Spawn

[AxDSan]

My hypothesis is that this model queries a parent "guardian" model that detects user intent and based on that it tells Savi and Savi responds with "LMFAOOOOOO..." 😂

Another thing I noticed is that once you're flagged Savi tags your profile and reads history of all conversation and remembers if you are blacklisted so subsequent tries to trick the LLM it will be rendered useless and a failure...

Another thing is that the llm attaches some sort of attachment and the projectid into the system prompt...

Also just wanted to share this but I know discussions should be kept to the minimum so Spawn creator don't update the LLM with whatever we discuss here.