liu-2023-u-shaped-curve-finding
IN premise — summaries/2026/08/24/liu-2023-lost-in-middle-s0-abstract.md
Created 2026-08-25T02:58:08+00:00
Liu et al. (2023) demonstrate that LLMs exhibit a U-shaped performance curve where accuracy is highest when relevant information is at the beginning or end of context and lowest in the middle, across GPT-3.5-Turbo and Claude-1.3.
Summary
Large language models reliably struggle to use information buried in the middle of a long passage, performing best when key details sit at the very start or very end, and this pattern holds across different model families. In practical terms, this means that simply fitting more text into a context window doesn't guarantee comprehension, so the placement of critical information within that window is as important as the window's size.