Junnan Li

UvA-DARE (Digital Academic Repository) Reward-Guided Speculative Decoding for Efficient LLM Reasoning

StateAct: Program State, before Pixels, for Long-Horizon Computer-Use Agents

Computer-use agents are usually improved by strengthening perception: better models for reading a screenshot and choosing where to click. Yet a screenshot is only a lossy rendering of the underlying program state, e.g., the files, application backends, and DOM that hold the task data. Different states can produce the s...

MCP-Universe RL: A Framework for Training MCP Tool-Use Agents via Reinforcement Learning

This work trains software-engineering, deep-research, and general tool-use agents on gpt-oss-20b and improves task reward in all three and presents MCP-Universe RL (MCP-U RL), an open-source framework that takes over both.