Love the fact, that is wasn't just "I asked ChatGPT". Way better video then expected (if I read AI in the title the video is usually crap, but not in this case)
This is better but I would also be interested in a video about getting a *useful* answer out of ChatGPT on a problem like this, because usually what it gives is nonsense for technical stuff.
This is probably closer to real AI than chatbots or what people call AI nowadays. This is just machine learning, otherwise known as numerical optimisation.
Funnily enough, you stumbled on quite some AI techniques even though the first part is stats. Your optimization technique at the end resembles gradient descent, but you ended up searching linearly and refining based on the sign of the gradient instead of steering with the value of the gradient. Additionally, if you did not already do this, a symmetric kill chamber allows you to lump TTK values of spawning spaces. In this case, you could use 4-axis symmetry to cut your farm in 8, replicating a block removal in the 7 equivalent positions. Even with your initial off-centered kill chamber with one entrance, you could already cut the farm in half code-wise.
I wondered about this exact problem multiple times and ran some rough tick warp tests but never bothered to code anything or study this in this capacity. Really good video, very easy to understand and the "Welcome to UA-cam" was perfect.
I think this is actually AI. Most people just think of neural networks, chat bots and such when hearing AI, but this os definitely a kind of machine learning. You had a look at your data and figured out what kind of model would best suit this. But AI doesn’t even need to be machine learning. I don’t think the title is misleading, people just forgot what the term actually means. (Someone from my local university actually told me once they have people being very surprised when they need to actually program a regular non-machine learning AI like pathfinding or similar.
tbf i wouldnt classify most pathfinding as ai as there isnt iterative improvement. for something like the tsp where you need to optimise for a 'good enough' solution, definitely ai if youre being fancy. but if its just get from a to b, id say most algos just do what you tell em im curious as to where youd draw a line between iterative optimisation and ai , im sure its a pointless semantic question (bc it really doesnt matter as long as it works!), but it seems to be an increasingly relevant one
The google definition of AI is pretty much just a computer doing something a human normally would. Of course not the most reliable source but still, I would tend to agree, this is AI, as is just about any other algorithm.
as mentioned in another comment his optimization kind of resembles gradient descent to find a local minimum (or in this case maximum) which basically all machine learning algorithms do as well. so this actually is a rudimentary machine learning algorithm imo.
@@user-bj7yu2ju3e Machine Learning basically just requires iterative improvement according to some kind of evaluation algorithm. We have such an algorithm and it is used to iteratively approach an optimum. Don’t see how this even could be not a kind of machine learning. AI isn’t really about just feeding stuff into a model, it’s basically data science and statistics to figure out how to normalise your data, identify and remove and how you could best process all that data in the end.
As much as I was surprised to hear about the usage of AI, I was more shocked to see you using actual statistics and analysis to find the best solution. Great job! 👍🏻
I really appreciate the time you take to explain such concepts in relatively simple terms so the average player can learn from your expertise and testing! I haven't developed the skills to _do_ technical work, but I'm good enough to appreciate it when I see it, and I'm looking forward to getting this schematic to play with!
it would be very interesting to see traditional gradient optimizing algorithms facing this, if you tune the hyperparameters right, you can probably consistently converge in under 10 steps!
Niko you are the goat!!! I learn in your videos to program, understand how math works, how u can do things in minecraft more efficient and you also show how youtube works. Congrats, thats how you get in rank 1 in my youtube.
been watching your videos, and what a coincidence has it been today! I just finished learning gradient descent and linear regression, and this video pops up in my recommendations. 11:26 what you did here is not exactly gradient descent but its very very similar. Excellent video man! This is what i hope to do one day, and this is one of the main reasons why I am learning ML or even Python lol.
I actually really love this approach! You could have made a symetrical measurement of ttk by mapping the data on each eighth slice of the platform to just one eight. That way you also have way more data per block and you get a symmetrical result. And because this slashes the total of variables to one eighth this also makes the optimal search way faster. It would also be interesting to know whether the shape of the platform affects the ttk in other ways because of the gang up mechanics of the piglin.
i dont know about the video, but for some reason, something in your patreons triggered my brain. i was looking for a word the whole effining week and my brain ultra-hard blocked even the slightest hint at getting close to it. so... thanks i guess! on the topic of gold farms, i would like to see a "harvester" design with cycling flying machines clearing the platforms faster (at least no stuck pigmen and in theory faster speeds than walking, while sacrificing a "row" of spawning spaces. even though it might require more killing setup, i think it would be an interesting take on the standard aggro farms.
Insane effort on the video; expected this to be some AI slop and came out pleasantly surprised! Love the math visualizations as well; saddening that this doesn't have more views...
I'm glad you didn't actually use a LLM to design a farm for shits and giggles and clickbait. What you did is actual data science and engineering its far more valuable. MATH FTW. Great job :D A tangent, this type of practical usages should be used in school to teach math/engineering. I remember my uni days when nobody in class understood what fourier transform actually is and why its needed and "good" students just memorized equations verbatim to gain cgpa. Years later I slowly understood the relevance and brilliance of FT by learning from internet/side projects how groundbreaking tool it is.
this is 1000% better than AI, great vid !!! can't even fathom the amount of time and head banging you put into this, but the result is awesome Edit : people that watched the video after the title change : "why everyone is talking about ai in the comment? 🧐"
1:05 actually it count to spigot, bukkit or any paper and purpur. but doesn't count to ordinary vanilla or servers with solo portal farms plugin installed
I’m curious whether this is dependent on the seed or the actual coordinates of each block. And as some ppl have already said, this is definitely a form of machine learning. The simplest form of machine learning is a simple regression, so derivative-based optimization does fit the bill
Intro is omission by huma-huma, and outro is ebb and flow by pandaraps. The rest of the music is custom commissioned music I bought for the channel ;-)
I was just wondering if having a drop between every few blocks you go wider is worth it to have some type of fast slime block launcher to get them to the player quicker. Or on the walk to the player?? And as always thanks for the great vid.
I do have one very important question about the farm as im trying to recreate it for my realm, what biome does it have to be in? Nether wastes? I have a area I would like to put it in, but it doesn't line up to where all parts of the farm are in the nether wastes. also on the topic of optimization, wouldn't it make sense to speed up the TTK by having a kill chamber at every level? I'm assuming you use entity cramming via the boats so that the spawn rate is faster since pigmen sitting in boats don't contribute to the entity count, so with that in mind, couldn't you lurer the pigmen with turtle eggs and by aggro but having killing chambers on every level instead of the floor of the farm? and if there are killing chambers on every level, that means every platform on every level has the same shape instead of being tapered off towards the top?
I wonder: Now that zombified piglins spawn reinforcements of their own kind instead of regular zombies, would allowing those reinforcements to spawn in the glassed area increase the rates in a meaningful way? Those reinforcements would likely have a shorter TTK, after all. (I guess you'd have to lower the floor in those lines where the boats are and make them magma blocks.)
ChatGPT would require so many specifics that you may as well just design it yourself. If I ask it for gold farm ideas it’ll probably just give me a generic portal-ticking design,
The only 100% optimized farm is spawning an infinite amount of zombified pigmen that die in the first tick and the items to be transported into an infinite amount of double chests in the second tick and for every gold sword to be smelted instantly and made into nuggets, then into golden ingots also instantly.
i very curious on where you were going with the whole "ai" in the tittle until you started showing graphs like you usually do and "oh yeah, if water bottles can have ""AI"" then this is very much an ai too"
Couldn't you make several boat looting chambers closer to the spawning platform (for example one in each cardinal direction and one in each diagonal) to reduce ttk even more? Having the piglins traverse all the way to the middle seems inefficient.
OK so funny story -- I uh did this and it didn't work so super well. A single trapdoor would have pigment sometimes getting stuck on the edge, and double trap doors made some weird thing happen -- they pigmen would fall then jump around for like 5 seconds, I think due to pathfinding reseting. I actually originally included this in the video but cut it, but think the solution presented is the best.
tbf, while this isn't AI, in the sense it's usually thought about, it leverages much of the same principles AI does: lots of statistics and gradient descent in an abstract space
A good example of how AI should be used… as a tool to assist where gaps of knowledge exist or long repetitive tasks so us humans don’t have to, the rest is still down to us
I love how this is not about AI. I prefer it like that
agreed
Nobody tell him how AI (specifically machine learning) works
you have ayche eye vee
@@uropig you have level 3 gyatt btw my skibidi rizz cannot stand being ohiomesticated anymore
@@ironfoot1938it’s funny how people who hate AI know the least about it. Though the people who love AI aren’t too far behind them…
Love the fact, that is wasn't just "I asked ChatGPT". Way better video then expected (if I read AI in the title the video is usually crap, but not in this case)
ong those videos are so lazy this is wayyyyy better
So true
This is better but I would also be interested in a video about getting a *useful* answer out of ChatGPT on a problem like this, because usually what it gives is nonsense for technical stuff.
"But Neco, that wasn't AI."
"You're right, welcome to UA-cam."
I've just been clickbated and I'm not even angry.
This is probably closer to real AI than chatbots or what people call AI nowadays. This is just machine learning, otherwise known as numerical optimisation.
Funnily enough, you stumbled on quite some AI techniques even though the first part is stats. Your optimization technique at the end resembles gradient descent, but you ended up searching linearly and refining based on the sign of the gradient instead of steering with the value of the gradient. Additionally, if you did not already do this, a symmetric kill chamber allows you to lump TTK values of spawning spaces. In this case, you could use 4-axis symmetry to cut your farm in 8, replicating a block removal in the 7 equivalent positions. Even with your initial off-centered kill chamber with one entrance, you could already cut the farm in half code-wise.
Minecrafts pathfinding sometimes prefers directions, i suspect you could lose a percent of efficiency for some mobs
I wondered about this exact problem multiple times and ran some rough tick warp tests but never bothered to code anything or study this in this capacity. Really good video, very easy to understand and the "Welcome to UA-cam" was perfect.
seeing the uplift after optimizing gnembon's original farm will be a good test for the farm design methodology too
Yeah it's been stuck on my mind for a while, was fun to play with!
After 8 million years he returned
I think this is actually AI. Most people just think of neural networks, chat bots and such when hearing AI, but this os definitely a kind of machine learning. You had a look at your data and figured out what kind of model would best suit this. But AI doesn’t even need to be machine learning. I don’t think the title is misleading, people just forgot what the term actually means.
(Someone from my local university actually told me once they have people being very surprised when they need to actually program a regular non-machine learning AI like pathfinding or similar.
tbf i wouldnt classify most pathfinding as ai as there isnt iterative improvement. for something like the tsp where you need to optimise for a 'good enough' solution, definitely ai if youre being fancy. but if its just get from a to b, id say most algos just do what you tell em
im curious as to where youd draw a line between iterative optimisation and ai , im sure its a pointless semantic question (bc it really doesnt matter as long as it works!), but it seems to be an increasingly relevant one
The google definition of AI is pretty much just a computer doing something a human normally would. Of course not the most reliable source but still, I would tend to agree, this is AI, as is just about any other algorithm.
@
Thing is, you don’t need iterative improvement for AI. AI is a pretty broad term.
as mentioned in another comment his optimization kind of resembles gradient descent to find a local minimum (or in this case maximum) which basically all machine learning algorithms do as well. so this actually is a rudimentary machine learning algorithm imo.
@@user-bj7yu2ju3e
Machine Learning basically just requires iterative improvement according to some kind of evaluation algorithm. We have such an algorithm and it is used to iteratively approach an optimum.
Don’t see how this even could be not a kind of machine learning.
AI isn’t really about just feeding stuff into a model, it’s basically data science and statistics to figure out how to normalise your data, identify and remove and how you could best process all that data in the end.
Forgot mid vid it was supposed to be about AI lol, great vid!
Same hahaha
9 months and nico is BACK with another banger of a video, great job!
Nicole isn't lost. This is his home n we r his family. Love ya bro.
This is very well done! I’m very happy that you used algorithms and not “just use AI bro”
Banger, the amount of detail in every farm from you, i dont want to build farms anymore without a infepth explanation from you 😂
i love the idea of this being applicable to other farms. thanks for sharing
As much as I was surprised to hear about the usage of AI, I was more shocked to see you using actual statistics and analysis to find the best solution. Great job! 👍🏻
i love that this video teaches concepts of calculus with a practical use case
I really appreciate the time you take to explain such concepts in relatively simple terms so the average player can learn from your expertise and testing! I haven't developed the skills to _do_ technical work, but I'm good enough to appreciate it when I see it, and I'm looking forward to getting this schematic to play with!
12:22 Learned about this method (and its drawbacks) in class. Interesting usage!
2:31 what a swede beacon
wow this is really cool. been putting off making a gold farm in my world, but this is definitely the one
Love this. I actually understood everything as an engineering grad and am glad I went to uni.
Amazing content! Thank you for all the effort and beautiful Manim animations! :)
YEESSS FINALLY NICO UPLOADED
Very intersting and well explained video, so inspiring. Keep up the good work👍👍
it would be very interesting to see traditional gradient optimizing algorithms facing this, if you tune the hyperparameters right, you can probably consistently converge in under 10 steps!
Love your videos NicoIsLost, will surely make this farm in my server!❤
Really cool work !
After 32,387,198 seconds... he has returned with a banger.
i'm so Happy you are still alive
yooo he’s back
Niko you are the goat!!! I learn in your videos to program, understand how math works, how u can do things in minecraft more efficient and you also show how youtube works. Congrats, thats how you get in rank 1 in my youtube.
It might just end up becoming the perfect one !
I loved it!
I knew when you said the farm was functional that it wasn’t ai and your actual approach was really interesting
Well well well look who’s alive
nice to see you again, your videos are really interesting
been watching your videos, and what a coincidence has it been today! I just finished learning gradient descent and linear regression, and this video pops up in my recommendations. 11:26 what you did here is not exactly gradient descent but its very very similar. Excellent video man! This is what i hope to do one day, and this is one of the main reasons why I am learning ML or even Python lol.
Fantastic video!! 🔥
Good to see you again Nico :))
Thanks! :-)
dope farm niko, love your ideas
HES BACK
I actually really love this approach!
You could have made a symetrical measurement of ttk by mapping the data on each eighth slice of the platform to just one eight. That way you also have way more data per block and you get a symmetrical result.
And because this slashes the total of variables to one eighth this also makes the optimal search way faster.
It would also be interesting to know whether the shape of the platform affects the ttk in other ways because of the gang up mechanics of the piglin.
i dont know about the video, but for some reason, something in your patreons triggered my brain. i was looking for a word the whole effining week and my brain ultra-hard blocked even the slightest hint at getting close to it. so... thanks i guess!
on the topic of gold farms, i would like to see a "harvester" design with cycling flying machines clearing the platforms faster (at least no stuck pigmen and in theory faster speeds than walking, while sacrificing a "row" of spawning spaces. even though it might require more killing setup, i think it would be an interesting take on the standard aggro farms.
He's baaack ❤❤
Perfect. Thats the only word to describe this video.
Insane effort on the video; expected this to be some AI slop and came out pleasantly surprised!
Love the math visualizations as well; saddening that this doesn't have more views...
forget AI taking over the world, AI can make minecraft farms
I feel like the title kind of does the video a disservice cause the amount of effort put into the content of the video and the video itself was great
Yeah this comment hit me. Thanks and trying to adapt :-).
Crazy good🔥
Great job, crazy amount of work, commenting just to boost video 😂.
the legend is Back
He is back
the legend is back
why do you let the pigmen drop? why not have a kill chamber on each layer in this design? doesn't the drop increase ttk?
I love the beacon colors
I'm glad you didn't actually use a LLM to design a farm for shits and giggles and clickbait. What you did is actual data science and engineering its far more valuable. MATH FTW. Great job :D
A tangent, this type of practical usages should be used in school to teach math/engineering. I remember my uni days when nobody in class understood what fourier transform actually is and why its needed and "good" students just memorized equations verbatim to gain cgpa. Years later I slowly understood the relevance and brilliance of FT by learning from internet/side projects how groundbreaking tool it is.
You didn't use an AI, but you coded your own, which makes it even better.
I hope all is better ... missed your videos
Nico is an AI confirmed!
Ahhh Nico has been found!
A new Video! Yeah
amazingngngngngnngng i never thought changing the circle to a weirder circle could give better rates
not the clickbait but i love the shape
He has risen baby girl.
no way im early on this goats video first one in 9 months
Love your Videos :))
this is 1000% better than AI, great vid !!! can't even fathom the amount of time and head banging you put into this, but the result is awesome
Edit : people that watched the video after the title change : "why everyone is talking about ai in the comment? 🧐"
cant wait for ai to be able to do this by itself XD without the hours of self testing
Finally! Love your videos! Can you do a 100% emerald farm?
great vid!
1:05 actually it count to spigot, bukkit or any paper and purpur. but doesn't count to ordinary vanilla or servers with solo portal farms plugin installed
OMG BABE WAKE UP Nico is LOST JUST UPLOADED
The man, the myth, the legend remembered his password
Thank you. love your Videos.
great job
I know this is a bit late, but would it be possible to take the idea of self-optimization and apply it to, like, iron or regular mob farms?
Are we gonna get another vid about the iron farm then and is that project getting continued?
I’m curious whether this is dependent on the seed or the actual coordinates of each block.
And as some ppl have already said, this is definitely a form of machine learning. The simplest form of machine learning is a simple regression, so derivative-based optimization does fit the bill
Yo what music did you use? It bops!! I was nodding in agreement the whole video
Intro is omission by huma-huma, and outro is ebb and flow by pandaraps. The rest of the music is custom commissioned music I bought for the channel ;-)
@@NicoisLOST Dang :D well they did a great job. Good choice! It really helped keep me engaged the whole time.
I was just wondering if having a drop between every few blocks you go wider is worth it to have some type of fast slime block launcher to get them to the player quicker.
Or on the walk to the player??
And as always thanks for the great vid.
I do have one very important question about the farm as im trying to recreate it for my realm, what biome does it have to be in? Nether wastes? I have a area I would like to put it in, but it doesn't line up to where all parts of the farm are in the nether wastes.
also on the topic of optimization, wouldn't it make sense to speed up the TTK by having a kill chamber at every level? I'm assuming you use entity cramming via the boats so that the spawn rate is faster since pigmen sitting in boats don't contribute to the entity count, so with that in mind, couldn't you lurer the pigmen with turtle eggs and by aggro but having killing chambers on every level instead of the floor of the farm? and if there are killing chambers on every level, that means every platform on every level has the same shape instead of being tapered off towards the top?
I wonder: Now that zombified piglins spawn reinforcements of their own kind instead of regular zombies, would allowing those reinforcements to spawn in the glassed area increase the rates in a meaningful way? Those reinforcements would likely have a shorter TTK, after all. (I guess you'd have to lower the floor in those lines where the boats are and make them magma blocks.)
ChatGPT would require so many specifics that you may as well just design it yourself. If I ask it for gold farm ideas it’ll probably just give me a generic portal-ticking design,
you forgot about the most important factor: how cool is it
and that goes to the minecart version
The only 100% optimized farm is spawning an infinite amount of zombified pigmen that die in the first tick and the items to be transported into an infinite amount of double chests in the second tick and for every gold sword to be smelted instantly and made into nuggets, then into golden ingots also instantly.
Can you optimize for the minimum amount of wood farms to get all wood types?
I don't want to build 1 for each type😨.Also as automatic as possible!
i very curious on where you were going with the whole "ai" in the tittle until you started showing graphs like you usually do and "oh yeah, if water bottles can have ""AI"" then this is very much an ai too"
NICO IS FOUND
Couldn't you make several boat looting chambers closer to the spawning platform (for example one in each cardinal direction and one in each diagonal) to reduce ttk even more? Having the piglins traverse all the way to the middle seems inefficient.
7:38 I might be silly, but wouldn't trapdoors just work, since mobs think open trapdoors are walkable full blocks?
OK so funny story -- I uh did this and it didn't work so super well. A single trapdoor would have pigment sometimes getting stuck on the edge, and double trap doors made some weird thing happen -- they pigmen would fall then jump around for like 5 seconds, I think due to pathfinding reseting.
I actually originally included this in the video but cut it, but think the solution presented is the best.
Awesome work. WhiteStoneJazz is missing from people mentioned section btw.
Thanks! Will add!
Hello, great video, can i use soul soil with fire instead of magma blocks in this farm?
tbf, while this isn't AI, in the sense it's usually thought about, it leverages much of the same principles AI does: lots of statistics and gradient descent in an abstract space
Hello. Nico i had a question sort of what if we add a zombie or something else to anger the iron golem , won't it help in efficiency
I thought you would run something like a genetic algorithm to find the best spawning platform size. Genetic algorithms are some sort of AI right?
Bless non generative AI content! Any ideias for other farms this pipeline would be useful? I might give it a try
Listened to this video while multitasking, thought it was about pikmin 😭
A good example of how AI should be used… as a tool to assist where gaps of knowledge exist or long repetitive tasks so us humans don’t have to, the rest is still down to us
human AI design 👍
Any tutorial for exactly how much it’s each platform and how to make the killing zone to replicate it in my survival
Honestly waiting for that fortune8diamond gold farm
What if you use a mob conveyor to make the mobs move faster?