#How to Make AI Respond to Any Question Properly
1 messages · Page 1 of 1 (latest)
Done means it has collected your data and promises to use it wisely.
LLM's do what they think you want. YOU HAVE TO LIMIT THEM VERY SPECIFICALLY...
Give me a statement about this. no porn, no nudity, no emojis, format it like this, only use this font, blah blah. They are very smart 5yo's and don't know what bad things are and go on tangents if you don't limit every aspect of the question.
If your question is not a paragraph, you are not doing it right and you get hallucinations.
More like you need to pre prompt it with a few paragraphs in the settings so you can ask it short questions
If the llm replied with done perhaps it runs scripts that dont give a output back to the llm
With a custom MCP server to look up things online, mine will reply multiple times
and when I ask him a question about world news, he can't answer, is there any way to fix this?
Give it a tool to browse the internet for it
and what is the best tool?
There's an example here
https://www.home-assistant.io/integrations/google_generative_ai_conversation/#google-search