MEMAUDIT: Optimizing Budgeted Long-Term LLM Memory Writing
Discover MEMAUDIT, a precise evaluation protocol for efficient, budgeted long-term memory writing in large language models (LLMs).
T2PO: Stable Multi-Turn RL with Uncertainty-Guided Exploration
Discover T2PO, an uncertainty-guided framework that stabilizes multi-turn reinforcement learning by optimizing exploration at token and turn levels.
Intervention Complexity: A New Measure of AI Intelligence
Discover intervention complexity, a novel AI intelligence measure refining reward functions for better evaluation of agent competence and learning efficien...
Evaluating LLMs on 1M-Token Contexts for Classical Chinese
Explore LLM performance on 1M-token context windows with retrieval and multi-hop reasoning in classical Chinese texts. Key insights on model accuracy and d...
Efficient Multi-Agent Framework for Long-Horizon Planning
Discover a novel multi-agent framework that boosts long-horizon planning efficiency by prioritizing planner roles and reinforcement learning.