Hello everyone, this will be a bit of a long read, i have a lot of context to provide so i can paint the full picture of what I’m asking, but i’ll be as concise as possible. i want to start this off by saying that I’m not an AI coder or engineer, or technician, whatever you call yourselves, point is I’m don’t use AI for work or coding or pretty much anything I’ve seen in the couple of subreddits I’ve been scrolling through so far today. Idk anything about LLMs or any of the other technical terms and jargon that i seen get thrown around a lot, but i feel like i could get insight from asking you all about this.
So i use DeepSeek primarily, and i use all the other apps (ChatGPT, Gemini, Grok, CoPilot, Claude, Perplexity) for prompt enhancement, and just to see what other results i could get for my prompts.
Okay so pretty much the rest here is the extensive context part until i get to my question. So i have this Marvel OC superhero i created. It’s all just 3 documents (i have all 3 saved as both a .pdf and a .txt file). A Profile Doc (about 56 KB-gives names, powers, weaknesses, teams and more), A Comics Doc (about 130 KB-details his 21 comics that I’ve written for him with info like their plots as well as main cover and variant cover concepts. 18 issue series, and 3 separate “one-shot” comics), and a Timeline Document (about 20 KB-Timline starting from the time his powers awakens, establishes the release year of his comics and what other comic runs he’s in [like Avengers, X-Men, other character solo series he appears in], and it maps out information like when his powers develop, when he meets this person, join this team, etc.). Everything in all 3 docs are perfect laid out. Literally everything is organized and numbered or bulleted in some way, so it’s all easy to read. It’s not like these are big run on sentences just slapped together. So i use these 3 documents for 2 prompts. Well, i say 2 but…let me explain. There are 2, but they’re more like, the foundation to a series of prompts.
So the first prompt, the whole reason i even made this hero in the first place mind you, is that i upload the 3 docs, and i ask “How would the events of Avengers Vol. 5 #1-3 or Uncanny X-Men #450 play out with this person in the story?” For a little further clarity, the timeline lists issues, some individually and some grouped together, so I’m not literally asking “_ comic or _ comic”, anyways that starting question is the main question, the overarching task if you will. The prompt breaks down into 3 sections. The first section is an intro basically. It’s a 15-30 sentence long breakdown of my hero at the start of the story, “as of the opening page of x” as i put it. It goes over his age, powers, teams, relationships, stage of development, and a couple other things. The point of doing this is so the AI basically states the corrects facts to itself initially, and not mess things up during the second section. For Section 2, i send the AI’s a summary that I’ve written of the comics. It’s to repeat that verbatim, then give me the integration. Section 3 is kind of a recap. It’s just a breakdown of the differences between the 616 (Main Marvel continuity for those who don’t know) story and the integration. It also goes over how the events of the story affects his relationships. Now for the “foundations” part. So, the way the hero’s story is set up, his first 18 issues happen, and after those is when he joins other teams and is in other people comics. So basically, the first of these prompts starts with the first X-Men issue he joins in 2003, then i have a list of these that go though the timeline. It’s the same prompt, just different comic names and plot details, so I’m feeding the AIs these prompts back to back. Now the problem I’m having is really only in Section 1. It’ll get things wrong like his age, what powers he has at different points, what teams is he on. Stuff like that, when it all it has to do is read the timeline doc up the given comic, because everything needed for Section 1 is provided in that one document.
Now the second prompt is the bigger one. So i still use the 3 docs, but here’s a differentiator. For this prompt, i use a different Comics Doc. It has all the same info, but also adds a lot more. So i created this fictional backstory about how and why Marvel created the character and a whole bunch of release logistics because i have it set up to where Issue #1 releases as a surprise release. And to be consistent (idek if this info is important or not), this version of the Comics Doc comes out to about 163 KB vs the originals 130. So im asking the AIs “What would it be like if on Saturday, June 1st, 2001 [Comic Name Here] Vol. 1 #1 was released as a real 616 comic?” And it goes through a whopping 6 sections. Section 1 is a reception of the issue and seasonal and cultural context breakdown, Section 2 goes over the comic plot page by page and give real time fan reactions as they’re reading it for the first time. Section 3 goes over sales numbers, Section 4 goes over Mavrel’s post release actions, their internal and creative adjustments, and their mood following the release. Section 5 goes over fan discourse basically. Section 6 is basically the DC version of Section 4, but in addition to what was listed it also goes over how they’re generally sizing up and assessing the release. My problem here is essentially the same thing. Messing up information. Now here it’s a bit more intricate. Both prompts have directives as far as sentence count, making sure to answer the question completely, and stuff like that. But this prompt, each section is 2-5 questions. On top of that, these prompts have way, way more additional directives because it the release is a surprise release. And there more factors that play in. Pricing, the fact of his suit and logo not being revealed until issue #18, the fact that the 18 issues are completed beforehand, and few more stuff. Like, this comic and the series as whole is set to be released a very particular type of way and the AIs don’t account for that properly, so all these like Meta-level directives and things like that. But it’ll still get information wrong, gives “the audience” insight and knowledge about the comics they shouldn’t have and things like that.
So basically i want to know what can i do to fix these problems, if i can. Like, are my documents too big? Are my prompts (specifically the second one) asking too much? For the second, I can’t break the prompts down and send them broken up because that messes up the flow as when I’m going through all the way to 18, asking these same questions, they build on each other. These questions ask specifically how decisions from previous issues panned out, how have past releases affected this factor, that factor, so yeah breaking up the same prompt and sending it in multiple messages messes all that up. It’s pretty much the same concept for the first but it’s not as intricate and interconnected to each other. That aside, i don’t think breaking down 1 message of 3 sections into 3 messages would work well with the flow I’m building there either way.
So yeah, any tips would be GREATLY appreciated. I have tried the “ask me questions before you start” hack, that smoothes things a bit. Doing the “you’re a….” Doesn’t really help too much, and pretty much everything else I’ve seen i can’t really apply here. So i apologize for the long read, and i also apologize if this post shouldn’t be here and doesn’t fit for some reason. I just want some help
[link] [comments]
