Despite the rapid growth of machine learning research, corresponding code implementations are often unavailable, making it slow and labor-intensive for researchers to reproduce results and build upon prior work. In the meantime, recent Large Language Models (LLMs) excel at understanding scientific documents and generating high-quality code. Inspired by this, we introduce PaperCoder, a multi-agent LLM framework that transforms machine learning papers into functional code repositories. PaperCoder operates in three stages: planning, where it constructs a high-level roadmap, designs the system architecture with diagrams, identifies file dependencies, and generates configuration files; analysis, which focuses on interpreting implementation-specific details; and generation, where modular, dependency-aware code is produced. Moreover, each phase is instantiated through a set of specialized agents designed to collaborate effectively across the pipeline. We then evaluate PaperCoder on generating code implementations from machine learning papers based on both model-based and human evaluations, specifically from the original paper authors, with author-released repositories as ground truth if available. Our results demonstrate the effectiveness of PaperCoder in creating high-quality, faithful implementations. Furthermore, it consistently shows strengths in the recently released PaperBench benchmark, surpassing strong baselines by substantial margins.
Robots need task planning methods to achieve goals that require more than individual actions. Recently, large language models (LLMs) have demonstrated impressive performance in task planning. LLMs can generate a step-by-step solution using a description of actions and the goal. Despite the successes in LLM-based task planning, there is limited research studying the security aspects of those systems. In this paper, we develop Robo-Troj, the first multi-trigger backdoor attack for LLM-based task planners, which is the main contribution of this work. As a multi-trigger attack, Robo-Troj is trained to accommodate the diversity of robot application domains. For instance, one can use unique trigger words, e.g., "herical", to activate a specific malicious behavior, e.g., cutting hand on a kitchen robot. In addition, we develop an optimization method for selecting the trigger words that are most effective. Through demonstrating the vulnerability of LLM-based planners, we aim to promote the development of secured robot systems.
Optimization plays a vital role in scientific research and practical applications, but formulating a concrete optimization problem described in natural language into a mathematical form and selecting a suitable solver to solve the problem requires substantial domain expertise. We introduce \textbf{OptimAI}, a framework for solving \underline{Optim}ization problems described in natural language by leveraging LLM-powered \underline{AI} agents, achieving superior performance over current state-of-the-art methods. Our framework is built upon four key roles: (1) a \emph{formulator} that translates natural language problem descriptions into precise mathematical formulations; (2) a \emph{planner} that constructs a high-level solution strategy prior to execution; and (3) a \emph{coder} and a \emph{code critic} capable of interacting with the environment and reflecting on outcomes to refine future actions. Ablation studies confirm that all roles are essential; removing the planner or code critic results in $5.8\times$ and $3.1\times$ drops in productivity, respectively. Furthermore, we introduce UCB-based debug scheduling to dynamically switch between alternative plans, yielding an additional $3.3\times$ productivity gain. Our design emphasizes multi-agent collaboration, allowing us to conveniently explore the synergistic effect of combining diverse models within a unified system. Our approach attains 88.1\% accuracy on the NLP4LP dataset and 71.2\% on the Optibench (non-linear w/o table) subset, reducing error rates by 58\% and 50\% respectively over prior best results.
Target-Oriented Dialogue (TOD) remains a significant challenge in the LLM era, where strategic dialogue planning is crucial for directing conversations toward specific targets. However, existing dialogue planning methods generate dialogue plans in a step-by-step sequential manner, and may suffer from compounding errors and myopic actions. To address these limitations, we introduce a novel dialogue planning framework, DiffTOD, which leverages diffusion models to enable non-sequential dialogue planning. DiffTOD formulates dialogue planning as a trajectory generation problem with conditional guidance, and leverages a diffusion language model to estimate the likelihood of the dialogue trajectory. To optimize the dialogue action strategies, DiffTOD introduces three tailored guidance mechanisms for different target types, offering flexible guidance towards diverse TOD targets at test time. Extensive experiments across three diverse TOD settings show that DiffTOD can effectively perform non-myopic lookahead exploration and optimize action strategies over a long horizon through non-sequential dialogue planning, and demonstrates strong flexibility across complex and diverse dialogue scenarios. Our code and data are accessible through https://anonymous.4open.science/r/DiffTOD.
Text-based games provide valuable environments for language-based autonomous agents. However, planning-then-learning paradigms, such as those combining Monte Carlo Tree Search (MCTS) and reinforcement learning (RL), are notably time-consuming due to extensive iterations. Additionally, these algorithms perform uncertainty-driven exploration but lack language understanding and reasoning abilities. In this paper, we introduce the Monte Carlo planning with Dynamic Memory-guided Large language model (MC-DML) algorithm. MC-DML leverages the language understanding and reasoning capabilities of Large Language Models (LLMs) alongside the exploratory advantages of tree search algorithms. Specifically, we enhance LLMs with in-trial and cross-trial memory mechanisms, enabling them to learn from past experiences and dynamically adjust action evaluations during planning. We conduct experiments on a series of text-based games from the Jericho benchmark. Our results demonstrate that the MC-DML algorithm significantly enhances performance across various games at the initial planning phase, outperforming strong contemporary methods that require multiple iterations. This demonstrates the effectiveness of our algorithm, paving the way for more efficient language-grounded planning in complex environments.
Intelligent agent systems based on Large Language Models (LLMs) have shown great potential in real-world applications. However, existing agent frameworks still face critical limitations in task planning and execution, restricting their effectiveness and generalizability. Specifically, current planning methods often lack clear global goals, leading agents to get stuck in local branches, or produce non-executable plans. Meanwhile, existing execution mechanisms struggle to balance complexity and stability, and their limited action space restricts their ability to handle diverse real-world tasks. To address these limitations, we propose GoalAct, a novel agent framework that introduces a continuously updated global planning mechanism and integrates a hierarchical execution strategy. GoalAct decomposes task execution into high-level skills, including searching, coding, writing and more, thereby reducing planning complexity while enhancing the agents' adaptability across diverse task scenarios. We evaluate GoalAct on LegalAgentBench, a benchmark with multiple types of legal tasks that require the use of multiple types of tools. Experimental results demonstrate that GoalAct achieves state-of-the-art (SOTA) performance, with an average improvement of 12.22% in success rate. These findings highlight GoalAct's potential to drive the development of more advanced intelligent agent systems, making them more effective across complex real-world applications. Our code can be found at https://github.com/cjj826/GoalAct.
Automation of code reviews using AI models has garnered substantial attention in the software engineering community as a strategy to reduce the cost and effort associated with traditional peer review processes. These models are typically trained on extensive datasets of real-world code reviews that address diverse software development concerns, including testing, refactoring, bug fixes, performance optimization, and maintainability improvements. However, a notable limitation of these datasets is the under representation of code vulnerabilities, critical flaws that pose significant security risks, with security-focused reviews comprising a small fraction of the data. This scarcity of vulnerability-specific data restricts the effectiveness of AI models in identifying and commenting on security-critical code. To address this issue, we propose the creation of a synthetic dataset consisting of vulnerability-focused reviews that specifically comment on security flaws. Our approach leverages Large Language Models (LLMs) to generate human-like code review comments for vulnerabilities, using insights derived from code differences and commit messages. To evaluate the usefulness of the generated synthetic dataset, we plan to use it to fine-tune three existing code review models. We anticipate that the synthetic dataset will improve the performance of the original code review models.
Can we build accurate world models out of large language models (LLMs)? How can world models benefit LLM agents? The gap between the prior knowledge of LLMs and the specified environment's dynamics usually bottlenecks LLMs' performance as world models. To bridge the gap, we propose a training-free "world alignment" that learns an environment's symbolic knowledge complementary to LLMs. The symbolic knowledge covers action rules, knowledge graphs, and scene graphs, which are extracted by LLMs from exploration trajectories and encoded into executable codes to regulate LLM agents' policies. We further propose an RL-free, model-based agent "WALL-E 2.0" through the model-predictive control (MPC) framework. Unlike classical MPC requiring costly optimization on the fly, we adopt an LLM agent as an efficient look-ahead optimizer of future steps' actions by interacting with the neurosymbolic world model. While the LLM agent's strong heuristics make it an efficient planner in MPC, the quality of its planned actions is also secured by the accurate predictions of the aligned world model. They together considerably improve learning efficiency in a new environment. On open-world challenges in Mars (Minecraft like) and ALFWorld (embodied indoor environments), WALL-E 2.0 significantly outperforms existing methods, e.g., surpassing baselines in Mars by 16.1%-51.6% of success rate and by at least 61.7% in score. In ALFWorld, it achieves a new record 98% success rate after only 4 iterations.
Large language models (LLMs) with different architectures and sizes have been developed. Serving each LLM with dedicated GPUs leads to resource waste and service inefficiency due to the varying demand of LLM requests. A common practice is to share multiple LLMs. However, existing sharing systems either do not consider the autoregressive pattern of LLM services, or only focus on improving the throughput, which impairs the sharing performance, especially the serving latency. We present SeaLLM, which enables service-aware and latency-optimized LLM sharing. SeaLLM improves the overall sharing performance by (1) a latency-optimized scheduling algorithm utilizing the characteristics of LLM services, (2) a placement algorithm to determine the placement plan and an adaptive replacement algorithm to decide the replacement interval, and (3) a unified key-value cache to share GPU memory among LLM services efficiently. Our evaluation under real-world traces and LLM services demonstrates that SeaLLM improves the normalized latency by up to $13.60\times$, the tail latency by up to $18.69\times$, and the SLO attainment by up to $3.64\times$ compared to existing solutions.
The current study describes a cost-effective method for adapting large language models (LLMs) for academic advising with study-abroad contexts in mind and for application in low-resource methods for acculturation. With the Mistral-7B-Instruct model applied with a Low-Rank Adaptation (LoRA) method and a 4-bit quantization method, the model underwent training in two distinct stages related to this study's purpose to enhance domain specificity while maintaining computational efficiency. In Phase 1, the model was conditioned with a synthetic dataset via the Gemini Pro API, and in Phase 2, it was trained with manually curated datasets from the StudyAbroadGPT project to achieve enhanced, contextualized responses. Technical innovations entailed memory-efficient quantization, parameter-efficient adaptation, and continuous training analytics via Weights & Biases. After training, this study demonstrated a reduction in training loss by 52.7%, 92% accuracy in domain-specific recommendations, achieved 95% markdown-based formatting support, and a median run-rate of 100 samples per second on off-the-shelf GPU equipment. These findings support the effective application of instruction-tuned LLMs within educational advisers, especially in low-resource institutional scenarios. Limitations included decreased generalizability and the application of a synthetically generated dataset, but this framework is scalable for adding new multilingual-augmented and real-time academic advising processes. Future directions may include plans for the integration of retrieval-augmented generation, applying dynamic quantization routines, and connecting to real-time academic databases to increase adaptability and accuracy.
While large language models (LLMs) have been widely applied to code generation, they struggle with generating entire deep learning projects, which are characterized by complex structures, longer functions, and stronger reliance on domain knowledge than general-purpose code. An open-domain LLM often lacks coherent contextual guidance and domain expertise for specific projects, making it challenging to produce complete code that fully meets user requirements. In this paper, we propose a novel planning-guided code generation method, DLCodeGen, tailored for generating deep learning projects. DLCodeGen predicts a structured solution plan, offering global guidance for LLMs to generate the project. The generated plan is then leveraged to retrieve semantically analogous code samples and subsequently abstract a code template. To effectively integrate these multiple retrieval-augmented techniques, a comparative learning mechanism is designed to generate the final code. We validate the effectiveness of our approach on a dataset we build for deep learning code generation. Experimental results demonstrate that DLCodeGen outperforms other baselines, achieving improvements of 9.7% in CodeBLEU and 3.6% in human evaluation metrics.
Compositional text-to-video generation, which requires synthesizing dynamic scenes with multiple interacting entities and precise spatial-temporal relationships, remains a critical challenge for diffusion-based models. Existing methods struggle with layout discontinuity, entity identity drift, and implausible interaction dynamics due to unconstrained cross-attention mechanisms and inadequate physics-aware reasoning. To address these limitations, we propose DyST-XL, a \textbf{training-free} framework that enhances off-the-shelf text-to-video models (e.g., CogVideoX-5B) through frame-aware control. DyST-XL integrates three key innovations: (1) A Dynamic Layout Planner that leverages large language models (LLMs) to parse input prompts into entity-attribute graphs and generates physics-aware keyframe layouts, with intermediate frames interpolated via trajectory optimization; (2) A Dual-Prompt Controlled Attention Mechanism that enforces localized text-video alignment through frame-aware attention masking, achieving the precise control over individual entities; and (3) An Entity-Consistency Constraint strategy that propagates first-frame feature embeddings to subsequent frames during denoising, preserving object identity without manual annotation. Experiments demonstrate that DyST-XL excels in compositional text-to-video generation, significantly improving performance on complex prompts and bridging a crucial gap in training-free video synthesis.
Planning is central to agents and agentic AI. The ability to plan, e.g., creating travel itineraries within a budget, holds immense potential in both scientific and commercial contexts. Moreover, optimal plans tend to require fewer resources compared to ad-hoc methods. To date, a comprehensive understanding of existing planning benchmarks appears to be lacking. Without it, comparing planning algorithms' performance across domains or selecting suitable algorithms for new scenarios remains challenging. In this paper, we examine a range of planning benchmarks to identify commonly used testbeds for algorithm development and highlight potential gaps. These benchmarks are categorized into embodied environments, web navigation, scheduling, games and puzzles, and everyday task automation. Our study recommends the most appropriate benchmarks for various algorithms and offers insights to guide future benchmark development.
Existing LLM-enabled multi-agent frameworks are predominantly limited to digital or simulated environments and confined to narrowly focused knowledge domain, constraining their applicability to complex engineering tasks that require the design of physical embodiment, cross-disciplinary integration, and constraint-aware reasoning. This work proposes a multi-agent autonomous mechatronics design framework, integrating expertise across mechanical design, optimization, electronics, and software engineering to autonomously generate functional prototypes with minimal direct human design input. Operating primarily through a language-driven workflow, the framework incorporates structured human feedback to ensure robust performance under real-world constraints. To validate its capabilities, the framework is applied to a real-world challenge involving autonomous water-quality monitoring and sampling, where traditional methods are labor-intensive and ecologically disruptive. Leveraging the proposed system, a fully functional autonomous vessel was developed with optimized propulsion, cost-effective electronics, and advanced control. The design process was carried out by specialized agents, including a high-level planning agent responsible for problem abstraction and dedicated agents for structural, electronics, control, and software development. This approach demonstrates the potential of LLM-based multi-agent systems to automate real-world engineering workflows and reduce reliance on extensive domain expertise.
Large Language Models (LLMs) exhibit substantial promise in enhancing task-planning capabilities within embodied agents due to their advanced reasoning and comprehension. However, the systemic safety of these agents remains an underexplored frontier. In this study, we present Safe-BeAl, an integrated framework for the measurement (SafePlan-Bench) and alignment (Safe-Align) of LLM-based embodied agents' behaviors. SafePlan-Bench establishes a comprehensive benchmark for evaluating task-planning safety, encompassing 2,027 daily tasks and corresponding environments distributed across 8 distinct hazard categories (e.g., Fire Hazard). Our empirical analysis reveals that even in the absence of adversarial inputs or malicious intent, LLM-based agents can exhibit unsafe behaviors. To mitigate these hazards, we propose Safe-Align, a method designed to integrate physical-world safety knowledge into LLM-based embodied agents while maintaining task-specific performance. Experiments across a variety of settings demonstrate that Safe-BeAl provides comprehensive safety validation, improving safety by 8.55 - 15.22%, compared to embodied agents based on GPT-4, while ensuring successful task completion.
Recent Computer-Using Agents (CUAs), powered by multimodal large language models (LLMs), offer a promising direction for automating complex desktop workflows through natural language. However, most existing CUAs remain conceptual prototypes, hindered by shallow OS integration, fragile screenshot-based interaction, and disruptive execution. We present UFO2, a multiagent AgentOS for Windows desktops that elevates CUAs into practical, system-level automation. UFO2 features a centralized HostAgent for task decomposition and coordination, alongside a collection of application-specialized AppAgent equipped with native APIs, domain-specific knowledge, and a unified GUI--API action layer. This architecture enables robust task execution while preserving modularity and extensibility. A hybrid control detection pipeline fuses Windows UI Automation (UIA) with vision-based parsing to support diverse interface styles. Runtime efficiency is further enhanced through speculative multi-action planning, reducing per-step LLM overhead. Finally, a Picture-in-Picture (PiP) interface enables automation within an isolated virtual desktop, allowing agents and users to operate concurrently without interference. We evaluate UFO2 across over 20 real-world Windows applications, demonstrating substantial improvements in robustness and execution accuracy over prior CUAs. Our results show that deep OS integration unlocks a scalable path toward reliable, user-aligned desktop automation.
Multi-agents has exhibited significant intelligence in real-word simulations with Large language models (LLMs) due to the capabilities of social cognition and knowledge retrieval. However, existing research on agents equipped with effective cognition chains including reasoning, planning, decision-making and reflecting remains limited, especially in the dynamically interactive scenarios. In addition, unlike human, prompt-based responses face challenges in psychological state perception and empirical calibration during uncertain gaming process, which can inevitably lead to cognition bias. In light of above, we introduce PolicyEvol-Agent, a comprehensive LLM-empowered framework characterized by systematically acquiring intentions of others and adaptively optimizing irrational strategies for continual enhancement. Specifically, PolicyEvol-Agent first obtains reflective expertise patterns and then integrates a range of cognitive operations with Theory of Mind alongside internal and external perspectives. Simulation results, outperforming RL-based models and agent-based methods, demonstrate the superiority of PolicyEvol-Agent for final gaming victory. Moreover, the policy evolution mechanism reveals the effectiveness of dynamic guideline adjustments in both automatic and human evaluation.
Social simulation through large language model (LLM) agents is a promising approach to explore and validate hypotheses related to social science questions and LLM agents behavior. We present SOTOPIA-S4, a fast, flexible, and scalable social simulation system that addresses the technical barriers of current frameworks while enabling practitioners to generate multi-turn and multi-party LLM-based interactions with customizable evaluation metrics for hypothesis testing. SOTOPIA-S4 comes as a pip package that contains a simulation engine, an API server with flexible RESTful APIs for simulation management, and a web interface that enables both technical and non-technical users to design, run, and analyze simulations without programming. We demonstrate the usefulness of SOTOPIA-S4 with two use cases involving dyadic hiring negotiation and multi-party planning scenarios.
This work defends the 'Whole Hog Thesis': sophisticated Large Language Models (LLMs) like ChatGPT are full-blown linguistic and cognitive agents, possessing understanding, beliefs, desires, knowledge, and intentions. We argue against prevailing methodologies in AI philosophy, rejecting starting points based on low-level computational details ('Just an X' fallacy) or pre-existing theories of mind. Instead, we advocate starting with simple, high-level observations of LLM behavior (e.g., answering questions, making suggestions) -- defending this data against charges of metaphor, loose talk, or pretense. From these observations, we employ 'Holistic Network Assumptions' -- plausible connections between mental capacities (e.g., answering implies knowledge, knowledge implies belief, action implies intention) -- to argue for the full suite of cognitive states. We systematically rebut objections based on LLM failures (hallucinations, planning/reasoning errors), arguing these don't preclude agency, often mirroring human fallibility. We address numerous 'Games of Lacks', arguing that LLMs do not lack purported necessary conditions for cognition (e.g., semantic grounding, embodiment, justification, intrinsic intentionality) or that these conditions are not truly necessary, often relying on anti-discriminatory arguments comparing LLMs to diverse human capacities. Our approach is evidential, not functionalist, and deliberately excludes consciousness. We conclude by speculating on the possibility of LLMs possessing 'alien' contents beyond human conceptual schemes.
Large language models (LLMs) have demonstrated strong capabilities in code generation, underscoring the critical need for rigorous and comprehensive evaluation. Existing evaluation approaches fall into three categories, including human-centered, metric-based, and LLM-based. Considering that human-centered approaches are labour-intensive and metric-based ones overly rely on reference answers, LLM-based approaches are gaining increasing attention due to their stronger contextual understanding capabilities and superior efficiency. However, the performance of LLM-based approaches remains limited due to: (1) lack of multisource domain knowledge, and (2) insufficient comprehension of complex code. To mitigate the limitations, we propose CodeVisionary, the first LLM-based agent framework for evaluating LLMs in code generation. CodeVisionary consists of two stages: (1) Multiscore knowledge analysis stage, which aims to gather multisource and comprehensive domain knowledge by formulating and executing a stepwise evaluation plan. (2) Negotiation-based scoring stage, which involves multiple judges engaging in discussions to better comprehend the complex code and reach a consensus on the evaluation score. Extensive experiments demonstrate that CodeVisionary achieves the best performance for evaluating LLMs in code generation, outperforming the best baseline methods with average improvements of 0.202, 0.139, and 0.117 in Pearson, Spearman, and Kendall-Tau coefficients, respectively. Besides, CodeVisionary provides detailed evaluation reports, which assist developers in identifying shortcomings and making improvements. The resources of CodeVisionary are available at https://anonymous.4open.science/r/CodeVisionary.
Large Language Models (LLMs) have shown tremendous potential as agents, excelling at tasks that require multiple rounds of reasoning and interactions. Rejection Sampling Fine-Tuning (RFT) has emerged as an effective method for finetuning LLMs as agents: it first imitates expert-generated successful trajectories and further improves agentic skills through iterative fine-tuning on successful, self-generated trajectories. However, since the expert (e.g., GPT-4) succeeds primarily on simpler subtasks and RFT inherently favors simpler scenarios, many complex subtasks remain unsolved and persistently out-of-distribution (OOD). Upon investigating these challenging subtasks, we discovered that previously failed expert trajectories can often provide valuable guidance, e.g., plans and key actions, that can significantly improve agent exploration efficiency and acquisition of critical skills. Motivated by these observations, we propose Exploring Expert Failures (EEF), which identifies beneficial actions from failed expert trajectories and integrates them into the training dataset. Potentially harmful actions are meticulously excluded to prevent contamination of the model learning process. By leveraging the beneficial actions in expert failures, EEF successfully solves some previously unsolvable subtasks and improves agent tuning performance. Remarkably, our approach achieved a 62\% win rate in WebShop, outperforming RFT (53. 6\%) and GPT-4 (35. 6\%), and to the best of our knowledge, setting a new state-of-the-art as the first method to surpass a score of 0.81 in WebShop and exceed 81 in SciWorld.
Recent advancements in large language models (LLMs) have enabled their use as agents for planning complex tasks. Existing methods typically rely on a thought-action-observation (TAO) process to enhance LLM performance, but these approaches are often constrained by the LLMs' limited knowledge of complex tasks. Retrieval-augmented generation (RAG) offers new opportunities by leveraging external databases to ground generation in retrieved information. In this paper, we identify two key challenges (enlargability and transferability) in applying RAG to task planning. We propose InstructRAG, a novel solution within a multi-agent meta-reinforcement learning framework, to address these challenges. InstructRAG includes a graph to organize past instruction paths (sequences of correct actions), an RL-Agent with Reinforcement Learning to expand graph coverage for enlargability, and an ML-Agent with Meta-Learning to improve task generalization for transferability. The two agents are trained end-to-end to optimize overall planning performance. Our experiments on four widely used task planning datasets demonstrate that InstructRAG significantly enhances performance and adapts efficiently to new tasks, achieving up to a 19.2% improvement over the best existing approach.
Industrial dashboards, commonly deployed by organizations such as enterprises and governments, are increasingly crucial in data communication and decision-making support across various domains. Designing an industrial dashboard prototype is particularly challenging due to its visual complexity, which can include data visualization, layout configuration, embellishments, and animations. Additionally, in real-world industrial settings, designers often encounter numerous constraints. For instance, when companies negotiate collaborations with clients and determine design plans, they typically need to demo design prototypes and iterate on them based on mock data quickly. Such a task is very common and crucial during the ideation stage, as it not only helps save developmental costs but also avoids data-related issues such as lengthy data handover periods. However, existing authoring tools of dashboards are mostly not tailored to such prototyping needs, and motivated by these gaps, we propose DashChat, an interactive system that leverages large language models (LLMs) to generate industrial dashboard design prototypes from natural language. We collaborated closely with designers from the industry and derived the requirements based on their practical experience. First, by analyzing 114 high-quality industrial dashboards, we summarized their common design patterns and inject the identified ones into LLMs as reference. Next, we built a multi-agent pipeline powered by LLMs to understand textual requirements from users and generate practical, aesthetic prototypes. Besides, functionally distinct, parallel-operating agents are created to enable efficient generation. Then, we developed a user-friendly interface that supports text-based interaction for generating and modifying prototypes. Two user studies demonstrated that our system is both effective and efficient in supporting design prototyping.
Manual relevance judgements in Information Retrieval are costly and require expertise, driving interest in using Large Language Models (LLMs) for automatic assessment. While LLMs have shown promise in general web search scenarios, their effectiveness for evaluating domain-specific search results, such as educational resources, remains unexplored. To investigate different ways of including domain-specific criteria in LLM prompts for relevance judgement, we collected and released a dataset of 401 human relevance judgements from a user study involving teaching professionals performing search tasks related to lesson planning. We compared three approaches to structuring these prompts: a simple two-aspect evaluation baseline from prior work on using LLMs as relevance judges, a comprehensive 12-dimensional rubric derived from educational literature, and criteria directly informed by the study participants. Using domain-specific frameworks, LLMs achieved strong agreement with human judgements (Cohen's $\kappa$ up to 0.650), significantly outperforming the baseline approach. The participant-derived framework proved particularly robust, with GPT-3.5 achieving $\kappa$ scores of 0.639 and 0.613 for 10-dimension and 5-dimension versions respectively. System-level evaluation showed that LLM judgements reliably identified top-performing retrieval approaches (RBO scores 0.71-0.76) while maintaining reasonable discrimination between systems (RBO 0.52-0.56). These findings suggest that LLMs can effectively evaluate educational resources when prompted with domain-specific criteria, though performance varies with framework complexity and input structure.
Evaluating the capabilities of Large Language Models (LLMs) has traditionally relied on static benchmark datasets, human assessments, or model-based evaluations - methods that often suffer from overfitting, high costs, and biases. ZeroSumEval is a novel competition-based evaluation protocol that leverages zero-sum games to assess LLMs with dynamic benchmarks that resist saturation. ZeroSumEval encompasses a diverse suite of games, including security challenges (PyJail), classic games (Chess, Liar's Dice, Poker), knowledge tests (MathQuiz), and persuasion challenges (Gandalf, Debate). These games are designed to evaluate a range of AI capabilities such as strategic reasoning, planning, knowledge application, and creativity. Building upon recent studies that highlight the effectiveness of game-based evaluations for LLMs, ZeroSumEval enhances these approaches by providing a standardized and extensible framework. To demonstrate this, we conduct extensive experiments with >7000 simulations across 7 games and 13 models. Our results show that while frontier models from the GPT and Claude families can play common games and answer questions, they struggle to play games that require creating novel and challenging questions. We also observe that models cannot reliably jailbreak each other and fail generally at tasks requiring creativity. We release our code at https://github.com/facebookresearch/ZeroSumEval.
Personal development through self-directed learning is essential in today's fast-changing world, but many learners struggle to manage it effectively. While AI tools like large language models (LLMs) have the potential for personalized learning planning, they face issues such as transparency and hallucinated information. To address this, we propose PlanGlow, an LLM-based system that generates personalized, well-structured study plans with clear explanations and controllability through user-centered interactions. Through mixed methods, we surveyed 28 participants and interviewed 10 before development, followed by a within-subject experiment with 24 participants to evaluate PlanGlow's performance, usability, controllability, and explainability against two baseline systems: a GPT-4o-based system and Khan Academy's Khanmigo. Results demonstrate that PlanGlow significantly improves usability, explainability, and controllability. Additionally, two educational experts assessed and confirmed the quality of the generated study plans. These findings highlight PlanGlow's potential to enhance personalized learning and address key challenges in self-directed learning.
To what extent do LLMs use their capabilities towards their given goal? We take this as a measure of their goal-directedness. We evaluate goal-directedness on tasks that require information gathering, cognitive effort, and plan execution, where we use subtasks to infer each model's relevant capabilities. Our evaluations of LLMs from Google DeepMind, OpenAI, and Anthropic show that goal-directedness is relatively consistent across tasks, differs from task performance, and is only moderately sensitive to motivational prompts. Notably, most models are not fully goal-directed. We hope our goal-directedness evaluations will enable better monitoring of LLM progress, and enable more deliberate design choices of agentic properties in LLMs.
Emotional support conversation (ESC) aims to alleviate the emotional distress of individuals through effective conversations. Although large language models (LLMs) have obtained remarkable progress on ESC, most of these studies might not define the diagram from the state model perspective, therefore providing a suboptimal solution for long-term satisfaction. To address such an issue, we leverage the Finite State Machine (FSM) on LLMs, and propose a framework called FiSMiness. Our framework allows a single LLM to bootstrap the planning during ESC, and self-reason the seeker's emotion, support strategy and the final response upon each conversational turn. Substantial experiments on ESC datasets suggest that FiSMiness outperforms many baselines, including direct inference, self-refine, chain of thought, finetuning, and external-assisted methods, even those with many more parameters.
Large language model (LLM)-based inference workloads increasingly dominate data center costs and resource utilization. Therefore, understanding the inference workload characteristics on evolving CPU-GPU coupled architectures is crucial for optimization. This paper presents an in-depth analysis of LLM inference behavior on loosely-coupled (PCIe A100/H100) and closely-coupled (GH200) systems. We analyze performance dynamics using fine-grained operator-to-kernel trace analysis, facilitated by our novel profiler SKIP and metrics like Total Kernel Launch and Queuing Time (TKLQT). Results show that closely-coupled (CC) GH200 significantly outperforms loosely-coupled (LC) systems at large batch sizes, achieving 1.9x-2.7x faster prefill latency for Llama 3.2-1B. However, our analysis also reveals that GH200 remains CPU-bound up to 4x larger batch sizes than LC systems. In this extended CPU-bound region, we identify the performance characteristics of the Grace CPU as a key factor contributing to higher inference latency at low batch sizes on GH200. We demonstrate that TKLQT accurately identifies this CPU/GPU-bound transition point. Based on this analysis, we further show that kernel fusion offers significant potential to mitigate GH200's low-batch latency bottleneck by reducing kernel launch overhead. This detailed kernel-level characterization provides critical insights for optimizing diverse CPU-GPU coupling strategies. This work is an initial effort, and we plan to explore other major AI/DL workloads that demand different degrees of CPU-GPU heterogeneous architectures.
Large Language Model (LLM)-powered agents have unlocked new possibilities for automating human tasks. While prior work has focused on well-defined tasks with specified goals, the capabilities of agents in creative design tasks with open-ended goals remain underexplored. We introduce GraphicBench, a new planning benchmark for graphic design that covers 1,079 user queries and input images across four design types. We further present GraphicTown, an LLM agent framework with three design experts and 46 actions (tools) to choose from for executing each step of the planned workflows in web environments. Experiments with six LLMs demonstrate their ability to generate workflows that integrate both explicit design constraints from user queries and implicit commonsense constraints. However, these workflows often do not lead to successful execution outcomes, primarily due to challenges in: (1) reasoning about spatial relationships, (2) coordinating global dependencies across experts, and (3) retrieving the most appropriate action per step. We envision GraphicBench as a challenging yet valuable testbed for advancing LLM-agent planning and execution in creative design tasks.