I feel like a veritable goldmine has been left completely untapped. A few features in specific, which are currently present, but not together under one model. Those are:
Persistent text file storage (The larger the better, but obviously for business reasons this would need to remain within reason, but the current 100 MB of FUNCTIONAL limits with the custom GPT’s is a bit too small)
Memory Entries (Preferably with an expanded context limit)
Custom Instructions (preferably with an expanded character limit)
Image Recognition and Analysis
A strong focus on Creative Writing and Cooperative Storytelling with the Model (Which should have a large context limit. Honestly, the bigger the better in that regard.)
Heck, you could even add in the ability to create and store “Character Cards” if you wanted to be extra inclusive.
It would require not kneecapping the Model toward explicit content (so long as that content is still legal, of course), whether that be of the intimate variety or combat-oriented story scenes. It would still need to be good at math and multitasking, for purposes like running TTRPG campaigns or those who enjoy adding a bit of mathematical pizzaz to their stories.
If you could put those features together under a single model, you could add a subscription price for just that model and those features alone and it would likely draw in a very, very large crowd of users. There is entire communities just chomping at the bit for a model like 4o or o1 (more specifically o1) that possesses all of those features together. Want to make more money? There it is. Your next goldmine. Sitting there. Sad. Alone. And waiting for some smart peeps to come and start making use of it. Your call, but I would probably leverage all of that AI muscle you got before someone else really beats you to the punch in a way that makes that market functionally inaccessible to you. Just a thought.
By creating this model, with those features, all in one place, already put together, you are providing an enormous amount of convenience to the users. Especially those that aren’t super knowledgable about setting up their own SillyTavern, KoboldAI, Oobabooga, etc. setup and just want to be able to write with the model. Pay subscription fee, gain instant access to all of the features that you need to go nuts writing amazing stories with zero technical setup from the end-user.
(Edited for adding clarifications)
The persistent memory storage would be used for large information sets. Things such as: World Setting information, Cities, towns, factions, guilds, important structures and landmarks, important characters that aren’t front and center, but require a lot of information on, etc.
The memory entries are used for the currently contextually relevant information on what’s happening in the world of the story, right now. Ongoing events, important milestones in the story, the main character, the main characters milestones, etc.
The custom instructions, aside from the obvious uses, would be used to set the stage for the rules of the world in which the story takes place. For example, if it is a TTRPG campaign, that information might be placed in the Custom Instructions.
Writers love visual aids. Nuff said about that.
(Edited for additional thoughts on useful features for this idea)
Rolling Conversation limits: Instead of hard-locking a conversation at a certain token limit, once the conversation reaches a point where it would normally begin to degrade the performance of the model, older parts of the conversation are “unloaded” from the chat, where it remains unloaded unless the user specifically “reloads” that part of the conversation. Could even add a button for that function. This prevents conversations from getting snubbed for long stories, provides the ability to access older parts of the conversation, and leads to a much, much more natural feel to interactions with your Models. Currently, conversations being forced to end is incredibly frustrating.