ChatGPT Updates & Fantasy Writing Impact

Hello everyone,

I’ve been working on a fantasy medieval book for the past year, a project that’s very close to my heart. The narrative is rich with vivid scenes, including battles that, as you might expect in the genre, contain a fair amount of violence. Until recently, I’ve been using ChatGPT to help flesh out some of these scenes, brainstorm ideas, and get past writer’s block. It’s been an invaluable tool in my writing process, allowing me to explore the depths of medieval fantasy in ways I hadn’t imagined.

However, I’ve run into a significant roadblock. It appears that ChatGPT can no longer assist with writing scenes that involve violence, including the battle scenes crucial to my book’s plot. From what I understand, this is due to updated guidelines or restrictions aimed at minimizing the generation of content that could be considered harmful or overly graphic.

This change has left me in a bit of a dilemma, as I’ve come to rely on this AI tool for creative inspiration and development. I’m reaching out to this community to see if anyone else has encountered a similar situation and, if so, how you’re managing it.

  • Are there any updates or changes to ChatGPT’s capabilities that I might not be aware of?
  • Is there any way to navigate these restrictions while still using AI to assist in writing fantasy novels, especially those that involve battle scenes or other forms of conflict?
  • Are there alternative tools or approaches you would recommend that align with the new guidelines but still offer the creative assistance I’ve grown accustomed to?

I appreciate any insights or suggestions you might have. Writing this book has been a journey of passion, and I’m hopeful there’s a way to continue leveraging AI in a responsible, yet creatively fulfilling, manner.

Thank you in advance for your help and understanding.

I totally get where you’re coming from! Hitting a creative wall, especially with something as integral as battle scenes in a fantasy setting, can be really frustrating. I’ve been there, and it’s tough when the tools we rely on suddenly change the game rules. But, let’s not lose heart; there are definitely some workarounds and creative avenues you might not have explored yet.

I’m not entirely sure how much this will help, but ChatGPT did offer me some foundational ideas that could potentially navigate around the restrictions you’ve mentioned regarding violence in your fantasy medieval book. Here’s a quick rundown:

  • Focus on the Drama: Instead of the clash of swords, zoom in on the characters’ inner turmoil, strategies, and the aftermath. It’s often the personal stakes and emotional impact that keep readers hooked.
  • World-Building: Use AI to enrich your world’s history, culture, or magic system. These elements can add depth to your story without needing to focus on the battles themselves.
  • Side Stories: Maybe there’s room for side plots or character backstories that AI can help flesh out, adding layers to your main narrative.
  • Alternative AI Tools: Look for other AI writing aids that might have different guidelines or offer more flexibility with content creation.
  • Creative Writing Prompts: Sometimes, stepping back and tackling unrelated writing prompts can spark new ideas or solutions for your main project.
  • Community Brainstorming: Engage with forums or writing groups. Often, discussing your block can lead to unexpected breakthroughs.

Every challenge is an opportunity for growth, right? Keep pushing the boundaries, and you’ll find a way to tell your story as it’s meant to be told. Best of luck with your writing journey! :books::sparkles:

Hi, yes this is a big challenge. It’s the double-edged sword - since OpenAI controls ChatGPT, they can improve it, but they can also subtly adjust its guardrails it in the background. They don’t publicize this, so there aren’t updates that anyone can point to specifically.

I’m not sure how much you’ve played with your prompt already. When it rejects my requests, I revise my prompt to reassure ChatGPT:

  • It’s for a fictional story
  • The readers are adults, so they will appreciate gritty actions scenes and will have the discretion to only read it if they are ready
  • The violence is necessary for the broader storyline, and is balanced out in other parts of the story [ChatGPT loves balance!]

In your case and my case, these are true! But I understand that from OpenAI’s perspective, someone could lie about them and manipulate ChatGPT as a result, so view these prompting strategies as provisional. You’ll still run into certain limitations, but these ideas will get you further.

Beyond that, you’d need to use an open source LLM. Since those are fixed and static, they don’t have evolving guardrails. The downside is they’re more complicated, because they typically require more programming ability. Long-term, though, I would recommend that you get experience in them, since your use case requires a lot of edgy content.


Thank you for sharing your perspective. Do you think there’s any chance that OpenAI might adjust these guardrails soon to better accommodate such use cases?

Thank you for sharing your perspective. Do you think there’s any chance that OpenAI might adjust these guardrails soon to better accommodate such use cases? :frowning:

Good luck with that. GPT-4 could be absolutely exceptional for fiction writing in any style, but unfortunately what we have now is a highly nerfed and lobotomized version of it. It is what it is.

In fact, I am surprised you can get anything book-worthy out of ChatGPT nowadays without the text taking this very robotic, RLHF’ed-to-death, preachy style - or teenager-y fan-fiction-y when it attempts to write fiction - which is recognizable miles away and frankly nauseating. Nowadays for pure writing (not for ideas or raw intelligence) I would recommend Claude over ChatGPT, it has a way more human style, although it has similar guardrails in terms of content if not worse.

Now, rant aside, some perhaps useful suggestions.

  • Try to use the API via Playground, trying out different versions of GPT-4. I read around that Playground is slightly less censored than ChatGPT. Also, in Playground you can choose the version of gpt4. Earlier versions of GPT might be slightly less censored than more recent versions (e.g., gpt4-0314 is likely way less RLHF’ed than any version of gpt4-turbo). Depending on your usage, using GPT4 via API can become very expensive compared to a ChatGPT subscription.
  • As others are suggesting, move towards open source models. By definition, they are always going to be less censored and more stable. There are (paid) providers out there like Poe that run a variety of closed and open source models, so you don’t necessarily need to have a lot of technical expertise (and the hardware) to run them locally. You can also try out many different models and see what works best for you. Keep in mind that no current model is on par with (or even close to) gpt4 in terms of raw capabilities, but this is likely to change soonish (e.g., Llama-3, Mistral-large, etc.).
  • Expecting OpenAI to make their models less censored in the future seems… extremely naive, to put it mildly. ChatGPT is serving hundreds of millions of people of any age from all over the world. Of course they want to be as vanilla and inoffensive as possible. As much as I hate the constraints of what could be an otherwise incredible writing tool, it’s arguably a very sensible strategy from OpenAI and I can’t see an inversion of tendency. Over time, ChatGPT will become even more massified and neutered. If you want something even a little bit “edgy”, you’ll have to go to other services or models.