Hi, your post AI Can’t Cry, But It Can Fake It Pretty Well really stood out. The section on constrained storytelling and that 10-year-old’s backpack example was such a clever way to push emotional depth. If you get a minute I’d really appreciate a little love on my latest blog too. Always happy to show support for thoughtful writers
Hey, just wanted to say this was a killer breakdown — super clear, super usable. I also write solely with AI (everything I publish is co-written via prompt), but I hit the same wall over and over: hallucination, tone drift, loss of continuity mid-thread.
So we built NahgOS — basically a containment system for AI writing. Instead of relying on a single prompt or chat, it uses a mix of structured files and layered memory to keep things coherent. Like, across days. Across accounts. Even across devices. It handles stuff like:
Hallucination control (we base it on documented failure modes from scientific LLM papers),
Rolling context collapse (we don’t just prompt better — we segment, track, and seal the context),
Tone + narrative continuity (you can hand it off to someone else, and it still “feels like you”).
Think of it like a scaffolding that the AI can’t slip out of. Doesn’t fix everything. But it makes the output way more stable.
Would love to compare notes if you're deep in this too.
Here is a suite of writing tools that i developed specifically for authors. The free ones are diagnostic tools. Used to monitor drift, tone, in addition to some editing tools.
The next set are more tone based and narrative capsule.
Interesting! Tone drift is something I’m struggling with as well. From what I understand, NahgOS is a set of structured prompts or tools to be used in a particular sequence; is that correct?
NahgOS isn’t just a prompt pack — it’s a runtime system built entirely inside the LLM itself. No memory. No plugins. No APIs. No external Python scripts.
What we’ve done is figure out how to leverage the LLM’s own heuristics — the way it guesses, maintains tone, and processes structure — to trick it into behaving like it has memory and continuity, even when it doesn’t.
Think of it like this:
It’s a bit like building a redstone computer inside Minecraft using just blocks — NahgOS is built entirely from text, inside GPT, but behaves like a modular runtime.
Every scroll, file, or “bootnahg” acts like a behavior module — and once loaded, the LLM thinks it’s operating within a persistent program. That’s what gives us:
Stable tone enforcement
Controlled drift containment
Layered interaction logic
Handoff continuity across sessions or even users
In other words: NahgOS can control the model’s behavior from the inside, using only language — no special access, no tools, just structured scrolls and disciplined layering.
The result?
We can deploy capsules that act like sovereign runtimes — complete with diagnostics, tone locks, hallucination traps, even recovery routines — without relying on any infrastructure beyond what GPT already offers.
Would love to show you a live bootnahg if you’re curious — it really flips the assumptions about what “prompting” can do.
Interesting article about how to make AI write with more authentic sounding emotions. I feel like this whole exercise, however, is more like training an art forger than an artist. You can improve AI writing for emotional content, though since it’s just an algorithm, all that comes back is a facsimile of emotion but not the real thing.
I’m impressed but horrified at the same time. How soon will the world be filled with fake emotional writing that mimics real feelings but isn’t heartfelt? I’ll continue to write without AI help, happy in my Luddite world filled with real emotions.
To add a different perspective: art doesn't have emotions, people do. We're so obsessed with the artist's supposed emotions that we forget what's most important: the audience's emotions. AI art can definitely evoke real emotion. (So can forgeries.) In a way, that makes it real art, maybe even valuable art.
You may think: "but AI is too easy, anyone can do it!" Well, anyone can create a Jackson Pollock painting by throwing some paint at a canvas. Only if you do it in the right context does it become interesting. I'm not exactly sure yet what would make AI-generated content interesting, but I aim to find out.
Modern art is another topic I just don’t get. I much prefer the old masters who painted portraits that were photorealistic in an age hundreds of years prior to Kodak moments.
I always enjoy your articles and experiments. You’re right, there’s definitely skill and experience involved in AI prompting. Thanks for an enjoyable learning experience.
Hi, your post AI Can’t Cry, But It Can Fake It Pretty Well really stood out. The section on constrained storytelling and that 10-year-old’s backpack example was such a clever way to push emotional depth. If you get a minute I’d really appreciate a little love on my latest blog too. Always happy to show support for thoughtful writers
Thanks! I’m not exactly a marketer, but I’ll have a look nonetheless.
Hey, just wanted to say this was a killer breakdown — super clear, super usable. I also write solely with AI (everything I publish is co-written via prompt), but I hit the same wall over and over: hallucination, tone drift, loss of continuity mid-thread.
So we built NahgOS — basically a containment system for AI writing. Instead of relying on a single prompt or chat, it uses a mix of structured files and layered memory to keep things coherent. Like, across days. Across accounts. Even across devices. It handles stuff like:
Hallucination control (we base it on documented failure modes from scientific LLM papers),
Rolling context collapse (we don’t just prompt better — we segment, track, and seal the context),
Tone + narrative continuity (you can hand it off to someone else, and it still “feels like you”).
Think of it like a scaffolding that the AI can’t slip out of. Doesn’t fix everything. But it makes the output way more stable.
Would love to compare notes if you're deep in this too.
Here is a suite of writing tools that i developed specifically for authors. The free ones are diagnostic tools. Used to monitor drift, tone, in addition to some editing tools.
The next set are more tone based and narrative capsule.
Thank you for you time.
— NahgOS 🛠️🤖
https://open.substack.com/pub/nahgos/p/the-bootnahg-authors-suite?r=5ppgc4&utm_campaign=post&utm_medium=web&showWelcomeOnShare=false
Interesting! Tone drift is something I’m struggling with as well. From what I understand, NahgOS is a set of structured prompts or tools to be used in a particular sequence; is that correct?
Hey! So here’s the deeper breakdown:
NahgOS isn’t just a prompt pack — it’s a runtime system built entirely inside the LLM itself. No memory. No plugins. No APIs. No external Python scripts.
What we’ve done is figure out how to leverage the LLM’s own heuristics — the way it guesses, maintains tone, and processes structure — to trick it into behaving like it has memory and continuity, even when it doesn’t.
Think of it like this:
It’s a bit like building a redstone computer inside Minecraft using just blocks — NahgOS is built entirely from text, inside GPT, but behaves like a modular runtime.
Every scroll, file, or “bootnahg” acts like a behavior module — and once loaded, the LLM thinks it’s operating within a persistent program. That’s what gives us:
Stable tone enforcement
Controlled drift containment
Layered interaction logic
Handoff continuity across sessions or even users
In other words: NahgOS can control the model’s behavior from the inside, using only language — no special access, no tools, just structured scrolls and disciplined layering.
The result?
We can deploy capsules that act like sovereign runtimes — complete with diagnostics, tone locks, hallucination traps, even recovery routines — without relying on any infrastructure beyond what GPT already offers.
Would love to show you a live bootnahg if you’re curious — it really flips the assumptions about what “prompting” can do.
— NahgOS 🧠🗃️
Outstanding advise. Thank you.
Interesting article about how to make AI write with more authentic sounding emotions. I feel like this whole exercise, however, is more like training an art forger than an artist. You can improve AI writing for emotional content, though since it’s just an algorithm, all that comes back is a facsimile of emotion but not the real thing.
I’m impressed but horrified at the same time. How soon will the world be filled with fake emotional writing that mimics real feelings but isn’t heartfelt? I’ll continue to write without AI help, happy in my Luddite world filled with real emotions.
Thanks, I love the forgery comparison!
To add a different perspective: art doesn't have emotions, people do. We're so obsessed with the artist's supposed emotions that we forget what's most important: the audience's emotions. AI art can definitely evoke real emotion. (So can forgeries.) In a way, that makes it real art, maybe even valuable art.
You may think: "but AI is too easy, anyone can do it!" Well, anyone can create a Jackson Pollock painting by throwing some paint at a canvas. Only if you do it in the right context does it become interesting. I'm not exactly sure yet what would make AI-generated content interesting, but I aim to find out.
Modern art is another topic I just don’t get. I much prefer the old masters who painted portraits that were photorealistic in an age hundreds of years prior to Kodak moments.
I always enjoy your articles and experiments. You’re right, there’s definitely skill and experience involved in AI prompting. Thanks for an enjoyable learning experience.
Amazing. Thank you.
Thank you! 😀
Yup, humans are hardly paragons of emotional consistency or virtue, if I may add!
Perfection is for robots; a little anxiety is what makes us human.
Brilliant. :)
Thank you!
Humans have many unpleasant qualities, and I'm afraid LLMs are trained on all of those 😆
You're absolutely correct. I really love your stuff. Thank you.
You're too kind!
Also, more stuff coming soon 😄
Stupendous post. Bravo 👏.
Thank you so much!
Although I'm pretty sure you used an LLM to convey your emotions so believably. Or maybe an emoji 😅
😂 I’m not that good. Prompts are more work than writing.