QuCo-RAG: Quantifying Uncertainty from the Pre-training Corpus for Dynamic Retrieval-Augmented Generation

In three linesQuCo-RAG proposes dynamic RAG grounded in pre-training corpus statistics rather than model-internal signals. It identifies low-frequency entities and verifies their co-occurrence in 4 trillion tokens using Infini-gram. On multi-hop QA benchmarks, it gains 5–12 EM points over baselines with OLMo-2, and up to 14 points on Llama-3, Qwen2.5, GPT-4.

Summary generated by Claude — human-verified