Reinforcement Learning with Model-rewarded Thinking (RLMT), a new technique that trains language models to explicitly generate detailed Chain-of-Thought reasoning before answering, making performance on open-ended chat and creative tasks better.
ARK-V1, a novel LLM agent designed to iteratively explore Knowledge Graphs to answer complex natural language queries, particularly those requiring a blend of factual and commonsense knowledge.
This research introduces PDDL-INSTRUCT, a novel instruction tuning framework that dramatically enhances Large Language Models’ (LLMs) ability to perform structured symbolic planning by integrating explicit logical Chain-of-Thought (CoT) reasoning and external verification feedback.
NO MATCHING ENTRIES FOUND