This has always been my chief criticism of LLM, aside and apart from my observation that much of the output from LLM is so distinctively flat that it would not pass a Turing test. LLM is by its very nature a source without attributions.
6 Likes
“I’ve also had enough of trying games with AI written description that I can’t interact with properly (e.g. unimplemented referenced scenery). Plus over flowery descriptions.”
I’m curious what games you are referring to. Other than AI Dungeon, I haven’t seen all that much experimentation with this.
Hebe from last year’s IFcomp was a very typical example. The room descriptions were all generated with an LLM, so it became clear quickly that anything mentioned in the first paragraph of the room description, however interesting it sounded, wasn’t worth trying to interact with because it wouldn’t actually be implemented in the game, just the result of AI embellishment.
(Plenty of other games with LLM text have this issue too, that’s just the first one whose name I could call to mind.)
8 Likes
Mike wandered into and explored an entirely illusionary hut recently with Last Audit of the Damned - Details .
8 Likes
etc.
6 Likes
So if almost everyone boycotts the AI games in IFComp, will the AI games win because the only people who rate them are AI enthusiasts? The ranking is determined by average rating, with no minimum number of ratings IIRC.
5 Likes
I doubt it, honestly; enough people vote in IFComp that I’m sure entries with LLM-generated text will get plenty of negative votes.
6 Likes
That’s a very fair point! And as much as I think it’s a good idea to have some sort of venue for AI experiments that’s not IFComp, I wouldn’t be especially interested in judging it, either.
5 Likes
IFComp has a wider voter reach. I don’t think the community centered here has as much weight there as smaller venues. Games will get played and rated either way.
8 Likes
I’m planning to spend at least 30 minutes on every AI-tagged game. I won’t write reviews, which will reduce the effort.
I may have to switch gears if there are a very large number of entered works.
6 Likes
The main reason for my vitriol is that, as a creative in the film and TV sector, my career has already been negatively impacted by generative AI. I’m a motion graphic designer, and currently, AIs cannot do what I do, so my livelihood has not been as badly affected as, for example, illustration. But this very morning I checked LinkedIn and was greeted with an offer to train AIs to do the kind of work I do. I’d be part of a team “identifying edge cases that AI models can’t currently solve”, in other words training AIs to make myself redundant.
The whole creative industry has been devalued, because businesses are thinking, why should we pay a creative to do that when we can get a machine to do it, for free and in a fraction of the time? The machines, of course, have been trained on work that human artists have spent decades of their lives perfecting, and entirely without their permission!
I come to this forum to read about Interactive fiction, so it’s aggravating to see that every other thread has something to do with AI. I don’t want to play a game written by a computer. If you can’t be bothered to learn the craft of writing, find a different hobby! Nor can I see a lot of point in AI assisted parsers - to interact with a parser game you type a verb and then a noun. If you can’t get your head around that, go and play Fortnite instead!
I haven’t said anything up till now because I generally just avoid any thread that has AI or LLM in the title. This one didn’t. @Angstsmurf’s post summed up my feelings towards AI and made me smile, so I felt like it deserved a response.
18 Likes
These seem to emphasize the use of AI generated narration, rather than just using AI to deal with unpredictable input. I still haven’t seen many examples of the latter.
There’s a Star Trek: Voyager episode called Worst Case Scenario where someone writes a holo novel about a Maquis mutiny on the ship. If you watch the episode you can see that whoever wrote the holo novel used mainly human written dialogue for the story. The problem is the program has to deal with different responses to Chakotay when he reveals his plan for the mutiny. And then the story has to diverge into different paths depending on whether the player decides to join the mutiny or not. That’s more of the type of experimentation I’d like to see.
Other folks have replied before me, and covered things pretty comprehensively 
2 Likes
Cheree and The Fortuna, which I linked to above, do that, as did Last Audit of the Damned, which Pinkunz mentioned (as well as the author’s other game, Mystery Academy – though I’m not linking to those since those are the ones at the heart of the ParserComp irregularities I mentioned; I have a three year old so I’ve finally learned not to reinforce bad behavior). There are also lots of other examples that have cropped up on the forum if you go looking – part of the weariness you’re picking up from the regulars here is that barely a week goes by without someone talking up a new project/platform to do exactly this.
I haven’t played them all, but I have engaged with a pretty strong sample in depth, and seen the discussion threads around the others, and they pretty much all fail exactly the way that you think they would: the LLM BSing gets in the way of the actual human-written content, contradicting it or elaborating it in ways the author didn’t intend and which wind up undermining proceedings. Cheree is probably the best of the bunch, aided by some careful design to cabin how off-piste the player can go, so it’s worth playing that one if this is something you’re interested in (fair warning that there’s incongruous half-naked anime girls).
8 Likes
A note for the newcomers: Mike Russo is one of the more famous and prolific IF reviewers in the community, so he has a much wider range of experience with these than most of us.
11 Likes
I am definitely telling my wife that I am famous based on this
15 Likes
Not many people regularly play and review every entry in a comp!
5 Likes
So if almost everyone boycotts the AI games in IFComp, will the AI games win because the only people who rate them are AI enthusiasts? The ranking is determined by average rating, with no minimum number of ratings IIRC.
What is actually the current practice for IF Comp score calculations?
The site says each score is “simply the average of all scores [the game] has received,” but looking at 2024’s numbers, AFAICT it isn’t just ‘individual scores ÷ votes cast.’ It looks like it’s weighted or adjusted somehow.
1 Like
I think it just looks that way because whatever code or markup is producing the “X judges rated this game Y” tooltip is counting from 0 when it should be counting from 1 (as indicated by the numbers across the bottom of each graph). If you use the actual score range of 1 to 10 instead of the tooltip’s mistaken range of 0 to 9, the scores add up/average out correctly.
6 Likes
I am also surprised at how strong the aversion to AI is here, but I can partly understand it, because I am also not a fan of stories that were generated exclusively with AI. It seems to me that there is a strict division into categories here: AI/non-AI. But the extent and the way in which AI is used in the stories also plays an important role, in my opinion. Is the story written exclusively by an LLM, as is currently done on some platforms, or only in a few places?
For my part, I like to read exciting, well-written stories with depth and mystery elements, and the purely AI-generated stories I’ve read so far haven’t really captivated me. That’s why I wanted to do something different and started this hobby project about a year ago. I pursued it alongside my full-time job as a senior consultant in the field of data science/AI and saw it as an opportunity to use and hone my skills while also creating something that I myself would enjoy using to create interactive stories (with or without AI).
But I see that I need content / stories to show the site and I think I will create one completely without AI.
3 Likes