ReContext: Recursive Evidence Replay as LLM Harness for Long-Context Reasoning

Understanding and reasoning over long contexts has become a key requirement for deploying large language models (LLMs) in realistic applications.

Understanding and reasoning over long contexts has become a key requirement for deploying large language models (LLMs) in realistic applications. Although recent LLMs support increasingly long context windows, they often fail to use relevant evidence that is already present in…