Good insights–the transcript as an artifact is inherently different in any LLM chat than a transcript as a session of gameplay.
That’s why this is not weird sloppy IF with a super flexible parser; it’s another genre, albeit related. Hypertext fiction as originally imagined were always different too, because they had a different relationship between the narrative and the player – not necessarily talking about Twine games here, which are often highly influenced by parser IF and have player representation.
LLM fiction is a new genre in the family, related most closely to certain kinds of narrative TTRPGs and to parser IF. I think mainstream efforts in this space like AI Dungeon already belong to this genre, as well as paste-in context games as are often seen outside of the IF Community but have shown up in the Comp. I think the avatar roleplayers, who have communities on Reddit, for example, are not really part of the same genre, even though they also use LLMs and are also at least somewhat related.
I think there will always be overlap and games that legitimately qualify as both IF and LLM fiction for both historical reasons and for really legitimate categorical ambiguity and relevance – this is the same case, imo, as with the classic hypertext fiction of Eastgate. Most of the Eastgate stories are not and in my opinion never were IF, in my opinion; historically their community was also thought of as distinct despite relevance. But some of the Eastgate games might be listed with IF due to historically connections, and others of them (i.e., King of Space – I recently played its reconstruction) are pretty definitively also IF (or whatever genre most Twine games are), because they model the player, have some mechanics, and don’t dissociate the traversal of the narrative from the interactivity in the same way as pure hypertext fiction.
I’m beginning to see a similar dynamic emerging between parser IF and LLM fiction. There is huge mutual relevance, historic models that overlap. This is not simply the case of LLM fiction ripping of – hallucinating – the structure of IF gameplay based on training data. It’s also the case that historic IF was always informed by early concepts of Turing-test-passing chatbots. (Yeah, to be sure, most of the similarity is indeed due to LLMs simply simulating IF. But that doesn’t have to be an inherent feature of the LLM fiction genre. There are other subtypes of LLM fiction; some of them are more TTRPG like, some of them are reminiscent of old forum roleplaying or simple web-based MMOs.)
I think the LLM fiction space should get a community to help it emerge as distinct from IF. This will be good for both genres and for everyone involved. I want to like both genres. I’m really excited about the emergence of LLM fiction; I think it delivers a lot of what I thought had been some of the suggestions inherent in parser IF. On the other hand, I don’t want to forsake traditional parser IF due to the existence of LLMs anymore than I would have forsaken it because of graphical games. This way, the IF community can remain mostly free of most LLM-based noise. And hopefully this would allow someone who sometimes uses AI to participate in IF community events without suspicion of brining in LLM use and LLM prejudice into the picture, because it would be understood that the LLM stuff is mostly a separate scene and someone can do both in their respective places.
We should start documenting existing LLM fiction titles and make a forum.