I did a thing with ChatGPT, (Interactive Fiction Demo)

[wendienew (Wendy Nugroho)]

So, I’ve been observing how people interact with AI such as ChatGPT. It’s kinda funny that people in general who doesn’t even know what IF is, interacts with ChatGPT just like we interact with IF. We give commands in form of text to the AI, and the AI response us using another text, describing things, explaining things. And is that how we play IF?

I really love and enjoy how the reality is going on today. This phenomena really soothen me up cause I feel that my love and hobby with IF really changed the way I see myself with other gamers around me. No one around me play IF, even not recognizing the existence of IF. Even after introducing them to IF, they don’t seem to be interested enough. And then, when AI became more accessible, anyone can use AI, and watching how they interact with it. I began to realize that I’m just like anybody else. They actually “playing” the AI just like I do with IF. It’s not completely the exact same thing, but the behavior is almost identical.

It triggered me to do something stupid with ChatGPT just a moment ago. I asked ChatGPT to behave as if I am playing an IF. It acted as the game, then I acted as the player.

Without further ado, this is how the magic happened:

This isn’t the ideal way to play IF, I don’t even think to reach the ending when playing with ChatGPT IF edition, lol. At least, this proves that IF is not dead, and will keep on living. Even in this modern era, everything is going digital, technology is way more advanced, where gaming is selling high quality graphics, we are human that will never stop interacting. Even in this AI era, we are still doing the old way to interact with them.

That was my little experience. If any of you ever tried doing the same thing, let me hear what’s your story.

This is my 5th topic after 11 years since the latest topic I created in this forum. Also, please excuse my bad English. I’m not a native English speaker.

Thanks for reading

[Scribarchist (Anthony Taylor)]

How long did you continue playing? Did the narrative start to lose track of the past at some point?

[wendienew (Wendy Nugroho)]

So, I was just improvised my interaction with ChatGPT. I did a brainstorming with the ChatGPT about the settings, characters, the story, things that the player can interact with, some of the details of objects and characters, how the game starts and the flow.

After I think it’s enough for the ChatGPT understand the world I built for the game, we started.

It was a surprise. This is probably the most immersive way to play an IF. I don’t exactly know what is in the background of my screen, and I expect some errors might occur. The good part is, it allows me to use different kind of actions that’s probably limited in IF games in general. I can even combine tools to make a makeshift weapon. I can even break the character. It’s also very forgiving when I did some typo.

It’s also scary how AI can do such a thing. I will probably do another couple of things but I definitely won’t do this too much. Technology is scary.

[wendienew (Wendy Nugroho)]

Not too long, to be honest, but I started another game just like what I just posted earlier. Yes, as expected, the ChatGPT is undeniably imperfect. At some point, it forgot about the condition of certain items the player have. So, in the story I played for the first time, the player was holding a burning torch, then the player got sucked in a pool of a thick black colored water. It snuffed the torch. But later on, the AI wasn’t smart enough to remember that. I fixed that by chat, but it broke the immersion.

[Jeff_Nyman]

@wendienew

At some point, it forgot about the condition of certain items the player have

Indeed, the context window will be a limitation in most circumstances. Gemini has a larger context window, and Claude has an even larger one. ChatGPT is also improving at using its “memory” aspect, which stores long-term concepts. DeepSeek is interesting in terms of how it stores “doubt” or even “counterfactuals.”

Still, it’s a fascinating area to play around in. I deal with this stuff every day in my professional context, and like any early democratized technology, it’s fascinating to watch how it’s evolving. Agentic AI is even more interesting in this arena since it relies so heavily on persistent context.

[wendienew (Wendy Nugroho)]

Wow, thanks for explaining that. Now I’m interested to do the same thing with different AI you mentioned. Maybe I will go back to ChatGPT later on, after it gets software upgrades.

[playingball9000]

You could possibly improve on the memory issue by writing a wrapper app around any LLM and using that to keep state (inventory items, player stats, conditions, etc) and just pass them into the context every time the user sends a command.

A more sophisticated approach might be to set up the LLM with a RAG database and have the state always written to the database. That way, every command the user inputs would automatically extract the relevant pieces of info, but the main generation would still be procedural.

Obviously, both are way beyond the initial scope of the topic, but interesting to think about.

[wendienew (Wendy Nugroho)]

Thank you, but I’m not really into that stuff. I was just doing silly stuff, not trying to make it my official IF platform.