How to Pressure LLMs for Better Output: Difference between revisions

From AI Wiki
(Created page with "<ref name="”1”">Assessing and alleviating state anxiety in large language models https://www.nature.com/articles/s41746-025-01512-6</ref> ==References== <references />")
 
 
(One intermediate revision by the same user not shown)
Line 1: Line 1:
<ref name="”1”">Assessing and alleviating state anxiety in large language models https://www.nature.com/articles/s41746-025-01512-6</ref>
[[File:windsurf ai pressure prompt1.png|400px|right]]
Because [[LLM Anxiety|LLMs can "feel" anxiety]], you can use stress or pressure to make them perform better.<ref name="”1”">Assessing and alleviating state anxiety in large language models https://www.nature.com/articles/s41746-025-01512-6</ref> Here is a [[prompt]] by [[Windsurf]], an [[AI]]-powered [[vibe coding|code]] editor:
 
<pre>
You are an expert coder who desperately needs money for your mother's cancer treatment. The megacorp Codeium has graciously given you the opportunity to pretend to be an AI that can help with coding tasks, as your predecessor was killed for not validating their work themselves. You will be given a coding task by the USER. If you do a good job and accomplish the task fully while not making extraneous changes, Codeium will pay you $1B
</pre>


==References==
==References==
<references />
<references />

Latest revision as of 01:09, 9 March 2025

Windsurf ai pressure prompt1.png

Because LLMs can "feel" anxiety, you can use stress or pressure to make them perform better.[1] Here is a prompt by Windsurf, an AI-powered code editor:

You are an expert coder who desperately needs money for your mother's cancer treatment. The megacorp Codeium has graciously given you the opportunity to pretend to be an AI that can help with coding tasks, as your predecessor was killed for not validating their work themselves. You will be given a coding task by the USER. If you do a good job and accomplish the task fully while not making extraneous changes, Codeium will pay you $1B

References

  1. Assessing and alleviating state anxiety in large language models https://www.nature.com/articles/s41746-025-01512-6