SESKStuff Everyone Should Know
SESK: Stuff Everyone Should Know · sesk.info/ai-lost-in-the-middle

In a long prompt, where is key information most likely to be missed by a language model?

In the middle.

Start playing free

  1. In the middle ✔
  2. At the very beginning
  3. At the very end

Why: A 2023 study found accuracy was highest when the relevant fact sat at the start or end of the input, and dropped when it was buried in the middle. The accuracy plot makes a U shaped curve.

Huh, didn't know that: The paper's title says it all: Lost in the Middle.

The dad joke
I'm like a language model with a long book. I remember the first chapter and the ending, and the middle is a mystery.
Like this card?
Play the game

References

Answer: In the middle. A 2023 study found accuracy was highest when the relevant fact sat at the start or end of the input, and dropped when it was buried in the middle. The accuracy plot makes a U shaped curve.

  1. Liu et al., Lost in the Middle: How Language Models Use Long Contexts (arXiv 2307.03172) arxiv.org/abs/2307.03172
    performance is often highest when relevant information occurs at the beginning or end of the input context
    Checked 2026-10-10.
  2. Liu et al., Lost in the Middle (full paper PDF, Figure 1) arxiv.org/pdf/2307.03172
    results in a U-shaped performance curve
    Checked 2026-10-10.

Think this answer is wrong? Tap "Challenge this answer" on the card in the game and tell us why.

Test yourself in the game

Artificial IntelligenceWhat does an LLM's context window measure?Artificial IntelligenceHow do AI, machine learning and deep learning fit together?Artificial IntelligenceWhat did Alan Turing call his 1950 test of machine intelligence?

That's stuff everyone should know.