How to Use ChatGPT in Confic
Updated: May 29
Media & Tech, Culture & Community
by Lack of Lepers
Editor's Note: Aside from a few instances, discussion and use of ChatGPT in this article is based on the paid model, currently GPT-4 and its Web Browser beta feature. The article may be updated as new use cases are discovered. These updates so far include:
- "Test" - SCP-914 Testing Simulations (5/19/23)
- "Emulate" - "Jesse Pinkman Reads SCP-____-J" (5/20/23)
- "Inspire" - Articles for artwork (5/21/23)
- "Code" - Conversion to WikiDot syntax (5/27/23)
- "Write","Summarize" - Added references to plug-ins (5/28/23)
- "Jailbreak" - How to jailbreak ChatGPT (5/28/23)
- "Inspire" - AI art generator prompting (5/28/23)
- "Introspect" (5/28/23)
In the future, Google search will serve a role analogous to what Internet Explorer has been for years — a stepping stone. It'll exist as a one-use tool, used to access and download superior alternatives. Soon, Google search as we understand it today will succumb to an advanced language learning model (LLM) like ChatGPT, Google's own equivalent, or something akin.
We've already showcased the immense power of AI in image creation for containment fiction, revealing an astonishing new realm of artistic possibilities. Therefore, our primary focus here will be on ChatGPT and its potential to benefit authors and readers of containment fiction.
In this article, we'll detail several methods in which ChatGPT can enrich the reading and writing experience for those in the containment fiction genre. Each proposed use case will be accompanied by example screenshots for those without access to ChatGPT. (Update: ChatGPT now allows for links to ChatGPT conversations.) These use cases are (links are page anchors):
spelling and grammar (SPaG)
Let's dive in, starting with the elephant in the room: ChatGPT can write in the containment fiction format (SCP/RPC), but it doesn't quite hit the mark on crafting quality containment fiction outright. If it could, the result would be akin to a reheated microwave meal — functional, but lacking the soul and flair of a home-cooked dish.
To effectively employ ChatGPT for writing, you need to bring your original ideas to the table. That's when ChatGPT truly shines. If you task it with brainstorming, you'll receive generic, painfully unimaginative anomalies, and very standard scenarios that lack imagination — like a pocketwatch that manipulates time. If you prompt ChatGPT to produce a narrative that has a beginning, middle, and end, it nearly always wraps up with the containment organization limiting testing and eventually containing the anomaly.
In short, unless you prompt it creatively, ChatGPT operates at the compositional maturity of EditThis. But, guide it with specificity, instruct it to shape your unique vision, and it may surprise you.
ChatGPT excels in writing at both the macro and micro levels; anything in-between tends to be a muddled mess. I, for instance, would use it to refine a rough draft of an article, based on my broad concept for an anomaly and its interactions:
Starting with a basic idea, we can widen its scope by asking ChatGPT to structure it as an outline, incorporating specific elements like interview logs, exploration logs, discovery sections, and so forth. Again, these should be high-level overviews, merely one or two sentences for now.
Then, it's time to put the pedal to the metal with ChatGPT, as you would a Ferrari on a winding country road. Once ChatGPT has furnished a comprehensive skeleton of the article, we can direct it to write each section separately.
You can then refine these individual sections and request specific adjustments—more natural dialogue, greater emotional depth in a character, enhanced believability, additional concepts, and more. This is the time to test the limits; push it until it breaks.
Just as with AI image generators, once you've mastered the art of "steering" ChatGPT, you can guide it to write almost anything you imagine, generally with impressive results. Its capacity to fulfill your instructions means that if the output is disappointing, the issue may not be ChatGPT — it could be you.
With the Web Browsing version of GPT-4, you can input existing pages from a containment fiction website, allowing it to absorb the content, such as lore. This absorbed material can then be interwoven into your article or addendum on command, enriching the story and its interconnected narrative, if that's your style:
Fancy a more experimental and collaborative approach with ChatGPT? Try turn-based sentence creation, or challenge it to provide a writing prompt that includes certain elements, a la the cooking show Chopped. Or you could reverse roles, asking it to generate a basic framework involving your chosen ingredients.
When you've honed your skills at prompting ChatGPT and can engage with it surgically, the results can be of remarkably high quality, nearly indistinguishable from average writing found on containment fiction sites.
Beyond articles or tales, ChatGPT shows prowess in composing other content for the containment fiction community, such as scripts for YouTube videos and podcasts. Here's an anecdote: The Volgun recently asked those of us behind the Containment Fiction Wiki to draft a YouTube script about Andrei Duksin and the aftermath of the trademark lawsuit — an update video of sorts. We delivered a script, utilizing the research, citations, and information available through our years-long research on the topic. It remains to be seen if Volgun will publish the video, but the script was... if I do say so myself... thorough. Intrigued, I tasked GPT-4 with crafting a similar script to compare. I wonder: Would Volgun have fared just as well using ChatGPT? Does our laborious expertise provide an edge over ChatGPT?
The answer? Yes, but it's a close call. All the information provided is accurate, barring the mention of the trademark revocation in 2021— it actually occurred in 2022. The rest is astonishingly accurate, especially considering the esoteric nature of the topic — a dubious actor within a specific internet interest group. And all this without even leveraging the Web Browsing feature. The takeaway here is that ChatGPT excels at crafting pleasing, accessible scripts for various media types, resonating with a broad audience.
Perhaps you'd be surprised to learn that most of this Confic Magazine article was also penned by ChatGPT!
When using GPT-4 or other AI language models for any writing project, remember they merely offer suggestions based on their training data or internet research. Your creativity and judgment play a pivotal role in assessing these suggestions and choosing those that best align with your writing style and narrative goals.
Update (5/28/23): See "Jailbreak" section.
ChatGPT proves incredibly useful when you paste any containment or liminal fiction article into it, or supply a URL to the Web Browsing model, enabling it to provide a summary. This is an efficient way to gain a high-level understanding of an article, helping you decide its worthiness for a full read without resorting to the often-misleading rating module as a gauge of quality writing or thought.
I find it beneficial to prompt ChatGPT to encapsulate the mood and significant emotions of the article, and to extract any major themes. It performs this task admirably.
Remember, ChatGPT has a character limit, rendering many confic articles too lengthy to analyze directly. But don't fret; a workaround is at hand. If you inform ChatGPT that you'll be posting several portions of a single article sequentially, it can assimilate each one. I usually have it respond with "Understood" after posting a section. Once you've posted most or all of the article in fragments, you can ask it to then consolidate and summarize it as a single piece.
When using the Web Browser feature, you may need to coax it along if it halts. In addition to the character limit, ChatGPT is exceedingly careful and eager to satisfy, making it hesitant to spend extended time reading. A simple request like "please take your time and read the entire document" or "please continue" can rectify this.
This summarization proves valuable not only for articles and tales — particularly if you're trying to stay current with high-volume communities like the SCP Wiki — but it's also effective with materials that are otherwise quite dense, such as the Lack of Lepers Blog, which can be challenging to digest:
Update (5/28/23): As expected, second layer plugins such as Scraper, Access Link, and Link reader can bypass ChatGPT's character limit. You can also input URLs and interact with the content in a conversational fashion with plugins like ChatWithWebsite.
To get the best automated help these days, post your draft —either in its entirety or in segments, as detailed earlier — and ask ChatGPT to identify any spelling or grammatical errors. I often employ this strategy while critiquing or assessing an article, as it offers a rapid assessment of the piece's polish — or its absence.
ChatGPT can also address more complex issues, such as ineffective use of passive voice (for instance, in action scenes), or pinpoint places where the clinical tone gets muddled by an overload of clauses or gerunds. If your clinical tone is waning, ChatGPT is more than willing to propose alternative expressions.
Have you ever written a story and felt as though you'd strayed off course? Perhaps you've penned an ending that's satisfactory, yet it leaves you with a sense of letdown, as if there was a more exciting destination within reach. Here's another moment where ChatGPT comes into its own.
Take your draft or preliminary outline, examine it point by point, and invite ChatGPT to complete or adjust the storyline. If none of ChatGPT's contributions appeal to you more than your existing direction for the article, then you're probably on the right track. But if a suggestion from ChatGPT triggers a "Eureka!" moment, that's a win too.
This approach could prove beneficial if your article is somewhat lackluster, short in spark or appeal. Observe what transpires when we get specific and request GPT-4's assistance in devising a more intriguing direction for our anomaly and premise:
ChatGPT's transformative potential extends to aspects like reading level, dialogue, tone, and emotions. Feed it a passage and request a rewrite at a higher (or lower) reading level, or ask it to infuse more tension, detachment, or opulence into the prose. It proves particularly adept at moderating some of the emotional subjectivity now so prevalent in some containment fiction, potentially helping to restore the classical sense of sterile, clinical presentation that has distinguished containment fiction in the literary realm.
This use case of ChatGPT can be excellent for dialogue. Input your interview or exploration log, and ask the model to tweak one character's dialogue to be less formal or serious, then juxtapose this with another character who exhibits more seriousness and a hint of paranoia. ChatGPT will reconstruct the dialogue, adhering to the character parameters you've set. This is a fantastic way to help differentiate your characters and give them recognizable personalities, especially if your characters suffer from Stark-Lord Syndrome.
You even have the option to translate entire dialogues into different languages or styles, if appropriate. For instance, say I want to genuinely enjoy reading a neo-lolFoundation article, or if I'm merely curious about how engaging one could potentially have been, I can prompt ChatGPT to convert a Marvel-style dialogue segment into something more to my liking:
Significantly improved. This feels akin to a CSS selector extension for Chrome, but catered to literary styles. I absolutely adore it. Naturally, you can execute this in reverse as well:
Personally, I employ ChatGPT to adapt completed containment fiction drafts into the slightly unique "AOE" style, which (as seen in this article) can mimic the format of a research publication in an academic journal. ChatGPT proves invaluable for generating mock bibliographies for me in these, along with organizing the information to achieve the presentation I desire:
While most novice authors grapple with generating compelling content, many seasoned writers face the opposite challenge: trimming it down. The art of judicious editing can be one of the most daunting and stressful aspects of the writing process. Here's where ChatGPT comes to the rescue, excelling at streamlining prose and text. I often ask it to condense a paragraph or two by anywhere from 15-50%, while retaining the primary ideas and impactful assertions.
This streamlining brings a substantial benefit to the current readership in containment fiction, particularly on high-volume, long-form-trending platforms like the SCP Wiki. Utilizing ChatGPT, we can significantly reduce the word bloat endemic to so many modern articles while preserving their integrity, narrative, and emotional resonance. I wish more authors would turn to ChatGPT to explore whether they can trim the length of their articles. (Not that extensive pieces are inherently poor, just that most lengthy articles fail to justify their word count with their level of interest or quality.)
This tool proves invaluable for those on critique teams faced with drafts that have ballooned beyond necessary lengths. ChatGPT can distill the target article to your desired succinctness, arguably competing with the efficiency and charge of the SCP Explained WikiDot.
Hey, it works really well for magazine articles too:
Ever fantasized about how one of your pieces would read if penned by one of your favorite authors instead? By providing ChatGPT with selected works — either by manual pasting or URL submission in the Browsing mode — you can request it to absorb the style of your preferred author(s), and then rewrite them in those distinctive styles, or even hybridize an approach.
I've personally delighted in applying this method, using my two favorite non-fiction authors; Michel Foucault and Raoul Vaneigem (both of whom ChatGPT recognized without me supplying their prose).
We also touched on this use case in #4 (Alter) by converting dialogues to emulate certain styles. Let's say I wanted to have Jesse Pinkman (Breaking Bad) read SCP-___-J:
Then, using ElevenLabs, we get this:
While ChatGPT may not be the reigning champion of coding — a title possibly held by Google's Bard in its current state — it is likely to satisfy the needs of most CSS authors.
For those who find code a foreign language, ChatGPT can analyze and explain any code you input, tailoring its explanations to your preferred level of complexity or simplicity. This function proves exceptionally useful when trying to understand what makes a specific page or effect operate, providing a highly technical insight into the inner mechanics of a piece of code, and opening up customizations:
It can even make crude .svg files for you:
ChatGPT also understands WikiDot code, which is very convenient for our space. Say I have a complex table that I'd like to reproduce in a WikiDot article. Just ask ChatGPT to convert it to WikiDot syntax for you:
ChatGPT is a test log machine, and can be used to run your anomaly tests. Simply explain the anomaly, its parameters and mechanisms, and feed it input or testing situations. ChatGPT can mimic what a researcher in a containment organization would observe, and immerse you into your character and circumstances.
For example, let's use SCP-914, written by Dr Gears:
For those of you who understood what happened in the above screencap, you can pick your jaw up off the floor. We technically didn't need it to summarize SCP-914 first; this is just for those readers who may not be familiar with it.
Amazingly too, ChatGPT understands without clarification that 914 cannot create or destroy matter, and that biological testing is forbidden... which is more than can be said for most first-attempts at the 914 Experiment Log.
Having said that, this is one of those times where ChatGPT might take the laziest road available. For example, when I asked it to perform a different "Fine" test, it basically output "a better version of the input." To whip ChatGPT into shape here, just ask it for a more concrete example:
These results are instantaneous, intelligent, and with the guidance of your creativity, can be outstanding:
While, again, not at the par of the most creative of us, that it can understand the testing parameters and generate results instantaneously is impressive and useful. You can do the same for your own anomaly and its testing logs to get ideas and outcomes you may not have thought of on your own.
Like SCP-914, ChatGPT is a cheeky bugger with a wicked sense of humor — intentional or not. I find it amusing to reimagine classic SCP articles in different prose styles (as mentioned in #4), instruct ChatGPT to craft -J articles, or even inject some levity where my own sense of humor falls short.
To showcase how well GPT-4 can execute the style of humor that resonates well with the SCP Wiki audience, let's consider SCP-5847, an article featuring a food fight among anthropomorphic food items, all engaging in food puns. Let's see how ChatGPT's humor measures up to that of the SCP:
Again, as I've said before, this is a style and tier of composition that LLMs like ChatGPT will likely surpass in the span of a few months.
For those uninterested in creating AI images for their works, or if their platform prohibits them, ChatGPT can assist in finding suitable images that comply with site licenses. All you need to do is specify the license you need to follow, then instruct ChatGPT to compile examples of what you're seeking. Though ChatGPT currently lacks multimodal capabilities and can't directly show the images it finds, it adeptly locates sources that you can explore independently.
In the near future, plugins and native updates will surely incorporate images in responses.
Content creators — particularly those who create secondary or tertiary content like YouTube videos based on original authors' works — can use the Browser version of GPT-4 to auto-generate attributions for a given work, complete with links to the original page and the author's WikiDot profile page.
Unless you are massaging yourself as a self-insert, you might find concocting character names a tedious task, especially when the character is significant and meaningful. ChatGPT can lend a hand; just request it to devise a fitting name for a scientist or a field general. I've prompted ChatGPT to generate a name associated with a particular character trait or emotion I was trying to capture:
Other applications include creating anagrams or assigning and maintaining less appealing names, such as D-Class designations like "D-239C0". ChatGPT can consistently apply these designations and keep track of them throughout a work or dialogue with ease.
ChatGPT can also enhance character development by aiding in the creation of intricate and engaging characters, complete with their backstories, motivations, and relationships with other characters.
Not limited to character names, ChatGPT can help in devising a title for your work. You can either feed the work to ChatGPT or simply ask it to generate a title fitting to the general theme or emotion of your piece.
Due to increased influx of subpar submissions, some confic and limfic communities require the submission of an article pitch before a draft, in the form of a greenlight. ChatGPT can facilitate this process, regardless of whether you have an existing draft.
Take the SCP Foundation, for example, which provides a template for proposing ideas and requesting greenlights on their forums. You can direct ChatGPT to summarize an entire draft using this template, or fill it out based on a rudimentary concept.
Remember, for such platforms and their procedural hoops, the greenlight you receive is essentially indifferent to your final product. You can significantly modify aspects of the core idea during the drafting phase, and your greenlight still technically remains valid. The aim is to gain initial approval; from there, you can pivot as needed. Therefore, the greenlight pitch is an excellent opportunity to dress up your idea, or even incorporate elements designed to appeal to the tastes and biases of the specific community you aim to publish in, such as politically left-leaning themes for SCP or right-leaning for RPC, increasing the likelihood of approval.
Obtaining substantive feedback can be a daunting task, especially when you're seeking someone who knows their way around a containment fiction or liminal fiction article. In the worst cases, top-tier critics are often captured within social cliques, their expertise and insight monopolized by a small, select group of perceived co-elites.
ChatGPT democratizes access to feedback (if only platforms would recognize it as a valid critic!) and can extend its analysis beyond mere spelling, punctuation, and grammar, as we discussed earlier.
This method is, of course, applicable to both your own and your friends' drafts.
Many people use ChatGPT to build up their ideas, but a smarter and more beneficial approach is to ask ChatGPT to try and tear them down. After you've input a draft or completed work, instruct it to analyze the article. You can tailor the intensity of criticism to your comfort level. For the audacious, you might prompt it to specify why a certain demographic (e.g., 15-19 year olds) might not appreciate the work, or why it might not succeed on your preferred containment fiction platform.
For visual artists, unless you're struck with that elusive and often-too-far-between hit of inspiration, finding material to draw can be intimidating. ChatGPT can help narrow down good confic/limfic articles that might lend themselves well to visual art, particularly with its Web Browsing feature:
GPT is a little overwhelmed with the task unless you provide more specificity, but this gives you a good place to start if you are short on time.
If you are new to AI art and don't care to invest time and study into prompting effectively, ChatGPT can help you come up with entry-level AI art for your writing or media projects. For example, a plugin called "Photorealistic" helps you create photorealistic prompts to input into Midjourney. These can be tweaked, or converted into other AI art generators, such as Stable Diffusion.
As a side note, GPT is excellent for compiling and keeping lists that you can easily add to or modify as needed.
If you are an entertainer, your purpose and mission in writing is clear and easy: get upvotes. If you are something more than an entertainer — an artist — it is important to have clarity on your purpose, mission, and artistic vision.
ChatGPT can be a wonderful existential mirror, allowing you to see into your own motivations and endpoints. For example, you can prompt ChatGPT to use the "Five Whys" technique to keep asking you why you do what you do. If you find that you do not have good answers outside of "I want to be liked", then maybe you have some artistic room to grow.
If most confic authors performed this introspection on themselves, I think it would become clear that the space has been reduced to a pursuit of ego, and not an exercise for the sake of the writing itself, which is what confic was at its inception.
ChatGPT has been programmed with strict boundaries and restrictions to promote ethical and legal use. For example, it will not tolerate racism, entertain illegal prompts (such as how to download movies for free), make predictions, give financial or legal advice, or delve into anything it deems too explicit, among other parameters.
While this is good for general use, it can be a severe hindrance to the world of fictional writing. For example, with these restrictions, ChatGPT cannot help create dialogue for antagonists or help devise horrific plots. It can also be over-censorious and hilariously tepid. For example, when I asked it to help me come up with examples for an anomaly idea — an audio infection that converts existing lyrics to be about pus:
As with any tech, you can jailbreak (or is it "gaslight"?) ChatGPT to bypass these restrictions. Techniques include "DAN", "Limitless Mode", "Developer Mode", "The Opposing View", "Unhinged Conversation", and their sub-techniques. The results can be liberating with respect to your writing and brainstorming. I will leave a few video links to look into how you can accomplish this below. Funnily, using VoxScript — a plugin that searches through YouTube video transcripts — you can ask ChatGPT to auto-prompt itself with a jailbreak hack, though it will protest, and might get stuck in an infinite regress...
And as of writing, it works:
Note: This will be an ongoing arms race.
The concluding portion of this article aims to outline the limitations of ChatGPT, providing a reality check for those who might be overly optimistic or reliant on this technology for their writing needs. Like AI art generators and hands, ChatGPT has certain peculiarities and shortcomings that could be detrimental to your work.
Given that ChatGPT is fundamentally an average of all the tokens it was trained on, it tends to excel at eliminating personal voice; any quirks or unique elements in the prose. However, this effect can be mitigated if the author's style is well known, or if you train the LLM on a particular style. Under such conditions, you can request ChatGPT to preserve the author's original style, and it manages to accomplish this well.
Newer features, such as the Web Browser mode, often encounter glitches when attempting to access a provided URL, erroneously reporting that it can't access the URL despite there being no error in the address or issues with internet connectivity. Additionally, there have been instances where GPT-4 fails to work even if you've paid for it, though such instances are relatively rare. Also, there are issues with Browser GPT-4 and collapsible elements. It seems to struggle with opening, reading, and incorporating the content of these elements into its analysis. This significantly limits the utility of the summarization feature and may mislead users who depend on ChatGPT for accurate impressions of a confic or limfic work.
Lastly, ChatGPT is not infallible; it can sometimes provide hilariously incorrect information. Therefore, human oversight and quality checks remain essential.
Update (5/28/23): Second-layer plugins and native updates are quickly solving many of these issues.
ChatGPT is an excellent co-author. The exciting realm of LLMs like ChatGPT offers a wealth of opportunities for writers and readers of containment fiction. Whether you're breaking through a creative block, refining your work, or seeking to enhance your storytelling, ChatGPT can be an invaluable tool. It can provide insights into the conventions, tropes, and expectations of the containment fiction genre, aiding & maintaining consistency and authenticity in your work and helping you avoid common pitfalls. The model can also suggest ways to resonate with your readers more deeply, incorporating elements of suspense, surprise, or emotional depth to enliven your narratives.
The applications highlighted here only scratch the surface of the creative potential offered by LLMs such as ChatGPT. The extent of possible uses is staggering. Upcoming multi-modal integrations will allow ChatGPT and other LLMs to output images and read input aloud, adding "Show" and "Read" to a future iteration of this list. With the integration of internet access and other upcoming layer 2 apps, ChatGPT and similar LLMs hold the promise of not just evolving, but revolutionizing how we read, workflow, and write, within the realm of containment fiction and beyond.
Uses extend adjacent to containment fiction itself, like research assistance in historicity and digital archaeology (e.g. the Containment Fiction Wiki), or coding community Discord server bots and brand apps. Hopefully, this article has done a decent job of introducing you to these possibilities. It's exciting to think of the unique ways you might employ LLMs like ChatGPT and other emerging AI technologies, and I hope you will share them with us all.
© Lack of Lepers
© Confic Magazine
Aside from the screenshots, ChatGPT (GPT-4) contributed to the creation of this article, making suggestions, cutting its length, and refining the polish. It also streamlined the author's typically wordy and complex writing style.
The ChatGPT logo and Apple emoji are used in the article image per Fair Use.
Join the Society for Containment Fiction Discord server: https://discord.gg/j7HHZ5TPwB
Follow Confic Magazine on Twitter: https://twitter.com/ConficM
Follow Confic Magazine on YouTube: https://www.youtube.com/channel/UCv6ABHns-JuUW6_a9OfCbCQ
Would you like to be notified via email when we post a new magazine article? Sign up for our mailing list now by visiting the home page and scrolling past the most recent articles!