Yisi Sang

Evo-Harness: Context-to-Harness Skill Compilation for Self-Evolving Agents

Tianxin Wei, Zhan Shi, Minhua Lin, Bing He et autres

Learning from experience is critical for developing capable, self-improving large language model (LLM) agents. Existing methods typically extract knowledge from accumulated trajectories via reflection, memory, rules, or skills. However, agents in realistic environments continuously encounter novel tasks, often offering only a one-shot …

SENTINEL: Failure-Driven Reinforcement Learning for Training Tool-Using Language Model Agents

Ziyi Wang, Yuxuan Lu, Yimeng Zhang, Qun Liu et autres

Language model agents are increasingly effective in solving realistic tasks through multi-turn tool use. However, training reliable tool-using agents remains challenging in practice. While reinforcement learning provides an on-policy paradigm for improving agents from their own environment interactions, its effectiveness depends heavily …

SENTINEL: Failure-Driven Reinforcement Learning for Training Tool-Using Language Model Agents

Ziyi Wang, Yuxuan Lu, Yimeng Zhang, Qun Liu et autres

Language model agents are increasingly effective in solving realistic tasks through multi-turn tool use. However, training reliable tool-using agents remains challenging in practice. While reinforcement learning provides an on-policy paradigm for improving agents from their own environment interactions, its effectiveness depends heavily …

A-Evolve-Training: Autonomous Post-Training of a 30B Model

Zhan Shi, Bing He, Yisi Sang, Hüseyin Ozan Ҫirkinoğlu et autres

Post-training a frontier model is normally weeks of human work: proposing data and recipe changes, launching runs, reading evals, deciding what to keep. We report an autonomous system that runs this loop with no human in the loop, post-training a 30B Nemotron …

A-Evolve-Training: Autonomous Post-Training of a 30B Model

Zhan Shi, Bing He, Yisi Sang, Benoit Dumoulin et autres

Post-training a frontier model is normally weeks of human work: proposing data and recipe changes, launching runs, reading evals, deciding what to keep. We report an autonomous system that runs this loop with no human in the loop, post-training a 30B Nemotron …

us (code pays fourni par la source)

Can LLM Agents Simulate Multi-Turn Human Behavior? Evidence from Real Online Customer Behavior Data

Association for Computational Linguistics 2026, Bennett Bei, Yan Han, Qi He et autres

Recent research shows that LLM Agents can generate ``believable'' human behaviors via prompt-only methods, and such agents have been increasingly adopted in downstream applications. However, existing evaluation of these agents only focuses on qualitative believability (whether human raters think they are accurate), …

us, cn, mx (code pays fourni par la source)

Trajectory2Task: Training Robust Tool-Calling Agents with Synthesized Yet Verifiable Data for Complex User Intents

Association for Computational Linguistics 2026, Pei Chen, Ziwei Dong, Jiri Gesi et autres

Tool-calling agents are increasingly deployed in real-world customer-facing workflows. Yet most studies on tool-calling agents focus on idealized settings with general, fixed, and well-specified tasks. In real-world applications, user requests are often (1) ambiguous, (2) changing over time, or (3) infeasible due …

us, cn, mx (code pays fourni par la source)

LLMs for Now, Fine-Tuning for Later: An Ensemble Approach to Data Drift in Domain-Specific Tasks

Association for Computational Linguistics 2026, Hansu Gu, Toby Li, Tun Lu et autres

Deploying machine learning models in real-world domain-specific scenarios is challenged by the scarcity of expert annotations and by data drift, where the statistical properties of incoming data continuously evolve. Active Learning (AL) iteratively improves compact models with expert annotations but suffers from …

us, cn, mx (code pays fourni par la source)

Adaptive Auto-Harness: Sustained Self-Improvement for Agentic System Deployment on Open-Ended Task Streams

Zewen Liu, Zhan Shi, Yisi Sang, Bing He et autres

Auto-harness systems such as A-Evolve, GEPA, and Meta-Harness improve LLM agents by optimizing prompts, skills, tools, memories, and supporting infrastructure from execution feedback, but they are typically evaluated on fixed offline benchmarks. Real deployments instead present open-ended task streams: histories grow without …

Adaptive Auto-Harness: Sustained Self-Improvement for Agentic System Deployment on Open-Ended Task Streams

Zewen Liu, Zhan Shi, Yisi Sang, Bing He et autres

Auto-harness systems such as A-Evolve, GEPA, and Meta-Harness improve LLM agents by optimizing prompts, skills, tools, memories, and supporting infrastructure from execution feedback, but they are typically evaluated on fixed offline benchmarks. Real deployments instead present open-ended task streams: histories grow without …

us, de, jm, mx (code pays fourni par la source)

Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents

Minhua Lin, Juncheng Wu, Zijun Wang, Zhan Shi et autres

LLM agents are increasingly deployed as systems built around editable external harnesses, including prompts, skills, memories and tools, that shape task execution without changing model parameters. Harness self-evolution adapts such agents by updating these harnesses from execution evidence. Yet it remains unclear …

Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents

Minhua Lin, Juncheng Wu, Zijun Wang, Zhan Shi et autres

LLM agents are increasingly deployed as systems built around editable external harnesses, including prompts, skills, memories and tools, that shape task execution without changing model parameters. Harness self-evolution adapts such agents by updating these harnesses from execution evidence. Yet it remains unclear …

us, de, jm, mx (code pays fourni par la source)