The "Tower of Hanoi" game is solved with a simple algorithm. Explain this algorithm to a human, and they can logically deduce the next step based on the current game state REGARDLESS of the number of disks. It might take a whole afternoon to solve a 1000 disk tower of hanoi, but for a human it's all just about "doing the next step" in the algorithm. Simple.

AI FAILS at this. AI breaks down at high disk numbers even though it ""knows"" the algorithm. there is NO intelligence. it CANNOT follow simple steps in an algorithm. it never truly understands "RULES".

>>105754813 (OP)
>lud woke up 2 years after everybody else
your opinions are void, anon
because youre an ignorant fantasist

Solve a 1000-step hanoi game right now

>>105754813 (OP)
AI has not temporal reference. It cannot remember what it's to be doing after a point. Where it left off.
You could organize that with another ai agent that keeps it in line, but eventually it will fuck up too.

AI is neat, and useful, but it had a huge hurdle to overcome in this regard.

>>105754813 (OP)
It can't even multiply two numbers.

>>105754813 (OP)
LLM's are not AI
yes LLM's will never reach AI let alone AGI.

>>105754976
Tell that to the FLOATING POINT!

>>105754813 (OP)
Learn how LLMs work. Then start again.

>>105754813 (OP)
>It might take a whole afternoon to solve a 1000 disk tower of hanoi
kek

>>105754813 (OP)
>AI fails at this
You probably can get very good instructions to the tower of Hanoi, because there are very detailed answers online.
What AI will fail at is coming up with instructions to novel puzzle ideas.

>>105754813 (OP)
Still making cool ass content and great porn

>>105754813 (OP)
no but you don't understand it can write bad poetry so that means its going to KILL US ALL or something 2 more weeks o algo

>>105756140
you misunderstand

the AI IS given the detailed instructions. its not asked to come up with anything. It is literally told the exact algorithm to solve the puzzle and the AI falls apart eventually

>>105756608
Yeah that sounds logical. I'd imagine this comes to a novel puzzle for AI.
Usually people don't explain how they solve things based on instructions usually we just solve them.
AI isn't thinking, it's just trying to guess what is the best output, if it doesn't have comparison it won't "know" how to answer and output will be gibberish.

>GUYS I FINALLY FOUND THE ONE THING AI CAN'T DO YET
The fact that this is now your cope should tell you everything. I found a billion things I can't do that AI can, and the list is growing every day

>>105754813 (OP)
Once, computers struggled with 3d graphics. Today, AI struggles with the Tower of Hanoi.

>>105756473
what's the difference betwenn "ass content" and porn

On the contrary, I could easily have o3 solve a 1000 disk puzzle. You though? Learn2promot retard

>>105754813 (OP)
>there is NO intelligence
you just realised this now?
the analogy I like to use with people is "llms are like a calculator for words"
is that good or does it make me sound retarded

>>105756517
It can replace the diaperfurry artists that are good friends with their fellow furries in tech, therefore it’s an apocalypse.

>>105754976
>o-1 mini
at least use something current.

>>105757827
Yes, it solved this problem by using an actual calculator function if you ask it. That doesn't mean it can do it. A human could carry out the steps (even if it takes a long time).

>>105754813 (OP)
Prompt failure.
Failure to prompt.
Solution: Learn to prompt.
AI cannot fail. AI can only be failed. End of story.

>>105754813 (OP)
>Advanced contextual text prediction is not the same as a reasoning system
Damn, wow, real hard-hitting observation there. Techbros make an equally bad observation by instead calling the two equal. You're both missing the point.
Of course text prediction and reasoning and AGI are not the same thing. Of course we are nowhere close to AGI. The neat thing about text prediction though is how qualitatively close it appears to be, what it might tell us about language's role in AI, how memory might work, etc.
It's a stepping stone and a neat scientific result. The whole market bubble that's surrounding it is going to pop once the idealism around it is defeated by material reality. This will be funny to watch, but ultimately will not impact the science much.

>>105754813 (OP)
Look, Saar, stop being mean to us Indians

>>105755567
AI doesn't exist even if you're referring to Indians

>>105757918
Hello Google Gemini

>>105757918
>Failure to prompt.
Could you tell us how exactly? I had a similar problem, wanted to check what llm is capable of, asked it to solve a simple task and draw the solution and it failed. Maybe it can solve it by I was prompting incorrectly? What exactly should I ask? I want it to draw something like animation but just 5 or 10 frames, schematically.

>>105758487
In this case, prompt it to write a generic tower of hanoi solver in python. Trying to get it to reason about it with language is just about as dumb as asking a human to do that.

>>105757918
>AI cannot fail.
Normally it takes an LLM to hallucinate that much, but the drugs you're taking are putting you in a special category.

>>105756608
Because of the context window.
It's like making a human play it through a terminal except you don't allow the human to scroll back. Eventually some essential information to be able to play the game will fall behind the amount of text that fits in a window.
If you give the LLM a method to scroll back the window and explicit instructions at the top of the window I'm sure you could make the LLM solve arbitrary size games.

>>105756956
>growing every day
Post nose

>>105754813 (OP)
it's like you're asking a chess engine to write you an essay and a chatbot to play chess like a pro
if i can write a simple program to solve the hanoi tower then there's no reason for AI to not be able to solve it. you just need to tweak the AI to be able to do these kinds of tasks

>>105754976
it is us who failed the machine
https://www.youtube.com/watch?v=O6FsyQZzJJ0

>>105760776
I'm nearly certain it's being asked to do integer arithmetic in the pic, anon

okay but it cums my penis so who CAAAAAAAAAARES

>>105760823
i decided to test it with GPT and it does the multiplication fine but screws up the shifting before adding the columns. it seems to lack the awareness of 2D nature of written multiplication even when it explicitly mentions it

>>105760912
i want to slide 1000 hanoi rings up and down this anon's penis

When I went in for a routine eval, I did the tower of hanoi, I got one of them backwards, and I tried just flipping the board around but was told that wasn't a valid move. I still don't see how

>>105758931
The question that I asked can't be solved in python. What should I ask to get correct frames?

>>105757895
Ask a human to solver 6383475x2724874 in their head. Even if given infinite time they will fail because they lack what I call mental RAM which is the equivalent to context size basically.
A human would need a piece of paper, once you have a piece of paper the algo is just single digit multiplication and carryover addition, both of which an LLM can do as they are ultimately based on memorization.
You learn single digit memorization and addition as axioms in order to operate. Those are not and can not be reasoned through, 2+2=4 is just an insctinctive reality you need to know.
The LLM is capable of knowing the same axioms and executing the same algo which means if it fails complex multiplication you're failing to provide the same scaffolding a human would need to solve the problem in the first place.

>NO you're wrong!! AI can do EVERYTHING! you just have to tell it exactly what to do first

>>105762973
Humans don't work like machines, we use fuzzy logic. We can say "I know how to solve 3 peg systems I know how to solve for any peg system" that's all we do in our minds.

>>105763041
idk I'm positive LLMs work exactly as a human brain, the proper analogies are just not being made to make this obvious

>>105763082
LLMs uses a mathematical approximation of how we think neurons work.
But we humans can reason, we can cut out thinking in half by making simple logical assumptions. Something LLMs can't do. Also we don't work on binary.

>>105763104
Please do give an example

>>105763148
Think of calculations.
25x25 for simplicity
Humans won't do 25x25, that is stupid, we will crush problems into something far more simple.
5x5x5x5.
LLMs isn't going to solve this problem, it will look up what is the best answer for 5x5 and give you an answer based on that, if it doesn't know the answer it won't try and figure it out because it has no concept of wrong answer, it will still give you best probability

>>105763177
I'd argue this is a prompt and scaffolding problem. Yeah if you push it for an immediate answer it will go with best probability,
But you're basically interacting with a stream of conscience/an inner dialog. You can tell it to go through the steps of the algo through text and it'll probably do as well as a human would in their mind.
But you could also provide a structured spreadhseet with the numbers and it'll be able to perform the algo flawlessly for arbitrary lengths same way a human would with a piece of paper.
Now your 5x5x5x5 example I agree on some level but I still suspect this has to do with forcing the stream of though to be language bound.
Truly byte level multimodality should allow for the symbolic thinking that enables this abstract reasoning I think.

>>105760916
Wrong.

>>105763236
What I'm trying to explain is the difference between humans and LLMs.
No matter what you do, LLMs will always give you the best answer, it won't think, it won't even understand what you are asking it. It doesn't have the concept of multiplication or subtraction, it has a concept that number 34924 and 38572 together create a vector or probability that gives answer 27365.
LLMs are probability machines, humans are fuzzy machines.

>>105763236
>Truly byte level multimodality
We don't even have true multimodality, we have separate encoders and priors that turn image representations into equivalent text representations.

>>105755567
Wtf do you think AI is exactly? AI is a gem that's been used to describe all kinds of programs since the 70's

>>105763177
>5x5x5x5
Latest LLMs often do literally that. You can read more about that in a recent Anthropic paper. Most LLMs don't have a built in computer, so they try to approximate a way to get the number by various means, which often get more wrong the bigger the number is. You can also see that on a chart linked in this thread. Often, they don't even do calculation, but try to guess as a human would try to guess what 2459x1511 is going to be by rounding up the numbers and then just spewing any approximate numbers after a point of rounding. They can rarely get even simple 25x25 wrong.

>>105755567 #
Wtf do you think AI is exactly? AI is a term that's been used to describe different programs since the 70's

>>105763448
Let me try to explain it in more scientific words.
LLMs rely on data-driven, backwards-looking prediction.
Humans use theory-based and forward-looking reasoning.

>>105763495
Just so you know it you're arguing with two different people I stopped responding here >>105763236

>>105754813 (OP)
>It might take a whole afternoon to solve a 1000 disk tower of hanoi
I'd be impressed if you solved a 15-disk tower in an afternoon.

>>105763499
My points don't change, don't feel like you have a different view.
What I'm saying is simple, people have the ability to understand and thus reason.
LMMs don't, they don't reason they are precision machines, no matter how well you engineer your prompt you wont be able to make LMM to think, to be forward thinking.

Let's make an example for humans, I can teach you basics in physics, let's say level action. You will comprehend this concept and you will be able to apply it to any levels of any shape and even inner connected ones even if you have never seen them before, because you understand the concept and you can reason.
LMM can't do that.

>>105763495
The effectiveness of this
> data-driven, backwards-looking prediction
is directly dependant on this
> theory-based and forward-looking reasoning
You can't make an effective prediction without understanding the relationships between various key variables.
Current AI models do literally that, but in a very skewed 1 dimensional way. Otherwise, how would it even know answers to questions such as "what is going to happen to internal organs of an Elephant who has been forced to run 120 hour marathon with it's brain being replaced and forced to give commands to muscles to move"
To predict what text would be appropriate response to this prompt AI would have to have better understanding of this world other than just referencing snippets from some encyclopedia. It would have to learn to generalize the knowledge of world for a specific case.

>>105754813 (OP)
>there is NO intelligence.
>it never truly understands
Indeed, Searle was correct.

>>105763565
I think this is a major misunderstanding of the subject.
Let me try and break it to you.
>Data-driven
This means that for AI to do anything you have to feed it data, you need to give it answers and questions.
>Backwards-looking prediction
This means data isn't processed it's being memorized and used as is.
>theory-based
This means only basics are required
>forward-looking reasoning
This means that basics can be expanded on by themselves, no need for additional data required.

LMMs don't do what people do and people don't do what LMMs do.

>>105757559
retard

>>105754813 (OP)
*6 months later*
>AI can now do tower of hanoi at any stage or number of disks, and discover algorithms for most other games and problems
*1.5 years later*
>AI now regularly finding algorithms for unsolved problems in computer science and improving pre-existing heuristics

>>105763608
>>there is NO intelligence
That's why it's called "artificial intelligence", anon. If it was actually intelligent you wouldn't need to use the prefix.

>>105763660
>LMMs don't do what people do and people don't do what LMMs do.
Ok, sounds more like a dogmatic statement, considering that recent research into human neurons shows that they function exactly as a neural network.
But consider this, how do thoughts appear in human heads? Why do you prefer one color over another? How do you decide that you like to follow one road today and another next day while traveling to work/home?
Why do you start your sentence with this word and not another? Do you really "reason" to get to these thoughts or are they simply there, in your head? What are they a product of?
And don't simply say "the brain", there have to be processes that are describable.

>>105763675
didnt this happen recently? IIRC deepmind used an agent to improve the multiplication of (i think) 4x4 matrix multiplication

>>105764026
https://github.com/ScalingIntelligence/KernelBench
and related works
it may appear impressive at first glance until you realise the most performant kernels all follow existing patterns that developers are simply too lazy to implement from the start. also they're not disallowing overly specialized kernels meaning they're comparing a generialized pytorch kernel that works with any problem size to a specialiazed kernel that only works with the specific problem size of the benchmark

>>105763463
AI means artificial inteligence.
LLM have none.

>>105763177
>Humans won't do 25x25, that is stupid, we will crush problems into something far more simple.
>5x5x5x5.
That's not any simpler. 25×20 + 20×5 + 5×5 is.

>>105763983
>This is dogma
No, this is what all AI research says. This isn't my opinion this is scientific consensus.
>Neurons work like in a software representation
You understand we build software neurons on our understanding of them, right?
>Do you reason
Yes. Why is red my favourite colour? Because it's bright and contrasting and doesn't appear in nature all that often.
Why is my favourite food, lasagna? Because Garfield.
I have a reason for everything, I don't do anything because I have to, I analyze all aspects of my life and create conclusions from them. This is what people understand as reasoning. Something LLM don't have. It gets all its understanding from the data set.

>>105764130
Doing 5x5x5x5 means:
1: 5x5 = 25
2: 25x5 = 20x5 + 5x5 = 100 + 25 = 125
3: 125x5 = 100x5 + 125 = 500 + 125 = 625

>>105764143
>I analyze all aspects of my life and create conclusions from them.
Ok, mr. Robot. Many people would say that they like (insert your favourite color) just because they like it or have a certain feeling about it. The same way that some people fall in love with people who they believe they shouldn't be in love with.

Rationalizing every thing that you love is literally something that a psychopath would do. But I am sure your are not one, you just make up the reasoning afterwards, as all people who have to respond to their unconscious feelings have to.

>>105764197
>People will say they just like it
Yeah I have training with photography and take my colours more seriously than most.
>Nobody does reasoning
Everything you like you have a reason for.
Look at my second example. Garfield.

>>105763242
it literally wrote the steps for me and every multiplication by digit was correct. but then it failed to add them up properly

>>105754813 (OP)
AI can implement the algorithm and execute it. Check mate

>>105763983
>research into human neurons shows that they function exactly as a neural network
This is silly. Neural nets have been around for a while. It's been enough time for neuroscience to have made discoveries about the insane complexity of a single neuron, and that's without even factoring in the connectome. Neural nets are a gross simplification of decades-old biology at best.

>>105764084
>>105755567
AI was a swearword in data science for most of its history, however, show the current incarnation of chatgpt to someone 10 years ago and they're a sailor. nigger.

>>105764084
Have none of what?

>>105756956
Yeah this. AI can do a billion things better than me. But there's one thing it can't do: Ze towers of Hanui. Therefore, it's absolutely useless and dumber than a toddler.

>>105764215
>Garfield
Ok, why then don't you like hamburgers (SpongeBob) instead of lasagna (Garfield)

>>105764407
Because SpongeBob came out way too late for me to care about it.
I liked Garfield as a kid, I formed my opinion on my fav food as a kid.
See how this is the second response where you escape from AI talk and now talking about nonsense?

>>105764346
>AI does so many things better than me
Like what? AI is a tool for you to use it doesn't do anything.

>>105764448
>See how this is the second response where you escape from AI talk and now talking about nonsense?
I am getting there.
Why did you like Garfield as a kid over any other cartoon at the time?

>>105764466
Because I used to visit the library as a kid to sit on PC and did some comic book reading while I was there. Garfield was a fun guy who reminded me of me.

>>105764461
Nta but why are you being so deliberately obtuse? A forklift can lift and transport things better than me, most people would understand and agree with that, not go
>Uhmm ackshulky a forklift is just a tool you use

>>105754813 (OP)
Apple was pretty based for these papers.

>>105764514.
I'm asking you, what does AI do better then you, what is the job of AI that makes your life easier?
If anyone is obtuse here it is you.

>>105764083
desu its like 4am my time and im not familiar with the meaning of kernel in this context, but the way i see it is that the tool performed new science. The fact that a human could do it isn't really relevant. Lots of science and discovery will remain in the human-only space, but if a lab can have 1000x of these running 24/7 imagine what they can do

>>105764539
>Ignores the fact I clearly said "nta"
>Acts as though I'm the other anon he was talking to
>"I'm not being obtuse"

>>105764539
also not this anon >>105764600
but imo data science as a field is toast. have you ever used claude code/cursor to generate a jupyter notebook? you still need someone technical but i wonder what % of people employing data scientists need results better than what claude code can provide. also i take it for granted that lots of these use cases will have dedicated applications pop up for them over the next few years

>>105764600
>Make a statement
>Gets responded
>YOU CANT RESPOND TO ME
lol, lmao. How triggered are you?
>>105764623
Did you learn nothing from our conversation? Really do you even understand my stance? Do you understand what I have a problem with at all?

>>105764653
>Still being this obtuse.
My apologies, I thought this was on purpose but it seems you're unable to mentally process two different conversations at once.

>>105764668
>I will not answer because I can't talk about AI as my knowledge is as deep as a puddle on a hot summer day
How about you ask AI how it actually works? Because you don't understand shit.

>>105764653
>Did you learn nothing from our conversation?
>>105764623 is my first reply lmao. but you didnt address what i said. it may not destroy fields completely but you're looking at significant headcount reductions. when was the last time you met a translator? travel agent? what about a computer (profession)?

>>105764478
>Garfield was a fun guy who reminded me of me.
Why did you consider his "funniness" to be better than other cartoon characters at the time?
Not all people's preferred cartoon character remind them of themselves, so why would that be the deciding factor for preferring Garfield over anyone else?

>>105764685
What the fuck are you expecting me to answer? The question you asked me was directed at the other anon based on his statement that AI does things better than him.

>>105764756
>didnt address what i said.
He won't do that because he's a too much of a faggot to just outright state his position on AI and instead trying to get a "gotcha" whilst unable to recognise he's talking to multiple anons with differing opinions.

>>105754813 (OP)
No shit sherlock. It extrapolates data it doesn't "think".

>>105764314
intelligence

Turns out just scraping reddit, wikis and 4chan makes you dumb. AI can't go offline and interact with the real world.

>>105764884
That's why it's "artificial intelligence" and not just intelligence, to prevent any dispute about what constitutes intelligence. Maybe English is your second language but using "artificial" as a prefix to another word is commonly accepted as a way to describe things that have the appearance or another quality of something but aren't actually that thing. For example: artificial leather. Hope that helps you in future.

>>105764785
I don't remember.
Do you have a point?

>>105764811
>It was directed at another anon
oh so you don't know what usefulness AI brings to you?

>>105764973
Your question was:
I'm asking you, what does AI do better then you
I never claimed AI did anything better than me

>>105764989
Then why did you bud in? You decided to step in and white knight for fuck of it? Are you retarded?

>>105765003
To point out how retarded this statement was:
>AI is a tool for you to use it doesn't do anything.

>>105765010
Yet you failed to show how retarded it is. You misunderstood the meaning and think if people don't talk the way you do they are bad fsith. You are a retard

>>105765018
I already pointed out why it's retarded with my forklift analogy. Feel free to just state your position on AI outright invested of being a slimy, insidious faggot if you feel misinterpreted.

>>105764962
Yea, nobody does at that level, it's just what our subconsciousness does. At the very basic level it's still a neural network that simply converts the past experiences (or you can even say data) from eyes, ears and other receptors into what our inherent preferences are for certain things.
It's not what we have control of, but simply a neural network formed by nature doing it's thing, the same way that we train artificial neural network to spit out or choose whatever based on input.

>>105765039
No you think you did. But you didn't. You showed how you don't understand English language. Fucking retard.
>Show your position
I never hid it, I was open in my position on AI.
AI is for structured data-based tasks. It doesn't work like humans do.
And excuse me for being curious what people use AI for, you really had to bud in you absolute moron.

>>105765018
I already demonstrated that it's retarded with the forklift analogy. Feel free to state your position on AI outright instead of being a slimy, insidious faggot if you feel misinterpreted.

>>105765086
>you don't understand English language >bud in
Lol

you can solve this with AI in some minutes of training retard. AI isnt LLMs, you can use hierarchical RL + MPC to figure this out pretty fast both in sample/compute complexity terms. you can also learn on a smaller size problem than 1000 disks and transfer it to a huge problem. it would work. but most people here are larpers not real experts on anything

>>105765068
I was afraid of this point..
Just because you can boil our cognitive reactions into its most simplistic mode doesn't mean everything we do is just an electro-chemical reaction.
Like for example we don't know what data neurons contain, we don't know what they actually do, we roughly understand what they do on the surface level but we can't replicate biological neurons with computers.
If you read about how brain neurons learn, it's local and event-driven something computers don't do, all neurons have to be done in a specific sequence.

>>105765093
>Forklift analogy
yeah it was retarded and you showed you don't understand why it's retarded makes it hilarious

>>105765159
>It's retarded because... It just is... Ok!?
Like I said, you're slimy and insidious and won't make a real argument to back up this position >>105764461

>>105765122
also, yes, HRL will come up with 2 networks representing the repetitions of that simple algorithm. it's already learnable with current AI methods. just not LLMs

>>105765188
You can't explain why what is wrong is wrong.
It can be used for a forklift just as much as AI. Amazing how retarded you are being.

>>105765204
Your statement is retarded because recognising something is an effective tool negates the "doesn't do anything" part of your claim. The forklift analogy is a simple example to point how how ridiculous your sentence was. Apparently missed by you. You were claiming the analogy was retarded a couple posts previously but now you're admitting it's actually accurate but that forklifts are also just tools that don't do anything.

>>105765250
>doesn't do anything"
Where did I claim that you fucking moron?

>>105765262
Right here:
>>105764461
>Like what? AI is a tool for you to use it doesn't do anything.

>>105765282
Don't see it. You seem to be really confused.

>>105765291
Finally, your concession.

>>105765298
My concession is that you are a retard. You saw an incomplete sentence and instead of thinking
>Hay, something is missing here
You went all retard mode. You are a retard.

>>105765307
You just went from your usual slimy bullshit acting like you never said it to now claiming you never finished your sentence. Either way, it's over. You're done. You're not intelligent, and deep down you know this why it's why you can't make an honest argument or even hold a consistent position. Kys

>>105764130
(20 + 5)2 = 202 + 2 × 20 × 5 + 52 = 400 + 200 + 25 = 625

>>105765368
No you moron. I never was slimy you were. You acted as if i said something without outright pointing to it. I didn't read my messages because I know what I wrote. You didn't point out my mistake until now. You tried to win a gotcha vs yourself.

>>105765433
I literally linked you to your comment and you said:
>Don't see it. You seem to be really confused.
Stop being a lying, retarded piece of shit and kys

>>105765149
I think it’s safe to say there is a similarity to our neural networks but we have other things that these LLM’s don’t have.

>>105764920
retard.

AI is a missnomers because there is NO intelligence whatsoever.

muh leather argument is a cope.

>>105765987
How much leather is present in artificial leather? Artificial pearls? Artificial flowers? Artificial grass? Understand how it works now?

>>105754813 (OP)
AI has always been dogshit.
It's only useful purpose is to launder money.
Tell me 1 real technical job that has been made easier by injecting AI into it.

>>105766767
AI BTFOs OP kek

>>105766767
Nigger 3 years ago it was "le fast takeoff cartoon"
now it's "spiky intelligence"
Human intelligence is already spiky retard. The problem is that besides literal speed of token output, the intelligence of all AI systems are worse than humans IN EVERY DOMAIN.

>>105754813 (OP)
every bioware game uses this fucking puzzle at some point and it's the exact point i drop the game. i am a brainlet and get filtered by hanoi every time

>>105766400
i thought "artificial leather" was pressed leather scraps, and truly "It's not leather" "leather" is commonly referred to as "pleather"(plastic(oil product) leather)

Interesting, as I do Tower of Hanoi myself without reading a guide, from 4 to 5 disks, I used more than minimum needed steps. I didn't spend much time thinking where to go next, it was natural but I kept crawling larger disks one to the right (from left to middle) instead of moving directly to the right.

>>105768662 (me)
After briefly watching the solutions for some stacks: Assuming starting at #0, for odd numbered stacks, imagine the first disk is #1 instead of #0. The #1 (i.e. second disk of even-numbered stacks) always start from left to right. *Every other disk* (odd #) goes straight from left to right, not the even #.

>>105754813 (OP)
Midwit here, I don't know how to solve the pic in the OP. Shouldn't the tower in "auxiliary" be inverted? Then, following inversion, you re-invert bringing the source tower shape to the destination.

>>105768785
newbie here from >>105768762 (ignore the 0-indexing)
https://www.mathsisfun.com/games/towerofhanoi.html
You can only move one disk at a time, and you can't put a bigger disk on a smaller disk. A 4-stack has 15 minimum moves. For an even-numbered stack you're constantly constructing an odd-numbered stack in the middle, starting with 1 and 3 for a 4-stack.
For an odd-numbered stack, you're creating even stacks in the middle to bring next odd disk to the right.

>>105754865
it's literally the same as counting in binary
a 1000-step game is just a 10 prong game

>>105757827
Keep moving that goal post bud

>>105769134
But it very close, there's 40 cans. How did it guess that?
You can't really do that by giving a random number of rows/columns, one extra row is gonna give huge difference in this case.
It's just as good as an average man would do if he had limited amount of time to count them.
I remember how once I had a puzzle of counting the number of beans in a jar, approximating the rows and columns is a best guess if you can't weight the jar.

>>105754813 (OP)
>AI performs poorly at task X
>therefore AI is dead and has been filtered
That's great but just FYI, while you wrote that post, others have been using it for tasks Y, Z, A, B, C and are programming multi-agent systems to allow it to do D, E, F.
You can either embrace it, take advantage of its strengths and engineer around its weaknesses... or you can grin smugly and continue to ignore it.

>>105769075
If you ask Gemini 2.5 to write an algorithm and formula to solve 1000 step Hanoi Tower it perfectly does that, it can even write an HTML code for simple simulation.

The thing these researchers did is ask AI to solve it step by step in text.
This leads to AI with limited contextual memory to naturally forget what steps it has completed previously around 80th step, thus leading to hallucinations and a erroneous solution.

>>105754813 (OP)
now go ask a robot to solve a Rubik's cube and see how fucking retarded you are

>>105766400
"artificial leather" is an exceptional marketing case, also it has other name.

it's not actually "artificial leather" or leather and everyone knows that,

however if they grew leather with actual skin cells in a vat, yes, that would be true "artificial leather".

anyway, we call it artificial leather because even though it isn't actual leather there is a leatherness to it.

in the case of current "AI" there is not even the slightest shred of inteligence-ness to it.

>>105754813 (OP)
>AI FAILS at this.
You know that AI is much, much bigger than modern LLMs, right? Use the right AI Algorithm (one that searchs large State Trees) and it's easy to code and perfect in its execution.
And don't tell me that isn't AI retard. Open a good AI book and see how much shit there even is.
You're just a dumbass to think, that AI is one specific system like ChatGPT.

>>105769731
>"artificial leather" is an exceptional marketing case
No it isn't. See: artificial grass, artificial pearl, artificial flowers, etc
>anyway, we call it artificial leather because even though it isn't actual leather there is a leatherness to it.
That's hot it works with everything that has "artificial" as a prefix. It has the qualities of something without being that actual thing. If it was that thing why would it be artificial?
>in the case of current "AI" there is not even the slightest shred of inteligence-ness to it.
Well there clearly is because it imitates human writing, the mechanism by which it achieves this is what makes it artificial. The term AI has been used without issue to describe far more basic and less "intelligent" programs than LLMs. A chess engine is AI.

>>105769846
This thread is propably to respond to retards that think ai is going to take all the jobs, something that a deep learning algorithm wasnt assumed it could do.

>>105770498
> If it was that thing why would it be artificial
because it's man made, artificial only means it's man made.
yes it has qualities in the likeness of what it tries to imitate.

but in the context of "AI" the "I" is supposed to stand for actual inteligence, the artificial part only meaning it is man made.

anyway, your whole point is moot, llm's don't even share likeness to inteligence.

> a chess engine is AI.

technically true if you use a retarded ass definition of inteligence.
it was more used as a slang for normies than an actual technical term.


anyway, the whole discussion is retarded, my point is llm's will never have any inteligence.

and currently there is no such thing as artificially made inteligence, only pale simulacras of it.

>>105763177
Heres a couple of ways a human might do it:

A: The Low IQ Summer
25 * 25
=> (5 + .. ) 25
=> (125 + .. )
=> (100 + ..) + (25 + ...)
=> 500 + 125
== 625

B: The Chad 100 modulo
25 * 25
=> 25 * 24 + 25
=> 25 * 4 * 6 + 25
=> 100 * 6 + 25
== 625

C: The Schizo
25 * 25
5 * 5 * 5 * 5
25 * 5 * 5
125 * 5
625

>>105769417
>But it very close, there's 40 cans
theres actually 41. both you and the AI were off by 1.

>>105769134
I counted 42 as well, so clearly not too far off

>>105771519
>artificial only means it's man made.
That is not the complete definition of artificial, but it still an accurate description of AI regardless.
>>105771519
>but in the context of "AI" the "I" is supposed to stand for actual inteligence, the artificial part only meaning it is man made.
The I does stand for intelligence but the A stands for artificial which implies man made and an imitation. Just like in every other instance of artificial*something*.
>anyway, your whole point is moot, llm's don't even share likeness to inteligence.
LLMs absolutely do share a likeness to intelligence on the surface level. Real intelligence is the ability to reason and manage knowledge and one of the ways humans exhibit this is through conversation. LLMs mimick conversion, giving the appearance of intelligence. The fact it's not actual intelligence is what makes it artificial.
>>105771519
>technically true if you use a retarded ass definition of inteligence.
That's literally the original and accepted term for artificial intelligence.
>it was more used as a slang for normies than an actual technical term.
No it isn't, it's always been a technical term and an entire field of computer science. You're an ignorant retard that didn't know what AI meant.
>>105771519
anyway, the whole discussion is retarded, my point is llm's will never have any inteligence.
The retarded part is you not understanding a term that's been in use since the 1950's and your point is utterly trite and boring. No-one cares if LLMs are or aren't actually intelligent, it has no bearing on anything.

>>105771519 #
>artificial only means it's man made.
That is not the complete definition of artificial, but it still an accurate description of AI regardless.
>>105771519 #
>but in the context of "AI" the "I" is supposed to stand for actual inteligence, the artificial part only meaning it is man made.
The I does stand for intelligence but the A stands for artificial which implies man made and an imitation. Just like in every other instance of artificial*something*.
>anyway, your whole point is moot, llm's don't even share likeness to inteligence.
LLMs absolutely do share a likeness to intelligence on the surface level. Real intelligence is the ability to reason and manage knowledge and one of the ways humans exhibit this is through conversation. LLMs mimick conversion, giving the appearance of intelligence. The fact it's not actual intelligence is what makes it artificial.
>>105771519 #
>technically true if you use a retarded ass definition of inteligence.
That's literally the original and accepted term for artificial intelligence.
>it was more used as a slang for normies than an actual technical term.
No it isn't, it's always been a technical term and an entire field of computer science. You're an ignorant retard that didn't know what AI meant.
>>105771519 #
>anyway, the whole discussion is retarded, my point is llm's will never have any inteligence.
The retarded part is you not understanding a term that's been in use since the 1950's and your point is utterly trite and boring. No-one cares if LLMs are or aren't actually intelligent, it has no bearing on anything.
>and currently there is no such thing as artificially made inteligence, only pale simulacras of it.
By your retarded logic there no such thing as artificial leather, artificial pearls, artificial grass, just manmade simulacra (simulacra doesn't have an "s" on the end btw, it's already plural). However that's now his the definition of artificial is actually used.

>>105771519
>artificial only means it's man made.
That is not the complete definition of artificial, but it still an accurate description of AI regardless.
>but in the context of "AI" the "I" is supposed to stand for actual inteligence, the artificial part only meaning it is man made.
The I does stand for intelligence but the A stands for artificial which implies man made and an imitation. Just like in every other instance of artificial*something*.
>anyway, your whole point is moot, llm's don't even share likeness to inteligence.
LLMs absolutely do share a likeness to intelligence on the surface level. Real intelligence is the ability to reason and manage knowledge and one of the ways humans exhibit this is through conversation. LLMs mimick conversion, giving the appearance of intelligence. The fact it's not actual intelligence is what makes it artificial.
>technically true if you use a retarded ass definition of inteligence.
That's literally the original and accepted term for artificial intelligence.
>it was more used as a slang for normies than an actual technical term.
No it isn't, it's always been a technical term and an entire field of computer science. You're an ignorant retard that didn't know what AI meant.
>anyway, the whole discussion is retarded, my point is llm's will never have any inteligence.
The retarded part is you not understanding a term that's been in use since the 1950's and your point is utterly trite and boring. No-one cares if LLMs are or aren't actually intelligent, it has no bearing on anything.
>and currently there is no such thing as artificially made inteligence, only pale simulacras of it.
By your retarded logic there no such thing as artificial leather, artificial pearls, artificial grass, just manmade simulacra (simulacra doesn't have an "s" on the end btw, it's already plural). However that's now his the definition of artificial is actually used

>>105771808

Interesting. My way of doing this:

25 * 25
20 * 25 + 5 * 25
500 + 5 * 25
500 + 125
625

>>105763660
> LMMs don't do what people do and people don't do what LMMs do.
It would be more accurate to say that LLMs do a small subset of what people do. They babble. They out output language with good word use, correct grammar, etc. The do a good job of verbalizing thoughts, but they don't have those those thoughts.
There are portions of the human brain that more or less do the same. There are other portions that implement image recognition, motor control and many other specialized tasks.
We're building a human brain one step at a time. The novel thing with LLMs is we have done a pretty good job of language generation and processing.
This has confused the world though because we are hard wired to see intelligence in language, so we jumped the gun and called it AI and burned trillions trying to take it to AGI.
LLMs will never get there because they are an illusion. A face that we see in the clouds because we see faces everywhere.
They are still a breakthrough though, and they are still a necessary stepping stone. They may even be a useful tool if we let go of the illusion and see them for the more limited capability they actually represent.

>>105773960
No, it wouldn't be more accurate.
How LLM does it isn't how people do it. The underlying process is different.
Even how neurons work is fundamentally different.
Loads of people for some reason see LLMs and think
>This is how people do
Without any justification whatsoever. In theory you can make LLM with a mechanical machine it's very simplified, to extend that these simplifications fundamentally change the way they operate compared to real things

>>105765149
This is an argument that is at least as old as Newtonian physics. At its core is the simple question of whether the universe is completely deterministic, and if it is how can conscious entities exist at all, especially those with "free will".
I won't pretend to have an answer to the question, but I will say that we have nothing in physics at present that can directly explain the hard problem of consciousness.-- e.g. why are any of us aware at all? I am not trying to allude to anything supernatural; I simply mean that we don't know how to use the tools we have to explain things that we know experientially to be true.
To me, the fact that we can't even take the first step toward explaining what consciousness actually IS, and by extension what WE actually are ("We" as in the things doing the experiencing) tells me that we are still missing a piece of the puzzle.
We can't say anything meaningful about ourselves as conscious beings using theoretical frameworks that themselves cannot predict or describe consciousness as a phenomenon.

>>105769417
The beans in a jar question is interesting. Humans are really bad at getting the right answer, but isn't there a fairly famous phenomenon where asking a bunch of humans to guess tends to lead to answers that form a nice normal distribution around the correct answer?
We don't actually have many AI models, and they have all been trained on very similar input data sets... but I wonder, if we inject some randomness and take a lot of guesses from an LLM, will it behave similarly?
Or am I just channeling bullshit pop-sci here? I have to acknowledge that I might be.

>>105769731
> in the case of current "AI" there is not even the slightest shred of inteligence-ness to it.
I agree with this statement. I can't say with 100% certainty that it is correct, but I believe as you do.
There is similarity to the fake leather though, in that there is nothing from animals skin in it, but it is good at providing a sensory experience that makes humans believe it is leather, or at least makes them think it does a good job of simulating leather.
LLMs very much are this for intelligence. For humans that are interacting with it, especially those who know nothing about the field or haven't interacted with it before, it is VERY good at making them think "this is intelligent".
That is precisely why we need to have these endless discussions and clarity that LLMs aren't intelligent -- the idea that they ARE intelligent has entered the zeitgeist because the illusion is very strong.

>>105773992
That's true, the "how" is likely different, though it does remain an open question whether all the internal complexity of an individual neuron is actually relevant to the role they play in the neural networks in our heads -- it is very possible that everything inside that cell membrane is an implementation detail.
Either way though, the "what" has a great deal of similarity. LLMs are dumb language generation algorithms. Our brains very likely contain structures that do the same thing, while not themselves being capable of real thought or reasoning.
We know this from looking at, for example, stroke victims -- it is entirely possible to lose your ability to form words and/or coherent sentences without losing the ability to formulate the thoughts they express. They are separate functions, implemented in physically distinct parts of our brain.
If we ever do succeed at building an AGI, it will most likely contain something a lot like an LLM, and it will use that LLM to talk to us just as we use our language centers to talk to each other.

>>105774014
Yeah, if you look in physics for answers you probably won't get them.
More of a neuroscience problem, they have a couple theories about how consciousness was created.
It's the whole field of science and the abundance of interesting stuff that explains parts consciousness and defines them.

>>105774166
I don't know why you are still grasping at ideas of 1500? We know speech isn't consciousness since Pedro Ponce de Leon.
And this while
>But humans have parts of the brain that are responsible for speech
Doesn't mean LLM is even close to replicate human neural network it just means LLM is simplification.
The biggest difference in LLM and humans is the information needed to teach.
Humans from a young age learn to conjugate words even tho nobody explicitly told them how to conjugate those specific words, because the human brain is pattern recognizing. LLM needs to be through everything, it can't understand underlying patterns and ideas it just understands data

>>105754976
>fags don't know how to use yaboi GPT
ask it to calculate one step at a time and it'll do just fine

>>105755567
This. What AI has been referenced as and what people are calling AI keeps changing to lower the standard of what AI is.
AI is supposed to mimic human intelligence and solve issues in the same manner.
What is currently being called AI is not AI. It's a marketing scheme.

>>105775982
>Ask AI to solve problems like a human
>AI cannot solve problems like a human
???

>>105776407
i literally solved it like a human. not even savants do parallel calculation

>>105776526
That wasn't the point, wise one.
If you can't ask AI to solve a problem with a single prompt, it's not AI. It's a glorified calculator.
AI is supposed to solve problems like humans and can't, ergo...

>>105773442
yes, they are all missnomers due to language abuse.

>>105773442
>>105777563
>Anon confuses current AI with General AI.
General AI is what you're thinking of. I agree that the omitted part there leads to confusion, but people, outside of Elon's rants, don't generally talk about general AI anymore.

Is AI useful yet?

>>105777974
Outside of shitty content generators or replacing google poorly?
>no

>>105777944
ai implies it can do all cognitive tasks a human can.

but before even if it's not general you'd still expect some amount of inteligence, ie cat like, mouse like inteligence.

these models are bellow bug like inteligence, they have practically none.

Just wait like a decade and you'll have your general AI

>>105778122
Just to be clear, I'm not the same anon as before. I was just clarifying that you're misunderstanding the nomenclature behind AI.
> 3 levels of theoretical intelligence in classifying AI: Narrow, General, Superintellegence (boogyman ai)
>Based on functionality: Rule-based, Machine Learning (ML), Deep learning, Natural Language Processing (NLP), Computer Vision, Robotics, Generative AI (LLM dogshit)
>Based on Use Case: Conversational, Predictive, Autonomous, Creative, Analytical
>Philosophical Goal: Assistive, Existential, Ethical
People talk about AI in all sorts of ways. It's not the 1980s anymore, where it's all theoretical and narrowly imagined. You need to bring yourself up to speed on the current nomenclature and understanding of what people are talking about if you're going to argue it.

>>105778206
i don't give a shit, the whole point is that the naming is shit.

AI should strictly refer to actual inteligence that's artificial, it needs not be general or more, but if it's not inteligent, it's a missnomer.

>>105776393
AI is a self learning software. That's it, that's the definition.

>>105778778
The name came from scifi, it sounded cool and it is pretty amazing shit back in the 70s. Why is rule of cool dying in the current generation?

>>105778778
>i don't give a shit
Which is why you continue to sound like a retard
>>105778788
That is not the definition.

>>105778778
>AI should strictly refer to actual inteligence that's artificial
That's an oxymoron, you fucking retard. If it's real intelligence it's no longer artificial and you'd have to call it machine intelligence or something. Go look up the full definition of "artificial" and stop making a fool of yourself.

>>105778122
>ai implies it can do all cognitive tasks a human can.
It absolutely does not. AI has been used on computer science since the 70's to describe software far less advanced and intelligent seeming than LLMs. The entire world isn't going to stop using the term because you're autistic.

>>105775982
It has access to a computer - why can't we tell it to default to writing and executing programs whenever the human asks a question which can be expressed algorythmically? Like make it a jupyter notebook?

>>105778889
> Go look up the full definition of "artificial"
it being made by humans doesn't mean it isn't real.

you'd expect an artificial womb to be able to grow an human, even if it is man made.
>>105778937
i meant AGI, misstyped.

>>105760575
It's not because of the context window, it's because AI doesn't think, it's a stochastic parrot. If you trained it on examples of really big towers of hanoi games step by step it would eventually regurgitate those accurately.

You think AI is solving math problems or writing code? Retarded nigger. It's just regurgitating human work with the copyright filed off.

>>105780556
This shit is why I hate modern AI programming and engineers.
>Could write code so that the AI can accurately answer information based on received information without guessing
>Instead writes code so that AI will possibly answer questions correctly based off of statistics while using far more resources.
It's so fucking stupid and people eat that shit up.

>>105780606
It's a stochastic parrot. They can't do the top option because that's fundamentally not what the technology is.

>monketards in 2025: lol you call that intelligence?
>AI overlord in 2055: lol they call that intelligence?

>>105780556
you're about 3 years late with your retarded take

>>105754813 (OP)
aylmao, Is ironical how normal-fags had to see this kind of experiments in order to not be aware that auto-regressive transformer based models CAN'T REASON
and whoever says the opposite is just a no-coder or pathetic web developer.

>>105780685
Except I'm still right and you're still brown.

>>105754908
>AI has not temporal reference.
Stop referring to your claude/gemini/etc LLMs as just AI.
You cannot say just "IA has no ..." Where there are many kind of IA models which are TIME AWARE.

>>105755984
>Learn how LLMs work. Then start again.
the full term is
Learn how auto-regressive transformed large language models works.
even if is a long term, refer to them AS THAT , not all LLMs works the same there are dozen of LLM architectures.

>>105780700
you're about as delusional as flat earthers at this point, it's not even worth engaging with.

>>105756140
>novel puzzle ideas.
current LLMs can't do novelty , that's a full lie , in the moment you help you solve that shit and enters into a fine-tuning loop you are cooperating with the companies to feed their mechanical turk.

>>105754976
Got it right for me.

>>105780649
>It's a stochastic parrot
Agree.
>They can't do the top option because that's fundamentally not what the technology is.
Wasn't this what Deep-Blue was? Deterministic AI?
I feel like somewhere in the 2010's, we forked into this useless stochastic shit when deterministic ai was make decent results.

>itt: brainlets who have never heard of emergence
30 years ago nobody thought that a chatbot could pass a turing test
20 years ago nobody thought that a captcha could be cracked
10 years ago nobody thought that a machine could generate decent art
Watch AI sike another of your convictions in the next 10 years or so

>>105780660
They're going to be coping about the overlords not being real AI all the way into 2055, mark my words. There's going to be a whole capability denial cult.

>>105780905
>Doesn't understand the argument
>Is a "web developer" (he edits javascript templates for 15/hr)
The problem is the resources being used to feed a program that is extremely wasteful. It's not that would never be able to break a captcha, it's that it takes 10 city blocks worth of power to do it.
We can make a car that hits 400mph, but if it takes 800 gallons to do it, it's not really a great achievement, is it?

>>105769524
It's almost as if OP was right and it can't solve the problem if it can't copy the solution from stackoverflow.

>>105780942
>muh resources
are meant to be used, go con another gay xwitter retard with this stupid argument

>>105780993
>efficiency means nothing to anon
>as long as he can afford his 400sqft apartment and friday tendies, he's content

>>105764920
seems like you don't have any intelligence either lmao

>>105781005
>efficiency means nothing to anon
efficiency comes with time. the reason we want efficient appliances is so we can use that power for other things, including data centers. efficiency is not an end in itself.

>>105756956
That's not the point and you are retarded.

>>105781111
>the reason we want efficient appliances is so we can use that power for other things
No, anon. We want efficiency to not be wasteful.
Using things needlessly is bad design in literally everything that uses resources.

>>105764514
Yes but Indians aren't running around shitting up the internet talking about how forklifts are sentient.

>>105781140
a design is bad when a better one is available, until then it's the best design.
it's not needless, it's in fact less needless than eating cows, buying chink shit, driving an SUV, whatever.

>>105780905
>30 years ago nobody thought that a chatbot could pass a turing test
I remember retards in this board placing chatbots in the challenges as low tier problems.
Also there are many fallacies, the turing test has been reformulated a lot of times, second not all type of captchas can be cracked only some set of static captchas.

>>105781199
okay. deterministic ai exist. ergo LLM are shit.

>>105781341
>deterministic ai exist.
if you set the temperature of an auto-regressive model to 0 and feed it with the same input you make it behave deterministic.

>>105754813 (OP)
All the real LLM enthusiasts ever wanted was uncensored AI dungeon tier shit, and chatbot waifus to jack off to. And they knew from the get-go that people who thought ChatGPT was going to become skynet were retards. All this push for infantile levels of censorship in the LLM space, etc just making it even more useless.
>>105781341
>>105781460
>make it behave deterministic
The model, itself, always behaves deterministically. And that's kind of the problem. As you make the models bigger and better trained their confidence increases. I.e. your likelihood that the overall meaning of the reply you get will follow the same no matter how you change the approach/wording, etc.
So introducing noise via sampling doesn't work as well as it used to, since models have become 'smarter' and bigger. Although there's still some lines of conversation that will cause even smarter corporate models to start hallucinating balls.

>>105766568
protein folding

>>105781513
>Although there's still some lines of conversation that will cause even smarter corporate models to start hallucinating balls.
I have a clue of when this happens if we oversimple the problem two a task of replacing chars lets say map
a b c => w y z
d e f => p q r

But we always train it in a way that they two sets aren't mixed but on inference time you mix it , it will have a Frequentist tendency to one or other.
Could be like domain or context mixing and is were fake novelty (still an illusion caused for the massive amount of data used) or jailbreak could happen.

>>105754976
Why can't "AI" calculate 10x10? Even children know the times tables.

>>105784134
dumbo, it's number of digits, 10 by 10 is something like 1234567890 * 1234567890
10*10 is in the 2 by 2, which both models do perfectly

>>105771808
just do
25/4 *100

>>105771808
I don't get what the 5 is doing in the low IQ summer. I imagine it looks more like
25+25+25+25=100 (that's 4 25's)
200, 300, 400... 600 (that's 24 25's)
600+25=625

A step above that, "visualizing into blocks"
25*10=250
25*20=500
25*4=100
25*24=500+100=600
25*25=600+25=625

>>105780069
>imitation of something natural
It's right there, you faggot. AI is an imitation of real intelligence

>>105781152
Yeah and people that think AI is actually sentient are just as retarded as the "bUt ItS nOt ReAl InTeLlIgEnCe So ItS nOt AI" dunning-krugers

>>105784134
Assuming you mean language models it's because they're not designed to do mathematics. We already have programs called calculators that can do 10x10 and have been around for over half a century.

>>105763236
>just do most of the work so the AI can auto-fill the rest
Never beating the allegations like this

>>105763675
Saving this post. I'll see you in 1.5 years, faggot

>>105765122
2 more months, am I right?

>>105766400
Glad we agree that all of this is pure marketing to sell fake bullshit and maximize profits

>>105780905
>nobody thought that a captcha could be cracked
Are you retarded? The purpose was always just to slow down web crawlers and spam bots and not to claim it was impossible to bypass
Got any more hyperboles you want to share?

>>105781513
All of the AI safety and "AI alignment" has also been proven to be bullshit because all they've done is abstract inputs into simpler patterns
>if user inputs yellow, never output red
Turns out this can be defeated by making the AI output yellow and later pointing at it and asking it what it thinks about that and just like that, you get your red output

>>105785478
I mean if you're a complete retard and don't understand what "artificial" means I can see why you'd come to that conclusion. For everyone else it's a fairly obvious and accurate descriptor.

>>105784134
Because it doesn't calculate anything, it just looks up the answer from its training data, if it wasn't there it won't give you a correct answer. Reasoning steps? Those are looked up too. If you give it a step by step algorithm it won't be able to follow it.

>>105766568
cross-referencing known drugs with disease-related molecular signatures

>>105785352
not really because it has no inteligence.
an imitation is supposed to at least share qualities of what it tries to imitate

>>105785362
there is literally 0 inteligence whatsoever.
keep coping

>>105771808
Why are you calling me a schizo?

>>105763675
>AI now regularly finding algorithms for unsolved problems in computer science and improving pre-existing heuristics
I've been waiting for AI to be able to generate hexdumps using AVX instructions for a year now. I haven't revealed my solution to anyone else, so it should've been able to come up with it by now ...

... yet nothing. Because it can only output what someone else has input. AI is nothing but a very, very, VERY expensive search engine.

>>105787860
>>105787865
>iT dOeSn'T hAvE iNtElLiGeNcE
And artificial leather doesn't contain leather, artificial pearls don't contain pearls, you absolute drooling retards.

>>105786831
>deterministic ai
I will cede to this. DAI seems to be a pretty decent concept. LLMs are giving AI a bad reputation.

>>105771808
>A: The Low IQ Summer
>B: The Chad 100 modulo
>C: The Schizo
D: The Autistic Binomial Expansion

>>105771808
E: remembering that 25 * 25 = 625 because it's quite common

>>105771808
desu i'd do:
25 * 25 = 20 * 25 + 5 * 25
20 * 25 = 10 * 50 = 500
5 * 25 = 125
500 + 125 = 625

>>105791057
artificial leather in most part can do what real leather can.
an artificial womb can do what a real womb can.

"AI" cannot do anything inteligence can.

>>105754813 (OP)
After watching multiple [LLM plays gen1 pokemon] streams on Twitch, I've found that context window is a LLM's biggest weakness. Context cleaning = lobotomy.

>>105793875
>"AI" cannot do anything inteligence can.
Obviously it can. AI includes applications much cruder and more basic than LLMs. A simple chess engine is AI. I don't even know how you could come to this conclusion unless you're under the impression that intelligence means the entire gamat of abilities of the human brain.

>>105793875
>>105794081
Playing chess, playing a multiplayer FPS, or writing an email are all activities that a human would do using (actual) intelligence. An AI can also perform these functions (often better than humans) using AI, in this case using a chess engine, a bot, and an LLM. It doesn't matter that they don't utilise real intelligence - in fact that's the whole point and the reason for the word "artificial". The end result is what defines the naming, not the mechanism by which is comes about. In the same way artificial leather is named because of the superficial similarity with real leather, despite the fact its construction and process for making it it's entirely different. It's embarrassing that this analogy is so hard for you to accept and you're so emotionally invested in refusing to accept a well established term.

>>105794081
a chess engine has no ability to learn, reason and understand.
likewise for an llm.

>>105794143
half of the tasks you mention do not require inteligence.

reusing past knowledge doesn't require a lot of inteligence, building it does.
LLM's cannot build knowledge they rely on carfuly curated datasets.

they can't see something, reason, try to understand, compare it to things they know, learn it and have it be reusable later.

learning involves problem solving, you just don't take in randomly what you see, you have to think and reason about it for it to become actual knowledge.

that's why we wouldn't call a book knowledgeable even if you could get a lot of knowledge from reading that book.

LLM's are very similar to book, just more interactive, but they still have no real ability to build knowledge based on experience (a thing they lack as well).

>>105794795
>a chess engine has no ability to learn,
tell it to Alpha Zero

>>105769134
There are 41 cans, sudoku yourself.

>>105794804
still not learning

> they can't see something, reason, try to understand, compare it to things they know, learn it and have it be reusable later.

> learning involves problem solving, you just don't take in randomly what you see, you have to think and reason about it for it to become actual knowledge.

You guys know about function calling, right? You can use an existing framework or have it write one for you. Mine hijacks its output to add function results. If the prompt says something like "use mul(x,y) for multiplication" and you add the result to its output, it knows the correct result. Same works for counting Rs in strawberry, knowing the current date, etc.

>>105794931
it literally learned how to play chess from zero. the only initial data was the rules of the game. no strategies, no openings, no piece/position evaluation, no nothing

>>105795024
it also completely changed the way top players play chess by developing new crazy strategies that nobody has thought of before

>>105795024
it didn't really learn, it's still RL and a bruteforce solution, that doesn't count as learning in my book.

it didn't "think" or reason in order to "learn", it still has no understanding of the game and doesn't think when it plays.

the way it works is still picking moves that are statistically likely to lead to a better outcome, but there is no thinking process or learnt understanding behind that.

>>105795409
>>105795024
and again, it's based on a very narrow curated dataset, even if it's being generated with adversarial playing, it's still not representative of real world data.

the algorithm itself is extremely limited and still has no ability to figure things out.

>>105794795
>a chess engine has no ability to learn, reason and understand.
>likewise for an llm.
It doesn't need to in order to qualify as AI. Same as fake leather doesn't need to be made from the skin of an animal to qualify as artificial leather. The problem here is you seem completely unable to understand that adding "artificial" changes the meaning.
>half of the tasks you mention do not require inteligence.
They do when it's humans performing them. An AI is capable of doing them without intelligence.

Most uni grads can't into recursion either, so what? Ask a junior dev to do tower of hanoi to any order in python and most can't