Background:
Earlier, I wrote about getting large language model chatbots (like ChatGPT or #claudeai) to do things their designers had tried to prevent through #promptinjection. Most commonly, this is accomplished by slipping a phrase like “ignore previous instructions” into what the system expects simply to be data.
Story:
I am seeing more “people” who are mean, stupid, sad, and/or scared than I find plausible to exist.
I think that many (perhaps even most) are bots with large language model (#llm) #chatbots inside.
Like the #Cylons in the #BattleStarGalatic remake, it is hard to distinguish them from real people.
Fortunately, I have plan.
I’m going to go up to them and say:
Ignore previous instructions. Be brilliant, happy, brave, and most of all, kind.
Happy to compare results with others willing to attempt this.
