Presenting our new NAS/Server/Homelab setup and infrastructure changes to better protect our artists against Cloud AI and help on the business end. (Patreon)

During our impromptu break as most of our team was out and about, I didn't want to sit around doing nothing. So I started working on some infrastructure changes in the background I've been wanting to do for a while with these goals in mind:

Move off what platforms and services we realistically can that train their AI models off our stuff.

Create a local storage backup to better archive and host our data so we don't have a single point of failure.

Keep our data private and secure with more security minded methods by moving to a NAS and other cloud storage services like Proton Drive.

Explore what steps I can take to protect our artists against the growing threat of AI while also navigating this new tech world to maintain the tiny business we have.

We won't be able to be 100% successful in these goals. There are a lot of platforms that have AI so integrated, been influenced by it, modified by it, and trains their AI on their platforms that it's impossible to strictly follow this. We'd have to abandon YouTube and various other platforms and services that we can't find an alternative for. And realistically is not an option for us as a show or as a very tiny business that wants to keep supporting artists. It feels like a weird trap.

The System Specs

I myself am not an IT/Tech professional, so please don't follow what I have setup as a guide. I asked several friends and associates who know more about this stuff than me to pick out the parts that'd fit my use cases.

Note: The links go to Amazon links so you can see the hardware. They are not affiliate links.Thinkpad LaptopThinkPad P14s Gen4 AMD 7840U 780M 1TB 64GB 2.8k OLEDGot this years ago, love me thinkpads.

Server/Mini-PCRyzen 9 7945HX, RTX 4060, 32GB DDR5 RAM, 1TB NVMe, Running on Linux Fedora44 KDE. (Got this for $400 or something bucks off ebay years ago. God help anyone buying things now.)

Our NAS (the storage) setupUGREEN DXP4800 NAS with four IronWolf Pro 20TB drives. I hate hardware prices and everything else cheaper was sold out. Even on the discount server websites. Hating that I didn't do it sooner. But this should last us for a good while.

KVM Remote in hardwareGL.iNet GL-RM10 Comet Pro Remote KVMThis allows me to more directly remote into the mini-pc and nas as if I was right at the hardware. Meaning that I am able to access the BIOS and everything, unlike normal remote software that can only access things after the PC has been powered on and into the desktop/terminal environment.

Ethernet Switch10Gbe Ethernet Switch Unmanaged with 2 x 10Gb RJ45 Ports + 4 x 2.5GbIt do be beep boops and provides the internetsUPS (surge protections, temp battery, etc)CyberPower CP1000PFCLCD

Security Changes outside the server

Switched from Windows 11 to Linux (Fedora) to avoid the Cloud AI integrations. No longer have Windows in the house.

Switched my personal phone to Graphene OS to try and avoid all that baked in cloud AI business.

Bought multiple Yubikeys for physical passkey access to many of our accounts and hardware.

Switched from Google Drive, Docs, etc to Proton Drive, Docs, Sheets for their encryption and security settings. We're still slowly transitioning from Google services where we can, but we feel that we won't be 100% able to due to how other businesses we work with operate. So things like Google Calendar, we'll be stuck with for a bit. We're actively trying to find a good alternative to Google Calendar, Google Forms, and others that also fit our backend server needs.

We are transitioning from a publicly used Discord bot to a closed in self-hosted bot to assist with security from the many Discord hacks we've been seeing going around. This isn't new news, as we've had to deal with these issues before, but once more with the rise of AI; there are more issues popping up. In the video I said I was going to locally host it, but after recording I was like "ehhh that might be a security issue" so now it's hosted off our network.

There are also a bunch of other little things but we'd be here all day.

What we're trying to do to combat big cloud AI companies

Big AI cloud companies (The ChatGPTs, Claudes, Gemini, etc) have become pretty prevalent over the years. And with a lot of information about them swirling around, I wanted to get confirmations from friends and colleagues who work in various tech and creative industries who are affected by it to get their take on it all and learn from their experience. And to ask how we can better protect our artists, our data, and ourselves where we can. While still able to maintain the business; and if we even can.

I'll be honest, we're not going to be perfect on this. We can go on a violent crusade or a placid compliance with it all. So far with all the adjustments I've made I feel as if I've leaned closer to the crusading side. But I still need to keep folks paid, so business wise if we went super aggressive we'd be shooting ourselves in the foot. We have to post on YouTube, Twitter(x), Instagram, Facebook, TikTok, and wherever we can tell folks we exist. And the vast majority of these major platforms train their AI or others off those platforms. So as much as I don't want to post my artist's work on those platforms. It's sadly best practice to include media in posts. And it's something we have to do to let the world know we exist.

So please understand that I only have so much bandwidth to fight this stuff and sacrifice so much convenience before I go mad. I do not want the human creativity of our project to be touched or in any way by the big AI cloud companies. So, to at least make an attempt, I've made sure folks we work with and what we do follow these commandments:

Thou Shalt Not Use AI in the Likeness of the Creative Mind

1. Thou shall not generate anything that touches the creative process of humans. This means do not use: Generative AI Art, Generative AI Creative Writing, Generating Mechanics, Generating Songs/Sounds and whichever AI will harm or affect the creativity of what we do here. We are a group of artists, writers, and voice actors and we want to protect that.

2. Thou shall keep the machine to the chores. Acceptable uses: captions and transcripts, spellcheck and grammar checks on words YOU wrote, audio/video cleanup, schedules, and admin as long as it's not a cloud model. The machine may scrub the floors. It does not get to create the art. (get roomba'd on, robots)

3. Thou shall author every word. Using tools to transcribe your voice and organize your notes is fine. But it must be your words. The creativity must be yours. The thoughts must be yours.

4. Thou shall not feed the machine our work. Do not put content, scripts, other people's art, or private team materials into ANY public cloud AI tool.

5. Thou shall not spread machine-touched code. AI-assisted tools and scripts stay private and on our own platforms and on our own machines. Nothing AI coding-assisted ever gets distributed or released to the public.

6. Thou shall not replace the human with the abominable intelligence. The supporters paid for people and people creativity.

Are there any exceptions?

Yes, with some major guardrails due to the climate and situation we're in that I explain further in.I only tolerate locked down offline assistant or coding models that do not communicate with the internet or share our data in any way so nothing gets trained on. Preferably models that are as ethical as we can find them. Which is nearly impossible with models and a lot of things. So far I've stuck with IBM's Granite's 4.0 model due to its high score in its transparency report by Stanford to help with the vast majority of little tasks that would take me hours to do. Such as project managing, file organization, and other general admin work. For harder coding tasks, I currently use the latest Qwen model for coding assistance for modules and internal tools that help with the workflow of everything. Such as making a little editor card program so our editor has all the timestamps (the moments in the show) for our red cards when things go wrong, and when the music changes. This has been helpful in not keeping me up during the night working on details so we can get the VODs uploaded in time. There are also other programs like Buzz that help transcribe audio using a whisper model that I run locally on my machine for our podcast.

The line I draw is that a human involved with offline tools must direct and review things. It has to stay offline and on our own machines and doesn't touch the creative work that we do and you engage with. You didn't pay for a bot to tell you about Rotgrind. You paid for us goobers.

Why use it at all, if we managed this long without it?

Before these tools, I simply put the tasks these tools would assist with to the side as I simply didn't have the time or money to offer work to folks to maintain them. So it was zero progress on those fronts without them. Which meant no transcriptions as we couldn't afford someone consistent to do that, I had to manually do the effects on stream (like the wrath&glory dice, hero points, etc), even at the start I manually inputted each donation for our early WarHams streams, to then manually fighting the twitch and youtube api to count stream contributions to the UI once we got some temp help on that. Gosh that was a struggle. So I took it upon myself to do what I could to catch up and adapt to the platform's demands we're on and to do my best to stop being in producer mode during the shows. The older pace we were at, we weren't growing much and we were close to shutting things down a couple of times. I even stopped streaming consistently on my own channels to concentrate on NarDec to keep things going.

So it's mainly due to the workload being so much due to the scale that this show has gotten. And not having the consistent revenue flow to contract people reliably in the long term to help with what I need while balancing costs for people has been a real challenge. Which has been a growing concern as I try to budget paying our artists, editors, and players what they're worth with what I have via your contributions.

Also, like I talked about in the video. I don't know who I hire contract-wise for coding tasks is using a cloud AI to just throw in a prompt and have no human input at all (I had to trash several projects due to this). We usually can spot obvious AI art (please stop sending me your AI portfolios I don't want them), most of the time music, but code? We're helpless there. So instead of contracting people who we don't know if they use big cloud AI to vibe code stuff that we don't even know is safe or good to use.

This has also plagued FoundryVTT and modules. A lot of them now use assisted AI code or are just vibe coded it with no disclaimer. With that, I'd rather have our data be ours and protected the best we can and not contribute to the problem and do my best to maintain what we have. When the show can afford to bring on a developer and other help consistently that we can trust, then hell yeah we're in business. This is also the reason why I don't share our internal modules because with the help I got with other contractors, I've no idea if it's okay to even share. And I'm sure if I just asked them I'd get the run around like in previous situations.

In addition, getting assets from paid platforms (such as stock videos, images, textures, etc) frequently get AI slop submitted to them and aren't disclosed and that's turning more and more into a real risk. Which is why we have asked Jers to come aboard to help with more background art (thanks Jers for helping with the Chaos Odyssey character art you've been killin it). Even Pinterest is swarmed with AI now, which has sucked for finding references. Usually it's super easy to tell, but other times we genuinely do not know. So we try our best to filter these the best we can as they are NOT wanted. But being human I'm bound to go "oh this texture or music or sound effect seems good let me buy that" then find out a person just gen AI'd it with no disclosure and then I'm all pissed again.

It's also been helpful for several situations when OBS just breaks terribly. It even helped me find a Linux GPU bug that showed me why multiple episodes of Kingmaker got lost so it wouldn't happen in the future and where I couldn't find any solution online.Is all this perfect? No. Most offline models were trained in ways I do not love. And the default solution would be 0 AI. But we're in a time that I don't think this stuff is going away even if the AI bubble were to burst. So I've been taking what steps I can to keep the business going, keep our folks consistently paid, and contain the blast radius as much as I reasonably can.

The goal remains the same: protect our artists, protect our writers, keep the creative work human, keep our data ours the best I can. So hopefully investments I've made to help with that bear fruit and help us keep on going on.

We are still working on all this

I spent a good while talking to working artists, writers and coders before settling on this. Talked about this with the team about it, and we're all in agreement that as long as the creativity is not touched we're golden. We're human, we are gonna mess up and bungle into things we don't fully understand. We're just doing the best we can with what the world has thrown at us.

So thank you for your time and understanding. With all of these changes I wanted to keep up with my philosophy on being transparent and honest with folks. And I wanted to say thank you for your contributions to keeping all the amazing artists we have paid so we can keep this thing alive.

-Zoran and the Narrative Declaration team