So I recently watched a YouTube series where a guy had ChatGPT play Detroit: Become Human, and I thought, hey, that would probably work for CoG.
For anyone else interested in doing this on their own games, what I do is explain that it is playing a game, tell it the stats that will affect the gameplay, let it know it can ask me for its current stats, and then copy-paste the description for the book.
I then copy-paste whatever the book says, and when it gets to a choice, I say:
Your choices:
1: Option 1
2: Option 2
3: Option 3
etc.
Finally, at the end of each chapter, Iām giving it updates on its current updates.
Iām starting it off with Life of a Space Force Captain. I just finished chapter 2, and itās interesting to see not only its choices, but it usually tells you what itās thinking while choosing it.
It has made a few choices that surprised me. For instance, it decided to be human rather than synth, which is part of why I started with this book.
Anyway, Iāll keep posting updates as I continue through the book. I also plan on having it play Choice of Robots and The Lost Heir in the future.
Let me know if you have any questions.
9 Likes
Ai dungeon or novelai might be better for this type of thing.
AI like that doesnāt work by āthinkingā- it is incapable of anything resembling āthoughtā, nor was it designed to āthinkā. It produces an output based on previous data and current data. It can make a good approximation of what a thought process might look like, but itās basically just choosing the option that is most like the data itās been previously fed, not making any considerations.
5 Likes
Iām a bit confused about whatās interesting about doing this rather than actually playing the game - isnāt this more effort for lower quality? Or I may have misunderstood - I originally read this as you giving the LLM a prompt to which it outputted text. Is it that you instruct the LLM to pick an option? In which case surely thatās the same as running RandomTest and reading the output?
3 Likes
The art of doing random stuff, without doing random stuff.
Iām not simply giving it the options. Iām making sure it knows whatās going on in the story as well as keeping it aware of its current stats. Admittedly, I have yet to get to the point where itās choices will matter, but thatās the whole point of my experiment. Can a chat AI understand its abilities? Who it can rely on? Will it choose to fall in love? AI has always fascinated me, so Iām interested in seeing how far this one has developed.
I watched the same YouTube video series 
. I started playing Mage Reborn on Dashingdon the same way. I wrote a JavaScript script to scrap the text from the page and the options and format it in a way friendly to ChatGPT. I didnāt feed it any stats though, ChatGPT canāt do math anyway and I wanted it to focus on the story aspect and not the game.
@HarrisPS, the interesting part is reading the " thought process" behind the choice. You can also ask the AI to criticize the available options and suggest more. Thatās like having an AI beta reader.
@bertilak, sure, by now everyone knows what an AI is and what it isnāt, itās still interesting to see whatās the statistically more probable answer. Even if itās somewhat random you can still draw insight from it. Itās like reading a tarot card, it has only as much value as you yourself give it.
I personally did that purely for amusementās sake and I was also surprised by some choices, not in the sense that I was surprised by how the AI āthinksā, but that it made really good arguments to defend its choice. And some actually gave me food for thought.
5 Likes
