GURU: A Reinforcement Learning Framework that Bridges LLM Reasoning Across Six Domains

Recent advancements in reinforcement learning (RL) for large language models (LLMs) have shown promising improvements in reasoning capabilities, particularly in specialized domains such as mathematics and coding, exemplified by systems like OpenAI's GPT-3 and DeepSeek-R1. However, the predominant focus on narrow, well-defined tasks has limited the generalizability of these models, as applying RL to broader reasoning domains remains challenging due to the scarcity of reliable reward signals and curated datasets for open-ended tasks. The development of GURU, a new RL framework, aims to bridge this gap by enabling LLMs to reason