Hangcheng Zhu

The Relative Surprisal Index reveals that the interplay between token probability and entropy is crucial for optimizing reinforcement learning in language models, leading to substantial performance gains.

LLM memory failures are systematic, stemming from operation-level issues like information loss and retrieval misalignment, and can be automatically corrected with prompt optimization guided by fine-grained error tracing.

LLM agents can achieve state-of-the-art performance in dynamic environments by treating memory as a continuously evolving graph, rather than a static repository.