12 papers
Revisiting the 16x Pretraining Compute Gap between Diffusion and Autoregressive Language Models
Guanghan Wang, Bing He, Zhan Shi +3
Diffusion language models (DLMs) have attracted wide attention for their fast, parallel decoding, but they are believed to be far less efficient to train than autoregressive langua…
Scaling Automatic Research Agents via World Models
Xiyuan Yang, Sheikh Sarwar, Jingru Cheng +8
Automating empirical research is a long-standing direction of AI. Recent automatic research (AutoResearch) agents bring this goal within reach, as modern LLMs show the capability t…
TRACE: Training Reasoning Agents for Causal Exploration with Synthesized Rewards
Rui Sun, Zhan Shi, Bing He
Reinforcement learning with verifiable rewards (RLVR) has advanced language-model reasoning in domains such as mathematics and code, where objective answers are inexpensive to chec…
Evo-Harness: Context-to-Harness Skill Compilation for Self-Evolving Agents
Tianxin Wei, Zhan Shi, Minhua Lin +14
Learning from experience is critical for developing capable, self-improving large language model (LLM) agents. Existing methods typically extract knowledge from accumulated traject…
@skills: Attention is all you have
Li Yin, Zhi Li, Zhan Shi +4
There are 56,804 public agent skills today, and teams write many more privately. The dominant delivery model is installation: once installed, a skill's description remains in the s…
Adaptive Auto-Harness: Sustained Self-Improvement for Agentic System Deployment on Open-Ended Task Streams
Zewen Liu, Zhan Shi, Yisi Sang +7
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 feedba…
A-Evolve-Training: Autonomous Post-Training of a 30B Model
Zhan Shi, Bing He, Yisi Sang +2
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 sys…
Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents
Minhua Lin, Juncheng Wu, Zijun Wang +14
LLM agents are increasingly deployed as systems built around editable external harnesses, including prompts, skills, memories and tools, that shape task execution without changing…
Controllable and Verifiable Tool-Use Data Synthesis for Agentic Reinforcement Learning
Siyuan Xu, Shiyang Li, Xin Liu +9
Existing synthetic tool-use corpora are primarily designed for offline supervised fine-tuning, yet reinforcement learning (RL) requires executable environments that support reward-…
Position: Agentic Evolution is the Path to Evolving LLMs
Minhua Lin, Hanqing Lu, Zhan Shi +11
As Large Language Models (LLMs) move from curated training sets into open-ended real-world environments, a fundamental limitation emerges: static training cannot keep pace with con…
How Do Latent Reasoning Methods Perform Under Weak and Strong Supervision?
Yingqian Cui, Zhenwei Dai, Bing He +7
Latent reasoning has been recently proposed as a reasoning paradigm and performs multi-step reasoning through generating steps in the latent space instead of the textual space. Thi…
Seeing but Not Believing: Probing the Disconnect Between Visual Attention and Answer Correctness in VLMs
Zhining Liu, Ziyi Chen, Hui Liu +9
Vision-Language Models (VLMs) achieve strong results on multimodal tasks such as visual question answering, yet they can still fail even when the correct visual evidence is present…