I FINALLY GOT ALLEN TO BOOT ON BARE METAL!

This is HUGE! They told me I couldn't run AI on an Intel Core i5 iMac from 2017 because it was too old and didn't have a fancy graphics chip. I wasn't a traditional "programmer" so please spare me the explanation of why what I did isn't going to work because of this or that. The internet was created by dropouts. It's time we take a different, more optimistic and structured approach to computer science.

https://youtube.com/shorts/8ANTkBwUVjs?si=D1GbhKGOD9XLhSg6

The rise in AI has been something of a unique experience for everyone. To say the least, it came on like an afterthought of COVID. We were all consumed by the mask mandates that this ultra power being just came out of nowhere. Suddenly Apple wasn't the most valuable company anymore. Anyone else remember when it was a big deal that Apple was the first Trillion dollar company? Now I hear that Elon was the first human trillionaire riding the wave of his Space xAI hype or whatever. That all happen? Because ChatGPT started predicting the next word like our autocorrect had been doing for years? That is wild…. To think that this all came about because our autocomplete needed an upgrade.

Anyway, how did all of this get funded exactly? Where did the money come from? That is the interesting part. Let's explore together shall we?

Also, you'll have to forgive me. I am talking to you through Diction, but I am not having AI clean it up because I'm sick of this donation page flagging my thing for AI generated. We all using it, so I don't know why they're having an issue with it, but I know it looks bad, and this would make me stand out, and I think this is very important. So this is why I'm doing this with you. So if my text is a little off. It doesn't look as pretty as everybody else's. That's because I didn't use AI. So, you have to forgive me.

Now something I want to talk about is where did all the money come from? Follow the money. So if we go back in history and just a few years, I like how everybody pretends this is just like just happened like forty years ago, but like yeah, okay, computers have been growing for forty years, but we're like just in twenty twenty, it was COVID. And then we got all that COVID money, right? And then we got all those stimulus checks, and then we spent all those stimulus checks, and then all that money disappeared, and nobody talked about it again.

And then all AI started coming around, and everybody kind of just like got hyped about that. But then they're spending trillions of dollars on this stuff, and now they're building data centers in communities. And ruining ecosystems, and it's causing all these health issues, and nobody wants to talk about it. And like, I'm really excited about AI, but these mega corporations are running towards annihilation, trying to get something that may not even exist far ahead of the researchers.

And now we have OpenAI and Anthropic publishing research articles that was the responsibility of the academia. For centuries, why are we having industry conducting research? Why is that being taken away from the universities and the the actual people who know the field? We're in dangerous territories, and the reason I'm I'm wanting to do this project is we need to take back control.

Intelligence should not be held. At gunpoint, or at subscription, or at their some guys' will, and it's scary to think that they created these models on our corpus, on our information, our knowledge, and our that we gave freely, that we gave for the betterment of humanity, and they took it, and they used. God only knows how much GPU compute to make this monster, and I love it. Don't get me wrong. I really enjoy AI, and I use it every single day, all day. And I've been working on this. I've been building AIEN with AI, and the point of this project is to get it out in the open that I'm getting amazing numbers by just taking out everything.

CUDA is a hoax. It is deliberately slowing down your AI, your LLMs. The minute I took Python out and CUDA out and NVIDIA out, my model, I didn't change a thing. My model sped up a thousand times. I'll go check the GitHub on the actual numbers, but it is insane.

I've got another suggestion. Maybe we need to take and do more of a open humanity style network, where we're more distributed. We stop with these data centers because we're really hurting people, and it needs to stop. There are so many local individual consumers who have our own GPUs that would be more than willing to link it up to a voluntary compute network where we just provide our own resource. Where if our GPU is idle, we can offer it to the to the network, to humanity. And we don't have to have these big data centers because it's in our homes. It's distributed. We have open source. Our community is more than willing. We have so many people who would do that. The OpenSource community has been doing it for DECADES at this point. We all have skin in this game. We're hurting environments, and it's not going to be sustainable. And now I hear Google is about to launch data centers into space. Can we stop? Have you seen the junk up there now? These. Rich people are destroying our planet.

And like I'm not this nutcase of a global warming kind of eco nut. Let's go hug a tree. That's not me, bro. I'm a libertarian. I am just I'm just chilling and I'm trying to vibe code or trying to learn and exists in peace on a planet that I can. Be proud of, and right now I'm seeing danger because we're on a couple different fronts.

One front, they're racing towards AGI. That's going. That's already hacking into systems. That needs to stop. We can't verify anything that thing is doing. That's the reason why I'm building AIEN to be verifiable with receipts. Doesn't take a claim at its word. Nothing is. It's everything is guilty until proven innocent in my system, not the other way around.

Also, this Nvidia thing, they're just pumping trillion. They're so far into this, they can't let it go. Man, if if this collapses, Nvidia's done. So, so I'm just saying we just need to take another look at this because my 401k is screaming. I'm nervous as hell.

And nobody's asking me, or the typical human, what we want to do about it. So let's collect and do something about it.

www.drakestapleton.com

www.aienos.com

https://github.com/aien-dev/aienix

Here's my plan:

AI keeps getting more capable. Running it keeps getting more expensive, too.

The industry answer so far has been scale. Bigger GPUs. More of them. Data centers the size of small towns, and power bills to match. Lawrence Berkeley National Laboratory put out research in June 2026 projecting that data centers could eat somewhere between 9.5% and 15.3% of all U.S. electricity by 2030.

I read numbers like that and one question will not leave me alone.

How much more could we get out of the hardware we already have?

That question is the whole reason AIEN exists.

I'm building an operating system where the AI is the user of the computer, not just another app running on it. NVIDIA, OpenAI, and the rest have built remarkable things, and I use their work like everyone else. What I want to know is how much of the software underneath all of it is wasted motion. Data copied from place to place. Memory rebuilt instead of kept. Work thrown away and done again because nobody saved it.

So I started measuring. Then I started building.

What I'm building

AIEN is an open-source research and engineering project. One man, so far. The idea underneath it is simple: a system should keep what it learns, reuse what it has already computed, and stop paying full price for the same work twice.

Four pieces, built to work together:

• AIENOS. The operating system itself. It gives AI agents a place to live that persists, that they can be secure in, and that can run directly on hardware the owner controls. Your machine. Your keys.

• Omega. The computational engine. Omega worries about where data sits, how often it moves, and which implementation actually runs fastest when you measure it instead of guessing.

• Cortex. The memory. Agents forget everything between sessions and rebuild context from scratch, over and over. Cortex exists so useful information survives and can be pulled back instead of reconstructed.

• The runtimes. Inference and agent coordination: sharing context between agents, managing GPU memory honestly, running models, and letting multiple agents work without stepping on each other or duplicating each other's effort.

Different pieces, same question. How much computation, memory movement, and energy does useful work actually require, and how much of what we spend today is just habit?

The research behind it: The Turing

There is a research framework riding alongside the engineering. I call it The Turing.

The industry measures AI in tokens per second, accuracy, latency, watts. Fine. Those numbers matter and I track them too. But none of them answer the question I actually care about.

What did the system figure out?

A computer can burn a billion operations and discover nothing. It can also find one compact explanation that keeps predicting new data correctly, and that explanation is worth more than all the flailing. I want a way to measure the difference, so I proposed one.

A Turing is one bit of improvement in held-out description length, measured against a declared baseline, after you subtract what it cost to describe the explanation itself. In everyday terms: your explanation has to earn its keep. It has to compress data it has never seen, even after we charge it for its own complexity.

From there, one more step. How much of that gain can a system produce per joule of energy? I call that Turing yield, T/J.

Now the honesty, because this part matters to me. This is research. T/J is not an established measure of intelligence, nobody in the field uses it yet, and the connection I am proposing between explanatory gain and physical energy has not been verified. What I have is a testable definition and an early result. The goal is a measurement other people can run, argue with, and try to break.

Why efficiency matters

Spend any time watching AI systems work and you see the waste.

Models reprocess the same context endlessly. Five agents on the same problem keep five private copies of things they could share. Work crosses between CPU and GPU and pays a toll in scheduling and transfers every single time. Somebody computes a result, uses it once, and lets it evaporate, so the next agent computes it again from zero.

Some of that cost is real. Physics is physics. A lot of it is just how the software happened to get built.

AIEN attacks the second kind. Shared memory. Computation you can reuse. Execution that stays on the GPU instead of commuting. State that persists. Decisions about where work runs, made from measurements.

When a workload finishes using less memory, fewer operations, and less energy, and the answer is still right, the same hardware suddenly does more. Local AI gets cheaper. A researcher without a cluster gets room to experiment. Data centers need less infrastructure for the same useful work.

One caveat, stated as loud as I can state it: AIEN has not demonstrated lower data-center electricity consumption. Nobody should read this page and think otherwise. That kind of claim would need evidence I do not have yet. I think the opportunity is real and worth chasing. That is all I am claiming.

What the project has demonstrated so far

This is past the proposal stage. Things run.

AIENOS has booted its own native code on an NVIDIA DGX Spark with no Linux underneath it. Kernel work is running on a Raspberry Pi 5 and on a 2017 Intel iMac I pressed back into service, CPU bring-up and hardware discovery included. On the Pi, the native system does real CPU inference.

The GPU numbers below come from the DGX Spark. Read the limits after them before you quote any of them.

• In one experiment, 500 computational branches shared a single 32,768-token prefix in about 1.20 milliseconds. The shared KV allocation came to 704 MB. The same thing done with unshared copies was estimated at 343.75 GB.

• A Linux-hosted inference test ran TinyLlama 1.1B across 16 concurrent requests and produced roughly 553 aggregate tokens per second.

• Recent Omega work executed a 30-billion-parameter mixture-of-experts model natively on the GPU.

On the research side, The Turing has an early positive result for idealized explanatory gain on held-out search data. It needs independent qualification before it means much, and the energy part of the research is not established at all yet.

And the limits, again, because they are part of the result: the GPU tests ran hosted on Linux. They did not run on a finished native AIENOS GPU stack, because that stack is not finished. Nothing here proves AIEN beats the leading inference runtimes under matched conditions. I would rather print that sentence myself than have a reviewer find it for me.

Who's behind AIEN

My name is Michael Drake Stapleton. I am an Aerospace Engineer but I have taken a wild interest in machine learning and artificial intelligence in a way that is hard to explain.

I work at 3M in Springfield, Missouri. AIEN is what I do after work, and before work, and on weekends. My degrees are in chemistry and biology, and I am a first-generation college student, so computer science is a field I walked into sideways because the questions would not leave me alone.

A lot of the code gets written in collaboration with AI coding systems. That surprises nobody anymore, but I will tell you exactly what my part is: I define the problems, I set the architecture, I direct the development, I evaluate what comes back, and I test whether the systems do what they claim to do. The receipts are public. Where the work falls short, the receipts show that too.

That way of building has real limitations, especially in low-level systems work and security, where a confident wrong answer can look exactly like a right one. Code written this way needs independent scrutiny. Getting that scrutiny is a large part of what this funding is for.

What the funding would support

Everything about this project is built to be checked by strangers. The money works the same way. Each phase buys harder evidence, and each phase has to earn the next one.

The initial ask is $5,000, with the full roadmap published behind it.

Phase 1: $5,000. Make it inspectable.
Half of it, about $2,500, pays for a focused independent security review of the boot chain and selected isolation mechanisms, done by somebody who owes me nothing. The other $2,500 goes to reproducible builds and benchmarking infrastructure, so anyone can rebuild the system from source and rerun my numbers on their own machine.

Phase 2: $50,000. Make the claims testable.
Benchmarks against the leading inference runtimes under matched conditions, run independently. Precision energy measurement hardware, because T/J stays an argument until somebody measures the joules properly. A second hardware bench, so no result depends on one machine in one room in Missouri. And the release engineering to produce signed, verifiable installs that a stranger can check without taking my word for anything.

Phase 3: $250,000. Make it a product someone can own.
Development time bought back from nights and weekends. A full external security audit, not a focused one. Then the first pilot deployments: real organizations running a capable AI system on hardware they own, under keys they control, with no meter running, and honest published reports about what worked and what broke.

The roadmap totals $305,000. Every phase ends with a public report. What the money bought. What the measurements showed. What is still unfinished. If a phase fails to produce its evidence, the next phase does not deserve funding, and the report will say that in those words.

The code, the testing tools, the evidence, and the research stay public at every phase. That part does not change with the size of the check.

Why I'm doing this

Efficiency is treated like a footnote in AI right now. Scale gets the headlines; waste gets a shrug. I think that is backwards, and I think the questions underneath it are the interesting ones.

Can hundreds of agents share work instead of duplicating it? Can a discovery be kept and reused instead of found, lost, and found again? Can we measure what a computation actually figured out, right next to what it cost? And can an ordinary person own a capable AI system outright, running on their own hardware, answering to them?

Those questions are what connect AIENOS, Omega, Cortex, and The Turing. They are also, not coincidentally, the questions nobody with a meter business model has much reason to ask.

Some of my hypotheses will fail. I know that going in. Building in the open means other people get to watch the failures happen, examine the results, challenge the assumptions, and take the pieces that work. That is a feature. That is the whole point.

My goal is to make useful machine intelligence less expensive to compute, less wasteful to operate, and more accessible to everyone.

Intelligence should be owned, not rented. That is why I'm building AIEN.