Shunyu Yao

Carlos E. Jimenez, John Yang, Alexander Wettig et al. · princeton, uw

Language models have outpaced our ability to evaluate them effectively, but for their future development it is essential to study the frontier of their capabilities. We find real-world software engineering to be a rich, sustainable, and challenging testbed for evaluating the next generation of language models. To this end, we introduce SWE-bench, an evaluation framework consisting of $2,294$ software engineering problems drawn from real GitHub issues and corresponding pull requests across $12$ popular Python repositories. Given a codebase along with a description of an issue to be resolved, a language model is tasked with editing the codebase to address the issue. Resolving issues in SWE-bench frequently requires understanding and coordinating changes across multiple functions, classes, and even files simultaneously, calling for models to interact with execution environments, process extremely long contexts and perform complex reasoning that goes far beyond traditional code generation tasks. Our evaluations show that both state-of-the-art proprietary models and our fine-tuned model SWE-Llama can resolve only the simplest issues. The best-performing model, Claude 2, is able to solve a mere $1.96$% of the issues. Advances on SWE-bench represent steps towards LMs that are more practical, intelligent, and autonomous.

R. Thomas McCoy, Shunyu Yao, Dan Friedman et al. · princeton

The widespread adoption of large language models (LLMs) makes it important to recognize their strengths and limitations. We argue that in order to develop a holistic understanding of these systems we need to consider the problem that they were trained to solve: next-word prediction over Internet text. By recognizing the pressures that this task exerts we can make predictions about the strategies that LLMs will adopt, allowing us to reason about when they will succeed or fail. This approach - which we call the teleological approach - leads us to identify three factors that we hypothesize will influence LLM accuracy: the probability of the task to be performed, the probability of the target output, and the probability of the provided input. We predict that LLMs will achieve higher accuracy when these probabilities are high than when they are low - even in deterministic settings where probability should not matter. To test our predictions, we evaluate two LLMs (GPT-3.5 and GPT-4) on eleven tasks, and we find robust evidence that LLMs are influenced by probability in the ways that we have hypothesized. In many cases, the experiments reveal surprising failure modes. For instance, GPT-4's accuracy at decoding a simple cipher is 51% when the output is a high-probability word sequence but only 13% when it is low-probability. These results show that AI practitioners should be careful about using LLMs in low-probability situations. More broadly, we conclude that we should not evaluate LLMs as if they are humans but should instead treat them as a distinct type of system - one that has been shaped by its own particular set of pressures.

Xun Liang, Hanyu Wang, Yezhaohui Wang et al.

In Natural Language Processing (NLP), Large Language Models (LLMs) have demonstrated high text generation quality. However, in real-world applications, LLMs must meet increasingly complex requirements. Beyond avoiding misleading or inappropriate content, LLMs are also expected to cater to specific user needs, such as imitating particular writing styles or generating text with poetic richness. These varied demands have driven the development of Controllable Text Generation (CTG) techniques, which ensure that outputs adhere to predefined control conditions--such as safety, sentiment, thematic consistency, and linguistic style--while maintaining high standards of helpfulness, fluency, and diversity. This paper systematically reviews the latest advancements in CTG for LLMs, offering a comprehensive definition of its core concepts and clarifying the requirements for control conditions and text quality. We categorize CTG tasks into two primary types: content control and attribute control. The key methods are discussed, including model retraining, fine-tuning, reinforcement learning, prompt engineering, latent space manipulation, and decoding-time intervention. We analyze each method's characteristics, advantages, and limitations, providing nuanced insights for achieving generation control. Additionally, we review CTG evaluation methods, summarize its applications across domains, and address key challenges in current research, including reduced fluency and practicality. We also propose several appeals, such as placing greater emphasis on real-world applications in future research. This paper aims to offer valuable guidance to researchers and developers in the field. Our reference list and Chinese version are open-sourced at https://github.com/IAAR-Shanghai/CTGSurvey.

John Yang, Akshara Prabhakar, Karthik Narasimhan et al. · princeton

Humans write code in a fundamentally interactive manner and rely on constant execution feedback to correct errors, resolve ambiguities, and decompose tasks. While LLMs have recently exhibited promising coding capabilities, current coding benchmarks mostly consider a static instruction-to-code sequence transduction process, which has the potential for error propagation and a disconnect between the generated code and its final execution environment. To address this gap, we introduce InterCode, a lightweight, flexible, and easy-to-use framework of interactive coding as a standard reinforcement learning (RL) environment, with code as actions and execution feedback as observations. Our framework is language and platform agnostic, uses self-contained Docker environments to provide safe and reproducible execution, and is compatible out-of-the-box with traditional seq2seq coding methods, while enabling the development of new methods for interactive code generation. We use InterCode to create three interactive code environments with Bash, SQL, and Python as action spaces, leveraging data from the static NL2Bash, Spider, and MBPP datasets. We demonstrate InterCode's viability as a testbed by evaluating multiple state-of-the-art LLMs configured with different prompting strategies such as ReAct and Plan & Solve. Our results showcase the benefits of interactive code generation and demonstrate that InterCode can serve as a challenging benchmark for advancing code understanding and generation capabilities. InterCode is designed to be easily extensible and can even be used to create new tasks such as Capture the Flag, a popular coding puzzle that is inherently multi-step and involves multiple programming languages. Project site with code and data: https://intercode-benchmark.github.io

Theodore R. Sumers, Shunyu Yao, Karthik Narasimhan et al.

Recent efforts have augmented large language models (LLMs) with external resources (e.g., the Internet) or internal control flows (e.g., prompt chaining) for tasks requiring grounding or reasoning, leading to a new class of language agents. While these agents have achieved substantial empirical success, we lack a systematic framework to organize existing agents and plan future developments. In this paper, we draw on the rich history of cognitive science and symbolic artificial intelligence to propose Cognitive Architectures for Language Agents (CoALA). CoALA describes a language agent with modular memory components, a structured action space to interact with internal memory and external environments, and a generalized decision-making process to choose actions. We use CoALA to retrospectively survey and organize a large body of recent work, and prospectively identify actionable directions towards more capable agents. Taken together, CoALA contextualizes today's language agents within the broader history of AI and outlines a path towards language-based general intelligence.

Baian Chen, Chang Shu, Ehsan Shareghi et al.

Recent efforts have augmented language models (LMs) with external tools or environments, leading to the development of language agents that can reason and act. However, most of these agents rely on few-shot prompting techniques with off-the-shelf LMs. In this paper, we investigate and argue for the overlooked direction of fine-tuning LMs to obtain language agents. Using a setup of question answering (QA) with a Google search API, we explore a variety of base LMs, prompting methods, fine-tuning data, and QA tasks, and find language agents are consistently improved after fine-tuning their backbone LMs. For example, fine-tuning Llama2-7B with 500 agent trajectories generated by GPT-4 leads to a 77% HotpotQA performance increase. Furthermore, we propose FireAct, a novel approach to fine-tuning LMs with trajectories from multiple tasks and prompting methods, and show having more diverse fine-tuning data can further improve agents. Along with other findings regarding scaling effects, robustness, generalization, efficiency and cost, our work establishes comprehensive benefits of fine-tuning LMs for agents, and provides an initial set of experimental designs, insights, as well as open questions toward language agent fine-tuning.

Shunyu Yao, Gary P. T. Choi

Tube-like surfaces are widely encountered in geometry processing, engineering structures, and medical anatomy, yet their intrinsic longitudinal and circumferential topology is not well preserved by conventional planar annular or rectangular parameterization domains. In this work, we propose a conformal parameterization framework for open tube-like surfaces with two boundary components. The proposed method first constructs a fixed-boundary tubular parameterization by cutting the input mesh, computing a disk-to-rectangle conformal map, and lifting the result to a three-dimensional tubular domain. To reduce residual distortion introduced near the cut seam, we further introduce a localized quasi-conformal correction scheme formulated on an annular domain, which improves conformality while leaving regions away from the seam unchanged. To handle noisy or irregular input boundaries, we also develop a free-boundary variant based on boundary extension and cycle-Laplacian smoothing, allowing the prescribed boundary constraints to be imposed on artificial outer rings rather than directly on the original surface. Finally, we derive two conformal toroidal bending maps that transform the tubular parameterization into toroidal geometries while preserving the underlying tube topology. Experiments on synthetic tube meshes and real vascular surfaces demonstrate that the proposed framework produces low-distortion parameterizations, effectively mitigates seam-induced artifacts, improves robustness for boundary-noisy inputs, and provides flexible tubular and toroidal target domains for downstream surface processing tasks.

Quan Shi, Carlos E. Jimenez, Shunyu Yao et al. · princeton

Recent advancements in AI reasoning have driven substantial improvements across diverse tasks. A critical open question is whether these improvements also yields better knowledge transfer: the ability of models to communicate reasoning in ways humans can understand, apply, and learn from. To investigate this, we introduce Knowledge Integration and Transfer Evaluation (KITE), a conceptual and experimental framework for Human-AI knowledge transfer capabilities and conduct the first large-scale human study (N=118) explicitly designed to measure it. In our two-phase setup, humans first ideate with an AI on problem-solving strategies, then independently implement solutions, isolating model explanations' influence on human understanding. Our findings reveal that although model benchmark performance correlates with collaborative outcomes, this relationship is notably inconsistent, featuring significant outliers, indicating that knowledge transfer requires dedicated optimization. Our analysis identifies behavioral and strategic factors mediating successful knowledge transfer. We release our code, dataset, and evaluation framework to support future work on communicatively aligned models.

Shunyu Yao, Rohan Rao, Matthew Hausknecht et al.

Text-based games present a unique challenge for autonomous agents to operate in natural language and handle enormous action spaces. In this paper, we propose the Contextual Action Language Model (CALM) to generate a compact set of action candidates at each game state. Our key insight is to train language models on human gameplay, where people demonstrate linguistic priors and a general game sense for promising actions conditioned on game history. We combine CALM with a reinforcement learning agent which re-ranks the generated action candidates to maximize in-game rewards. We evaluate our approach using the Jericho benchmark, on games unseen by CALM during training. Our method obtains a 69% relative improvement in average game score over the previous state-of-the-art model. Surprisingly, on half of these games, CALM is competitive with or better than other models that have access to ground truth admissible actions. Code and data are available at https://github.com/princeton-nlp/calm-textgame.

Quan Shi, Michael Tang, Karthik Narasimhan et al.

Computing olympiads contain some of the most challenging problems for humans, requiring complex algorithmic reasoning, puzzle solving, in addition to generating efficient code. However, it has been understudied as a domain to evaluate language models (LMs). In this paper, we introduce the USACO benchmark with 307 problems from the USA Computing Olympiad, along with high-quality unit tests, reference code, and official analyses for each problem. These resources enable us to construct and test a range of LM inference methods for competitive programming for the first time. We find GPT-4 only achieves a 8.7% pass@1 accuracy with zero-shot chain-of-thought prompting, and our best inference method improves it to 20.2% using a combination of self-reflection and retrieval over episodic knowledge. However, this is far from solving the benchmark. To better understand the remaining challenges, we design a novel human-in-the-loop study and surprisingly find that a small number of targeted hints enable GPT-4 to solve 13 out of 15 problems previously unsolvable by any model and method. Our benchmark, baseline methods, quantitative results, and qualitative analysis serve as an initial step toward LMs with grounded, creative, and algorithmic reasoning.

Yuqian Sun, Xingyu Li, Shunyu Yao et al.

Recent advances in Generative AI (GAI) have led to new opportunities for creativity support. However, this technology has raised ethical concerns in the visual artists community. This paper explores how GAI can assist visual artists in developing original characters (OCs) while respecting their creative agency. We present ORIBA, an AI chatbot leveraging large language models (LLMs) to enable artists to role-play with their OCs, focusing on conceptualization (e.g., backstories) while leaving exposition (visual creation) to creators. Through a study with 14 artists, we found ORIBA motivated artists' imaginative engagement, developing multidimensional attributes and stronger bonds with OCs that inspire their creative process. Our contributions include design insights for AI systems that develop from artists' perspectives, demonstrating how LLMs can support cross-modal creativity while preserving creative agency in OC art. This paper highlights the potential of GAI as a neutral, non-visual support that strengthens existing creative practice, without infringing artistic exposition.

Shunyu Yao, Noah Shinn, Pedram Razavi et al.

Existing benchmarks do not test language agents on their interaction with human users or ability to follow domain-specific rules, both of which are vital for deploying them in real world applications. We propose $τ$-bench, a benchmark emulating dynamic conversations between a user (simulated by language models) and a language agent provided with domain-specific API tools and policy guidelines. We employ an efficient and faithful evaluation process that compares the database state at the end of a conversation with the annotated goal state. We also propose a new metric (pass^k) to evaluate the reliability of agent behavior over multiple trials. Our experiments show that even state-of-the-art function calling agents (like gpt-4o) succeed on <50% of the tasks, and are quite inconsistent (pass^8 <25% in retail). Our findings point to the need for methods that can improve the ability of agents to act consistently and follow rules reliably.

John Yang, Carlos E. Jimenez, Alexander Wettig et al.

Language model (LM) agents are increasingly being used to automate complicated tasks in digital environments. Just as humans benefit from powerful software applications, such as integrated development environments, for complex tasks like software engineering, we posit that LM agents represent a new category of end users with their own needs and abilities, and would benefit from specially-built interfaces to the software they use. We investigate how interface design affects the performance of language model agents. As a result of this exploration, we introduce SWE-agent: a system that facilitates LM agents to autonomously use computers to solve software engineering tasks. SWE-agent's custom agent-computer interface (ACI) significantly enhances an agent's ability to create and edit code files, navigate entire repositories, and execute tests and other programs. We evaluate SWE-agent on SWE-bench and HumanEvalFix, achieving state-of-the-art performance on both with a pass@1 rate of 12.5% and 87.7%, respectively, far exceeding the previous state-of-the-art achieved with non-interactive LMs. Finally, we provide insight on how the design of the ACI can impact agents' behavior and performance.

Shunyu Yao, Binghui Peng, Christos Papadimitriou et al.

Despite their impressive performance in NLP, self-attention networks were recently proved to be limited for processing formal languages with hierarchical structure, such as $\mathsf{Dyck}_k$, the language consisting of well-nested parentheses of $k$ types. This suggested that natural language can be approximated well with models that are too weak for formal languages, or that the role of hierarchy and recursion in natural language might be limited. We qualify this implication by proving that self-attention networks can process $\mathsf{Dyck}_{k, D}$, the subset of $\mathsf{Dyck}_{k}$ with depth bounded by $D$, which arguably better captures the bounded hierarchical structure of natural language. Specifically, we construct a hard-attention network with $D+1$ layers and $O(\log k)$ memory size (per token per layer) that recognizes $\mathsf{Dyck}_{k, D}$, and a soft-attention network with two layers and $O(\log k)$ memory size that generates $\mathsf{Dyck}_{k, D}$. Experiments show that self-attention networks trained on $\mathsf{Dyck}_{k, D}$ generalize to longer inputs with near-perfect accuracy, and also verify the theoretical memory advantage of self-attention networks over recurrent networks.

Shunyu Yao, Karthik Narasimhan, Matthew Hausknecht

Text-based games simulate worlds and interact with players using natural language. Recent work has used them as a testbed for autonomous language-understanding agents, with the motivation being that understanding the meanings of words or semantics is a key component of how humans understand, reason, and act in these worlds. However, it remains unclear to what extent artificial agents utilize semantic understanding of the text. To this end, we perform experiments to systematically reduce the amount of semantic information available to a learning agent. Surprisingly, we find that an agent is capable of achieving high scores even in the complete absence of language semantics, indicating that the currently popular experimental setup and models may be poorly designed to understand and leverage game texts. To remedy this deficiency, we propose an inverse dynamics decoder to regularize the representation space and encourage exploration, which shows improved performance on several games including Zork I. We discuss the implications of our findings for designing future agents with stronger semantic understanding.

Shunyu Yao, Tzu Ming Harry Hsu, Jun-Yan Zhu et al.

We aim to obtain an interpretable, expressive, and disentangled scene representation that contains comprehensive structural and textural information for each object. Previous scene representations learned by neural networks are often uninterpretable, limited to a single object, or lacking 3D knowledge. In this work, we propose 3D scene de-rendering networks (3D-SDN) to address the above issues by integrating disentangled representations for semantics, geometry, and appearance into a deep generative model. Our scene encoder performs inverse graphics, translating a scene into a structured object-wise representation. Our decoder has two components: a differentiable shape renderer and a neural texture generator. The disentanglement of semantics, geometry, and appearance supports 3D-aware scene manipulation, e.g., rotating and moving objects freely while keeping the consistent shape and texture, and changing the object appearance without affecting its shape. Experiments demonstrate that our editing scheme based on 3D-SDN is superior to its 2D counterpart.