Reply to thread

Message: <blockquote data-quote="kermit4karate" data-source="post: 9767210" data-attributes="member: 7053643">You've got some inaccuracies here. First, it's trivial to create boundaries during play now. You just set a safe word and agree to the parameters you set for using it. You tell the AI not to allow you to break character or override a ruling unless you use the safe word first -- so, unless you say "banana," or whatever else you set. Instruct the AI that you want it to end the game immediately if you try to get it to change a ruling without using the safe word. Obviously, it's still just a tool (for the time being), but it's a tool that listens to you. One does have to voluntarily agree to follow their own rules. The AI isn't going to fight you for supremacy (yet). You do bring up a point about general purpose LLMs versus purpose-built LLMs with custom datasets. Those do exist and can be superior to general LLMs for certain tasks, but the general purpose LLMs like ChatGPT can do/play any game for which the rules are publicly available (published online like 5E is). It knows the rules and follows them at least as closely as the average human does.</blockquote>