I suspect Lemmy is going to dislike this, but local LLMs are great writing helpers.
When you’re stuck on a sentence or a blank paragraph, get them to continue it, and rewrite it once it jogs your mind. If you’re drafting ideas for characters or chapters, you can sanity check them or sometimes get ideas. They can reword and clean up your writing and improve it beyond what self experimentation can do… just keeping in mind that its like an idiot intern that tends to repeat itself and hallucinate.
And this is very different from an API model like ChatGPT because:
It won’t refuse you.
It’s formatted as a notebook you can continue at any arbitrary point, rather than a user/chatbot type format.
The writing isn’t so dry, and it isn’t filled with AI slop, especially with cutting edge sampling
All its knowledge is “internal,” with no reaching out to the web or hidden prompting under your nose
Along with all the usual reasons, namely being free, self-hosted, more ethically trained, fast and efficient with long context thanks to caching, and has nothing to do with Sam Altman’s nightmarish visions.
I mean, I kind of get the intention, but, personally, I don’t need or want that kind of help in my hobby. If I worked as a pro writer, and I had a deadline and I needed to be more “efficient” with my writing, maybe I would give LLMs a try, but I don’t need to be “efficient” with a hobby, since I enjoy the whole process, even the blockages. It’s just much more satisfying to come up with the ideas yourself as you write.
I don’t mean to be harsh or anything, if it works better for you to do things the way you do them, great for you. In the end, what makes one feel better is what matters.
For me it’s not efficiency, if anything it takes much longer running the LLM notepad because I will randomly ask it about passages/chapters and revise stuff.
It’s kinda fun having an assistant to just bounce ideas off of on a whim. You can’t get that with beta readers, as they don’t just sit there while you write (and the LLM is much faster), and I don’t feel like I’m being charged for every response with an API model, especially if it ingests the entire story every time.
It’s also “smart” beyond me. For instance, sometimes I wanna come up with a name for a character, city or whatever, and I can ask it “what’s an interesting name for this engineer character’s brother, from X city in the story, maybe something mythological and fire themed,” and it will crank out tons of examples and start a little conversation about it. It takes me places I never would have googled, much less known off the top of my head, because everything is on the top of an LLM’s head.
I admit that the Beta-reading stuff sounds interesting, although I have never felt the need for one. If what I wrote is bad beyond correction, then it’s bad and that’s it, it is what it is; and if it’s good, then great. My main objective is that I am satisfied with the result, and I have never had problems in that aspect.
And as for the names, I have the habit of researching everything that catches my attention, a taste inherited from my mother who is a university professor. I even have an encyclopedic collection of 15 volumes that was my Wikipedia when I didn’t have internet, inherited from my mother as well.
Seeing the origins, evolution and etymology of certain things and concepts is quite inspiring. If I have at least one concept, say: a character who is from Scotland or a similar place, I look up common Scottish names and their meanings, and from there choose the one I like best, and with what I get I can delve into the character’s personality and/or background based on their name. Even if I don’t find what I’m looking for, I can get information that I can use for something else, I always find more than what I’m looking for.
Yeah, it’s great for instant feedback. You can ask it “does this dialogue above feel out of character or stilted?” or ask it about character growth, and it will generally give an interesting critique in-line with your writing in the notepad.
You have to keep in mind that it can go off the rails (repeatedly querying it is good to check for this), and has a sycophancy bias, and is kinda dumb, but its still quite useful.
I suspect Lemmy is going to dislike this, but local LLMs are great writing helpers.
When you’re stuck on a sentence or a blank paragraph, get them to continue it, and rewrite it once it jogs your mind. If you’re drafting ideas for characters or chapters, you can sanity check them or sometimes get ideas. They can reword and clean up your writing and improve it beyond what self experimentation can do… just keeping in mind that its like an idiot intern that tends to repeat itself and hallucinate.
And this is very different from an API model like ChatGPT because:
It won’t refuse you.
It’s formatted as a notebook you can continue at any arbitrary point, rather than a user/chatbot type format.
The writing isn’t so dry, and it isn’t filled with AI slop, especially with cutting edge sampling
All its knowledge is “internal,” with no reaching out to the web or hidden prompting under your nose
Along with all the usual reasons, namely being free, self-hosted, more ethically trained, fast and efficient with long context thanks to caching, and has nothing to do with Sam Altman’s nightmarish visions.
I’d recommend: https://huggingface.co/nbeerbower/Qwen2.5-Gutenberg-Doppel-32B
And once the story gets long: https://huggingface.co/EVA-UNIT-01/EVA-Qwen2.5-32B-v0.2
I’d recommend LanguageTool (with a local server and the browser extension) for locally hosted spelling/grammar/style as well.
I have ADD, which may be why I find this setup to be so therapeutic.
I mean, I kind of get the intention, but, personally, I don’t need or want that kind of help in my hobby. If I worked as a pro writer, and I had a deadline and I needed to be more “efficient” with my writing, maybe I would give LLMs a try, but I don’t need to be “efficient” with a hobby, since I enjoy the whole process, even the blockages. It’s just much more satisfying to come up with the ideas yourself as you write. I don’t mean to be harsh or anything, if it works better for you to do things the way you do them, great for you. In the end, what makes one feel better is what matters.
For me it’s not efficiency, if anything it takes much longer running the LLM notepad because I will randomly ask it about passages/chapters and revise stuff.
It’s kinda fun having an assistant to just bounce ideas off of on a whim. You can’t get that with beta readers, as they don’t just sit there while you write (and the LLM is much faster), and I don’t feel like I’m being charged for every response with an API model, especially if it ingests the entire story every time.
It’s also “smart” beyond me. For instance, sometimes I wanna come up with a name for a character, city or whatever, and I can ask it “what’s an interesting name for this engineer character’s brother, from X city in the story, maybe something mythological and fire themed,” and it will crank out tons of examples and start a little conversation about it. It takes me places I never would have googled, much less known off the top of my head, because everything is on the top of an LLM’s head.
I admit that the Beta-reading stuff sounds interesting, although I have never felt the need for one. If what I wrote is bad beyond correction, then it’s bad and that’s it, it is what it is; and if it’s good, then great. My main objective is that I am satisfied with the result, and I have never had problems in that aspect.
And as for the names, I have the habit of researching everything that catches my attention, a taste inherited from my mother who is a university professor. I even have an encyclopedic collection of 15 volumes that was my Wikipedia when I didn’t have internet, inherited from my mother as well.
Seeing the origins, evolution and etymology of certain things and concepts is quite inspiring. If I have at least one concept, say: a character who is from Scotland or a similar place, I look up common Scottish names and their meanings, and from there choose the one I like best, and with what I get I can delve into the character’s personality and/or background based on their name. Even if I don’t find what I’m looking for, I can get information that I can use for something else, I always find more than what I’m looking for.
As someone who hates to worry about character growth and what not this sounds interesting
Yeah, it’s great for instant feedback. You can ask it “does this dialogue above feel out of character or stilted?” or ask it about character growth, and it will generally give an interesting critique in-line with your writing in the notepad.
You have to keep in mind that it can go off the rails (repeatedly querying it is good to check for this), and has a sycophancy bias, and is kinda dumb, but its still quite useful.
I’m not terribly interested in dialogue because I see a large part of any world building I do to be mainly encyclopedia entries more than a story