llm-lost-in-middle-reader-saturation-before-recall

IN premise — summaries/2026/08/24/liu-2023-lost-in-middle-sA-acknowledgments.md

Created 2026-08-25T02:58:09+00:00

In open-domain QA pipelines, the LM reader's accuracy saturates before the retriever's recall is exhausted, meaning adding more retrieved passages does not improve accuracy beyond a point.

Summary

In systems that search for relevant passages and then hand them to a language model to answer a question, there comes a point where the model simply cannot make use of any additional context, even though better retrieval could still find more relevant material. This means the practical bottleneck is the model's ability to process and integrate information, not the search step, so spending more effort retrieving extra passages yields no real gain in answer quality.