How to improve your LLM results by duplicating prompts

If you want to get the best results from LLM without paying for long outputs or finesse tuning, here a concrete, elementary trick:

Duplicate your prompt!

The researchers found that repeating exactly the same innings can dramatically improve performance (boost to 76% on specific tasks).

LLM processes text from left to right, each token can only look at the previous context, never forward.

So when you write a long prompt with a context at the beginning and a question at the end, the model can draw on that context for an answer, but the context was processed before the model even recognised the question.

This asymmetry is a basic structural property of the way in which things work LLM.

Prompt repetition helps circumvent this limitation by giving the model a second pass over the full context.

There is no calculation of new losses here and no convoluted prompt engineering.

It's just a structural hack that works on almost every major model they've tested.

Here's a piper: https://arxiv.org/pdf/2512.14982