Since we now have clarity on what the IFComp allows and that other comps have already explicitly banned GenAI, it’s my belief that this leaves a gap in GenAI stories and games. There are still interesting ideas being explored combining IF and GenAI and I believe it would be fun to see those works coalesce in a competition.
I’m not (at this time) volunteering to lead a new comp, but I’ll open the discussion. There are so many interesting ideas.
two panel web app where the game is on the left and GenAI commentary on the right
an NPC driven by GenAI
room descriptions with attributes that GenAI produces varied output
The problem is, people seem much more interested in generating games with LLMs than in playing games generated with LLMs. This kind of competition has been proposed a handful of times so far, but always by people who want to enter it, never by people who want to run it or judge it. And so it always fizzles out shortly after.
The LLM bans in Spring Thing and IFComp were put in place for logistical reasons rather than philosophical ones: judges said outright that LLM-generated entries were making them not want to participate, and without judges, there’s no comp. I have no objection with someone launching a new comp for this, but they’ll need to overcome those logistical issues first.
Speaking personally, I’m also curious about what applications LLMs could have in parser IF! They’re very good at text processing, and parser IF depends on a lot of text processing. But I haven’t yet seen an application that I really enjoyed or wanted to keep playing, so I’m also not volunteering to run or judge a comp like this.
I like this idea and would participate. But can AI not be mandatory?
Then games can be entered without being under the AI use microscope. I would like to enter games that (probably) do not use AI, while feeling assured that if I decided to, for example, use a cover made by AI or a logo, i won’t be marginalized.
I can’t imagine anyone would care to specifically ban non-AI games from such a competition. But as has been alluded to already in this thread, no-one actually wants to run this competition anyway.
While I share the concerns, I think there’s some potential to it - you’d want to get interested folks to commit ahead of time to judging, and I almost think the better approach is a competition of tech demos rather than games.
That’s a good idea. What if instead of everything being under wraps until the end, entrants were able to post WIP ahead of the deadline. Kind of, to test the water and gauge interest.
For example, to publish a “demo”, then later, publish the completed game?
I think that may change now that IFComp has ruled against GenAI content.
And I didn’t say I wouldn’t. I just said it’s not YET on my radar. If I did decide to run a comp it would be something Speed-IF like and more of an anthology or demo-room type thing and less of a competition. I think removing the voting aspect and directing it at innovation and creativity is important for the foreseeable future. If a reasonable pattern emerges from innovation, then maybe a new competition emerges with those guardrails.
So, I’ve experimented a fair bit with GenAI for purposes of creative writing. Even once got an app working which generates a 100+ page book with reasonable narrative consistency. However, while the prose had high points, the lows were very, very low
Then I experimented with an auto-GM able to run a reasonably consistent RPG adventure module, but the world state was about as consistent as eating soup with a spork and playing it felt like eating dry oatmeal. Experiment shelved, despite it roughly working.
However, I think there’s still potential. Parser systems are actually fairly good at maintaining a consistent world state. They only struggle when situations occur where neither the system nor author accounted for the user’s action.
I intend to run an experiment where I don’t let an LLM do the heavy lifting, because it’s clearly not very good at that (yet). Instead, I’m hoping to try using it as a sort of fallback “ghost in the machine”. A fallback Vox Machina (which is not a name I should use for this) to help update the world state if an edge case or particularly difficult command needs to be parsed.
However, the way LLMs function now, I wouldn’t want to play LLM based int fiction either. Not if the LLM is the primary author. I am curious what my experiment yields, however. I would be so happy if it could run a TTRPG that way.
I can see all your points here. I’ve been playing with AI stuff and, for the most part, it doesn’t make the grade. But that doesn’t mean it’s useless. I’ve come to realise, it will eventually find its niche as a tool to help people, but not to replace their work.
The best results I’ve had are working with a multi-step process involving manual work and AI “bits”. Anyone that thinks AI can do the hard parts on its own, are mistaken.
So, i would like to see more experimentation and results. Especially as things might improve over time. Or if they do.
Currently, there is no outlet here to show and review results for this, and i think this initiative talked about might be a way. I see you’ve been experimenting with AI in your system, and seeing a number of both pros and cons. This is all good.
I wonder if a competition is the right format, but yes, I see your point.
Frankly, LLM use has a bad reputation, which I am going to say it has earned. Not because it itself is a terrible technology (though the ethics surrounding their training data leaves much to be desired), but mostly because of their usage having been primarily with the intent to supplant creatives by commercial interests, rather than aid creatives.
This in turn has given birth to the very fitting title of “AI slop”, where the writing has clearly never seen an editing pass. If anything, I would enjoy a space within this community to at least figure out ways in which LLMs can aid creativity, rather than try and supplant it. I know it already massively aids me in programming, but it also occasionally helps when I have a writer’s block. There’s something about writing the prompt itself that just makes the pieces fall in place, at which point I just abort the LLM use and continue writing/programming.
If anyone is going to crack the problem of if and how AI can be implemented in IF, my money is on these guys. Great interview worth listening to which addresses all these problems.
