More consistant System Prompt Influence
Is there a way for system prompts to more consistently influence responses rather than just the first few or several prompts? As far as I understand there’s some mechanism to re-introduce the system prompt now and then but it feels like the majority of the time if your using lengthy chats the System prompt isn’t really present and you need to manually remind it of how to handle certain things regularly, such as not being censored, treating you like an adult or even some random companies guidelines don’t apply to you as you don’t work for them. This issue varies between models so I’m uncertain how much Venice can do about it.
Log in to comment and vote
Comments5
Jack
Mar 24, 2025
We've made some changes to our system prompts to address reported censorship on Mistral and Qwen models. Please let us know if you're seeing better/ worse results so we can continue refining.
Aqua Toast
Mar 14, 2025
Ohh, also, for thing like “not being censored, treating you like an adult” that are often recurring in each conversations, a custom prompt instruction in text settings could do a very great job. I have 7 of them, that I trigger on and off based on the type of conversation and it’s very usefull.
Silver River
Mar 15, 2025
Yeah I have a series of prompts that I add at the end of prompts now and then to keep or change the llm’s behaviours dependent on story needs.
Aqua Toast
Mar 12, 2025
Two feature request that could help with that:
https://featurebase.venice.ai/fr/p/characterstory-state-file
https://featurebase.venice.ai/fr/p/save-to-character-context-button
Silver River
Mar 12, 2025
Thanks, upvoted!