Pinlong Cai

REOPD's token-wise adaptability allows for more stable and reliable training, reducing the risk of reward hacking while enhancing performance across varied domains.

Shifting the focus from static state transitions to dynamic, agent-centric feedback could revolutionize how we train and evolve intelligent agents.

Forget static memory—MemHarness reconstructs past experiences to fit the present context, dramatically boosting decision-making performance in LLM agents.

Outperforming GPT-5.4, IndustryForge-27B achieves a remarkable 33.65 percentage point lift in CAD-specific tasks, setting a new standard for multimodal models in industrial applications.

A lightweight proxy model can efficiently discover high-reward behaviors, enabling significant performance boosts in stronger LLMs without the computational burden of traditional methods.

AssemCAD transforms the landscape of CAD assembly generation by ensuring that mechanical assemblies are not only generated but also validated against engineering principles, achieving unprecedented levels of physical consistency.

IterCAD achieves a significant leap in CAD generation by enabling iterative, interactive design processes that align with real-world manufacturing practices.

COM-based execution outperforms GUI interactions, achieving state-of-the-art results in professional software manipulation tasks where traditional methods fail.

EviProp achieves superior evidence-page retrieval by leveraging a novel graph-based approach that captures complex document structures, outperforming traditional methods.

IA-RAG reveals that modeling knowledge as dynamic time intervals can drastically improve temporal reasoning in language models, outperforming traditional static approaches.

AgentSchool offers a powerful new way to simulate educational environments, moving beyond simple role-play to model learning as a dynamic state transition and providing a testbed for long-horizon memory and multi-agent coordination.