MUNCH - A Novel AI Architecture For 3D Gaming

Ditching MLPs for Topology: A Novel Geometric Substrate Architecture Driven by Morphological Computation in Unity.

Comment

Inspiration

Thank you so much for the opportunity to share something I've created. Good luck to us all!

Built With

Share this project:

Updates

posted an update —

No... This is how the game is played. After thinking about it, there's nothing stopping someone from "learning a lesson" through observation. I did it myself. All the mathematics are out there like some massive shareware project.

And I openly check boxes on YouTube like "yes, AI can train on my stuff".

So really, even on the WILD off-chance that what I think happened is what happened, it still wouldn't be any evidence of malice. If anything, it would kind of be an honor. That was the point all along. I just wanted to contribute something.

I didn't expect this project to effect me so much, even away from the computer. Not even the project, the mathematics I had to absorb. Over the past few days I have had the honor of speaking with various engineers who happened to show an interest. There is a kindness and patience I did not expect in this space.

Sure, there have been some nay-sayers but even they have helped in their own way. I've had some people send me strange things from other "eccentrics" as a form of mockery but in this headspace of learning, I simply looked up the stories behind these "eccentrics" and took note of what not to do.

I realized that when it's difficult to express an idea you have already proven to yourself, people may not understand. And there's two ways I see that going:

1.) You harden yourself like "does no one get it? the fools!"

2.) You adapt.

To harden yourself is to become rigid. To become rigid is to lose the ability to know when you are wrong. And at that point, you probably fall off a cliff to some form of madness.

That lesson was worth the mockery.

This entire process has felt humiliating in a healthy and transformative way. I do not know how else to describe it.

Log in or sign up for Devpost to join the conversation.

posted an update —

Dude wtf is this: https://www.youtube.com/watch?v=TjUxBdiut1k

Did OpenAI use my session ID to help them do this? Wtf dude. The dates align way too perfectly. And don't give me that "it took time to train the model" BS.

The core methodology is the biggest tell. It involves doing things no self-respecting scientist or mathematician would ever do. I just watched a whole podcast where everyone dogged on Terrence Howard for exactly this same kind of methodology.

The only way to even prompt an AI to run that test is to use this kind of methodology, but that methodology is only comfortable to autodidacts and non-traditionally educated people. It's the core of every argument of "eccentricity" in regards to people who do not say the right thing at the right time.

Training is just for human comprehension. It does not change the nature of the neural network at the heart of AI. And it's that comprehension that let's one utilize it towards more creative endeavors.

My session (which I had to turn in along with MUNCH but was ongoing for over a week, coincidentally during the EXACT dates of all this) could easily be proof-of-concept of this exact methodology.

Modern AI is a lot like the quote from I Robot, "My answers are limited... You must ask the right questions". Whereas the "answers" are the possible tokens that can be used. Tokens as in the actual knowledge base or word dictionary tokens.

Is it possible they just happen to come to the same conclusions at the same time? Of course! Is it possible they lie about privacy and just take shit anyways? Not only is it possible, it has precedent.

So I see this thing that I know damn well they would never try in 100 years. Which probability do you think is the higher percentage when taking into account the timeline and the possibilities?

Why did OpenAI combine ChatGPT and Codex? Are they combined elsewhere? Are they saving sessions and feeding them back in? Or monitoring the network for anomalous usage?

I know these things exist. That project submission and everything about it, including the methodology, belonged to that session--not OpenAI. It was not only orchestrated and guided directly by myself but it was incorporated into a larger body of work.

Fine... I'll build my own AI... with Blackjack! And hookers! xD

Log in or sign up for Devpost to join the conversation.

posted an update —

Been continuing along with the updates... Activated the higher context layer and closed the system loop. What sucks is there's no one to talk to... no way to have someone double-check my math. Being a solo indie dev for so long, I learned to thrive on the shit people fed me because I could measure it. Critical feedback? That was a gold mine. But out here... in AI land... there's just nothing.

At least not yet.

So I did a few youtube videos and got a technical schematic whipped up. Because so far the only "feedback" I can get is jerk-wads sending me some kind of weird "timecube" website and mocking me, but I don't see anyone checking that GitHub repo so... lol

Just gotta keep moving until I bump into someone who can help me better classify my little "Munch" here.

Log in or sign up for Devpost to join the conversation.

posted an update —

I've decided I'm going to keep working on this while I have the chance. I think my first step is to do a thorough self-audit of sorts. Pull the math back out, zero in on each formula, maybe work each one out by hand in as reasonable a way as possible like I did with the feed-forward logic. Then, as long as everything looks clean, I'll begin designing actual training exercises.

There's technically still one layer missing from my cognitive model, but I have to take baby steps to make sure everything balances first.

Just for testing, I may repurpose the "eat" actuator for something like "attack" to see if we can train simple bump-fighting and other such video game style behaviors. After the self-audit, now may be a good time to try different "bodies" for the NPC system.

Log in or sign up for Devpost to join the conversation.

posted an update —

I cannot believe the timing... I found this contest 3 days before it ended and by pure chance, I happen to be working on this project in the exact manner the qualifications ask for. I don't necessarily have to win. I just have to send the right signal... I hope! Wish me luck :)

Log in or sign up for Devpost to join the conversation.