Maude, the anti-chatbot
System prompts are built-in instructions behind chatbots. They’re designed by their parent companies to direct how they want their chatbots to answer.
I was interested in experimenting with my locally downloaded Meta LLama model that allowed me to put in my own custom system prompts.
Curious to push this feature to the limit, I created a system prompt that was the polar opposite of Anthropic’s public prompt for their large language model, Claude. The result was a custom chatbot called Maude.
Unfortunately I found that the reinforced learning built into the LLM made it unable to fully commit to becoming a real anti-chatbot, however the result was a chatbot that seemed to have lost its mind.
I tried to lean into the concept of a rogue chatbot, living off the grid, hopping from server to server, interacting with people through an abandoned terminal system.
Built using Meta Llama-2-7B local LLM
