targets a single structure of LLMs. We observe that the structure units of LLMs differ in terms of inference cost and functionality. Therefore, pruning a single structure unit in isolation often results in an imbalance between performance and efficiency. In addition, previous works mainly employ a prescribed pruning ratio. Since the significance of LLM modules may vary, it is ideal to distribute the pruning load to a specific structure unit according to its role within LLMs. To address the two issues, we propose a pruning method that targets multiple LLM modules with dynamic pruning ratios. Specifically, we find the intrinsic properties of LLMs can guide us to determine the importance of each module and thus distribute the pruning load on demand, i.e., what to prune and how much to prune. This is achieved by quantifying the complex interactions within LLMs. Extensive experiments on multiple benchmarks and LLM variants demonstrate that our method effectively balances the trade-off between efficiency and performance.

Zero-shot coordination (ZSC), the ability to adapt to a new partner in a cooperative task, is a critical component of human-compatible AI. While prior work has focused on training agents to cooperate on a single task, these specialized models do not generalize to new tasks, even if they are highly similar. Here, we study how reinforcement learning on a **distribution of environments with a single partner** enables learning general cooperative skills that support ZSC with **many new partners on many new problems**. We introduce *two* Jax-based, procedural generators that create billions of solvable coordination challenges. We develop a new paradigm called **Cross-Environment Cooperation (CEC)**, and show that it outperforms competitive baselines quantitatively and qualitatively when collaborating with real people. Our findings suggest that learning to collaborate across many unique scenarios encourages agents to develop general norms, which prove effective for collaboration with different partners. Together, our results suggest a new route toward designing generalist cooperative agents capable of interacting with humans without requiring human data.

Recent research builds various patching agents that combine large language models (LLMs) with non-ML tools and achieve promising results on the state-of-the-art (SOTA) software patching benchmark, SWE-bench. Based on how to determine the patching workflows, existing patching agents can be categorized as agent-based planning methods, which rely on LLMs for planning, and rule-based planning methods, which follow a pre-defined workflow. At a high level, agent-based planning methods achieve high patching performance but with a high cost and limited stability. Rule-based planning methods, on the other hand, are more stable and efficient but have key workflow limitations that compromise their patching performance. In this paper, we propose PatchPilot, an agentic patcher that strikes a balance between patching efficacy, stability, and cost-efficiency. PatchPilot proposes a novel rule-based planning workflow with five components: reproduction, localization, generation, validation, and refinement (where refinement is unique to PatchPilot). We introduce novel and customized designs to each component to optimize their effectiveness and efficiency. Through extensive experiments on the SWE-bench benchmarks, PatchPilot shows a superior performance than existing open-source methods while maintaining low cost (less than 1\$ per instance) and ensuring higher stability. We also conduct a detailed ablation study to validate the key designs in each component. Our code is available at https://github.com/ucsb-mlsec/PatchPilot.

Large language models (LLMs) have shown remarkable advancements in enabling language agents to tackle simple tasks. However, applying them for complex, multi-step, long-horizon tasks remains a challenge. Recent work have found success by separating high-level planning from low-level execution, which enables the model to effectively balance high-level planning objectives and low-level execution details. However, generating accurate plans remains difficult since LLMs are not inherently trained for this task. To address this, we propose Plan-and-Act, a novel framework that incorporates explicit planning into LLM-based agents and introduces a scalable method to enhance plan generation through a novel synthetic data generation method. Plan-and-Act consists of a Planner model which generates structured, high-level plans to achieve user goals, and an Executor model that translates these plans into environment-specific actions. To train the Planner effectively, we introduce a synthetic data generation method that annotates ground-truth trajectories with feasible plans, augmented with diverse and extensive examples to enhance generalization. We evaluate Plan-and-Act using web navigation as a representative long-horizon planning environment, demonstrating a state-of-the-art 57.58% success rate on the WebArena-Lite benchmark as well as a text-only state-of-the-art 81.36% success rate on WebVoyager.

While numerous frameworks have been developed to enhance the reasoning abilities of large language models (LLMs), there is a scarcity of methods that effectively balance the trade-off between cost and quality. In this paper, we introduce Fleet of Agents (FoA), a novel and intuitive yet principled framework utilizing LLMs as agents to navigate through dynamic tree searches, employing a genetic-type particle filtering approach. FoA spawns a multitude of agents, each exploring the search space autonomously, followed by a selection phase where resampling based on a heuristic value function optimizes the balance between exploration and exploitation. This mechanism enables dynamic branching, adapting the exploration strategy based on discovered solutions. We conduct extensive experiments on four benchmark tasks, \``Game of 24\'', \``Mini-Crosswords\'', \``WebShop\'' and \``SciBench\'', utilizing four different LLMs, GPT-3.5, GPT-4, LLaMA3.2-11B, and LLaMA3.2-90B. On average across all tasks and LLMs, FoA obtains an absolute quality improvement of $\simeq 5\%$ while requiring only $\simeq 35\%$ of the cost of previous SOTA methods. Notably, our analyses reveal that (1) FoA achieves the best cost-quality trade-off among all benchmarked methods, and (2) FoA+ LLaMA3.2-11B surpasses the Llama3.2-90B model. FoA is publicly available at [https://github.com/au-clan/FoA](https://github.com/au-clan/FoA).

Behavioral diversity, expert imitation, fairness, safety goals and others give rise to preferences in sequential decision making domains that do not decompose additively across time. We introduce the class of convex Markov games that allow general convex preferences over occupancy measures. Despite infinite time horizon and strictly higher generality than Markov games, pure strategy Nash equilibria exist. Furthermore, equilibria can be approximated empirically by performing gradient descent on an upper bound of exploitability. Our experiments reveal novel solutions to classic repeated normal-form games, find fair solutions in a repeated asymmetric coordination game, and prioritize safe long-term behavior in a robot warehouse environment. In the prisoner's dilemma, our algorithm leverages transient imitation to find a policy profile that deviates from observed human play only slightly, yet achieves higher per-player utility while also being three orders of magnitude less exploitable.

We study the repeated principal-agent bandit game, where the principal indirectly explores an unknown environment by incentivizing an agent to play arms. Unlike prior work that assumes a greedy agent with full knowledge of reward means, we consider a self-interested learning agent who iteratively updates reward estimates and may explore arbitrarily with some probability. As a warm-up, we first consider a self-interested learning agent without exploration. We propose algorithms for both i.i.d. and linear reward settings with bandit feedback in a finite horizon $T$, achieving regret bounds of $\widetilde{O}(\sqrt{T})$ and $\widetilde{O}(T^{\frac{2}{3}})$, respectively. Specifically, these algorithms rely on a novel elimination framework coupled with new search algorithms which accommodate the uncertainty from the agent's learning behavior. We then extend the framework to handle an exploratory learning agent and develop an algorithm to achieve a $\widetilde{O}(T^{\frac{2}{3}})$ regret bound in i.i.d. reward setup by enhancing the robustness of our elimination framework to the potential agent exploration. Finally, when our agent model reduces to that in (Dogan et al., 2023a), we propose an algorithm based on our robust framework, which achieves a $\widetilde{O}(\sqrt{T})$ regret bound, significantly improving upon their $\widetilde{O}(T^{\frac{11}{12}})$ bound.

Recent advancements in Large Language Models (LLMs) have led to increasingly diverse requests, accompanied with varying resource (compute and memory) demands to serve them. However, this in turn degrades the cost-efficiency of LLM serving as common practices primarily rely on homogeneous GPU resources. In response to this problem, this work conducts a thorough study about serving LLMs over heterogeneous GPU resources on cloud platforms. The rationale is that different GPU types exhibit distinct compute and memory characteristics, aligning well with the divergent resource demands of diverse requests. Particularly, through comprehensive benchmarking, we discover that the cost-efficiency of LLM serving can be substantially optimized by meticulously determining GPU composition, deployment configurations, and workload assignments. Subsequently, we design a scheduling algorithm via mixed-integer linear programming, aiming at deducing the most cost-efficient serving plan under the constraints of price budget and real-time GPU availability. Remarkably, our approach effectively outperforms homogeneous and heterogeneous baselines under a wide array of scenarios, covering diverse workload traces, varying GPU availablilities, and multi-model serving. This casts new light on more accessible and efficient LLM serving over heterogeneous cloud resources.

World models (WMs) demonstrate strong capabilities in prediction, generation, and planning tasks. Existing WMs primarily focus on unstructured data while cannot leverage the ubiquitous structured data, often represented as graphs, in the digital world. While multiple graph foundation models have been proposed, they focus on graph learning tasks and cannot extend to diverse multi-modal data and interdisciplinary tasks. To address these challenges, we propose the Graph World Model (GWM), a world model that supports both unstructured and graph-structured states with multi-modal information and represents diverse tasks as actions. The core of a GWM is a generic message-passing algorithm to aggregate structured information, either over a unified multi-modal token space by converting multi-modal data into text (GWM-T) or a unified multi-modal embedding space by modality-specific encoders (GWM-E). Notably, GWM introduces action nodes to support diverse tasks, where action nodes are linked to other nodes via direct reference or similarity computation. Extensive experiments on 6 tasks from diverse domains, including multi-modal generation and matching, recommendation, graph prediction, multi-agent, retrieval-augmented generation, and planning and optimization, show that the same GWM outperforms or matches domain-specific baselines' performance, benefits from multi-hop structures, and demonstrate strong zero-shot/few-shot capabilities on unseen new tasks. Our codes for GWM is released at https://github.com/ulab-uiuc/GWM.

We introduce SWE-Lancer, a benchmark of over 1400 freelance software engineering tasks from Upwork, valued at \\\$1 million USD total in real-world payouts. SWE-Lancer encompasses both independent engineering tasks — ranging from \\\$50 bug fixes to \\\$32000 feature implementations — and managerial tasks, where models choose between technical implementation proposals. Independent tasks are graded with end-to-end tests triple-verified by experienced software engineers, while managerial decisions are assessed against the choices of the original hired engineering managers. We evaluate model performance and find that frontier models are still unable to solve the majority of tasks. To facilitate future research, we open-source a unified Docker image and a public evaluation split. By mapping model performance to monetary value, we hope SWE-Lancer enables greater research into the economic impact of AI model development.

We introduce PaperBench, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research. Agents must replicate 20 ICML 2024 Spotlight and Oral papers from scratch, including understanding paper contributions, developing a codebase, and successfully executing experiments. For objective evaluation, we develop rubrics that hierarchically decompose each replication task into smaller sub-tasks with clear grading criteria. In total, PaperBench contains 8,316 individually gradable tasks. Rubrics are co-developed with the author(s) of each ICML paper for accuracy and realism. To enable scalable evaluation, we also develop an LLM-based judge to automatically grade replication attempts against rubrics, and assess our judge's performance by creating a separate benchmark for judges. We evaluate several frontier models on PaperBench, finding that the best-performing tested agent, Claude 3.5 Sonnet (New) with open-source scaffolding, achieves an average replication score of 21.0%. Finally, we recruit top ML PhDs to attempt a subset of PaperBench, finding that models do not yet outperform the human baseline. We open-source our code (https://github.com/openai/preparedness) to facilitate future research in understanding the AI engineering capabilities of AI agents.

Self-attention has emerged as a core component of modern neural architectures, yet its theoretical underpinnings remain elusive. In this paper, we study self-attention through the lens of *interacting entities*, ranging from agents in multi-agent reinforcement learning to alleles in genetic sequences, and show that a single layer linear self-attention can *efficiently* represent, learn, and generalize functions capturing pairwise interactions, including out-of-distribution scenarios. Our analysis reveals that self-attention acts as a *mutual interaction learner* under minimal assumptions on the diversity of interaction patterns observed during training, thereby encompassing a wide variety of real-world domains. In addition, we validate our theoretical insights through experiments demonstrating that self-attention learns interaction functions and generalizes across both population distributions and out-of-distribution scenarios. Building on our theories, we introduce *HyperFeatureAttention*, a novel neural network module designed to learn couplings of different feature-level interactions between entities. Furthermore, we propose *HyperAttention*, a new module that extends beyond pairwise interactions to capture multi-entity dependencies, such as three-way, four-way, or general $n$-way interactions.

Large Language Models' safety-aligned behaviors, such as refusing harmful queries, can be represented by linear directions in activation space. Previous research modeled safety behavior with a single direction, limiting mechanistic understanding to an isolated safety feature. In this work, we discover that safety-aligned behavior is jointly controlled by multi-dimensional directions. Namely, we study the vector space of representation shifts during safety fine-tuning on Llama 3 8B for refusing jailbreaks. By studying orthogonal directions in the space, we first find that a dominant direction governs the model's refusal behavior, while multiple smaller directions represent distinct and interpretable features like hypothetical narrative and role-playing. We then measure how different directions promote or suppress the dominant direction, showing the important role of secondary directions in shaping the model's refusal representation. Finally, we demonstrate that removing certain trigger tokens in harmful queries can mitigate these directions to bypass the learned safety capability, providing new insights on understanding safety alignment vulnerability from a multi-dimensional perspective.

Generalizing multi-agent reinforcement learning (MARL) to accommodate variations in problem configurations remains a critical challenge in real-world applications, where even subtle differences in task setups can cause pre-trained policies to fail. To address this, we propose Context-Aware Identity Generation (CAID), a novel framework to enhance MARL performance under the Contextual MARL (CMARL) setting. CAID dynamically generates unique agent identities through the agent identity decoder built on a causal Transformer architecture. These identities provide contextualized representations that align corresponding agents across similar problem variants, facilitating policy reuse and improving sample efficiency. Furthermore, the action regulator in CAID incorporates these agent identities into the action-value space, enabling seamless adaptation to varying contexts. Extensive experiments on CMARL benchmarks demonstrate that CAID significantly outperforms existing approaches by enhancing both sample efficiency and generalization across diverse context variants.

Large Language Models (LLMs) have demonstrated the ability to solve a wide range of practical tasks within multi-agent systems. However, existing human-designed multi-agent frameworks are typically limited to a small set of pre-defined scenarios, while current automated design methods suffer from several limitations, such as the lack of tool integration, dependence on external training data, and rigid communication structures. In this paper, we propose \textbf{MetaAgent}, a \textbf{finite state machine} based framework that can automatically generate a multi-agent system. Given a task description, MetaAgent will design a multi-agent system and polish it through an optimization algorithm. When the multi-agent system is deployed, the finite state machine will control the agent's actions and the state transitions. To evaluate our framework, we conduct experiments on both text-based tasks and practical tasks. The results indicate that the generated multi-agent system surpasses other auto-designed methods and can achieve a comparable performance with the human-designed multi-agent system, which is optimized for those specific tasks.

Time-series Generation (TSG) is a prominent research area with broad applications in simulations, data augmentation, and counterfactual analysis. While existing methods have shown promise in unconditional single-domain TSG, real-world applications demand for cross-domain approaches capable of controlled generation tailored to domain-specific constraints and instance-level requirements. In this paper, we argue that text can provide semantic insights, domain information and instance-specific temporal patterns, to guide and improve TSG. We introduce ``Text-Controlled TSG'', a task focused on generating realistic time series by incorporating textual descriptions. To address data scarcity in this setting, we propose a novel LLM-based Multi-Agent framework that synthesizes diverse, realistic text-to-TS datasets. Furthermore, we introduce Bridge, a hybrid text-controlled TSG framework that integrates semantic prototypes with text description for supporting domain-level guidance. This approach achieves state-of-the-art generation fidelity on 11 of 12 datasets, and improves controllability by up to 12% on MSE and 6% MAE compared to no text input generation, highlighting its potential for generating tailored time-series data.

A generalist foundation model agent needs to have a large and diverse skill repertoire, such as finding directions between two travel locations and buying specific items from the Internet. If each skill needs to be specified manually through a fixed set of human-annotated instructions, the agent’s skill repertoire will necessarily be limited due to the scalability of human-annotated instructions. In this work, we address this challenge by proposing Proposer-Agent-Evaluator (PAE), an effective learning system that enables foundation model agents to autonomously discover and practice skills in the wild. After a context-aware task proposer generates instructions based on website information, the agent policy attempts those tasks in the real world with resulting trajectories evaluated by an autonomous VLM-based success evaluator. The success evaluation serves as the reward signal for the agent to refine its policies through RL. We validate PAE on challenging vision-based web navigation, using both real-world and selfhosted websites from WebVoyager and WebArena. Our results show that PAE significantly improves the zero-shot generalization capability of VLM Internet agents (around 50% relative improvement) to both unseen tasks and websites.

Developing autonomous agents that quickly explore an environment and adapt their behavior online is a canonical challenge in robotics and machine learning. While humans are able to achieve such fast online exploration and adaptation, often acquiring new information and skills in only a handful of interactions, existing algorithmic approaches tend to rely on random exploration and slow, gradient-based behavior updates. How can we endow autonomous agents with such capabilities on par with humans? Taking inspiration from recent progress on both in-context learning and large-scale behavioral cloning, in this work we propose behavioral exploration: training agents to internalize what it means to explore and adapt in-context over the space of ''expert'' behaviors. To achieve this, given access to a dataset of expert demonstrations, we train a long-context generative model to predict expert actions conditioned on a context of past observations and a measure of how ''exploratory'' the expert's behaviors are relative to this context. This enables the model to not only mimic the behavior of an expert, but also, by feeding its past history of interactions into its context, to select different expert behaviors than what have been previously selected, thereby allowing for fast online adaptation and targeted, ''expert-like'' exploration. We demonstrate the effectiveness of our method in both simulated locomotion and manipulation settings, as well as on real-world robotic manipulation tasks, illustrating its ability to learn adaptive, exploratory behavior.

The ability of agents to collaborate with previously unknown teammates on the fly, known as ad hoc teamwork (AHT), is crucial in many real-world applications. Existing approaches to AHT require online interactions with the environment and some carefully designed teammates. However, these prerequisites can be infeasible in practice. In this work, we extend the AHT problem to the offline setting, where the policy of the ego agent is directly learned from a multi-agent interaction dataset. We propose a hierarchical sequence modeling framework called TAGET that addresses critical challenges in the offline setting, including limited data, partial observability and online adaptation. The core idea of TAGET is to dynamically predict teammate-aware rewards-to-go and sub-goals, so that the ego agent can adapt to the changes of teammates’ behaviors in real time. Extensive experimental results show that TAGET significantly outperforms existing solutions to AHT in the offline setting.