Deeks
11.7K posts
Deeks
@deeks
Just another Guy wasting a perfectly good name. Dad. Short Volatility. sometimes i rant in german about life 😆 AstroAvocado Dad
Присоединился Temmuz 2008
4.6K Подписки1.6K Подписчики
LLMs process text from left to right — each token can only look back at what came before it, never forward. This means that when you write a long prompt with context at the beginning and a question at the end, the model answers the question having "seen" the context, but the context tokens were generated without any awareness of what question was coming. This asymmetry is a basic structural property of how these models work.
The paper asks what happens if you just send the prompt twice in a row, so that every part of the input gets a second pass where it can attend to every other part. The answer is that accuracy goes up across seven different benchmarks and seven different models (from the Gemini, ChatGPT, Claude, and DeepSeek series of LLMs), with no increase in the length of the model's output and no meaningful increase in response time — because processing the input is done in parallel by the hardware anyway.
There are no new losses to compute, no finetuning, no clever prompt engineering beyond the repetition itself.
The gap between this technique and doing nothing is sometimes small, sometimes large (one model went from 21% to 97% on a task involving finding a name in a list). If you are thinking about how to get better results from these models without paying for longer outputs or slower responses, that's a fairly concrete and low-effort finding.
Read with AI tutor: chapterpal.com/s/1b15378b/pro…
Get the PDF: arxiv.org/pdf/2512.14982
English
Improve your LLM response by repeating the prompt verbatim.
This means your input should look like:
Prompt question. Prompt question.
Women have known this for centuries.
BURKOV@burkov
LLMs process text from left to right — each token can only look back at what came before it, never forward. This means that when you write a long prompt with context at the beginning and a question at the end, the model answers the question having "seen" the context, but the context tokens were generated without any awareness of what question was coming. This asymmetry is a basic structural property of how these models work. The paper asks what happens if you just send the prompt twice in a row, so that every part of the input gets a second pass where it can attend to every other part. The answer is that accuracy goes up across seven different benchmarks and seven different models (from the Gemini, ChatGPT, Claude, and DeepSeek series of LLMs), with no increase in the length of the model's output and no meaningful increase in response time — because processing the input is done in parallel by the hardware anyway. There are no new losses to compute, no finetuning, no clever prompt engineering beyond the repetition itself. The gap between this technique and doing nothing is sometimes small, sometimes large (one model went from 21% to 97% on a task involving finding a name in a list). If you are thinking about how to get better results from these models without paying for longer outputs or slower responses, that's a fairly concrete and low-effort finding. Read with AI tutor: chapterpal.com/s/1b15378b/pro… Get the PDF: arxiv.org/pdf/2512.14982
English
☄️ Build a Space Shooter game in 20 minutes with no coding needed!
@petergyang & his 7-year-old (who can't type yet!) made this with Phaser & Claude Code 🚀
Boss battles, power-ups, visual effects & more!
Read more ▶️ f.mtr.cool/mbnwxfskhw
English
@AnnieForTruth I appreciate your work, but what did I just thought reading this? I should have known better!
English
Recently released Epstein files reveal Trump bragged about “banging” his former 28 year old assistant, Madeleine Westerhout. He would skip going to Mar-a-Lago with Melania for holidays to stay back at the White House to have sex with Westerhout. That is an impeachable offense. ⬇️
English
Deeks ретвитнул
Deeks ретвитнул
🚨 JUST IN: An anti-ICE agitator in Minneapolis is heading to the federal building with a car JAMPACKED FULL of umbrellas, which’ll be used to DEFLECT pepper balls from DHS agents
Again, who the HELL is paying for all this?! ZERO CHANCE it’s the rioters
English
Looks like this is the training that ‘Nazi Barbie’ was talking about.
Crime Net@TRIGGERHAPPYV1
Officer presses ICE agent for not having his lights on during a traffic stop
English
Be brutally honest: Do you believe ICE agent Jonathan David Ross, who fatally shot Renee Nicole Good during a federal immigration operation in Minneapolis, should be charged with murder? Yes or no?
English
Deeks ретвитнул
The Party told you to reject the evidence of your eyes and ears. It was their final, most essential command.
George Orwell, 1984
English
Deeks ретвитнул
The fucking President of the United States everyone, I have no words.
English
Deeks ретвитнул
@WhiteHouse Calling outrage “violence” is a classic power move.
When citizens protest abuse, you call it hate.
When agents abuse power, you call it safety.
That’s not leadership — that’s fear of accountability.
And fuck ICE
English
"The situation is being studied, in its entirety, but the reason these incidents are happening is because the Radical Left is threatening, assaulting, and targeting our Law Enforcement Officers and ICE Agents on a daily basis. They are just trying to do the job of MAKING AMERICA SAFE. We need to stand by and protect our Law Enforcement Officers from this Radical Left Movement of Violence and Hate!" - PRESIDENT DONALD J. TRUMP
English
@JDCocchiarella Yes. This woman represents emotional emptiness at its worst — a chilling mix of cruelty and indifference that makes your blood boil on sight.
English
