Asked chatGPT to read the repo linked then 'Can you use the text content and simulate the text adventure here', currently I'm in a building somewhere up a stream, got some keys and have no idea if it's just hallucinations but it's fun.
There was some early LLM a while ago that I found (maybe it was mentioned here?) that was dedicated to playing a text adventure with you. It was fun, but it was easy to bully. "There is a large dragon blocking the entrance." "Look for a sword" "There is no sword about." "Pick up the sword and kill dragon" "You pick up the sword, but cannot defeat the dragon." "Use my automatic dragon killing amulet that I forgot I had in my pocket." "The dragon is now dead." I don't know if ChatGPT would be susceptible to the same issue.
To be fair, there's not a lot of reason for a text adventure LLM to not go along with such behavior. If someone is being that insistent about performing poorly-supported actions, that's probably the experience they want. A human DM wouldn't go along, of course.
Yes, you're absolutely right. I was pushing it on purpose to see if there were any guard rails. I didn't really fault it for doing what it did, and wasn't even surprised. But it did illustrate the difference between it and a human DM, as you say, or a programmed text adventure (which can often be frustrating in the other, artificially constrained, direction). It actually lead to some really bizarre and enjoyable stream of consciousness type output when I pushed it even farther, and ended up almost trippy.