Google Research: Repeating the Prompt Twice Boosts LLM Accuracy from 21% to 97%
udmrzn · x · 2026-11-16
A Google Research paper, Prompt Repetition Improves Non-Reasoning LLMs, shows that simply duplicating the input prompt (<QUERY> → <QUERY><QUERY>) improves accuracy across popular non-reasoning models including Gemini, GPT, Claude, and DeepSeek.
- Why it works: LLMs are causal language models, so early tokens can't attend to later ones. When a question sits at the bottom of a long document, the model processes the context "blind." Repeating the prompt lets every token attend to the full question.
- Results: Consistent gains across 7 benchmarks × 7 models; one complex extraction task jumped from 21% to 97%.
- Cost: Zero — hardware processes inputs in parallel, so the second pass adds no generated tokens or latency.
A free, immediately testable trick for non-reasoning-mode calls.
More from coding & agent
- Claude Code ranks last on Terminal-Bench 2.0 as simpler harnesses beat it — OwariDa · 2026-11-16
- Dad Builds 3D Puzzle Game for His 5-Year-Old With Fable 5.2: 30 Levels in One HTML File — JohnnyFFM · 2026-11-16
- User cancels $147 of subscriptions in a week with Grokbot, which pays and enters credentials — DominiqueCAPaul · 2026-11-16
- Frustrated researcher codes his own procurement, accounting and HR suite for free — arjunrajlab · 2026-11-16
- Dev compresses 4 years of memories and books into a personal agent in under 600 lines of Python — UltraRareAF · 2026-11-16
- Sentry CEO mocks agent 'play-house' UX of stacking slash-command modes in one prompt — zeeg · 2026-11-16