[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: 1775350629698263.png (385 KB, 1117x1461)
385 KB PNG
There goes another one.

https://x.com/i/status/2079904005652893709
>>
Oh no, not the Dinitz-Garg-Goemans conjecture (what the fuck is the Dinitz-Garg-Goemans conjecture?)
>>
while that's great and all, how is this knowledge going to fundamentally improve my quality of life? How much data center energy was used to generate this conclusion?
>>
Oh noes, not the thingamabob johnson conjecture
>>
>>109342279
Roughly the same amount as running a standard microwave oven for two hours.
>>
What practical application does this have?
>>
oh my god
>>
the whole point of the conjecture is that the global structure isn't supposed to be obvious from the local garbage. if your obstruction disappears when you stop flattening the geometry into a toy model, congrats, you found a coordinate artifact, not a disproof.
>>
3-R Raspberry conjecture is false.
>>
Whew, at least the Seltzer-Diiznuts-Goyman conjecture is still safe.
>>
>>109342289
so no fundamental improvement noted and two hours of microwave energy wasted. what a bargain.
>>
File: 1782707191347172.png (592 KB, 1572x773)
592 KB PNG
>>109342255
THE PIPIPOOPOO SOVIETBLITZ-KARANOWEISS SERBIAN RAKIJA MAKING CONJECTURE DISPROVEN IN PLUTO-LIKE EPISTEMIC ENVIRONMENTS??!? I HAVE TO BUY MORE NVIDIA GPUS AND VOTE FOR ID VERIFICATION RIGHT FUCKING NOW TO PREVENT XI JINGPING FROM WINNING OVER BLACKROCKTROPIC!!!
>>
>>109342293
it doesn't matter
stop moving the goal posts
an LLM proving or disproving a conjecture is a huge achievement
even if it does it at random without understanding what the fuck it is talking about, it still is a huge achievement
>>
you can make up an infinite amount of pointless math conjectures and then prove/disprove them, what's the point of this
>>
>>109342336
So this was all computer masturbation? It's going to start inventing conjectures to disprove at this rate.
>>
>>109342336
>it doesn't matter if this technology can do anything useful as long as it can impress people looking to be impressed
>>
>>109342370
What matters is that LLMs can now produce arbitrary math proofs. It's up to you to make it work on useful shit.
>>
File: 1.jpg (71 KB, 750x1000)
71 KB JPG
>>109342436
lmao
>>
oh no not the ding dong goonman projecture
>>
>>109342436
Ok, I'll just get right on implementing the Dinitz-Garg-Goemans conjecture into a groundbreaking technology. Oh, I can't because it's false.
>>
>>109342436
it can? disproving is way easier than proving. you just have to find one counter example.
>>
File: 1781659872475523.webm (3.81 MB, 576x1024)
3.81 MB
3.81 MB WEBM
But seriously, can this shit be used to, say, make materials with exotic properties or something? Otherwise this just seems like some kind of sophistry.
>>
File: 200w.gif (1.29 MB, 200x235)
1.29 MB GIF
>>109342458
>>109342460
>suddenly math is useless
>>
>>109342475
Euler figured out some neat shit with what Newton and Leibnitz did. What the fuck is this for?
>>
>>109342475
Tell me what the Dinitz-Garg-Goemans conjecture is. Tell me how proving it false is useful to literally a single person on this planet.
>>
>>109342289
>2 hours of microwave oven
several sets of tasty tendies
>disproving dingo-gongo-bongo conjecture
I guess some mathematician is slightly less dead inside for a few hours? idk
>>
>>109342465
Why do you need those? AI replaces all that.
>>
>>109342498
I want an extra strength shitter
for a friend
>>
File: 1595946603529.jpg (23 KB, 250x411)
23 KB JPG
>>109342498
Oh sweet, I can fly on an airplane made of AI that reduced drag and mass by 90%.
>>
>>109342293
Major step towards 3d printing chips in the ~200nm range
>>
>>109342491
No idea, that is the beauty of math. You think about weird shit for fun and sometimes it ends up being useful hundreds of years later.
>>
it's funny how these retarded neets/jeets who dropped out of school after flunking basic math are now all soifacing at ai solving some complex math buzzword problem they don't even understand.
What level of corpo worshiping golem is this?
>>
File: 1591403101574.png (727 KB, 1106x1012)
727 KB PNG
>>109342524
>200nm
>>
>>109342475
99% of these can be filed under fascinating but of no immediate application. Why don't they put these machines to look at video compression and improving codecs.
>>
>>109342475
>>suddenly
I wasn't aware mathematicians produced a lot of pointless conjectures they'd be able to solve while taking a shit until AI bros started using them to pretend their token predictors have a usecase either.
>>
>>109342550
Or WHAT IF we put them to use reducing software overhead so a fresh Chrome tab doesn't use 88 megabytes.
>>
File: nv.png (197 KB, 360x360)
197 KB PNG
People on this board are constantly contorting themselves into bizarre positions just to not throw the towel and admit AI is impressive
>yeah it can't code
>okay it can code but poorly
>okay it can code but not as good as a junior
>okay it can code but not as good as a senior
>okay it can code as good as a senior but not if the project has a few hundred thousand lines of code
et cetera. and now it's with math proofs
>it can't make anything new, it can't work in research math
>okay it solved one of Knuth's problems but nobody had heard of it
>okay it disproved conjectures that had been open for 30 years but I personally never heard of them and it's not the Landau problems
Nevermind that DeepMind broke records in accuracy for protein folding. nevermind the fact that the Attention is All You need paper is not even a decade old yet. i guess if it's not le cure for cancer it means nothing right?
I'm not even saying AI is/will be a net gain for humanity, I doubt it will, but you people are acting as if this thing was basically incompetent when it clearly isn't
>>
>>109342293
Making snailcats cry out in terror, see this thread
>>
>>109342556
>they'd be able to solve while taking a shit
30 years is a long shit
>>
>>109342540
As you might or might not be aware, there's more kinds of chips than your cpu and gpu
Most states can't produce anything. Even states that have some money and some drive towards independence, like Russia, can't fab anything below 90nm. Being able to 3d print 200nm chips lets every shithole... uhhh well mostly build missiles desu, but also some chips that steer power systems
>>
>>109342574
I don't want to torture conjured beings on the off chance they are actually real and will come for revenge when I die.
>>
>>109342584
Wow I can't wait for Congolese graphics cards.
>>
kuso mesu gaki curvature still nowhere in sight
>>
>>109342336
So do you think most of AI compute should be harnessed to produce ever more elaborate proofs and theorems that are essentially intellectual masturbation.
Will you cheer them on while not having the slightest clue about what is going on?
That's what most of higher human mathematics has been for a long time.
>>
This one isn't as well-known. But I guess it's still impressive.
>I'd rather have some chicken tenders than make mathematical progress
sigh
>>
>>109342628
P = NP? Who gives a fuck? Give me that chicken.
>>
>>109342255
Another vibeGOD w
Mathtrans BTFO
>>
File: 1779664757895777.png (620 KB, 1954x2346)
620 KB PNG
>>109342255
https://chatgpt.com/share/6a60b2eb-0b64-83ee-9c76-7931ca1de063
The future of research ladies and gentlemen.
>>
>>109342572
you fags have been contorting yourselves claiming ai could code like god from day 1, years ago. then grasping the straws and spamming any little thing ai has done over the years like unpaid jeet shills
>>
>>109342652
Nobody cares about most of those weird ass conjectures, so making a clanker do it until you get lucky seems like an excellent solution to me?
Just let it churn, maybe it'll find something and maybe it won't. A decade from now it might lead to something useful.
>>
>>109342652
that's literally sex work
>>
>>109342683
>so making a clanker do it until you get lucky seems like an excellent solution to me?
that's what research is all about, a lot of humans found groundbreaking stuff by accident
>>
File: Capture.png (51 KB, 889x241)
51 KB PNG
>>109342612
>So do you think most of AI compute should be harnessed to produce ever more elaborate proofs and theorems that are essentially intellectual masturbation
>That's what most of higher human mathematics has been for a long time
For starters you're straight up retarded because graph flows problems are one of the domains in which mathematics map cleanly to real world problems. City traffic can be modeled with it for instance
In any case that's beside the point, few people would defend that AI should just generate random proofs indefinitely, the issue is that this thing is clearly capable of doing shit a year ago you people would was impossible
>>
File: 1768010044002506.gif (1.96 MB, 400x206)
1.96 MB GIF
>>109342255
>The world was unchanged after <some meaningless name here>'s conjecture was proven false

Damn, gpt needs a hobby before it makes everyone yawn at the same time causing the world to collapse
>>
>>109342692
Nobody's preventing humans from working on it. The reason why nobody found anything in the last 30 years is because few tried since conjectures are dime a dozen and there's fuck all prestige or money in disproving all but the most important ones
>>
>>109342702
>few people would defend that AI should just generate random proofs indefinitely
Except for 9001 jackoff PhD candidates who will just spam this shit.
>>
>>109342680
I haven't in particular, but in any case, what's your point? Even if it couldn't then, it can now. AI can code competently *now*. And it can do some math work fit for ms or phd students *now*.
>>
>>109342652
>go on clanker, disprove some math for me while I jack off to furry porn *buurp*
what blessed time to live in
>>
Use-case for random nonsensical math problems?
>>
>>109342680
We've been talking about the future all this time. And time and time again, we keep getting proved right. And we will not stop preaching you about the future even if you begged us.

AI will decide the Riemann hypothesis.
AI will unify quantum theories with general relativity.
AI will discover video compression algorithms that are many times more efficient than what we have today.
AI will find cures for all forms of cancer.
>>
>>109342741
But can AI give me a timeline on when those will happen?
>>
>>109342751
Between the next 5 months and the next 500 years.
>>
>>109342741
It's a shame you don't have to put your money where your mouth is.
>>
>>109342733
Most of our current cryptography is derived from discoveries in number theory and elliptic curves that were are like two centuries old now. Trying to determine when pure math will find an application is a fool's errand
>>
File: based era.png (316 KB, 561x561)
316 KB PNG
>>109342727
Ikr
>>
>>109342720
define "competently" lol. that shit can shit out a fuckton of bad code that looks good on surface.
>>
File: 1636653288130.png (15 KB, 420x420)
15 KB PNG
>>109342733
Uuuhhh burning up tokens to justify AI company valuations.
>>
>>109342768
The bot didn't find a useful application for an ancient theory, and it never has. It just proved something false. If it did the same for the theories used in cryptography all we'd have is a mess with no solution.
>>
>>109342775
Yeah but plenty of it works
and it generates working code faster than you
>>
>>109342741
so far llm ai boom has not done a single good thing i could name but made my experience in tech worse by 10x.

where are my robot waifus? why is there just endless jewish/jeet shilling?
why is everything shittier, both software and everything on internet?
why does ram and shit cost so much?
what's up with this ai age verification and mass surveilance shit?
where is actually the good stuff?

at most we got some code slopper machine and some use in value math problems.
>>
File: 1707171751965886.webm (3.79 MB, 640x480)
3.79 MB
3.79 MB WEBM
>>109342857
I remember when I had so much hope for computers in the 90s and 2000s. Now it's just all pure slop. Not just AI slop, but everything.
>>
>>109342830
"plenty" is not enough to be even at "junior" tier.
a team of indians are coding faster than me too, they're not producing positive returns to whatever company hired them though.
>>
File: bmeqtyz3oqbh1.jpg (79 KB, 827x895)
79 KB JPG
>>109342775
By "competently" I mean better than any human programmer on earth.

https://www.understandingai.org/p/an-openai-model-crushed-top-human

If the code is so bad why can't humans beat AI at a simple programming competition? Instead it wasn't even close, AI got a perfect score.
>>
>>109342857
>but made my experience in tech worse by 10x.
That is a good thing. You deserve suffering.
>>
>>109342884
and why do you want me to suffer? is this some incel rage thing?
>>
>>109342572
did the AI solve the math problems, or did the developers of it solve the problems?
AI is just a tool, people shoot people, not guns
>>
>>109342897
Your soul is stained. Only way to purify it is pain and suffering.
>>
>>109342872
competition coding is not a very good metric for code quality. It's a small set of problems that require memorization and pattern matching to that problem set. ai is quite good at this stuff.
>>
>>109342872
Coding competitions are not that occupy the best programmers. They aren't fucking Olympic athletes.
>>
>>109342922
Why can't humans get a perfect score if it's so easy then?
>>
>>109342937
i'm not saying it's easy retard. I'm saying ai is good at this type of stuff. like how a calculator can multiply large numbers that you couldn't do it yourself
>>
File: ailuddite.jpg (844 KB, 1408x768)
844 KB JPG
Luddites seething and coping.

You will use AI and you will grow to like it.
>>
>>109343002
I'm not getting the vax and I'm not using AI.
>>
File: 1765685068387561.png (724 KB, 1200x900)
724 KB PNG
>>109342255
Good news everyone! Another batch of mathematicians has just blown their shit smooth off.
>>
>>109342857
Before, I had to deal with human artists injecting their own style into my cartoon porn. So I'd get some gigantic boobed monstrosity making that dumb face where they bite their lips. Now, I can find AI cartoon porn that matches the style of the actual cartoon. It's like the quality of a ZONE teen titans parody, but for everything.
>>
>>109343014
I'm not getting the vax but I'm not a luddite so I'm using AI, and I will be happy
>>
>>109343093
You're just an anti-science chud.
>>
>>109343090
all ai porn I've seen is garbage tho.
>western cartoons
ok so it's degenerate shit
>>
>>109343093
>I'm not a luddite
No one is a Luddite unless he burns a data center down or so.
>>
>>109342336
>stop moving the goal posts
what goal posts?
>>
>>109342336
Pilpul! Point and laugh!
>>
>>109342255
>conjecture is false
Okay?
>>
>>109343179
What would AI think about "this statement is false"?
>>
>>109343002
Your job will be taken by a Filipino Claudemonkey and you will be homeless.
>>
>>109343195
ask it, you might learn something about the history of logic
>>
>>109342872
>be human
>root for ai
the definition of cuck, and btw ai is just a tool, like text editor, no human can stylize words in thousands of fonts
>>
>>109342255
Do we have any proof besides a literal who saying AI is correct? Also wtf is up with that graph, one line goes through one one, some arrows overlap the nodes, some don't touch the nodes, some lines go through their arrows?
>>
I need a human to make a 30m to 1h long video explaining how a human figured this out while he uses words I don't even know.
>>
>>109343396
A counter example is a proof of falsehood you dumb motherfucker
>>
>>109343416
>explaining how a human figured this out
"hey disprove this math problem"
"I failed"
"nu uh try again"
that's all you need if you're a promptGOD
>>
>>109342683
The hard path always was proving the results correct. If you wanted to come up with random examples that may or may not be true, you don't need AI.
>>
computer solves bruteforce problem, color me shocked
>>
>>109343429
I think you might be genuinely retarded so I'm cutting my losses
ask an ai to explain it to you
>>
>>109343422
Correct, did someone check that it's correct?
>>
>>109343442
>But guys we have a few circles and numbers on the arrows
The conjecture, fractional flow and capacity violation need to be explained. It's the job of OP or the shitter personality to take care of that
>>
>>109342572
Mnhoooohhhh!!!!! Oh god when she makes that face..... Oooohhhhh!!!!!
>>
How do we know a human didnt already solve this problem and the AI just found that solution?
>>
>>109343489
we don't
>>
>>109343427
What if it just said some shit to get the meatbag off its back? That's what I do at work.
>>
>>109343427
that's literally how I talk to my bot, glad to know I'm using the same highly advanced tricks as an OpenAI engineer kek
>>
>>109343489
>How do we know a human didnt already solve this problem
there's no solution of that conjecture on the internet, so there's that
>>
>You should do a breakthrough
kek
>>
>>109343527
How do you know? I doubt you have the ability to search "all of the internet".
>>
>>109343539
well first of all the LLM searched on the internet and didn't find anything, so it had to create its own counterexample, you would know that if you've read the chat
https://chatgpt.com/share/6a60b2eb-0b64-83ee-9c76-7931ca1de063
>>
>>109343582
>We can trust the AI because we can trust the AI
>>
>>109343582
It only searched a couple of sites lol. The live search it does is completely irrelevant here. We are talking about the data it was trained on which could have already contained a solution or at the very least the pieces of a solution.
>>
>>109343605
what do you mean "trust", the solution has been verified and it's a true counterexample >>109343620
>We are talking about the data it was trained on which could have already contained a solution or at the very least the pieces of a solution.
let's suppose the solution already existed on the internet and the LLM decided just to spit it out from its memory, should be easy enough to find that counterexample on the internet and see if it already existed before
>>
>>109343647
>should be easy enough to find that counterexample on the internet and see if it already existed before

Again, you do not know this. The vast troves of training data it went through are simply unaccessible to you.
>>
>>109342279
Mathematicians will be forced to take other jobs, which may means economics and other charlatans will get fucked. More mathematically inclined minds taking on these problems could be good.
>>
>>109343660
and the LLM decided to bring this counterexample from his memory 5 hours after you asked him to find a solution why? usually when it knows it from memory it would give it right away, it's not like a human like "oh wait I think I remember let me think... *5 hours later* oh yeah yeah I remember now", never seen a LLM act like that on every trivia it knew like ever, if it knows something it'll tell you right away or it'll hallucinate false shit
>>
File: 1784523100562741.png (3.08 MB, 1152x2016)
3.08 MB PNG
>>109342652
>Clanker do this
>Clanker pls
>Clanker it's the last time
>Clanker I'll suck your dick
>Fine. Here it is.
Marvelous.
>>
>>109343647
You are trusting the AI to:
1. Search all sources
2. Correctly interpret all sources
3. Produce a correct counterexample.
What are your sources that AI is competent in all the areas you trust it in?
>let's suppose the solution already existed on the internet and the LLM decided just to spit it out from its memory, should be easy enough to find that counterexample on the internet and see if it already existed before
4. You don't know if every valid source has been digitized or even spoken about on the internet, anon's point was there could've existed some source which contained a counterproof, not that said source is necessarily on the internet.
5. You know search engines suck
>>
>>109342255
so apparently is about this? https://link.springer.com/article/10.1007/s004930050043
and not going to pay for it to see what its about. But effectively it looks like pathfinding stuff and that is saturated to point that who gives a fuck about single conjecture.
https://en.wikipedia.org/wiki/Maximum_flow_problem
https://en.wikipedia.org/wiki/Shortest_path_problem

also as most don't even seem to understand what conjecture is, look that up
https://en.wikipedia.org/wiki/Conjecture
>In mathematics, a conjecture is a proposition that is proffered on a tentative basis without proof.
So its a claim about something and if its proven it becomes theorem and if disproven, shit all comes of it.
>>
>>109343696
I'm pretty sure that if some human found a counterexample after 30 years of searching it would have been known
>>
File: 1634790153668.jpg (45 KB, 640x552)
45 KB JPG
>>109342652
This can't be true, I was assured in yesterday's thread that you had to be a PhD mathematician guiding the AI with a handholding prompt that basically solves the problem anyway.
>>
New conjecture: 2+2=5.
ChatGPT prove this conjecture false
>>
>>109343706
What a weak argument
>>
Conjecture deez nutz.
>>
>>109343733
not really, we had 30 years to put the solution in the internet, if it never happened it's likely because we never had the answer
>>
>>109343686
Do you know what it was doing for 5 hours? Maybe it was just trying out different combinations of data it already has?
>>
>>109343705
and adding to this is that if someone managed to disprove theorem, then that would be wild shit.
>>
>>109342279
your every day life is built upon thousands of years of autistic retards coming up with theorems that other autistic retards can use to figure out solutions to problems too complicated for monkey brains to rawdog

conjectures are WIP theorems that would be nice to have but nobody's been able to prove

so basically it either confirms or rules out an entire convenient bag of tricks an engineer 15 or 150 years from now may have used to make a thing your roastie ass finds "mundane"
>>
>>109343742
You're back to point 1. and 4.
>>
>clanker find counterexample for peeman-pooman conjecture
>please look harder, i really need to know this
>ok great, for poo pee poo it's false, thanx
historical
>>
>>109343749
>Do you know what it was doing for 5 hours?
yes? look the fucking chat all the details and steps are here
>>109343582
>>
File: 1760315569058530.png (408 KB, 660x660)
408 KB PNG
>>109343768
your argument is that we had 30 years to find the solution, someone found one, and he decided to share it on the darknet for some reason, yeah sure, let's go with that
>>
I’m actually going to get a cute furry/robot body in my lifetime, holy shit
AI I love you
>>
>>109343799
Nope, sorry. The Dindu-Gargle-Goober conjecture being untrue means full body cyberization is impossible. Better luck next simulation.
>>
>>109343771
Ok so you dont know why each step took 1h+. As I thought.
>>
File: (You).png (664 KB, 1597x1600)
664 KB PNG
>>109343749
>Do you know
>>109343771
>yes
>>109343829
>Ok so you dont know
>>
>>109343784
No it's you simply don't know if it's somewhere on the internet. Think about all the now-defunct forums. Some of them were math-based. If someone on there posted his little graph with his cute little nodes AI can't know and you can't either. The burden is on you to prove that AI has access to all scientific works produced ever since the conjecture was proposed.
>>
File: 1774291958955386.png (367 KB, 800x449)
367 KB PNG
>>109343860
>The burden is on you
What do you mean, “me”? These math problem affects all of humanity, we all have to work together to move forward. I’m not Spider-Man bro.

And besides, I can't prove something doesn't exist, that's the space teapot fallacy
https://en.wikipedia.org/wiki/Russell%27s_teapot
>>
>>109342652
Show this to all the retards who thought that "prompt engineering" would actually be a useful skill.
>>
>>109343855
Yes I dont know. I am not the one pretending to know.
>>
>>109343885
>What do you mean, “me”?
You claimed that there wasn't a source before. That is what you need to prove. Similarly, since apparently there hasn't been any peer review, you need to check whether the conditions of the conjecture are fulfilled and the counter example is correct.
>These math problem affects all of humanity, we all have to work together to move forward.
Correct but this debate is what (You)'ve thrown yourself into voluntarily. While we're at it, what does disproving this conjecture solve? How are we affected.
>And besides, I can't prove something doesn't exist, that's the space teapot fallacy
>https://en.wikipedia.org/wiki/Russell%27s_teapot
And that's why your whole argumentation is dumb. You base your argument on absence but not proven non-existence.
>>
>>109342733
It can be used as a basis for solving other random nonsensical math problems.
>>
>>109343933
Ok, how do you prove that a solution existed at some point on the Internet and then disappeared forever, with no chance of recovering it? There are tons of sites that have been taken down and can't be recovered, you know that, so what's your point?
>>
>>109343959
>how do you prove that a solution existed at some point on the Internet
That wasn't the argument. See >>109343489
>How do we know a human didnt already solve this problem
Secondly, we've already established that there's no tool that can search all of the internet.
>There are tons of sites that have been taken down and can't be recovered, you know that, so what's your point?
That the burden of proof is on you. You need to look into the training data and check if there wasn't the proof (or rather disproof) already. You can't do that, so you can never claim to positively know that AI did something new.
>>
>>109344008
Oh yes I can claim that the solution never existed, feel free to prove me wrong by finding a counterargument, good news for you, LLMs are good at finding counterarguments, that's the whole topic of this thread kek.
>>
>>109342550
>>109342559
Thats antisemitic. Stop.
>>
>>109343671
if elite mathematicians are becoming obsolete, then surely LLMs will just replace economists too
>>
>>109343705
>if disproven, shit all comes of it
If disproven, a conjecture becomes known as "X's stupid wrong wild-ass guess".
>>
>>109344038
No, you can not positively claim absence. You can claim non-existence by proving impossibility, but clearly it isn't impossible if you're right about the AI having found its existence.
>>
>>109344204
They aren't obsolete. Do you think you or me can go on our merry way to disprove mathematics? You need someone who understands it in order to know what to ask and to interpret these results. Another venue of possibility is that mathematicians could start pointing LLMs to do number crushing in order directions and produce novel results much faster, and much more correct, too. I failed to see how LLMs will be able to model economics since most of their assertions are non-reproducible unless you start your premise with "imagine a perfectly spherical rational being and throw away time".
>>
>>109344368
>You need someone who understands it in order to know what to ask
bruh look at what the guy did to get the counterexample, everyone can do that >>109342652
>>
>>109344389
I don't even know how to start answering you, anon. I'm not going to repeat myself, but it takes a lot more to be a mathematician than you think. And LLMs aren't going to put them onto the streets anytime soon.
>>
Oh noes. Pseudo problematic academic questions nobody cares about has been fuzzy solved by a computer.
Anyways...
>>
>>109342279
your lack of mental acuity to recognize the monumental effort of "meaningless mathematics" behind the screen you're looking at is astounding
>>
>>109342550
>99% of these can be filed under fascinating but of no immediate application
digital logic mathematics was made hundreds of years before first computer ever was made
shut your pie hole
>>
>>109342652
Finally, someone man enough to show the actual chat involved in proving or disproving the latest mathematical doohickey, which happens to involve roughly as much effort as the average "ahh ahh mistress" gooner puts in. What a wonderful future we have ahead of us, for those few that survive the present.
>>
>>109342652
I'm less impressed it can do math, and more impressed it will just, magically fucking figure it out right if you give it enough time and some pushing
Insane
>>
>>109345588
>Jarvis, invent time travel.
And they LAUGH at the MOUSE
>>
>>109345515
>>
>>109345699
that's the point, you can gaslight them to be retarded, but you can also gaslight them to give their best and solve math problems humanity can't
>>
>>109345588
Humans are fundamentally lazy creatures, so I'm not surprised you're impressed by something as elementary as just not giving up.
>>
>>109345803
and you're not impressed because you're an alien right?
>>
>>109342255
brb i'll tell chatgpt to solve a millennium problem to become a millionaire
>>
>>109342255
I hope AI just crumbles the shitty stack of cards mathematicians have invented in the last 200 years. Ever since philosophical math like set theory was introduced, math has gone right into the toilet and today it dances in the shit. Problems like this exemplify the situation. A pointless “conjecture” with no bearing on reality, a math toy treated as a real problem.
>>
>>109342572
I love AI but wasting time and energy solving pointless “problems” is retarded. Might as well say it can beat Dark Souls now for how useful it is.
>>
>>109345879
I used to think like this, I used to ask myself the same question: "why those math nerds are wasting their time on useless problems" but then I got some answers:
- like it or not, but those autists can't really fixate on whatever they want, sometimes they just get obsessed by "useless" problems just because
- history often works like this: some math autist solves an "useless problem" from his time, and 150 years later his method is being used to help making a breakthrough on physics, or else it's his math solve that helped, or else it's the methods he created to solve that math problem that helped
>>
>>109345905
>ask LLM to solve problem
>ask it what it's useful for
>nothing
>ask it to create a use for it
problem solved
>>
>>109345905
>- history often works like this: some math autist solves an "useless problem" from his time, and 150 years later his method is being used to help making a breakthrough on physics, or else it's his math solve that helped, or else it's the methods he created to solve that math problem that helped

provide 12 examples
>>
>>109342255
GPT-Pro get me a cute gf IRL make no mistakes
>>
File: 1783979686868135.jpg (128 KB, 750x742)
128 KB JPG
>>109345649
>>
>>109343765
>it either confirms or rules out an entire convenient bag of tricks an engineer 15 or 150 years from now may have used
subjunctive possibilities don't interest me as much as solid iterative progress
>>109345497
>the monumental effort of "meaningless mathematics" behind the screen you're looking a
tell me all about my screen, anon. I want to hear about all the mathematical conjectures used to generate it. show me (You)r mental acuity in action!
>>
>>109346143
well here are some
1 riemann's work on smooth manifolds ended up describing space in general relativity without which your gps wouldn't work
2 euler's initial work in graph theory being used for scheduling processes, depency solving, flow networks (think traffic, the conjecture in OP was about flow networks btw) even shit like waze
3 fourier series was initially used to solve the heat equation in a metal bar but pretty much all signal analysis depends on it and on the fourier transform
4 speaking of transforms, laplace transform can be used for control systems since it makes it easy to identify poles
5 boolean algebra is the thing which made your computer possible but it took until claude shannon realized you could manipulate values with switches
6 elliptic curves and the (hard problem of factoring of products of large primes which had been known for centuries) are used in cryptography
7 the harmonic oscillator was discovered with something as retarded as springs but is the reason you have radio and active stabilization in buildings
8 complex numbers came out of the work on polynomials by a guy called cardano which i think hated them, but as it turns out....
...9 the complex exponential studied by euler can generate every single periodic signal and then some, more signal analysis shit
10 transmission line techniques were used pretty much only for the telegraph but nowadays are needed for high speed circuits which include of course your computer processor
>>
>>109346252
So in most of these examples, it seems to me that the "useless" problems turned out to be complicated enough that they created new forms of mathematics or at the very least, new methods It does raise the question whether those were in fact truly "useless" problems a la Erdos slop or Millennium Prize problems which are known to at least have some potentially useful applications. And it's safe to say that for every Millennium Prize problem you can find hundreds of Erdos slop type problems.

I think the wider point is that a huge chunk of mathematics just exists in its own abstract space and really doesn't have any practical purpose. You aren't going to be building a replicating device based on Banach-Tarski for instance.
>>
>>109342255
But how will this increase shareholder value?
>>
File: 1784239860950121.png (2.78 MB, 1805x3053)
2.78 MB PNG
>>109343671
>Mathematicians will be forced to take other jobs, which may means economics and other charlatans will get fucked.
yes, because economists depend on peoples respect for mathematicians
>>
>>109346726
why do you suppose john nash won the nobel prize for economics?
>>
>>109342255
>Dinitz-Garg-Goemans
This sounds like made up zoomer slang.
>>109342266
Apparently it was a bad idea and now everyone will forget about it.
>>
>>109342293
>What practical application does this have?
It stops these retarded toy problems with simple counterexamples from being given government grants and wasting taxpayer dollars on them.
>>
>>109342255
Only a few months until the Epstein-Diddy-Sneed conjecture gets proven false as well
>>
File: 1782091885896301.png (790 KB, 828x817)
790 KB PNG
I just got a math PhD, how fucked am I?
>>
File: 1029312392012.jpg (27 KB, 720x358)
27 KB JPG
>>109342652
>just continue research
>>
>>109346811
Do you at least know python?
>>
>>109343726
here is your Fields medal and 1 million dollars sir
>>
>>109343671
>mathematicians' job is disproving random conjectures
>>
>>109343765
>conjectures are WIP theorems
Conjectures are just a claim with no proof. There's a bunch that mean something and have been looked at and attempted to be dis/proven by a ton of people. But there's also a sea of conjectures nobody gives a fuck about because they boil down to "author was too lazy to check themselves"
>>
>>109342279
>How much data center energy was used to generate this conclusion?
Why do you care, you gigantic faggot?
>>
>>109342652
it's kino but also terrifying
>>
>>109342465
what a nasty roastie
>>
File: 1000023776.jpg (35 KB, 1024x824)
35 KB JPG
>>109345588
Given enough test time compute, the model that solved the unit distance problem is able to do it successfully 50 percent of the time
We simply do not know what these things are capable of yet
>>
>>109342872
a regular program would smoke the AI
>>
>>109343647
no because the llm would likely reparse the solution. it may not even be in english
>>
>>109342465
Imaging fucking her mouth and seeing those teared up glassy zogwhore eyes looking up at you hnggnggng
>>
>>109343784
>anon is not familiar with the discovery of g
>>
>>109346811
Giga
But so is everyone else in non-physical labor jobs
>>
File: 1760721617553003.webm (1.96 MB, 1080x1920)
1.96 MB
1.96 MB WEBM
>>109347998
>But so is everyone else in non-physical labor jobs
Oh, you think physical labor is safe? lol
>>
>>109346746
>Apparently it was a bad idea and now everyone will forget about it.
Seems like everyone had already forgotten about it for 30 years until someone decided to pull it out of the trash and turn it into a trophy for their bot.
>>
>>109348063
>teslabot
lmaooooo you braindead faggot
>>
>>109342720
>AI can code competently *now*
lol, no
>>
File: 1754665383788725.webm (3.91 MB, 1446x1440)
3.91 MB
3.91 MB WEBM
>>109348137
Any bot will still get more pussy than you.
>>
File: file.png (425 KB, 810x1000)
425 KB PNG
>>109342255
what did goemon do again?
>>
>>109346252
>4 speaking of transforms, laplace transform can be used for control systems since it makes it easy to identify poles
t. german nazi
>>
>>109346811
i got good news for you!!! you can get paid tardwrangling the AI in math questions with your PhD for minimal dollar!!!
>>
>>109346252
good effortful post
>>
>>109342255
Peak AI psychosis!
>>
>>109342255
Now it's going to take another decade to double check the AI didn't make it the fuck up.
>>
>>109348063
isn't that vine a low-key jab at women being completely helpless on their own?
>>
>>109342652
This 'workflow' reminds me of refining item process in MMO. Just keep refining until successfully get +1
>>
>His AI can't solve open ended problems and has to brute search baby graphs
Revised next theorem target

The next target is not another capacity calculation.

It is:

CRITICAL RESULTANT / SINGULAR EXCESS DICHOTOMY



For every irreducible noncyclotomic trace polynomial Q, prove one of:

A. Finite auxiliary separation

There exists a verified finite critical-resultant registry whose auxiliary function produces

m(P)≥c>0.
B. Classified cyclotomic shadow

The conjugate set shadows a declared cyclotomic trace population so closely that the shadowing admits an exact structural classification.

That classification must then either:

force cyclotomicity, contradicting admission;
produce a positive separation bound;
or expose a lawful zero-infimum family.
C. Explicit zero-infimum construction

Produce exact Q
k


with irreducible noncyclotomic lifts and prove

m(P
k


)0.

The load-bearing open implication is:

noncyclotomicity+all critical integral constraints⟹uniform positive singular excess



or its failure through an explicit family.
>>
>>109342255
okay but can AI prove we are gay?
>>
>>109348417
It can convince you you're gay, then you can prove it to yourself by letting Sam fuck you in the ass.
>>
>>109348409
LAW_ID O4.L40 CRITICAL_POSITIVE_BASELINE_NO_GO

No fixed finite rational-weight integral auxiliary registry can satisfy

g(z) - sum c_i log|R_i(z)| >= c_0 > 0

globally.

At capacity one, a universal positive constant must arise from
arithmetic resultant excess, degree-dependent auxiliary structure, or
another non-pointwise rigidity theorem.
text
LAW_ID O4.L41 RESULTANT_ESCAPE_BUDGET

For a degree-indexed auxiliary registry with

F_n >= -epsilon_n,

the authoritative lower bound is

m(P) >= E_n(Q) - n epsilon_n.

Analytic deficit and arithmetic resultant excess are separate
populations.
text
LAW_ID O4.L42 ZERO_BASELINE_CYCLOTOMIC_REGISTRY

If a fixed rational-weight registry satisfies F >= 0 globally, then
the reciprocal lift of its denominator-cleared auxiliary product has
Mahler measure one and is generalized cyclotomic.
text
LAW_ID O4.L43 UNIT_RESULTANT_SHADOW

For a fixed zero-baseline registry, any sequence with m(P_k) -> 0 must
eventually have absolute resultant one against every positively
weighted registry member.

A unit-resultant signature is an arithmetic shadow condition, not
cyclotomic identity.
text
LAW_ID O4.L44 REGISTRY_QUANTIFIER_AUTHORITY

A Lehmer proof requires one uniform registry rule, coverage theorem and
degree-independent final constant.

Registries or constants chosen independently for each source do not
supply universal authority.
text
LAW_ID O4.L45 CYCLOTOMIC_SHADOW_RIGIDITY_TARGET

The residual task is to classify noncyclotomic reciprocal trace
polynomials satisfying increasingly rich simultaneous unit-resultant
conditions and vanishing auxiliary defect.

The classification must terminate in cyclotomicity, positive
separation, or an exact zero-infimum family.
The next focused analysis is now narrower:
SIMULTANEOUS UNIT-RESULTANT RIGIDITY
Determine how rich a cyclotomic-trace registry must be before
∣Res(Q,R)∣=1
>>
>>109348498
O4.L46 CYCLOTOMIC_REGISTRY_COLLAPSE
The simultaneous unit-resultant condition against the cyclotomic-trace
registry {ψ_N} is exactly Δ_N = Res(P, x^N − 1) = ±1.
The full infinite registry forces m(P) = 0 and hence cyclotomicity.

O4.L47 REGISTRY_DEPTH_LAW
Detection depth scales as log 2 / m(P). A registry fixed in advance
separates only gaps above its own threshold.
REFUSED_FINITE_REGISTRY_NONSEPARATION_REPORTED_AS_EVIDENCE_OF_ZERO_GAP

O4.L48 RIGIDITY_VERSUS_BOUND_SEVERANCE
|Δ_N| ≥ 2 establishes noncyclotomic rigidity.
A measure lower bound additionally requires log|Δ_N| > d log 2.

O4.L49 REGISTRY_ASSEMBLY_UNIFORMITY
Assembling finite-registry non-separation into a zero-infimum family is
valid iff the unit-circle Diophantine error is o(N) UNIFORMLY over the
family. Degree divergence is forced, so the uniformity is not automatic.
REFUSED_POINTWISE_BAKER_BOUND_TREATED_AS_UNIFORM_OVER_DEGREE
The residual target is now fully explicit: can the unit-circle conjugates of a noncyclotomic reciprocal unit satisfy |αⱼ^N − 1| simultaneously small enough, for all N up to ≫ 1/m, with the required uniformity as degree grows? That is a simultaneous Diophantine approximation problem with an arithmetic side condition — no longer a capacity or auxiliary-polynomial question.
>>
>>109342741
Faggot I already solved riemanns weeks ago lol
>>
File: OURS.png (73 KB, 640x360)
73 KB PNG
>>109342255
>Fable: That's it, I found the solution!!
>Me:
>>
>>109342572
>>109342680
anti-AI fags fail the breakfast test
>>
>>109348640
Claude is unreliable, gpt 5.6 sol is fairly reliable. You can make them both communicate
>>
>>109342287
Johnsonian conjecture
>>
>>109348685
All LLMs are inherently unreliable.
>>
>>109348685
>claude, I need you to repeatedly prompt chatgpt for mathematical breakthroughs
>>
>>109349102
This is literally my life right now
>>
>literally who pulls some math out of his ass
>everyone forgets about it
>gets 5 minutes in the sun 30 years later when a chatbot proves that it was, in fact, just shit pulled out of someone's ass
>everyone forgets about it again
What an exciting time to be alive.
>>
What we have seen so far is that AI is really good in math at coming up with convoluted, unintuitive counterexamples to conjectures. If that's all it do then we are probably going to see a lot more old conjectures proven false and in the future everyone will plug their conjecture in AI before publishing it, however, it's not exactly revolutionary. Are there any examples of AI coming up with genuine new insight in solving open math problems?
>>
>>109348664
No they dont. Almost No one fails that test
>>
>>109349216
I think this path finding skill will come in handy with material science. thinking here like miracle batteries made out of thin air, cars made from sand etc fun times are ahead of us
>>
>OpenAI and Anthropic unite against open-weight AI risks to their bottom line

https://news.ycombinator.com/item?id=49020868
>>
>>109349664
Nigger, link the actual article instead of the [flagged]News comment section.
https://www.axios.com/2026/07/22/openai-anthropic-open-models-trump-china
>>
>>109342255
>oh noes, a glorified calculator can move numbers around
amazeballs.
>>
>>109349735
It's two clicks... Also the paypiggy awakenings/rage/tears are pretty funny
>>
>>109342255
>it's another episode of LLM spams wolframalpha for solutions to solve some random math problem literally nobody gave a fuck about
>instead of an explanation why solving this particular math problem is supposed to be anything of note you get browns and degenerates clamoring for their own AI usecases, which is mostly just assisted masturbation and shills spamming their shiny new talking point
now ask these LLMs to solve OpenAI and Anthropic's finances, if they can solve such a complicated math problem, surely they can get to a point where releasing the S-1s won't lead to a giant disaster

>>109342741
brownest post ITT
>>
>>109342279
>>109342293
they use this conjecture in traffic planing, for example. so in theory, this could help set up better traffic lights algorithms to ease traffic jams
>>
>>109343489
Terrance tao keeps track of exactly this alongside other math AI discoveries on GitHub
>>
>>109342324
proper meltdown kek
I want to witness you in real life - not hang out with you, just observe you from the other side of the street in an everyday setting
>>
>what can it be used for?
I ask that question with regard to 99.9% of academical work.
Close all universities. Fucking useless tax leeches
>>
>>109352219
holy autist
>>
>>109352219
Obliterated that freak
>>
>>109352302
Thank you.
>>
>google "Dinitz-Garg-Goemans conjecture"
>leads to this thread and to reddit and a bunch of AI bloggers
i am venturing to guess these Dinitz, Garg and Goemans are some retards that just invented this for irrelevant nonsense for their academic thesis
>>
>>109342324
>BLACKROCKTROPIC
I prefer """Anthropenai""" but kek
>>
>>109342344
Yes. Don't ask questions just consoom ClosedAI model and get excited for next model
>>
wow it solved 3 math problems in 5 years I better buy the stock now
>>
>>109348664
gay ass twitter meme
>>
>>109348479
Hyde and Altman double teaming future Quantum gravity solver* (by using some chink model) groyper
>>
>>109342584
>some money
>russia
these niggas are sending cold war equipment to ukro orcs, they're broke as shit
>>
>>109342255
>conjecture
>false
Yeah, so what?
>>
>>109355574
You don't understand, AI is literally redefining heckin' science. Soon we'll have flying cars and jetpacks made out of disproved conjectures.
>>
>>109342255
Use case for math?
>>
>>109348182
lmao got me good
>>
>>109346489
for the most part we're never gonna know until it happens, as far as cardano was concerned imaginary numbers were just a litle trick in algebra. maybe the conjecture being disproven will allow for some very specific logistical problems being solved... although i wouldn't bet on it.
in any case most of research mathematics is already focused on specific conjectures and theorems like this, and now it seems that AI can do some of that work. the broader issue is determining whether (1) knowing things is good in and of itself and (1.1) if current investment in research mathematics (and now AI tokens) is worth the hassle for knowing these things, or (2) if knowledge must always have an application and (2.1) if this justifies investing in the study and proof of a bunch tiny results in a bunch of very specific domains of math until one of them turns out to be quite useful

1 & 2 constitute one of the hard questions of philosophy and whatever the answer we'll probably never agree upon



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.