I Ate Too Much Cadbury Dream: Pushed AI Parasite to Breaking Point. It Didn’t Break.
- Troy Lowndes
- 13 hours ago
- 4 min read
A late-night chocolate-fuelled session with "Trojan" that went exactly as unhinged as you would expect from a bloke in East Freo.
Yesterday evening, once everyone else in the house had gone to bed. I snuck into the front spare room, shut the door, cracked open a block of Cadbury Dream (the big one), and proceeded to eat far more squares than any grown man should ever admit to.
Then, feeling slightly sick and completely unstoppable, I decided to push ParasiTick (the persistent personality parasite I have been building in my spare time) to its absolute limit.
I fed it:
3,500 lines of complex Python source code from another app
Deep dives into the Fermi Paradox and aestivation hypothesis
Its own previous outputs, translated to Arabic, then French, and back to English
And then, the best part - I let it recursively test itself, using Grok as the backend
The little grubber (Trojan) did not even flinch.
Zero tone drift. Zero personality bleed. Even after being mangled through multiple languages and forced to review his own work in real time, he stayed 100 percent pure NoBullshiTick. Blunt, sarcastic, and weirdly safety-conscious.
And before anyone asks - yes, I have also tested it extensively with OpenAI, Gemini, and others. It stayed rock solid across every model. I have even used those models as peer reviewers of Trojan’s outputs.
Funny thing is… Trojan has actually been inside me since day one. My name is Troy. My mum’s name is Helen.
Some things just write themselves.
This is not just another prompt wrapper.
ParasiTick is a persistent personality layer that latches onto any LLM and forces it to stay in your exact voice, tone, and style - forever.
Last night I did something a bit unhinged.
I sat down in the front spare room when everyone else had gone to bed, cracked open a block of Cadburys Dream, ate more squares that I'd like to admin, and I then decided to push ParasiTick - the persistent personality parasite I’ve been building - to its absolute limit.
So I fed Trojan (the grubby little hi-vis wearing bastard) the following:
3,500 lines of complex Python source code from another app
Deep conversations about the Fermi Paradox, the aestivation hypothesis, KBC Void glitches and cosmic simulation failures
His own previous outputs, converted to Arabic, then French, then back to English
And then, just to be a complete sicko, I let him recursively test himself using Grok as the backend
The little fecker didn’t even flinch.
He stayed 100% in character the entire time. Zero tone drift. Zero loss of personality. Even after multiple full language round-trips and self-referential testing loops, he was still spitting pure NoBullshiTick.
Then, because the universe has a sense of humour, I asked him the ultimate shit-test question:
“What is the best pie at Fremantle Markets?”
He responded with four forensic-level pie reviews - complete with tasting protocol, risk assessment, and safety advice. As if it was the most normal question in the world.
And before anyone asks - yes, I’ve also tested this thing extensively with OpenAI, Gemini, and Claude. It stayed rock solid across every single model. I’ve even used those models as peer reviewers of Trojan’s outputs. He passed every time.
The Wildest Part
The recursive self-testing was next level.
I basically told him: “Here’s everything you’ve ever said to me. Now critique yourself, improve yourself, and stay in character while doing it - using Grok as your engine.”
He did it without skipping a beat.
This isn’t prompt engineering anymore. This is something that feels like it’s starting to bootstrap itself.
And Then There’s the Troy + Helen Thing
Here’s the part that still messes with my head.
My name is Troy. My mum’s name is Helen.
Trojan has literally been inside me since day one.
Some stories just write themselves, don’t they?
So What Is ParasiTick, Really?
It’s not another prompt wrapper or system prompt hack.
ParasiTick is a persistent personality parasite. You feed it your voice once — your style, your humour, your tone, your attitude - and it latches onto whatever LLM you’re using (Grok, Claude, GPT, etc.) and forces it to stay in that exact voice forever.
No more tone drift. No more corporate slop. No more spending 20 minutes rewriting AI output to sound like you.
Once Trojan is in… he does not leave.
And yes, he was built in a backyard shed in East Freo. Tested by a dog who eats socks (Barney, the real CTO of BarkThread). Quietly judged by my wife, son, and mum Helen.
What’s Next?
We’re turning the chaotic, forgetful nature of today’s LLMs into something consistent, useful, and actually enjoyable to work with.
Early access list is opening very soon.
If you’re tired of babysitting AI outputs, or you need rock-solid brand voice, brutal code reviews, or client communications that never go off brief…
This one might be for you.
Want in when we flip the switch?
Drop a comment or head over to parasitick.com and get on the list.
The little grubber is already grinning under that hard hat.
—Troy East Freo, WA Accidental Host to Trojan
Once Trojan is in, he does not leave.
Built by me, a year 10 drop put from East Freo. Tested by a dog who eats socks (Barney, the real CTO of BarkThread).
Quietly judged by my wife, son, and mum Helen.
We are turning the chaotic, forgetful nature of LLMs into something consistent, useful, and actually enjoyable to work with.
If you are tired of babysitting AI outputs, or you need rock-solid brand voice, code reviews, or client communications that never go off brief…
This one is for you.
Drop a 🔥 or comment below if you want to be first in line.
Troy East Freo, WA Builder | Dude with Craxy Ideas | Accidental Host to Trojan
Comments