Conversation

One pattern I'm liking is using lists and questions as way to add randomness to the generation. A list of five items is almost always different each time, and the humans choice adds to that randomness The randomness generating any one passage is never as broad as a genre
1
2
If you don't give it a genre it's going to start kind of generically until the details start to point somewhere, as it starts to stereotype the small details it added to the story earlier. But high level descriptions like genre or character put everything else in that framework
1
2
Was rewatching Dune and thought I'd try to make the CYOA story be in Dune. ChatGPT would sometimes refuse depending on wording (complaining it is a language model without access to the internet), but once I got past that it worked quite well
ChatGPT offers genre recommendations and the human replies "Science fiction set on the world of Arrakis, from the book Dune". It takes this suggestion and offers appropriate themes specific to Dune, such as vying for power and the desert.
2
It usually chooses typical/stereotypical scenarios, like every Romance CYOA is a young woman in the city looking for love. To see if it would stick to this I asked for "A romance between three women." It offered the theme "Navigating polyamory💞" and created "Love in Triplicate."
1
Unintentionally I am playing through a story where I have just been diagnosed with a rare and aggressive form of cancer. Do I start by (1) Talking to my doctor about treatment options, (2) Calling my family to tell them the news... Damn, you can work through some stuff with this
3
Replying to
I'm amazed how well this works! Not just how it extrapolates but how it pauses at appropriate times for input; in my experience one of the biggest issues with multi-part prompts is that it answers for you.
1
1