How I Made PROGI the Redstone Machine that LEARNS!!! (kinda)

Discord: / discord
NumBot: • How I Made NumBot the ...

[@rodricrack1072]

PROGI is too good, but he doesn't learn, he memorizes.

[@floorbeardthepirate1141]

What is learning?

[@rodricrack1072]

@@floorbeardthepirate1141 When you learn smth, you get skills to do that activity in the cases you were trained and in new cases where you can extend it. For example: if you learn how to divide numbers, they give you some examples, you try to guess which is the method and in the exam with other cases they evaluate you if your method was correct until you can generalize it. That is how neural networks work for example.

[@vizzy1702]

@@rodricrack1072 At the end of the day...that's called memorization. Learning is just "effectively remembering"

[@TheBrainReal]

@@vizzy1702 you can remember that 1 + 1 = 2, but that doesn't mean you have learned how to add numbers together

[@QUBIQUBED]

@@vizzy1702 not really, to me learning means that you remmeber something you learned and figure out how to use that skill in any situation.

[@DiceRobo]

This isnt a AI, this is a FPGA! Its really impressive that's even possible in Minecraft

[@grithon5499]

ah yes, a combination lock that anyone can walk up to and change the combination. sekuity

[@iangustafson609]

Idk if you’re joking or not but… it’s not a combination lock… It’s a programmable logic gate… big difference

[@jeffreybaresic1452]

@@iangustafson609 no one asked

[@fralegend0152]

@@jeffreybaresic1452 I did

[@RoboticGames-ll4ul]

well cant you just remove the button?

[@geraldkenneth119]

I actually made something functionally identical to your neuron, which I named the pseudo-perceptron, a year or two ago. Although mine was way bulkier and It didn’t go anywhere because I didn’t realize what you could do with it. Now that I’ve seen this video I definitely plan to re-investigate, thank you

[@CraftyMasterman]

Glad I could spark your interest :D

[@secretgmg3916]

Jup built something like that too like a year ago, really cool. Mine is a bit smaller though ^ ^

[@J2ko]

14:52 my brain suddenly forgetting what I was going to get from another room

[@infinitygamer4991]

Lol

[@aldrnarienby4756]

You made a LUT (Look-Up Table), aka a programmable gate, and then put a bunch of them together to make an EEPROM that loads one bit of data in each memory address and the same data in the inverse address

[@enderbrother5774]

Nerd wait why do I understand nooooo

[@GoggledGecko]

No one : This guy : Yeah one redstone ai wasn't enough.... Let's make another, more complex one!!! You have some mad redstone skills dude

[@cwmd7651]

I have no words... I can barely make a 2x2 piston door and yet there are people doing crazy things like this. I really wish I could do things like this, but regardless, keep it up! This is so incredible!

[@morbid1134]

Have no worries. I can make computational redstone like this and am unable to make a 3x3. -_-

[@okboing]

I designed a somewhat programmable logic gate myself, it had 2 inputs A and B, and a 4 bit input C. The bits in c would be placed in a blank truth table, and A and B would be plugged into a gate with that truth table. the output of the whole operation is whatever the gate outputs.

[@strangebird5974]

I really like your videos, you are really talented with redstone and have some good ideas as to what use it can be put to. But I think I'll have to agree with the people saying that you are not using 'machine learning' in the same sense that it is used in that field. Also, your term 'neuron' kind of has a meaning already, in the brain and in neural networks, which I don't think your otherwise amazing module fits. But please don't be discouraged! I think you are doing amazing stuff and hope you continue to do so. But if you are looking for inspiration you could read up on neural networks and other stuff like that? I think a neural network should be possible in minecraft. (Think something like modules that connect to other similar modules with either variable signal strength or in varying numbers that then determine whether or not those modules will themselves fire. And the gates for firing or not within each module should be changable and in the optimal case selfchanging based on some response to the output. You could even take inspiration from real neurons and give each module a downtime-clock so that their own signals wouldn't immediately propagate backwards or something. Someone with a degree in computer science or similar, correct me please.) But anyway, as a lot of people are saying, your channel is criminally underrated. And I keep your video on small piston extenders in a tab often, because it is so handy. I am a complete redstone noob, but slowly learning. ...but... maybe think about pursuing a career in computer science or electrical engineering or something instead of a career as a youtuber, maybe? Not wanting to sound like your parents. Just a suggestion :p (Maybe you are already doing that. I'll shut up.)

[@eliasalbo1347]

Like i have 2 points, maybe 3. Ok this is amazing, I would have never thought this would be possible in minecraft. You sir are amazing i am surprised that you don't have more views/ likes on this i am subscribing right now. Also i kind of get what you are doing but are you programing it to recognise it like if you offset it like numbot or do you have to manually put in each combination and its logic circuits will find if its in its memory or not. Asking that made me kinda just realize that it's just making memory but whatever. This is truly amazing, great job

[@_Funtime60]

The PLG is actually a 2-bit EPROM which is identical in function IRL anyways.

[@TheNerdBird_]

It's like sethbling had a funny son :P You're amazing dude! If my PC wasn't broken, Id totally hop onto the server and check out your redstone.

[@wshockmc2445]

Wow great contraption, Lock that everyone can pick

[@Taumignon]

That's actually super useful for a lock

[@LanguageBLOX1_Alt]

it basically just memorizes not learns because you cant put in another similar shape and it still recognise it without pushing the button, still good work!

[@smolboye1878]

Small correction, the number of distinct input/output tuples of a logic circuit is 2^n, not n^2

[@Versuffe]

I knew something was wrong when he said that

[@rgbtryhardled636]

so why is it 36^2 and not 2^36 since there are 2^36 different images you can make.

[@orbismworldbuilding8428]

Now to make a computer with these, and then a self learning one

[@Polar_Onyx]

Progi when creeper: I rember 😁 Progi when reset: I fogor 💀

[@xandersfs2346]

I made a machine that tests inputs and records outputs to determine the type of logic gate it is attached to and displays it on a board

[@Eliasdbr]

My man just invented the redstone fpga

[@bwiebertram]

That's one way to make simple thing complicated

[@thegoheads]

Wow, impressive. Are progi and numbot the first of their kind? Or had other people done similar concepts already? If this is all your original idea & design, then you must be one of the world's top redstoners for sure.

[@CraftyMasterman]

probably not an original idea, but I made them without any previously used cicuits

[@deltamico]

yay i can set the combination of the lock i came to rob!

[@charlestonianbuilder344]

Your like earning 1k Subs every few days!

[@beachsam5713]

You're a calm guy 👍

[@rcpg248]

Hes canadian

[@debblez]

16:06 Actually 2^36, which is over 68 billion

[@hardiehero8987]

That’s what I was thinking too! It’s insane. 68 BILLION COMBONATIONS

[@deltamico]

yeah, but i'm sure if PROGI remembers two overlaping images, it evaluates the overlap (with optional pixels on positions of at least one of the images) as a stored image as well

[@vibaj16]

No, I'm pretty sure it's 36^2. 2^36 would be the number off different inputs you could give it, but 36^2 is the number of different inputs that it can remember

[@debblez]

@@vibaj16 you can make it remember whatever input you want... also there is no storage in minecraft that is base 36

[@vibaj16]

@@debblez That's not what I meant. I am sure it can't remember 68 billion different inputs. That would need way more space to store all that

[@averagecommenter4623]

PROGI 2.0 should display what it remembered.

[@sir_slimestone3797]

I call bs, this isn't machine learning, it's storing variables and with your combination lock your just hooking up your little variable modules up to a giant AND gate. There is no machine learning going on here, btw for those of you who would argue that the machine can change combos and that's learning, WRONG, it's changing the value in a variable, still not learning. Computers have been able to change the values in a variable for as long as they've existed, machine learning is not recognizing when an input and outputting a variable. It's just an if statement with a print function as the code block if you boil it down. PROGI is the exact same thing, just on a bigger scale, PROGI is just a system that checks if it has an input in it's stored list which is just memory, if the input is in it's memory out goes an output. There is no learning here you're just storing a number of variables and checking if an input matches any of them then outputting an output that is stored with the variable, this isn't learning. Anyone with any sense of computer science would know the difference. Stop with your click bait, it'd be cool if you called it a working input memory comparison system but you lied and called it machine learning, and all for what, views, clickbait isn't cool.

[@circuit10]

Well all machine learning is memory... maybe... I don't know

[@thatsnosquirrel7425]

Can I have a TL;DR please?

[@CraftyMasterman]

@@thatsnosquirrel7425 I mean they just said the same thing like a million times :p

[@J2ko]

ThatsNoSquirrel TL;DR: machine learning good, learning machine bad

[@ashikra]

Lmao

[@henryzhang3961]

Seems more like an image database to me... unless it can recognise a creeper face translated a few blocks over? still far better than I could do

[@CraftyMasterman]

It technically can do that :p but yeah there isn't rly any learning bit oof

[@MrEdrum]

I'm wondering. If you'd push the button when all inputs are off and again when all inputs are on, wouldn't it recognize every image you put in as correct? Based of the size of that build it looks like every pixel decides on it's own if that pixel's input is correct. so if every pixel is told that both it's on and off state is correct, then the and gate will always be true. To be able to actually have it recognize all combinations, you'd need either a huge amount of those cells chained together (so every pixel can check against every other pixel unrealistically huge) or have one of those programmable gates with 36 inputs and thus 36^2 memory cells (also unrealistic) If you used your two input design to cross check two pixels every time, you would need at least 4 patterns to tell every gate that every combination is correct which makes it a bit more reliable, but of course you need a tradeoff between size & speed, and functionality. The more inputs your gates have, the less false positive images will exist, but the slower it will be. For people who don't get what I am saying: Imagine this contraption with only 4 pixels so 2 of his programmable gates. If the image is 0 0, 0 0 in other words both rows off. both gates learn 0 0 is correct. Doing the same with 1 1, 1 1 will give both gates the truthtable of an XNOR (Both inputs are the same -> output is on) if you now input 0 0, 1 1 the first gate will see 0 0 and say thats correct, the second gate will see 1 1 and say thats correct. This leads to the final output being correct even though you never said that this image would be right

[@thomb.9013]

the possible combinations isn't inputs^2, it's 2^inputs.

[@upsilan_mitstrima]

I have a theoretical idea on how you can make a contraption that learns, so say you have a tic tac toe game, you make the machine has **every** Option in the game so it would be i think 9! Or 9^9 or something. And then each time it loses it destroys the option that made it lose, it might be complicated but i really think its a cool idea its like a real learning computer

[@vertexia9982]

This is a real thing, I believe you had the vsauce video pop into your head

[@upsilan_mitstrima]

@@vertexia9982 yes maybe, i saw that video, but you can make it in minecraft for sure you just need the piston randomizer and then you can remove options from it

[@Hyrum_Graff]

I was thinking about making a minecraft MENACE; the problem is in making a variable-weights randomizer.

[@thatonefoxxy]

you could also refunction it as an image storage Hard Drive, because it stores images.

[@thatonefoxxy]

because he showed that he can hold 16x16 images in his storage and match it, means if we could refunction it we could store images

[@kratix0469]

The number of options actually is 2 to the power of 36 and not 36 squared

[@vivi3_076]

Because PROGI only compare two by two inputs, when you input two or more images, there is a lot of others images that work, so that kinda doesn’t work neither as a lock or as a AI or whatever it is really named

[@H3artFang]

But what about: if I press the button when light is off telling that output is wrong?

[@7lain]

Well... Thanks! I like new brain. Now i know how to press like button.

[@_Funtime60]

isn't it 2^36? since it's base^bits?

[@ebrahimmuneer7238]

I wish Bedrock had block spitting. It would make my version numbot easier to make!

[@Benthehuman]

This vid looks like it's going to go viral.

[@timiempire2279]

Is the programmable logic gate doable in bedrock edition? Just checking before I spend an hour trying to do it only to find out it doesn't work.

[@CraftyMasterman]

Making one is possible but my design doesn't work on bedrock

[@timiempire2279]

@@CraftyMasterman Thank you!!

[@matthewfarquhar6962]

I understand it but I’m probably going to need to design my own because I play bedrock and sticky pistons can’t release a block

[@matthewfarquhar6962]

I forgot there was more of the video

[@joeblobberson8104]

what if you made something where you could give a name, draw an image, and if you put that same image in, it displays the name. or make an actual neural network that can figure out what something is, even if it's never seen it before

[@deltamico]

Is there a world download? I want to test my theory about overlaping images.

[@sohailaji4660]

Its on the synergy server.

[@deltamico]

figured by now...

[@bengineer8]

Number of states = 2^inputs. Make an FPGA.

[@bengineer8]

Also, it is not really thinking a all.

[@CraftyMasterman]

@@bengineer8 not thinking, learning :)

[@bengineer8]

@@CraftyMasterman More like memorizing. True learning would be if, upon seeing a creeper face, could detect it even with an offset it never saw before.

[@leonbierich9666]

Do you have a Server?

[@themovercell2318]

How do you go on the server?

[@Silverfi5h]

Make a flying machine that can dodge stuff

[@simi_says]

People, subscribe to this dude!

[@noobpro5674]

16:07 Why is it 36 squared? I thought it was 2 to the 36th?

[@comet1072]

Maybe I'm mistaken, but can't you start to 'mix' creeper faces. Lets say you've inputted two creeper faces. Each cell just keeps track of two inputs and when it sees the two inputs that it recognized from one of the faces, it would turn on. When you give one cell two correct inputs from the first creeper face, and the neighboring cell two correct inputs from the second creeper face, they would still both turn on. At least I don't see how the different cells communicate whether or not their input actually is from the same face. I might be completely wrong though. Would it be possible to provide us with a world download? Edit: as others pointed out, this is not at all like actual machine learning. I mean, I totally understand using that term, but I'd personally be more pleased if you weren't going for the more 'clickbaity phrasing'.

[@CraftyMasterman]

I don't think mixing will work it will not output the smiley. If you remember the really basic setup with only 2 inputs, you can make an XOR or an AND gate. The cells will just see mixed creeper faces as a different combo

[@CraftyMasterman]

It's on a server called synergy I said the ip in the video at some point

[@comet1072]

@@CraftyMasterman I'm not talking about mixing within a cell, I sent you a message on discord further specifying what I mean

[@ShubhamSingh-pv2yn]

Better than mumbo jumbo can do...

[@vihaanthebot429]

Is your server on bedrock?

[@hezeetaylor7471]

How do you spell the name of the server??😊

[@srseergi5271]

Yellow buttons be like: -Alzheimer time

[@Borobmw]

It's 2^36 combinations, not 36^2.

[@superbt2027]

How can it memorize 1296 things with only 6 memory cells? Is there something I'm missing?

[@heylloe9382]

I dare you to make this on bedrock. Ill give you sub

[@spencerfunk6697]

real gpt in minecraft seems feasible

[@Luckstar__]

I don´t understand how this first button thing is built. I guess I have to experiment a lot to make it work...

[@Luckstar__]

a few minutes later and I got it

[@eneaganh6319]

Isn't a nand gate a gate that outputs when it is not an and Is it !A&!B or !(A&B)

[@Judeity]

sever ip?

[@WildEngineering]

wanna collab on some builds?

[@CraftyMasterman]

oo checked out your channel, you sir have some pretty neato stuff

[@WildEngineering]

@@CraftyMasterman thanks man, what server do you play on?

[@CraftyMasterman]

@@WildEngineering synergy

[@L_S_Barros]

i dont have original minecraft tutorial pls

[@KightCheesAcademy]

Now at 50 likes

[@x35gaming]

this is really just a redstone FPGA

[@sebbog]

45 likes is alot? thats strange

[@-my-ssma9192]

so complex

[@Kraxel-North]

When the minecraft AI is too dumb so you ake your own

[@Benthehuman]

Could the "neuron" also work as a xor gate?

[@CraftyMasterman]

Yep whatever gate you please

[@Benthehuman]

Wow that’s insane

[@vibaj16]

@@Benthehuman not really, this is just memorizing. The version you might be thinking of (which can't be an XOR gete) is much better for actual machine learning

[@upstudio9708]

But craftymastermab

[@muhammadtahaali614]

No I don’t need another sethbling, actually i do you are Crazy. Better than mumbo

[@muhammadtahaali614]

You need more subs dude

[@CraftyMasterman]

@@muhammadtahaali614 thx :D

[@mangoalias608]

ok sure it doesn't technically "learn" but still this is a REDSTONE MACHINE that can REMEMBER 1000+ IMAGES why are you complaining????????

[@fine93]

pff cant give more likes...

[@anugrah1921]

Talking about likes😭

[@LionelChacon]

First!!!!!!!!!!!!!!!!!!!!!!!