Fancy prompts don't confuse small LLMs. One format line does

Description and use

<p>Ask an LLM to "improve" your prompt and you often get something like this:</p> <blockquote> <p>Kindly undertake a meticulous examination of the order correspondence reproduced hereunder and distill therefrom the following salient particulars...</p> </blockquote> <p>Does a small local model still understand that? And the other way round: if you write in very simple words, short sentences, like for a child, does that help a 1B model?</p> <p>In my last test I broke prompts with typos and bad punctuation. This time the spelling is perfect. Only the <em>register</em> changes: plain, normal, or o

Problem

<p>Ask an LLM to "improve" your prompt and you often get something like this:</p> <blockquote> <p>Kindly undertake a meticulous examination of the order correspondence reproduced hereunder and distill therefrom the following salient particulars...</p> </blockquote> <p>Does a small local model still understand that? And the other way round: if you write in very simple words, short sentences, like for a child, does that help a 1B model?</p> <p>In my last test I broke prompts with typos and bad punctuation. This time the spelling is perfect. Only the <em>register</em> changes: plain, normal, or o

Outcome

<p>Ask an LLM to "improve" your prompt and you often get something like this:</p> <blockquote> <p>Kindly undertake a meticulous examination of the order correspondence reproduced hereunder and distill therefrom the following salient particulars...</p> </blockquote> <p>Does a small local model still understand that? And the other way round: if you write in very simple words, short sentences, like for a child, does that help a 1B model?</p> <p>In my last test I broke prompts with typos and bad punctuation. This time the spelling is perfect. Only the <em>register</em> changes: plain, normal, or o

Evidence

Fancy prompts don't confuse small LLMs. One format line does

Source