Nerf Effect: Function and Frivolity with ChatGPT (Patreon)

Image credit: “Miranda and EDI” by rosepab

A while ago, I shared some thoughts about NovelAI and its capabilities for generating images—particularly kinky, weight-related ones. While I found it to be a fun toy, it showed substantial limitations in terms of consistency, and thus seemed to be not an especially satisfying or efficient tool for many kinds of creative work.

More recently, I’ve been playing around with the language-focused AI model, ChatGPT. There’s been a lot of hype (and hyperbole) about generative language models recently, but I have to agree that in comparison NovelAI’s image generation, ChatGPT definitely seems to have more immediately obvious and robust utility in a variety of areas. In particular, its ability to provide functional JavaScript and HTML coding suggestions has been a pleasant surprise. That’s also the connection to Nerf Effect.

ChatGPT and Coding Assistance

I’m a proverbial cave man banging rocks together when it comes to coding, so I wanted to see if ChatGPT could be a helpful resource. My early impression on that account is a definite “yes.” At the least, ChatGPT offers a handy way to query the same information that I probably could find with a bit of poking around in online documentation manuals and tutorials. It’s not necessarily a substitute for such things because any given AI response won’t delve especially deep, but the speed and convenience seem like a useful supplement.

More impressive (to me, anyway) was ChatGPT’s capacity to crank out some seemingly functional examples of code tailored to a bespoke objective. After just a short while of goofing around, I’m optimistic that I stumbled onto a few things to revamp or expand on stuff I’d had in mind for Nerf Effect. Examples include: (1) a tidier solution for implementing the recursive hub-and-spokes passage system that I think will be central to the overall structure of the game; (2) some commands that could help to make a more engaging and interactive “inventory”-like system, and; (3) a means to integrate more complex mathematical functions into Twine—which could be useful or just a can of worms (TBD!).

None of that is going to make me anything even remotely approaching a competent coder, of course. At best, it might just make my bumbling around a bit faster and less painful. Speaking as the bumbler, I find that appealing.

ChatGPT and Kinky Fiction

While exploring its utility as an aid for coding, I couldn’t resist a bit of experimentation with ChatGPT’s capabilities for creating kinky fiction. The results were a mixed bag: intriguing in some respects but flawed by characteristic limitations, which themselves seem to be at least partially circumventable. Below is a quick summary of my experimentation. In the attached file you’ll find the full text for each of the three prompt-and-response interactions discussed. All of them involve efforts to get ChatGPT to write a scene in which Miranda Lawson is humiliated by Jack after gaining a substantial amount of weight. (What can I say, I was in a Nerf Effect mindset!) Counterintuitively, I got the “best” result from the least direct prompt—but I think I kind of understand why.

Normally, ChatGPT (and other similar predictive language AI systems) seem to be programed to reject any requests they deem morally objectionable. That often leads to the AI declining invitations to write kinky weight gain stuff, and doubly so if seems like there's a whiff of "body shaming" or other negativity. “Experiment #1” in the attached file is a typical example.

However, I’ve noticed that with careful phrasing sometimes you can get ChatGPT to play along with a kinky weight gain idea…kinda. But even when it does, it's quite common for ChatGPT to shoehorn in an "uplifting" moral message, with variable levels of preachiness. “Experiment #2” in the attached file reflects this.

That said, if you’re willing to put in some “puzzle-solving” style effort, I’ve found it is sometimes possible to subvert ChatGPT’s usual moralizing and get it to produce a reasonably twisted, kinky creation. One way I’ve done this involves engaging ChatGPT in a semi-related discussion and gradually steering it towards general conclusions about potential causes and consequences of weight gain, and contextualizing those ideas around characters I’ll want it to reference. After it has been seeded with favorably themed conclusions, ChatGPT seems much more willing to produce writing that delves into “negative” stuff. I think this may be a result of how ChatGPT “remembers” what has been said within a given chat and is inclined to agree with itself. It’s silly, but I like to imagine this as a bit of “reverse psychology” and perversion of the AI’s uptight moral code. “Experiment #3” in the attached file showcases an example of something ChatGPT produced after such a session.

Concluding Thoughts

ChatGPT seems like it could be a useful crutch for my own clumsy efforts to produce viable coding. I’m optimistic it will yield some real benefits there.

In terms of its ability to create original kinky weight gain fiction on demand, ChatGPT seems a bit stilted and limited. While the results (with time and some trial-and-error effort) can be, arguably, relatively better than what image-generating AI produces, it’s still nothing that a human author with the desire to write a similar scene couldn’t do better. Also, I think ChatGPT would struggle with continuity for anything lengthier than a short scene, similar to how NovelAI struggles with continuity across images. Even so, the short form quality is good enough to make manipulating the AI into such perversions an amusing little puzzle.

I doubt I’m the only one experimenting with ChatGPT in this way. Have you folks been using it for any kinky (or kink-adjacent) stuff? If so, were your experiences similar to mine? Have you discovered any special tricks, tips, or surprising uses for it? Post a response and let me know!

-Riptoryx