arXiv subjects

Lyumanshan Ye

Publications and source records attributed to Lyumanshan Ye.

At least 19 recordsLinked to original sources

arXiv2026

MOPD-Router: Rethinking Teacher Routing in Multi-Teacher On-Policy Distillation

Multi-teacher on-policy distillation (MOPD) integrates specialized capabilities into a single student, but existing practice typically hard-routes each prompt to a domain-matched teacher for the entire rollout. This dependence on prompt-level domain labels restricts using unlabeled training mixtures and leaves complementary signals from other teachers unused. We introduce MOPD-Router, a framework that routes supervision over the full teacher pool at each token, without domain labels or training a separate routing model. Its plug-in interface supports different metrics for selecting and weighting teacher-specific OPD signals. Within this interface, we propose ExpertAlign, which scores each teacher by whether its correction to the student at the current token expresses the specialization that teacher acquired during post-training, and compare it against two reference metrics built on teacher confidence (Entropy) and teacher-student discrepancy (Novelty). Experiments on unlabeled and domain-labeled training mixtures under strong-to-weak and same-size distillation scenarios show that ExpertAlign achieves the strongest overall performance in all four settings. On unlabeled data, it improves the overall score by 5.88 (+12.3%) points over Mean aggregation; on domain-labeled data, it outperforms standard MOPD by 3.95 (+7.8%) points without using available domain labels. These results demonstrate token-level routing can exploit cross-domain complementary supervision, and reduce exclusive reliance on prompt-level domain assignment. Code is available at: https://github.com/TURLEing/MOPD-Router.

cs.LG↗

arXiv2026

Mind2Dialogue: Training Human-Aware Language Models by Simulating User Mental States

As language models become more capable, long-term collaboration in learning, reasoning, and decision-making calls for a deeper understanding of the people they serve. Yet training such human-aware language models faces a fundamental supervision gap because current datasets for LLM assistant training contain few if any well-informed responses explicitly grounded in users' unspoken beliefs and goals. Scaling such supervision is inherently constrained, as users' underlying states are not directly observable. We thus propose the Mind2Dialogue framework to mitigate this gap by simulating users' mental states and turning them into privileged supervision for human-aware training. Specifically, we first propose a psychology-guided simulator that preserves personal characteristics while updating mental states through interaction to generate coherent conversations. The key idea is to enforce a shared evolving mental state that drives user behavior and guides an Oracle assistant's responses. Our privileged distillation then trains models on the Oracle's well-informed responses to assist users without direct access to their mental states at deployment. Moreover, we propose to evaluate human-aware learning by combining personalization and theory of mind, examining how models understand people and act on that understanding. Training on the full Mind2Dialogue corpus improves every reported personalization metric over the corresponding Qwen, Llama, and OLMo instruction-tuned baselines, including gains of 26.6 to 40.9 percentage points in preference-following generation. The gains extend to belief and action reasoning on Qwen and Llama, beyond personalized assistance. Looking forward, Mind2Dialogue makes user simulation a foundation for genuine AI collaborators that understand beliefs and intentions behind people's words and support their long-term goals across education, work, and everyday life.

cs.CL↗

arXiv2026

ERPBench: Evaluating LLM Agents for Enterprise Decision-Making Across Competitive Market Ecologies

Large language model (LLM) agents are increasingly proposed for enterprise workflows, yet existing evaluations rarely test whether business-decision conclusions transfer across competitive market ecologies. We introduce ERPBench, an execution-instrumented benchmark for enterprise decision agents in a six-round Enterprise Resource Planning (ERP) simulation with coupled pricing, production, procurement, inventory, finance, and shared-market competition. ERPBench evaluates the same 100 fixed problems in two matched competitive market ecologies: Solo, where each evaluated LLM agent competes against fixed rule-based opponents, and Arena, where six evaluated LLM agents compete in a shared market. Across six model families, this yields 1,200 model-level trajectories spanning 7,200 decision rounds. Under the observed service configuration, the leading model differs between ecologies: DeepSeek leads in Solo (252.29M mean valuation; mean rank 1.67), whereas Gemini leads in Arena (263.95M; 1.76). The two ecologies identify the same task-level winner on only 21 of 100 problems, and Gemini's bottom-rank rate falls from 22 % to 0 % in Arena. ERPBench supports paired evaluation of whether enterprise-agent rankings transfer across competitive market ecologies, supplemented by aggregate execution-intervention analysis. Code and benchmark resources are available in our https://github.com/GAIR-NLP/erp-bench.

cs.AI↗

arXiv2026

Balancing Safety and Autonomy: Accessibility-Oriented Interventions in Generative AI for Cognitive Impairment

Generative AI systems are increasingly used by older adults with cognitive impairment for everyday tasks such as information seeking, health management, and communication. While these systems provide flexible, language-based support, their open-ended outputs introduce risks of over-reliance, misinterpretation, and inappropriate decision-making. Prior work has focused on usability and adoption, with limited attention to how system design shapes users' participation in decision-making and the distribution of agency in care contexts. We present a qualitative study of 45 individuals with cognitive impairment and their caregivers. We identify five accessibility-oriented mechanisms: AI Capability Constraint, Human Oversight Embedding, Cognitive Engagement Maintenance, Human-AI Relationship Regulation, and Risk Transparency and Control, through which systems structure interaction. These mechanisms both support and constrain users by redistributing decision-making across users and caregivers. We show that their effects vary by impairment level: while protective mechanisms support users with severe impairment, they can restrict autonomy for those with mild impairment. As impairment progresses, tensions become less visible as user participation diminishes. Our findings highlight the need for dynamic designs that balance safety and autonomy in AI-supported care.

cs.HC↗

arXiv2026

"Everyone Says Them": Deception Typologies, Probabilistic Trust, and Grassroots Safety Knowledge Among Gay Dating App Users in China

Gay dating applications have become critical platforms for sexual minority men to seek relationships and community, yet they also expose users to deceptive interactions that remain underexplored in HCI and CSCW research. This study examines how gay male users in China experience, identify, and respond to deception on dating applications. Through semi-structured interviews with 22 participants across platforms including Blued, Aloha, Fanka, and Soul, we make three contributions. First, we identify a typology of deceptive practices extending beyond profile misrepresentation to encompass relational, emotional, financial, and commercial forms of deception. Second, we document the layered, probabilistic verification strategies users develop through long-term platform use, showing that trust assessment operates as a multi-signal, provisional process rather than a binary judgment. Third, we demonstrate that risk recognition is a collaborative practice shaped by the circulation of experience, the abstraction of recurrent tactics, and the codification of shared rules within the community.

cs.HC↗

arXiv2026

AlphaEval: Evaluating Agents in Production

The rapid deployment of AI agents in commercial settings has outpaced the development of evaluation methodologies that reflect production realities. Existing benchmarks measure agent capabilities through retrospectively curated tasks with well-specified requirements and deterministic metrics -- conditions that diverge fundamentally from production environments where requirements contain implicit constraints, inputs are heterogeneous multi-modal documents with information fragmented across sources, tasks demand undeclared domain expertise, outputs are long-horizon professional deliverables, and success is judged by domain experts whose standards evolve over time. We present AlphaEval, a production-grounded benchmark of 94 tasks sourced from seven companies deploying AI agents in their core business, spanning six O*NET (Occupational Information Network) domains. Unlike model-centric benchmarks, AlphaEval evaluates complete agent products -- Claude Code, Codex, etc. -- as commercial systems, capturing performance variations invisible to model-level evaluation. Our evaluation framework covers multiple paradigms (LLM-as-a-Judge, reference-driven metrics, formal verification, rubric-based assessment, automated UI testing, etc.), with individual domains composing multiple paradigms. Beyond the benchmark itself, we contribute a requirement-to-benchmark construction framework -- a systematic methodology that transforms authentic production requirements into executable evaluation tasks in minimal time. This framework standardizes the entire pipeline from requirement to evaluation, providing a reproducible, modular process that any organization can adopt to construct production-grounded benchmarks for their own domains.

cs.CL↗

arXiv2026

Rubrics to Tokens: Bridging Response-level Rubrics and Token-level Rewards in Instruction Following Tasks

Rubric-based Reinforcement Learning (RL) has emerged as a promising approach for aligning Large Language Models (LLMs) with complex, open-domain instruction following tasks. However, existing methods predominantly rely on response-level rewards, introducing severe reward sparsity and reward ambiguity problems. To address these issues, we propose Rubrics to Tokens (RTT), a novel rubric-based RL framework that bridges coarse response-level scores and fine-grained token-level credit assignment. RTT introduces a Token-Level Relevance Discriminator to predict which tokens in the response are responsible for a specific constraint, and optimizes the policy model via RTT-GRPO, which integrates response-level and token-level advantages within a unified framework. Furthermore, when transitioning from one-dimensional, outcome-level reward to three-dimensional reward space in the token-level rubric-based RL, we propose a novel group normalization method, called Intra-sample Token Group Normalization, to accommodate this shift. Extensive experiments and benchmarks demonstrate that RTT consistently outperforms other baselines in both instruction- and rubric-level accuracy across different models.

cs.CL↗

arXiv2026

ASI-Evolve: AI Accelerates AI

Can AI accelerate the development of AI itself? While recent agentic systems have shown strong performance on well-scoped tasks with rapid feedback, it remains unclear whether they can tackle the costly, long-horizon, and weakly supervised research loops that drive real AI progress. We present ASI-Evolve, an agentic framework for AI-for-AI research that closes this loop through a learn-design-experiment-analyze cycle. ASI-Evolve augments standard evolutionary agents with two key components: a cognition base that injects accumulated human priors into each round of exploration, and a dedicated analyzer that distills complex experimental outcomes into reusable insights for future iterations. To our knowledge, ASI-Evolve is the first unified framework to demonstrate AI-driven discovery across three central components of AI development: data, architectures, and learning algorithms. In neural architecture design, it discovered 105 SOTA linear attention architectures, with the best discovered model surpassing DeltaNet by +0.97 points, nearly 3x the gain of recent human-designed improvements. In pretraining data curation, the evolved pipeline improves average benchmark performance by +3.96 points, with gains exceeding 18 points on MMLU. In reinforcement learning algorithm design, discovered algorithms outperform GRPO by up to +12.5 points on AMC32, +11.67 points on AIME24, and +5.04 points on OlympiadBench. We further provide initial evidence that this AI-for-AI paradigm can transfer beyond the AI stack through experiments in mathematics and biomedicine. Together, these results suggest that ASI-Evolve represents a promising step toward enabling AI to accelerate AI across the foundational stages of development, offering early evidence for the feasibility of closed-loop AI research.

cs.AI↗

arXiv2026

Speed by Simplicity: A Single-Stream Architecture for Fast Audio-Video Generative Foundation Model

We present daVinci-MagiHuman, an open-source audio-video generative foundation model for human-centric generation. daVinci-MagiHuman jointly generates synchronized video and audio using a single-stream Transformer that processes text, video, and audio within a unified token sequence via self-attention only. This single-stream design avoids the complexity of multi-stream or cross-attention architectures while remaining easy to optimize with standard training and inference infrastructure. The model is particularly strong in human-centric scenarios, producing expressive facial performance, natural speech-expression coordination, realistic body motion, and precise audio-video synchronization. It supports multilingual spoken generation across Chinese (Mandarin and Cantonese), English, Japanese, Korean, German, and French. For efficient inference, we combine the single-stream backbone with model distillation, latent-space super-resolution, and a Turbo VAE decoder, enabling generation of a 5-second 256p video in 2 seconds on a single H100 GPU. In automatic evaluation, daVinci-MagiHuman achieves the highest visual quality and text alignment among leading open models, along with the lowest word error rate (14.60%) for speech intelligibility. In pairwise human evaluation, it achieves win rates of 80.0% against Ovi 1.1 and 60.9% against LTX 2.3 over 2000 comparisons. We open-source the complete model stack, including the base model, the distilled model, the super-resolution model, and the inference codebase.

cs.CV↗

arXiv2026

ProjDevBench: Benchmarking AI Coding Agents on End-to-End Project Development

Recent coding agents can generate complete codebases from simple prompts, yet existing evaluations focus on issue-level bug fixing and lag behind end-to-end development. We introduce ProjDevBench, an end-to-end benchmark that provides project requirements to coding agents and evaluates the resulting repositories. Combining Online Judge (OJ) testing with LLM-assisted code review, the benchmark evaluates agents on (1) system architecture design, (2) functional correctness, and (3) iterative solution refinement. We curate 20 programming problems across 8 categories, covering both concept-oriented tasks and real-world application scenarios, and evaluate six coding agents built on different LLM backends. Our evaluation reports an overall acceptance rate of 27.38%: agents handle basic functionality and data structures but struggle with complex system design, time complexity optimization, and resource management. Our benchmark is available at https://github.com/zsworld6/projdevbench.

cs.AI↗

arXiv2026

daVinci-Dev: Agent-native Mid-training for Software Engineering

Recently, the frontier of Large Language Model (LLM) capabilities has shifted from single-turn code generation to agentic software engineering-a paradigm where models autonomously navigate, edit, and test complex repositories. While post-training methods have become the de facto approach for code agents, **agentic mid-training**-mid-training (MT) on large-scale data that mirrors authentic agentic workflows-remains critically underexplored due to substantial resource requirements, despite offering a more scalable path to instilling foundational agentic behaviors than relying solely on expensive reinforcement learning. A central challenge in realizing effective agentic mid-training is the distribution mismatch between static training data and the dynamic, feedback-rich environment of real development. To address this, we present a systematic study of agentic mid-training, establishing both the data synthesis principles and training methodology for effective agent development at scale. Central to our approach is **agent-native data**-supervision comprising two complementary types of trajectories: **contextually-native trajectories** that preserve the complete information flow an agent experiences, offering broad coverage and diversity; and **environmentally-native trajectories** collected from executable repositories where observations stem from actual tool invocations and test executions, providing depth and interaction authenticity. We verify the model's agentic capabilities on `SWE-Bench Verified`. We demonstrate our superiority over the previous open software engineering mid-training recipe `Kimi-Dev` under two post-training settings with an aligned base model and agentic scaffold, while using less than half mid-training tokens (73.1B). Besides relative advantage, our best performing 32B and 72B models achieve **56.1%** and **58.5%** resolution rates, respectively, which are ...

cs.SE↗

arXiv2025

A Longitudinal Study on the Attitudes of Gay Men in Beijing Towards Gay Social Media Platforms: Lonely Souls in the Digital Concrete Jungle

Over the past decade, specialized social networking applications have become a cornerstone of life for many gay men in China. This paper employs a longitudinal mixed-methods approach to investigate how Chinese men who have sex with men (MSM) have shifted their attitudes toward these platforms between approximately 2013 and 2023. Drawing on archival analysis of online discourses, a quantitative survey of 412 participants, and in-depth semi-structured interviews with 32 participants, we trace the complex trajectory of this evolution. Our findings reveal a clear pattern: from the initial embrace of these applications as revolutionary tools for community building and identity affirmation (2014--2017), to a period of growing ambivalence and critique centered on commercialization, ``hookup culture,'' and multiple forms of discrimination (2017--2020), and finally to the present era (2020--2023), characterized by pragmatic, fragmented, yet simultaneously critical and reconstructive uses. Today, users strategically employ a repertoire of applications -- including global platforms (e.g., Grindr and Tinder), domestic mainstream platforms (e.g., Blued), and niche alternatives (e.g., Aloha) -- to fulfill differentiated needs. We develop a detailed temporal framework to capture this attitudinal evolution and discuss its design implications for creating more supportive, secure, and community-oriented digital environments for marginalized groups.

cs.HC↗

arXiv2025

DeepPersona: A Generative Engine for Scaling Deep Synthetic Personas

Simulating human profiles by instilling personas into large language models (LLMs) is rapidly transforming research in agentic behavioral simulation, LLM personalization, and human-AI alignment. However, most existing synthetic personas remain shallow and simplistic, capturing minimal attributes and failing to reflect the rich complexity and diversity of real human identities. We introduce DEEPPERSONA, a scalable generative engine for synthesizing narrative-complete synthetic personas through a two-stage, taxonomy-guided method. First, we algorithmically construct the largest-ever human-attribute taxonomy, comprising over hundreds of hierarchically organized attributes, by mining thousands of real user-ChatGPT conversations. Second, we progressively sample attributes from this taxonomy, conditionally generating coherent and realistic personas that average hundreds of structured attributes and roughly 1 MB of narrative text, two orders of magnitude deeper than prior works. Intrinsic evaluations confirm significant improvements in attribute diversity (32 percent higher coverage) and profile uniqueness (44 percent greater) compared to state-of-the-art baselines. Extrinsically, our personas enhance GPT-4.1-mini's personalized question answering accuracy by 11.6 percent on average across ten metrics and substantially narrow (by 31.7 percent) the gap between simulated LLM citizens and authentic human responses in social surveys. Our generated national citizens reduced the performance gap on the Big Five personality test by 17 percent relative to LLM-simulated citizens. DEEPPERSONA thus provides a rigorous, scalable, and privacy-free platform for high-fidelity human simulation and personalized AI research.

cs.AI↗

arXiv2025

Can Intelligent User Interfaces Engage in Philosophical Discussions? A Longitudinal Study of Philosophers' Evolving Perceptions

This study investigates the evolving attitudes of philosophy scholars towards the participation of generative AI based Intelligent User Interfaces (IUIs) in philosophical discourse. We conducted a three year (2023--2025) mixed methods longitudinal study with 16 philosophy scholars and students. Qualitative data from annual interviews reveal a three stage evolution in attitude: from initial resistance and unfamiliarity, to instrumental acceptance of the IUI as a tool, and finally to a deep principled questioning of the IUI's fundamental capacity for genuine philosophical thought. Quantitative data from blind assessments, where participants rated anonymized philosophical answers from both humans and an IUI, complement these findings. While participants acknowledged the IUI's proficiency in tasks requiring formal logic and knowledge reproduction, they consistently identified significant shortcomings in areas demanding dialectical reasoning, originality and embodied understanding. The study concludes that participants do not see the IUI as a peer but rather as a sophisticated mirror whose capabilities and limitations provoke a deeper reflection on the unique and irreplaceable human dimensions of philosophical inquiry, such as intuition, value laden commitment and the courage to question fundamental premises.

cs.HC↗

arXiv2025

Colin: A Multimodal Human-AI Co-Creation Storytelling System To Support Children's Multi-Level Narrative Skills

Children develop narrative skills by understanding and actively building connections between elements, image text matching, and consequences. However, it is challenging for children to clearly grasp these multi level links only through explanations of text or the facilitator's speech. To address this, we developed Colin, an interactive storytelling tool that supports children's multi level narrative skills through both voice and visual modalities. In the generation stage, Colin supports the facilitator to define and review the generated text and image content freely. In the understanding stage, a question feedback model helps children understand multi level connections while co creating stories with Colin. In the building phase, Colin actively encourages children to create connections between elements through drawing and speaking. A user study with 20 participants evaluated Colin by measuring children's engagement, understanding of cause and effect relationships, and the quality of their new story creations. Our results demonstrate that Colin significantly enhances the development of children's narrative skills across multiple levels.

cs.HC↗

arXiv2025

Final Happiness: What Intelligent User Interfaces Can Do for The Lonely Dying

This study explores the design of Intelligent User Interfaces (IUIs) to address the profound existential loneliness of terminally ill individuals. While Human-Computer Interaction (HCI) has made inroads in "Thanatechnology," current research often focuses on practical aspects like digital legacy management, overlooking the subjective, existential needs of those facing death in isolation. To address this gap, we conducted in-depth qualitative interviews with 14 lonely, terminally ill individuals. Our core contributions are: (1) An empirically-grounded model articulating the complex psychological, practical, social, and spiritual needs of this group; (2) The "Three Pillars, Twelve Principles" framework for designing IUIs as "Existential Companions"; and (3) A critical design directive derived from user evaluations: technology in this context should aim for transcendence over simulation. The findings suggest that IUIs should create experiences that augment or surpass human capabilities, rather than attempting to simulate basic human connections, which can paradoxically deepen loneliness. This research provides a clear, user-centered path for designing technology that serves not as a "tool for dying," but as a "partner for living fully until the end".

cs.HC↗

arXiv2025

InnovatorBench: Evaluating Agents' Ability to Conduct Innovative LLM Research

AI agents could accelerate scientific discovery by automating hypothesis formation, experiment design, coding, execution, and analysis, yet existing benchmarks probe narrow skills in simplified settings. To address this gap, we introduce InnovatorBench, a benchmark-platform pair for realistic, end-to-end assessment of agents performing Large Language Model (LLM) research. It comprises 20 tasks spanning Data Construction, Filtering, Augmentation, Loss Design, Reward Design, and Scaffold Construction, which require runnable artifacts and assessment of correctness, performance, output quality, and uncertainty. To support agent operation, we develop ResearchGym, a research environment offering rich action spaces, distributed and long-horizon execution, asynchronous monitoring, and snapshot saving. We also implement a lightweight ReAct agent that couples explicit reasoning with executable planning using frontier models such as Claude-4, GPT-5, GLM-4.5, and Kimi-K2. Our experiments demonstrate that while frontier models show promise in code-driven research tasks, they struggle with fragile algorithm-related tasks and long-horizon decision making, such as impatience, poor resource management, and overreliance on template-based reasoning. Furthermore, agents require over 11 hours to achieve their best performance on InnovatorBench, underscoring the benchmark's difficulty and showing the potential of InnovatorBench to be the next generation of code-based research benchmark.

cs.AI↗

arXiv2025

Interaction as Intelligence Part II: Asynchronous Human-Agent Rollout for Long-Horizon Task Training

Large Language Model (LLM) agents have recently shown strong potential in domains such as automated coding, deep research, and graphical user interface manipulation. However, training them to succeed on long-horizon, domain-specialized tasks remains challenging. Current methods primarily fall into two categories. The first relies on dense human annotations through behavior cloning, which is prohibitively expensive for long-horizon tasks that can take days or months. The second depends on outcome-driven sampling, which often collapses due to the rarity of valid positive trajectories on domain-specialized tasks. We introduce Apollo, a sampling framework that integrates asynchronous human guidance with action-level data filtering. Instead of requiring annotators to shadow every step, Apollo allows them to intervene only when the agent drifts from a promising trajectory, by providing prior knowledge, strategic advice, etc. This lightweight design makes it possible to sustain interactions for over 30 hours and produces valuable trajectories at a lower cost. Apollo then applies supervision control to filter out sub-optimal actions and prevent error propagation. Together, these components enable reliable and effective data collection in long-horizon environments. To demonstrate the effectiveness of Apollo, we evaluate it using InnovatorBench. Our experiments show that when applied to train the GLM-4.5 model on InnovatorBench, Apollo achieves more than a 50% improvement over the untrained baseline and a 28% improvement over a variant trained without human interaction. These results highlight the critical role of human-in-the-loop sampling and the robustness of Apollo's design in handling long-horizon, domain-specialized tasks.

cs.AI↗