Fact · Voice and AI
In tests on long inputs, language model accuracy was highest when the relevant information sat at the start or end of the input, and dropped significantly when it sat in the middle.
Checked against its source on October 17, 2026
In the source's own words
“performance is often highest when relevant information occurs at the beginning or end of the input context, and significantly degrades when models must access relevant information in the middle of long contexts”
Lost in the Middle: How Language Models Use Long Contexts, Nelson F. Liu et al., arXiv (accepted to Transactions of the Association for Computational Linguistics), 2023, Abstract.
Open the original- What it measures
- The authors tested multi-document question answering and key-value retrieval, moving the one relevant document to different positions in the input.
- What it does not show
- The tasks are retrieval tasks in long inputs, not short chat prompts. The size of the effect depends on the model and the input length. The direction is the finding to rely on.