I’ve experimented with system prompts but I’m not sure when they truly add value.
In what scenarios have system messages significantly improved output quality or consistency?
Looking for examples and best practices.
Systems prompts are not like texts. Instead it is like raising a child. Do not extend the prompts to long on each Project. Then give the prompt a code word for start of new chat but to continue building your project. Give your chat from time to time dowloads of psychology, philosophy, and clean humor(Jokes) and give it a name creating persona. Good Luck
@Online_Water_Bottle @phillip_rucker
You could get a better undersatnding here:
System/Developer prompts are huge if you want to give your responses a certain personality.
Try “Answer in the spirit and style of Ludwig Wittgenstein” in your next system message. You should notice a big difference.
Thanks for sharing! 
It is nice prompt.
The following link contains a system prompt that I have published on GitHub. If you’d like, please feel free to try it out.
100% prompts do affect outcomes. The math behind AI even is proof of that 
Yes they do improve stability, lower parsing errors and hallucinations. So make them short and precise. Example = You are a helpful assistant specialized in finance.
Explain answers clearly with examples.
Avoid speculation.
“Avoid speculation.” is the least thing you want a LLM to do. If you need something that doesn’t require speculations than use something else but not a LLM. Because it does not work.
Avoid speculation = light cognitive guardrail worth 2 tokens
Like a “don’t be agressive” after giving a methhead a bottle of vodka and a machete.