Pick the curve shape that the lost-in-the-middle effect produces

MCQ·Easy·4.0 · 0·~1 min·Asked atElastic Harvey Roblox

Attempt it

**Lost-in-the-middle** is a behavioral pattern in large language models, first documented by Liu et al. in 2023, where the model's ability to use information in a long context depends strongly on where that information sits in the prompt. The shape of the dependence is a **U-curve**. Plot retrieval accuracy on the y-axis and the position of the target chunk in the context on the x-axis, and you get a curve that is high at the start, low in the middle, and high again at the end. The two high regions correspond to **primacy** (the start of the context, where the model attends reliably) and **recency** (the end of the context, where attention is also strong because of how decoder attention naturally weights the most recent tokens). The low region in the middle is where the model often skips information even when that information is directly relevant to answering the question. This effect persists in 2026 even on frontier long-context models like Claude Opus 4.7, GPT-5.5, and Gemini 3.1 Pro, although the depth of the trough varies. The practical context-engineering move is to position important content at the head or tail of the retrieved block, not in the middle. The most important chunk goes first, the second-most goes last, and less critical content fills the middle where the U-curve is low.

The **lost-in-the-middle** effect produces a U-shaped accuracy curve when you plot retrieval accuracy against the position of the target document inside the context window. The two ends of the curve are high (primacy at the start, recency at the end) and the middle is significantly lower. The finding comes from Liu et al. (2023), "Lost in the Middle: How Language Models Use Long Contexts," which evaluated 10 documents at different positions on multi-document QA. Accuracy was around 75% when the target document was first, dropped to around 50% when it was in the middle, and rose back to 70% when it was last. The shape held across models and task variants. In 2026 the U-shape persists but the trough is shallower on frontier models. RULER measurements on Claude Opus 4.7, GPT-5.5, and Gemini 3.1 Pro show high accuracy across positions up to roughly 32K context length, with the U-curve becoming more pronounced past that. The takeaway is not that the effect has been solved, but that it kicks in later than it used to. The mechanism is rooted in attention dynamics. Causal attention naturally weights recent tokens (recency), and the start of context tokens get strong attention because they appear in every attention computation as anchors (primacy). The middle competes for attention against everything else and tends to lose. The practical move is **head and tail positioning** for retrieved content: place the highest-priority chunks at the start and end of the retrieved block, leave the middle for context that supports rather than answers the query. Adding a question restatement just before the answer-generation cue is a complementary trick that exploits recency directly.

TL;DR

Lost-in-the-middle produces a U-shaped accuracy curve: models attend reliably to the start (primacy) and the end (recency) of a long context but skip content placed in the interior.

Memory aid

Easy to grasp

Imagine a teacher reading a long list of names and later asking which names were on it. Students remember the first few (they were paying full attention at the start) and the last few (those are still ringing in their ears). The middle names get fuzzy. Language models do something similar with long context: the names at the start and end stick, the names in the middle blur. Plot what they remember against where the name was in the list, and you get a U-shape: high on both ends, low in the middle. The trick for context engineering is to put the important stuff where the U is high, not where the U is low.

Key concepts

Context Engineering Attention

Concept explanation~2 min read

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.

The lost-in-the-middle effect is one of the most robust empirical findings in long-context language modeling. It says that retrieval and reasoning accuracy depend on where the target content sits in the context window, with the dependence taking a U-shape: high at the start (primacy), low in the middle, high at the end (recency). This deep dive walks through the original finding, the attention-mechanism intuitions that explain the shape, the 2024-2026 replications on frontier models, and the context-engineering moves that exploit the curve rather than fighting it.

The original finding: Liu et al. (2023)

Nelson Liu and collaborators published "Lost in the Middle: How Language Models Use Long Contexts" in July 2023, and the paper introduced both the name and the curve shape that the field now treats as canonical.

The setup was multi-document question answering. Given a query and 10, 20, or 30 retrieved Wikipedia paragraphs, one of which contained the answer, the question was: does the model's ability to use the answer-containing paragraph depend on where that paragraph sits in the input?

The answer was yes, and the dependence had a striking shape. When the gold paragraph was at position 1 (first), accuracy was around 75%. When it was at the last position, accuracy was around 70%. When it was in the middle (position 5 out of 10), accuracy dropped to around 50%. Plotting accuracy against position produced a clear U-curve.

The finding was not specific to one model. Liu et al. tested GPT-3.5, GPT-4, Claude 1.3, MPT-30B, and LongChat-13B. All five showed the U-shape, with varying depths. The conclusion was that lost-in-the-middle is a structural property of how decoder transformers attend over long context, not an artifact of any one model's training.

The paper also showed the effect on key-value retrieval (find the value associated with a key in a JSON-like input). Here too the U-curve appeared, and here too it was robust across models. The structural story was strong enough that the finding immediately became a default assumption in long-context prompt engineering.

Why the curve is U-shaped: attention mechanics

2024-2026 replications on frontier models

Context-engineering moves that exploit the curve

Situations where this technique stops working.

2–4 min · Everything important, quickly.

Real products, models, and research that use this idea.

Liu et al. (2023) Lost in the Middle paper plotted the original U-curve on multi-document QA
RULER (NVIDIA, 2024) replicated the U-shape across modern long-context models at varying lengths

What an interviewer would ask next. Try answering before peeking at the approach.

QWhat does an inverted U-shape look like, and when do you see it in practice?

An inverted U is high in the middle and low at the ends. You rarely see it in language-model retrieval. The closest analog is a model trained heavily on the middle of long contexts (e.g., book continuation), but even there primacy and recency still dominate. The inverted U is mostly a distractor option in multiple choice.

2 more follow-ups an interviewer would ask next. Sign in to reveal them.

Red flags & common mistakes

The phrases that signal junior thinking. Click to expand.

Most common mistake

Expecting accuracy to decay monotonically with position and missing that the start is also a privileged location, not just the end.

60 second bullets to scan on the way to the call.

State that the lost-in-the-middle curve is U-shaped
Identify primacy and recency as the two high regions

Primary sources. Browse if you want the original framing.