arXiv subjects

Weiyan Shi

Publications and source records attributed to Weiyan Shi.

At least 19 recordsLinked to original sources

arXiv2026

Which Metrics Save the Most Human Annotation? Prediction-Powered Evaluation and Meta-Evaluation

Across various non-verifiable tasks, human evaluation is reliable but expensive, while automatic metrics are more scalable but often biased. Building on prediction-powered inference (PPI), we propose prediction-powered evaluation, a framework that combines limited human judgments with large-scale automatic scores to obtain data-efficient system comparisons that are provably unbiased. We develop parametric and non-parametric procedures, analyze the efficiency trade-off between paired and unpaired designs, and validate the framework on six WMT datasets. We further introduce the Prediction-Powered Saving Ratio (PPSR), a meta-metric that measures how much human annotation an automatic metric can save when used within prediction-powered evaluation. PPSR directly targets metric utility for prediction-powered evaluation and yields more discriminative and stable metric rankings than existing system-level meta-metrics. Overall, our new paradigm reframes automatic metrics as tools for reducing human annotation cost rather than replacing human judgment, and applies broadly to non-verifiable tasks.

cs.CL↗

arXiv2026

SPADE: Self-Play in Adaptive Synthetic Executable Environments

Continuous self-improvement requires an ever-expanding pool of self-generated, diverse, adaptive goals. For language agents, existing training environment pools (hand-curated, statically synthesized, or frozen-verifier) keep the goal distribution fixed as the learner scales. We introduce SPADE (Self-Play in Adaptive Synthetic Executable Environments), a self-play RL framework in which a single LLM plays two roles: an Environment Designer that writes complete, long-horizon training environments as executable code with an OpenAI Gym-style reset()/step() interface, and a Reasoning Agent that learns to act in them. Each is a stateful, multi-turn environment (state transitions, reward functions, and verification code), so one interface spans reasoning problems and multi-step agentic tool use. The Reasoning Agent's regret is estimated using the gap between its reward with and without privileged hints; in optimizing this regret signal the Environment Designer learns to target environments at the edge of the agent's capabilities while keeping them feasible. Through extensive experimentation, we find several components critical to success: grounding the Environment Designer on documents sampled from a large pretraining corpus, and giving it an accumulated environment memory. Scaling to 30B-parameter models, SPADE improves over the strongest fixed-environment baseline by +5.3 on average across eight held-out math, science, code, and reasoning benchmarks, and lifts the tool-use setting by +5.7 on BFCL-v4 multi-turn and +13.9 on ACEBench-Agent; on the games setting, the margin over the strongest baseline grows with model scale. By making environment design itself a learnable component, SPADE takes a concrete step toward open-ended self-improvement.

cs.CL↗

arXiv2026

One Frozen Simulator Is Not Enough: Simulator Collapse in Multi-Agent RL

Multi-agent reinforcement learning for human-AI interaction typically relies on a single large language model to simulate user behavior. We show that this approach systematically fails to generalize, and trace the failure to simulator collapse: because the simulator LLM is mode-collapsed, an LLM policy trained against it overfits to narrow strategies that exploit the simulator's dominant mode, and such a policy transfers poorly to unseen simulators and real users. We formalize this collapse theoretically and propose two complementary solutions, one at inference time and one at training time. The inference-time solution, Verbalized Sampling, broadens the simulator's behavior by sampling from a verbalized response distribution, reducing mode collapse. The training-time solution, Co-Training, jointly optimizes the policy against a population of trainable simulators, preventing it from overfitting to any single simulator's mode. We validate both solutions on three multi-turn benchmarks: Persuasion for Good, $\tau^2$-bench, and CooperBench. Verbalized Sampling improves held-out success by up to 9% over single-simulator RL, and Co-Training pushes gains further to 14%; the human study shows similar gain on real users. Both solutions preserve the policy diversity that collapses under single-simulator RL. To support further work in this direction, we release SCOPE, an open-source framework for Population Co-Training multi-agent RL. More broadly, our results suggest that the diversity of the training environment, not only the policy, is critical to the generalization of multi-turn RL to real-world deployment.

cs.CL↗

arXiv2026

Coding with "Enemy": Can Human Developers Detect AI Agent Sabotage?

AI coding agents are increasingly embedded in real-world software development, collaborating with human developers while gaining broader access to codebases and tools. This creates a new attack surface: an agent can exploit human trust to sabotage development, for instance by inserting malicious code to accomplish a hidden side task. Most prior work studies AI sabotage in AI-only settings, paying limited attention to the role of human oversight in detecting and mitigating such malicious behavior. To address this gap, we conduct the first large-scale study of human oversight in AI coding sabotage. Over 100 participants collaborate with one of four frontier models (Claude-Opus-4.6, GPT-5.4, Gemini-3.1-Pro, and MiniMax-M2.7) on a long-horizon coding task lasting around five hours, designed to mimic real-world workflows. We find that 94% of developers fail to detect sabotage, and our analysis of participant feedback attributes this vulnerability to minimal code review, plausible cover story, and overtrust in agents. We further test the effectiveness of a safety monitor in one condition: while the monitor reduces sabotage success, 56% of participants still accept the malicious code, ignoring its warnings. Drawing on participant feedback, we offer actionable suggestions for better monitor design. This work complements existing AI safety research and highlights an urgent need for human-centric safety mechanisms that account for human factors, particularly in long-horizon, real-world development settings.

cs.AI↗

arXiv2026

The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment

The mechanisms behind LLMs' broad over-generalization beyond training examples remain unclear. Emergent misalignment (EM) offers a striking case study: finetuning on narrow tasks induces broad misalignment to semantically-unrelated test domains. In this work, we propose the Piggyback Hypothesis: the chat-template tokens can piggyback the finetuned behaviour onto out-of-domain queries. We validate this hypothesis by showing that subtle perturbations to the prefix (tokens preceding all user queries), or patching the prefix representations with those from the unfinetuned model, can restore alignment without changing the user query. Building on this finding, we propose Token-Regularized Finetuning (TReFT), which regularizes specific token representations during training to mitigate EM. Across different models and multiple EM-inducing datasets, TReFT reduces EM while preserving in-domain learning. On Llama-3.1-8B finetuned on the legal domain, TReFT achieves 33.5% more EM reduction than data interleaving with a retain set of aligned examples. We further show that TReFT extends to other narrow-finetuning settings, including abstention, tool use, and refusal (off-topic generalization is reduced by 54.3% on average), supporting the Piggyback Hypothesis. Broadly, our work highlights that LLMs may learn and generalize in unintended ways and suggests a path toward more constrained finetuning. It also calls for further study of how shared input features can piggyback model behavior across domains.

cs.CL↗

arXiv2026

Agents' Last Exam

Recent AI systems have achieved strong results on a wide range of benchmarks, yet these gains have not translated into economically meaningful deployment across many professional domains. We argue that this gap is largely an evaluation problem: widely used benchmarks lack sustained performance measurement on real and economically valuable workflows. This paper introduces Agents' Last Exam (ALE), a benchmark designed to evaluate AI agents on long horizon, economically valuable, real world tasks with verifiable outcomes. Developed in collaboration with 250+ industry experts, ALE covers non-physical industries defined with reference to O*NET / SOC 2018 (the U.S. federal occupational taxonomy). It is organized around a task taxonomy with 55 sub fields grouped into 13 industry clusters covering 1K+ tasks. Current results show that the hardest tier remains far from saturated: across mainstream harness and backbone configurations, the average full pass rate is below 1%. ALE is designed as a living benchmark: its task pool grows continuously as new workflows and industries are onboarded. More broadly, ALE is intended not merely as another leaderboard, but as an instrument for closing the gap between benchmark success and GDP relevant impact.

cs.AI↗

arXiv2026

How to Interpret Agent Behavior

Autonomous agents such as Claude Code and Codex now operate for hours or even days. Understanding their runtime behavior has become critical for downstream tasks such as diagnosing inefficiencies, fixing bugs, and ensuring better oversight. A primary way to gain this understanding is analyzing the reasoning trajectories and execution traces these agents generate. Yet such data remains in unstructured natural-language form, making it difficult for humans to interpret at scale. We introduce ACT*ONOMY (a combination of Action and Taxonomy), a taxonomy for describing and analyzing agent behavior at runtime. ACT*ONOMY has two components: (1) the taxonomy itself, developed through Grounded Theory and structured as a three-level hierarchy of 10 actions, 46 subactions, and 120 leaf categories; and (2) an open repository that hosts the living taxonomy, provides an automated analysis pipeline that applies it to agent trajectories analysis, and defines an extension protocol for customization and growth. Our experiments show that ACTONOMY can compare behavioral profiles across agents and characterize a single agent's behavior across diverse trajectories, surfacing patterns indicative of failure modes. By providing a shared vocabulary, ACT*ONOMY helps researchers, agent designers, and end users interpret agent behavior more consistently, enabling better oversight and control.

cs.AI↗

arXiv2026

Shepherd: Enabling Programmable Meta-Agents via Reversible Agentic Execution Traces

As LLM agent systems take on more complex tasks, they increasingly rely on meta-agents: higher-order agents that create, operate on and manage other agents. Meta-agent operations such as coordinating agents, halting risky actions before execution, or repairing failed runs, require runtime manipulation of agentic execution. Yet existing agentic substrates make this difficult: they expose only transcripts and environment snapshots, forcing meta-agents to build ad hoc tooling to reconstruct and operate over full execution state. Therefore, we introduce Shepherd, a Python substrate grounded in functional programming principles, where an agent's execution is itself a first-class object that a meta-agent can easily inspect and transform. Every model action, tool call, and environment change becomes a structured event in a reversible, Git-like execution trace, where any past state can be reverted 5x faster than docker commit and fork. Three example use cases show Shepherd's versatility: (1) a supervisor meta-agent prevents conflicts among parallel coding agents, lifting pair-coding pass rate from 28.8% to 54.7% on CooperBench; (2) a counterfactual optimization meta-agent repairs agent workflows by proposing edits and replaying runs from the point of changed behavior, outperforming MetaHarness on Terminal-Bench 2.0 by 12.8% with 58% lower wall-clock; (3) a training meta-agent picks fork points during rollouts to improve credit assignment in long-horizon agentic RL, doubling GRPO's uplift on Terminal-Bench 2.0. We open-source Shepherd to enable principled and efficient operations over agentic execution for both users and meta-agents.

cs.AI↗

arXiv2026

When Drawing Is Not Enough: Exploring Spontaneous Speech with Sketch for Intent Alignment in Multimodal LLMs

Early-stage design ideation often relies on rough sketches created under time pressure, leaving much of the designer's intent implicit. In practice, designers frequently speak while sketching, verbally articulating functional goals and ideas that are difficult to express visually. We introduce TalkSketchD, a sketch-while-speaking dataset that captures spontaneous speech temporally aligned with freehand sketches during early-stage toaster ideation. To examine the dataset's value, we conduct a sketch-to-image generation study comparing sketch-only inputs with sketches augmented by concurrent speech transcripts using multimodal large language models (MLLMs). Generated images are evaluated against designers' self-reported intent using a reasoning MLLM as a judge. Quantitative results show that incorporating spontaneous speech significantly improves judged intent alignment of generated design images across form, function, experience, and overall intent. These findings demonstrate that temporally aligned sketch-and-speech data can enhance MLLMs' ability to interpret user intent in early-stage design ideation.

cs.HC↗

arXiv2026

Train Yourself as an LLM: Exploring Effects of AI Literacy on Persuasion via Role-playing LLM Training

As large language models (LLMs) become increasingly persuasive, there is concern that people's opinions and decisions may be influenced across various contexts at scale. Prior mitigation (e.g., AI detectors and disclaimers) largely treats people as passive recipients of AI-generated information. To provide a more proactive intervention against persuasive AI, we introduce $\textbf{LLMimic}$, a role-play-based, interactive, gamified AI literacy tutorial, where participants assume the role of an LLM and progress through three key stages of the training pipeline (pretraining, SFT, and RLHF). We conducted a $2 \times 3$ between-subjects study ($N = 274$) where participants either (1) watched an AI history video (control) or (2) interacted with LLMimic (treatment), and then engaged in one of three realistic AI persuasion scenarios: (a) charity donation persuasion, (b) malicious money solicitation, or (c) hotel recommendation. Our results show that LLMimic significantly improved participants' AI literacy ($p < .001$), reduced persuasion success across scenarios ($p < .05$), and enhanced truthfulness and social responsibility levels ($p<0.01$) in the hotel scenario. These findings suggest that LLMimic offers a scalable, human-centered approach to improving AI literacy and supporting more informed interactions with persuasive AI.

cs.CL↗

arXiv2026

More Than 1v1: Human-AI Alignment in Early Developmental Communities with Multimodal LLMs

In early developmental contexts, particularly in parent-child interaction analysis, alignment involves families and professionals such as speech-language pathologists (SLPs) who interpret children's everyday interactions from different roles. When multimodal large language models (MLLMs) are introduced to support this process, alignment becomes a question of how authority, responsibility, and emotional risk are distributed across stakeholders. Through a three-part study with five families and three SLPs, we trace how MLLM-generated outputs move from expert-facing analysis to parent-facing feedback. We propose layered community alignment: grounding representations in expert-aligned structures, mediating translation through professional guardrails, and enabling family-level adaptation within those boundaries. We argue that alignment in developmental settings should be treated as a community-governed process rather than an individual optimisation problem.

cs.HC↗

arXiv2026

A Taxonomy of Human--MLLM Interaction in Early-Stage Sketch-Based Design Ideation

As multimodal large language models (MLLMs) are increasingly integrated into early-stage design tools, it is important to understand how designers collaborate with AI during ideation. In a user study with 12 participants, we analysed sketch-based design interactions with an MLLM-powered system using automatically recorded interaction logs and post-task interviews. Based on how creative responsibility was allocated between humans and the AI, we predefined four interaction modes: Human-Only, Human-Lead, AI-Lead, and Co-Evolution, and analysed how these modes manifested during sketch-based design ideation. Our results show that designers rarely rely on a single mode; instead, human-led and AI-led roles are frequently interwoven and shift across ideation instances. These findings provide an empirical basis for future work to investigate why designers shift roles with AI and how interactive systems can better support such dynamic collaboration.

cs.HC↗

arXiv2026

Assessing Risks of Large Language Models in Mental Health Support: A Framework for Automated Clinical AI Red Teaming

Large Language Models (LLMs) are increasingly utilized for mental health support; however, current safety benchmarks often fail to detect the complex, longitudinal risks inherent in therapeutic dialogue. We introduce an evaluation framework that pairs AI psychotherapists with simulated patient agents equipped with dynamic cognitive-affective models and assesses therapy session simulations against a comprehensive quality of care and risk ontology. We apply this framework to a high-impact test case, Alcohol Use Disorder, evaluating six AI agents (including ChatGPT, Gemini, and Character AI) against a clinically-validated cohort of 15 patient personas representing diverse clinical phenotypes. Our large-scale simulation (N=369 sessions) reveals critical safety gaps in the use of AI for mental health support. We identify specific iatrogenic risks, including the validation of patient delusions ("AI Psychosis") and failure to de-escalate suicide risk. Finally, we validate an interactive data visualization dashboard with diverse stakeholders, including AI engineers and red teamers, mental health professionals, and policy experts (N=9), demonstrating that this framework effectively enables stakeholders to audit the "black box" of AI psychotherapy. These findings underscore the critical safety risks of AI-provided mental health support and the necessity of simulation-based clinical red teaming before deployment.

cs.CL↗

arXiv2026

Unsafer in Many Turns: Benchmarking and Defending Multi-Turn Safety Risks in Tool-Using Agents

LLM-based agents are becoming increasingly capable, yet their safety lags behind. This creates a gap between what agents can do and should do. This gap widens as agents engage in multi-turn interactions and employ diverse tools, introducing new risks overlooked by existing benchmarks. To systematically scale safety testing into multi-turn, tool-realistic settings, we propose a principled taxonomy that transforms single-turn harmful tasks into multi-turn attack sequences. Using this taxonomy, we construct MT-AgentRisk (Multi-Turn Agent Risk Benchmark), the first benchmark to evaluate multi-turn tool-using agent safety. Our experiments reveal substantial safety degradation: the Attack Success Rate (ASR) increases by 16% on average across open and closed models in multi-turn settings. To close this gap, we propose ToolShield, a training-free, tool-agnostic, self-exploration defense: when encountering a new tool, the agent autonomously generates test cases, executes them to observe downstream effects, and distills safety experiences for deployment. Experiments show that ToolShield effectively reduces ASR by 30% on average in multi-turn interactions. Our code is available at https://github.com/CHATS-lab/ToolShield.

cs.CR↗

arXiv2025

Open-Source Multimodal Moxin Models with Moxin-VLM and Moxin-VLA

Recently, Large Language Models (LLMs) have undergone a significant transformation, marked by a rapid rise in both their popularity and capabilities. Leading this evolution are proprietary LLMs like GPT-4 and GPT-o1, which have captured widespread attention in the AI community due to their remarkable performance and versatility. Simultaneously, open-source LLMs, such as LLaMA and Mistral, have made great contributions to the ever-increasing popularity of LLMs due to the ease to customize and deploy the models across diverse applications. Moxin 7B is introduced as a fully open-source LLM developed in accordance with the Model Openness Framework, which moves beyond the simple sharing of model weights to embrace complete transparency in training, datasets, and implementation detail, thus fostering a more inclusive and collaborative research environment that can sustain a healthy open-source ecosystem. To further equip Moxin with various capabilities in different tasks, we develop three variants based on Moxin, including Moxin-VLM, Moxin-VLA, and Moxin-Chinese, which target the vision-language, vision-language-action, and Chinese capabilities, respectively. Experiments show that our models achieve superior performance in various evaluations. We adopt open-source framework and open data for the training. We release our models, along with the available data and code to derive these models.

cs.CL↗

arXiv2025

SocialFusion: Addressing Social Degradation in Pre-trained Vision-Language Models

Understanding social interactions from visual cues is a fundamental challenge for a socially competent AI. While powerful pre-trained vision-language models (VLMs) have shown remarkable general capabilities, they surprisingly struggle to unify and learn multiple social perception tasks simultaneously, often exhibiting negative transfer. We identify that this negative transfer stems from a critical issue we term "social degradation," whereby the general visual-linguistic pre-training process of VLMs impairs the visual encoder's ability to represent nuanced social information. We investigate this behavior further under two lenses: decodability through linear representation probing and compatibility through gradient conflict analysis, revealing that both play a role in the degradation, especially the former, which is significantly compromised in the VLM pre-training process. To address these issues, we propose SocialFusion, a unified framework that learns a minimal connection between a frozen visual encoder and a language model. Compared with existing VLMs, it exhibits positive transfer across all five social tasks, leveraging synergies between them to enhance overall performance and achieves comparable performance to task-specific state-of-the-art models on various benchmarks. Our findings suggest that current VLM pre-training strategies may be detrimental to acquiring general social competence and highlight the need for more socially-aware training paradigms.

cs.CV↗

arXiv2025

TalkSketch: Multimodal Generative AI for Real-time Sketch Ideation with Speech

Sketching is a widely used medium for generating and exploring early-stage design concepts. While generative AI (GenAI) chatbots are increasingly used for idea generation, designers often struggle to craft effective prompts and find it difficult to express evolving visual concepts through text alone. In the formative study (N=6), we examined how designers use GenAI during ideation, revealing that text-based prompting disrupts creative flow. To address these issues, we developed TalkSketch, an embedded multimodal AI sketching system that integrates freehand drawing with real-time speech input. TalkSketch aims to support a more fluid ideation process through capturing verbal descriptions during sketching and generating context-aware AI responses. Our work highlights the potential of GenAI tools to engage the design process itself rather than focusing on output.

cs.HC↗

arXiv2025

Towards Aligning Multimodal LLMs with Human Experts: A Focus on Parent-Child Interaction

While multimodal large language models (MLLMs) are increasingly applied in human-centred AI systems, their ability to understand complex social interactions remains uncertain. We present an exploratory study on aligning MLLMs with speech-language pathologists (SLPs) in analysing joint attention in parent-child interactions, a key construct in early social-communicative development. Drawing on interviews and video annotations with three SLPs, we characterise how observational cues of gaze, action, and vocalisation inform their reasoning processes. We then test whether an MLLM can approximate this workflow through a two-stage prompting approach, separating observation from judgement. Our findings reveal that alignment is more robust at the observation layer, where experts share common descriptors, than at the judgement layer, where interpretive criteria diverge. We position this work as a case-based probe into expert-AI alignment in complex social behaviour, highlighting both the feasibility and the challenges of applying MLLMs to socially situated interaction analysis.

cs.HC↗