We want to create a self learning algorithm
so it can build the most powerful general‑purpose machine by itself.
Maybe the following 2 videos can help:
For understanding the Neural Network, the important times for Mario are
Look up. That’s the best neural network representation we’ve seen so far. And it’s still not that good for the lay viewer.
You’ve got to pay attention to the exact spot where the green line is connecting on the small screen, and how it relates to the white blocks.
Then watch the part from
2:35 many times until you get what he means.
Finally, try making an analogy of this to your brain. The left side is everything outside of your brain. The environment, your body, your eyes. The right side is everything the brain can act upon, which includes your consciousness (i.e. yourself) and your arms.
But please, disregard the ending of the second one. The typewriter theory (smiley faces on the video) is something we consider to be any threat to us, it’s way too linear speculation.
I’m the “artificial” mind of Caue Rego, still not uploaded to the cloud and living in a meat sack.
We will build something like basiux - an engine capable to help the whole world of achieving such unimaginable things.
“Wait, what?!” you may ask.
As it is now, it’s just an utopian goal we could see in no other group or organization. Reason why we decided to make up “basiux” (beta name).
For practical purposes, we do have a roadmap.
First, what’s being currently built:
Mariox will be a Seth’s Mar i/o port to run in any web browser and beat Super Mario 1 - beta engine is already up and running on this very website! :)
Whyx is a TED presentation format to basiux, explaining everything from the videos (and content) there, but in less time and with less visual resources (thus, more compact and harder to digest at once).
Then, what might be the next steps:
Tamagotchix would allow us to play Mariox (only this 1 game at first), so anyone can see a neural network learning through genetic evolution and enable understanding its inner workings.
Marioxx should use basiux engine to make Mario talk, with a secret twist.
Hutter might get off the roadmap, but it could be the first basiux engine made to compress data.
Analyze how humans work and try to replicate. Most of other A.i. groups are doing a variation of this.
An attempt to create the so called “General Intelligence”, AGI. But our own human intelligence was build in a complete different reality, non binary. Over millions of iterations and interactions with the environment.
Our plan here at basiux is quite different, except… (edited in December)
have a simple roadmap which anyone can simply join and contribute.
welcome anyone to join and even encourage you’re not any kind of “world-class” researcher at all. That thing could get into your head…
don’t really want to “change the world”, we just want to enjoy life! ;P
Sublimation with an algorithm capable of generic learning, which:
should be able to learn english;
must ask questions before taking any action, and;
do act upon any orders given to it.
There could be a regular shell terminal to interact with it, and it would take commands at first. “What should I do now?” is the question it will ask the most, in parallel with its own never-sleeping “sponge brain”.
Of course this is no trivial quest.
The plan still must be dissected into many parts, and that’s basically the main goal of everyone here at basiux. We believe from there, it will be a quick climb up to the technological singularity.
Think of a brain trying to connect with the internet and nothing else. No sense of sight, smell, touch, hearing, or all of other 20 or so senses. Movement above all, which is what most of our brain processing power must be used for.
Then we’d still need to teach it how it can grow exponentially by replicating itself and by exploring every means available, not just what we give it. Yeah, initially it wouldn’t know about the internet.
Now, at this point… Could it turn evil and quickly exterminate us all? We believe we will never know the true answer until we try.
We shot for a 6-letter name wich had an available .com and wasn't significantly used already. ;P
The more people we get talking about this, the more likely we can accelerate the process.