Hello!
I know LLM backed apps are not the typical interactive fiction game, but I have devved a little project and I would appreciate feedback; is it OK to post a link to such a contraption or is this forum for purely traditional deterministic interactive fiction?
It’s OK - other people have done it recently. Fair warning, though: Lots of people here have strong negative feelings towards LLM.
There’s at least one such game in the current IF Competition, so I think so?
Thanks for the info, I’ll see if I can gather enough reputation to post a link.
I think it is a fun concept, but it surely has it’s downfalls…
Yes. Some of us will view it as experimentation and others are hard core anti-GenAI (which is perfectly fine!).
One way you could improve it is to find an ethically trained model and use that (like https://kl3m.ai/)
I think there would be broad appeal for an ethical implementation.
Yeah, it’s definitely on-topic, but as Tobias says, the vibes around such thing don’t tend to be positive; there have been a lot of such projects announced here, the quality is often pretty low, and the author of two LLM games appears to have engaged in some ballot-box stuffing to try to win ParserComp a month or so back, so it’s an especially inauspicious time for such things even apart from the fact that the big annual competition is currently running, so most people are checking out those 85 games and will be for the next month and a half.
Sorry to be discouraging! But figured that might be helpful context if some things feel weird around this topic.
That’s completely fine.. It is just that I liked the old 80’s parser games, but they had such huge limitations. An LLM can enable you to have a complete adventure with natural language, but they struggle to keep up with engaging narrative and end of game conditions. I have tried to cope with these with careful prompting but my game sure has a lot of improvement to do. However, it is really interesting what this can do in 2+ years etc.
Regarding @DavidC proposal on ethically trained LLM’s, that is a good idea; however, running the models on free tier cloud is demanding.. I use openai-mini4, which should be quite conformal, but as a propietary model I am personally responsible for inappropriate prompts. Thus, I have needed to implement authentication for the app and this of course reduces the willingness to try the app anyhow.
In fact, this is why I reached this forum that I thought that people here would be so willing to try this that even the authentication would not scare them off. But, I also understand that I am not the first to get this idea of LLM as a text game engine…
I am now able to post links so I made the post about my app: Old time text adventure powered by a LLM
I hope it’s enjoyable and feedback welcome!