Searcharxiv⌕ Search

arXiv subjects

Tongshuang Wu

Publications and source records attributed to Tongshuang Wu.

At least 19 recordsLinked to original sources

SWE-WhatIf: Simulating Real Users of Coding Agents from Recorded Session

Recorded sessions between users and coding agents offer a way to evaluate how agents complete real-world tasks and interact with users. However, these sessions cannot simply be replayed with new agents, because recorded user replies may no longer apply when an agent takes a different trajectory. We introduce SWE-WhatIf, which replays recorded sessions with new agents by modeling the user's intents, the criteria for satisfying them, and the conditions under which they arise. Using this model, a user simulator generates replies that adapt to unseen agent turns while remaining grounded in the recording. Across 89 real-world sessions, SWE-WhatIf generated the replies most similar to those of real users and matched the strongest baseline when sessions were rewritten. To test adaptation, we replay sessions with agents instructed to avoid or repeat actions that the recorded user pushed back on. SWE-WhatIf produced the highest win rate for avoiding these actions (66.3%), significantly above chance. We further demonstrate its utility by ranking five LLMs as coding agents, where the highest-ranked model depended on how users specified and developed their intents, and a study with 16 users, where replaying their own sessions changed 14 participants' rankings and informed their future agent use.

cs.CL↗

How Should Planning Agents ReSHAPE Their Search Space?

Planning agents and benchmarks treat a user's initial specification as fixed. Yet the requests users write exclude plans they would gladly take, since they rarely discover the possible trade-offs they would accept until seeing the options. Agents already revise requests through interaction, but whether this helps planning is unknown. We formalize this step as the ReSHAPE task, which judges a revision by whether its search space matches the acceptable plans and a planner finds one there. Using ReSHAPE Planter, which builds instances from planning benchmarks with a human-validated simulated user, we analyze 5,800 agent-revised specifications in travel and shopping. We find that reshaping matters: one correct trade-off makes the same planner reach an acceptable plan up to 4.6x as often, and search space quality predicts what the planner finds. Current agents, however, admit at most 32% of the acceptable plans, where that trade-off alone admits over 52%, because they ask about preferences instead of showing plans; in travel, showing plans raises F1 by 9.7 points over asking, with two model families. Planning agents should thus treat a request as a hypothesis to test against the user's reactions to concrete plans to recognize user trade-offs.

cs.AI↗

SpeechEQ: Benchmarking Emotional Intelligence Quotient in Socially Aware Voice Conversational Models

As multimodal conversational systems increasingly engage in spoken interaction, their ability to navigate paralinguistic social cues has become a critical bottleneck for natural human-AI communication. However, existing evaluations of machine emotional intelligence assess reasoning exclusively through isolated text or passive acoustic perception, overlooking the complex cross-modal reasoning required for active, multi-turn dialogue. We introduce \textsc{SpeechEQ}, a comprehensive framework designed to evaluate the sociolinguistic reasoning of Speech-Language Models (SLMs). The framework includes a validated dataset of 2,265 dialogues across 15 Emotional Quotient (EQ) subscales grounded in EQ-i 2.0 theory, along with a multi-turn evaluation protocol measured by our proposed Spoken EQ (SEQ) score inspired by human EQ assessments. Experiments show limitations in how both existing Speech Emotion Recognition and end-to-end Speech-Language Models understand and apply paralinguistic cues through speech. While end-to-end architectures outperform cascaded systems, \textsc{SpeechEQ} reveals that current multimodal models remain bottlenecked by a text-reliant ``modality shortcut,'' an alignment-induced ``safety trap,'' and ``contextual amnesia,'' highlighting the barriers to truly emotionally aware AI. Our benchmark can be accessed at https://huggingface.co/datasets/SpeechEQ/SpeechEQ and demo page at https://speecheq.github.io

cs.CL↗

Humans' ALMANAC: A Human Collaboration Dataset of Action-Level Mental Model Annotations for Agent Collaboration

Recent advances in LLM agents have enabled complex cognitive capabilities, such as multi-step reasoning, planning, and tool use, that increasingly position these agents as collaborative partners in human workflows. Effective collaboration, however, requires collaborators to continuously maintain and align mental models of their own reasoning, partners' intentions, and shared goals during the collaborative process. Today's agents rarely develop such capabilities since they are primarily optimized for task completion rather than for maintaining the underlying mental models required for collaborative processes. Moreover, the community lacks authentic human collaboration data with action-level mental model annotations that could guide agents toward collaborative competence. To bridge this gap, we present ALMANAC, a dataset of Action-Level Mental model ANnotations for Agent Collaboration built from the Map Task, a classic dyadic routing task from social science. ALMANAC contains 2,987 collaboration actions, each paired with theory-informed mental model annotations that record the participants' self-reasoning, perceived partner intent, and perceived team goal. We benchmark six LLMs on predicting humans' next-turn behavior and mental models. Our results demonstrate ALMANAC's utility in evaluating models' ability to simulate human collaborative behaviors and infer their underlying mental models. The project page is available at https://human-agent-collab.hailab.io/almanac.

cs.AI↗

3DCodeEdit: Benchmarking Instruction-Guided 3D Program Editing

Language models increasingly create 3D content through code, with practical workflows often relying on editing existing programs rather than regenerating them, for speed and precise control. Such edits require understanding what a program builds and foreseeing how a change reshapes it, yet prior benchmarks mostly evaluate generation or simple edits against a reference. In this paper, we introduce 3DCodeEdit, a benchmark of 2,592 instruction-guided editing tasks over 108 Blender programs that construct objects entirely in code. Its edits range from parameter to structural changes, style redesign, and scene composition, and are judged from renders against human-reviewed checklists. Across 17 models, the average success rate is only 37%, and even the strongest models complete fewer than 60% of structural and style edits. As models become more capable, their failures shift to geometric foresight, where the code runs but builds the wrong shape. We therefore study whether multi-turn feedback can compensate for this missing foresight. Feedback from a separate reviewer model that describes how the renders differ from the request helps most, more than maximum reasoning effort or an agent harness, and can let a 27B model surpass the single-shot rate of a 2.8T-parameter model.

cs.CL↗

Knowing When to Stop: The Evidence-Action Gap in Coding Agents

Coding agents are trained to complete tasks, and they keep trying even when the environment makes completion impossible. The evidence that only outside help can restore progress usually appears within a few steps, yet agents rarely act on it: they over-persist, attempt undesired workarounds, and claim success the record does not support -- a phenomenon we call evidence-action gap. To measure it, we introduce ImpasseBench, which injects twelve realistic environment blockers into otherwise solvable repair tasks and marks, in each trajectory, the step at which the collected evidence becomes sufficient. Scoring actions relative to that step separates appropriate effort from over-persistence and premature handoff. Across eight frontier models and 19,371 trajectories, we find that nearly every run that needs outside help collects sufficient evidence, yet a third still exhaust their budget, and every unsafe request we detect comes after that point. Fortunately, an escalation tool, or an external monitor for models that ignore the tool, already guards against much of this behavior: the tool cuts budget exhaustion from 34.0% to 6.7%, and the monitor nearly triples task success after environment restoration. Even then, a fired escalation is not yet a useful handoff: tickets reliably name the blocker but often omit the agent's own progress and verification status. Taken together, we argue that knowing when to stop therefore needs explicit support, both in the tool that lets an agent hand off and in the report it leaves behind.

cs.SE↗

Modeling of Therapist-Client Dynamics in Psychotherapy Using LLM-Based Assessments: An Empirical Study

This study aimed to develop and validate a computational framework for modeling therapist--client interaction processes and to examine how therapist behaviors and relational qualities relate to client self-disclosure and emotional expression in psychotherapy sessions. We developed automated assessment pipelines using large language models (LLMs) to measure therapist behaviors (empathy, exploration), relational quality (rapport), and client outcomes (self-disclosure, self-directed negative emotion, outward-directed negative emotion). Model-generated scores were validated against human annotations. We then applied these measures to approximately 2,000 hours of psychotherapy transcripts from the Alexander Street corpus. This yielded an estimation sample of 119,233 client utterances from 1,573 sessions and 37 clients. Automated measures showed good to excellent agreement with human annotations (ICC = 0.60 - 0.81), with rapport (ICC = 0.81), self-disclosure (ICC = 0.78), empathy components (ICC = 0.60 - 0.75), and emotion measurements (ICC = 0.66 - 0.80). SEM analyses revealed that therapist empathy and exploration were each associated with increased client disclosure and increased expression of self-directed negative emotions. Both behaviors were also associated with reduced expressions of enjoyment. Surprisingly, rapport in the prior session showed no association with client self-disclosure or negative emotion expression. This study demonstrates that LLM-based measurement combined with structural modeling can capture core therapeutic processes at scale.

cs.CY↗

Empirically Grounded Behavioral Adaptivity for LLM-Based Virtual Patients in Psychotherapy Training

Simulated patients offer a scalable way to train psychotherapy micro-skills such as empathic responding and exploratory probing, but current systems either follow fixed scripts or rely on LLMs that drift unpredictably over long sessions. We present the Adaptive Virtual Patient (AVP), which adapts its disclosure behavior in response to trainee skill, from guarded, through moderate openness, to full disclosure. The AVP is informed by a structural equation model fit to nearly 2,000 hours of real-world psychotherapy transcripts, which quantifies how therapist empathy and exploration shift a patient's openness over time. An LLM generates the AVP's utterances conditioned on a disclosure level that the dynamics module updates each turn. In an evaluation with 20 clinicians and trainees over 80 sessions (1,033 turns), the AVP's disclosure rises in response to therapist empathy and exploration, while a prompt-only baseline stays flat; ablations confirm that the empirically motivated parameterization outperforms alternatives, with exploration carrying most of the adaptive signal.

cs.CY↗

World of SWECraft: Agent Collaboration at Scale via Simulated Software Worlds

As teams of AI agents take on larger projects and long-standing open problems, agent collaboration offers another dimension for scaling inference-time compute. To understand what makes such collaboration effective and scalable, we develop Software World, a framework for simulating multi-agent coworking in realistic software ecosystems. It comes with a pipeline to create environments of real, interdependent Python packages of different topics, sizes, and dependency structures, which we call software worlds. In a simulation, agents maintain packages in the world with a high-level goal of improving software efficiency, and they are free to discover tasks to work on and collaborate with others. We evaluate agents' work in a world collectively by measuring speedup in hidden dependent packages. The framework allows us to study different elements of collaboration by varying communication, ownership, and resource allocation. We run simulations in three different worlds, totaling more than 10,000 hours of agent work across eight models. Overall, performance improvement scales with inference-time compute, with the best model improving downstream efficiency by up to 31%. We show that most agents can find meaningful tasks via collaboration, but some frontier models can prematurely stop looking for improvements and make little progress. We release the simulation framework and traces to support future studies of agent collaboration at https://szj.io/sw.

cs.AI↗

Align AI to Dynamic Human-AI Workflows

Current alignment approaches typically focus on emulating human behavior using static representations of human preferences, failing to capture the dynamic, context-dependent nature of real-world human-AI interactions. In this paper, we argue for a shift from static and emulative to interactive and complementary alignment, where preferences emerge through interaction and alignment is defined not by satisfying preferences alone. We first formalize this gap by contrasting existing alignment with a trajectory-level view in which human and model behavior co-evolve over time. Because these interaction dynamics have not been adequately captured within existing ML formulations, we ground this perspective in insights from an interdisciplinary workshop. We draw on lessons from social-science accounts of human-human collaboration and then argue that human-AI systems amplify these dynamics, introducing new asymmetries that make reasoning about uncertainty harder and introduce new coordination challenges. Based on these lessons and new challenges, we conclude by outlining a research agenda for developing AI systems that align with humans in interaction, requiring an interdisciplinary synthesis of machine learning and the social and decision sciences.

cs.AI↗

Impact of Expertise and Trust on AI-assisted Work

Artificial intelligence (AI) agents promise to extend what users can accomplish beyond their existing expertise, but they also raise the risk of producing plausible-looking work that users may struggle to evaluate. We ask how task-relevant expertise and trust in AI shape agent-assisted work. To isolate these factors, we conduct a think-aloud study in which 20 professionals with varying data-science expertise use the same agentic tool to complete the same data-science hiring assessment, used at a major U.S. company. We find that surface productivity can come at the expense of understanding: participants who delegated more completed more of the task, but were also more likely to produce invalid solutions. Experts more often resisted this tradeoff, sacrificing completion to decompose the task and scrutinize agent outputs. Our findings call for evaluating agentic assistance not only by how much it accomplishes, but by whether users remain able to evaluate the work they produce.

cs.HC↗

OdysSim: Building Foundation Models for Human Behavior Simulation

Large language models are increasingly deployed as human simulators for interactive evaluation and social simulation. Yet helpfulness-driven post-training pulls them toward a homogeneous, overly agreeable assistant register, creating a behavioral Sim2Real gap. We present OdysSim, the largest open systematic investigation of behavioral foundation models, i.e., models trained to simulate human behavior at scale. We propose SOUL, a taxonomy of five capability axes (CONV, SS, COG, ROLE, EVAL) that unifies 62 datasets and 23 benchmark tasks under one framework. Specifically, we curate the OdysSim corpus (21.4M interactions, 10B tokens, retrofitted with back-generated social contexts), construct the SOUL-Index benchmark, and develop an end-to-end training recipe combining midtraining, task-specific RL, and expert distillation. The resulting open 8B OSim model ranks first or tied-first on 8 of 23 tasks, outperforming any individual frontier model by this count, with the strongest gains on conversational and social tasks. Its outputs are also more human-like in length, formatting, and word choice, and it transfers zero-shot to out-of-distribution user simulation on $τ$-bench, nearly matching real users on reaction alignment (93.2 vs. 93.5). We further show that LLM-as-judge RL induces reward-hacking patterns, and that our detectors can mitigate them during post-training. Together, our findings suggest that behavioral foundation models require rethinking the LLM training paradigm. We release all artifacts to support future research.

cs.CL↗

Improving Attributed Long-form Question Answering with Intent Awareness

Large language models (LLMs) are increasingly being used to generate comprehensive, knowledge-intensive reports. However, while these models are trained on diverse academic papers and reports, they are not exposed to the reasoning processes and intents that guide authors in crafting these documents. We hypothesize that enhancing a model's intent awareness can significantly improve the quality of generated long-form reports. We develop and employ structured, tag-based schemes to better elicit underlying implicit intents to write or cite. We demonstrate that these extracted intents enhance both zero-shot generation capabilities in LLMs and enable the creation of high-quality synthetic data for fine-tuning smaller models. Our experiments reveal improved performance across various challenging scientific report generation tasks, with an average improvement of +2.9 and +12.3 absolute points for large and small models over baselines, respectively. Furthermore, our analysis illuminates how intent awareness enhances model citation usage and substantially improves report readability.

cs.CL↗

Better Harnesses, Smaller Models: Building 90% Cheaper Agents via Automated Harness Adaptation

Frontier LLM agents are automating many business tasks, but their high inference cost makes large-scale deployment unsustainable. Small language models (SLMs) offer a cheaper alternative, yet they typically fall short when swapped into a harness designed for a frontier LLM. We show that for many routine business tasks, SLM agents can match LLM performance at 90% lower cost, when paired with an adapted harness that can be automatically discovered by a meta agent. The key insight is that much of the task difficulty is shared across instances and can be lifted from the model into the harness via tailored instructions, tools, and orchestration loops. To study this systematically, we create a framework that maps agent failure modes to harness adaptation strategies, and build a harness optimizer that automatically discovers effective adaptations from failure trajectories. Across seven business-oriented agentic tasks and three SLM families, we found optimized harnesses significantly improve performance on 16 of 21 task-SLM pairs, with seven pairs closing the SLM-LLM performance gap and the best SLM agent recovering 89.7% of LLM performance at 4% of the cost. Our analysis further shows that adaptation works best for tasks with more repetitive workflows and for SLMs with sufficient base capabilities. Together, these results suggest that harness adaptation can expand the practical deployment range of SLM agents in routine business tasks.

cs.SE↗

"GenAI Defaults to Bias!" Gamify AI Literacy Through Reflections on Prompts

As Generative AI (GenAI) becomes widespread, it is increasingly important for the public to understand the model's behaviors and biases. However, existing AI literacy efforts miss opportunities to engage the general public to reflect on enduring GenAI bias and behaviors (e.g., how GenAI defaults to its internal bias in response to ambiguous or challenging prompts). In this work, we introduce ImaginAItion, a multiplayer game to help adults better reflect on GenAI bias and understand GenAI behaviors. ImaginAItion is grounded in reflective play to surface GenAI limitations by encouraging players to manipulate prompt specificity (e.g., an underspecified prompt "CEO" defaults to a white man). From ten sessions (n=30), we find that the game significantly improved players' understanding of GenAI behaviors by 35% in accuracy. Qualitative analysis showed how game mechanisms supported player reflections, including on prompting strategies to mitigate GenAI bias. Our work demonstrates a viable pathway to scale GenAI literacy through playful, social interventions resilient to rapidly evolving technologies.

cs.HC↗

Discretizing Reward Models

Despite their widespread use, the role of reward models in shaping reinforcement learning is poorly understood. Reward models offer a tempting promise: they automatically estimate response quality in the absence of verifiers or human judges. Unlike "verifiable rewards" which typically produce binary scores, reward models typically produce continuous scores, allowing them to be sensitive to fine-grained differences in responses. However, we show this apparent strength is a serious weakness: many popular reward models are oversensitive, assigning different scores to equally good responses. Theoretically, we show that seemingly perfect reward models can be highly oversensitive; empirically, this oversensitivity can lead to bad policies. In place of existing notions of "reward model accuracy," we propose evaluating reward models using distinct measures of "discriminative ability" and "specificity" (the complement of oversensitivity). As a solution, we describe a training-free algorithm that uses Monte Carlo dropout on any neural reward model to produce discrete reward clusters. Theoretically, we prove there exist discretizations that reduce oversensitivity at minimal expense of discriminative ability; empirically we show, in both controlled and natural RL settings, that discretizing rewards leads to less reward hacking and better policies than training on the original rewards.

cs.LG↗

Practice Less, Explain More: LLM-Supported Self-Explanation Improves Explanation Quality on Transfer Problems in Calculus

We conducted a between-subjects experiment (N=92) comparing three conditions in a calculus learning environment: no self-explanation (control), menu-based self-explanation, and open-ended self-explanation with LLM-generated feedback. All conditions showed positive learning gains within a fixed 60-minute practice session, with no significant between-condition differences in post-test performance. On transfer questions, the open-ended condition produced significantly higher-quality explanations than control on "Not Enough Information" (NEI) problems ($β$=+11.9 percentage points, $p$=.030), though the corresponding NEI multiple-choice accuracy advantage was not significant ($p$=.183). Moreover, across all post-test open-ended explanations, the open-ended condition showed a marginally significant advantage ($β$=+7.3%, $p$=.057). These findings suggest that LLM-supported open-ended self-explanation can improve explanation quality on NEI transfer problems, with weaker evidence across broader transfer explanation measures. Notably, these effects emerged even though learners in the open-ended condition completed substantially fewer practice problems within the same practice time.

cs.HC↗

DR Tulu: Reinforcement Learning with Evolving Rubrics for Deep Research

Deep research agents perform multi-step research to produce long-form, well-attributed answers. However, most open deep research agents are trained on easily verifiable short-form QA tasks via reinforcement learning with verifiable rewards, which does not extend to realistic long-form tasks. We address this with Reinforcement Learning with Evolving Rubrics (RLER), where rubrics are constructed and maintained to co-evolve with the policy model during training. This allows the rubrics to incorporate newly explored information from search and contrasting model responses, enabling better fact checking and more discriminative on-policy feedback. Using RLER, we develop Deep Research Tulu (DR Tulu-8B), the first fully open model that is directly trained for open-ended, long-form deep research. Across four long-form deep research benchmarks in science, healthcare, and general domains, DR Tulu substantially outperforms existing open deep research agents (by 15.6% over Tongyi DR on average) and matches or exceeds proprietary deep research agents (by 0.7% over OpenAI DR on average), while being significantly smaller and cheaper per query (1000x cheaper than OpenAI DR per query).

cs.CL↗