I'll give you a free (you) if you can come up with a math puzzle that an average college grad with a math degree can solve within 15 minutes, but the top AIs cannot (gemini-3-pro, gpt-5-high [1] ). No images. No spelling gotchas. Normal math that would be appropriate for a textbook.You can run these models for free on https://lmarena.ai (choose "Direct Chat") Post a screenshot of these models failing if you can come up with such a puzzle.[1] Seems to be smarter than gpt-5.2-high
>>16870981It's been chugging for about 5 minutes already, OP.Not looking good.
>>168709875.1 is dumber than both 5 and 5.2. I tend to trust 5 a bit more. Anyway, if it gets stuck (happens rarely -- it's a free service, low priority), click "New Chat".
30 seconds
>>16870993>>16870987
>>16870989It's too long to paste here, but 5 worked with the problem I gave it. It wasn't a very hard problem, but required a little bit of logical thinking. I'm impressed.
>>16870993The trick is to keep a spare slider at the end and increment it each time you do a move.
>>16870993H...hot
>>16871000These are really cool, you can also multiply and divide.
>>16871001chatgpt answerhttps://ctxt.io/2/AAD48bfXFQ
>>16871001AI makes these obsolete.
>>16871002Yeah it's definitely got the right ideaThe only thing I'd change is you can use the spare sliders themselves to keep track. As you keep incrementing, the number you're working on keeps diminishing and the result keeps taking up its place
>>16870981ai bros desperately want to be relevant
>open book>contains knowledge>bro knowledgists are cookedthis is you rn
>AI is fake>t. cannot even come up with 1 college-level math problem that AI cannot solveI'm never returning to this shit site
>>16871152Fuck off shill, I'm not training your chatbot.
>>16871152This goes both ways though , why hasn't anyone posted an example of it.That likes AI.
>>16870981>Calculate the number of r's in the word redistributioner.
>>16870988Average college grads would not solve this, but AI can. We need the opposite.T. Tao (the greatest active mathematician) already stated that o1 or o3 is like a "mediocre PhD student", which puts it above college level. And o1 and o3 are ancient history at this point.
>>16870995>I'm impressed.>>16871010>Yeah it's definitely got the right idea>>16871199>Calculate the number of r's"No spelling gotchas". Incidentally, some AIs will get this.Anyhow, I declare AI the winner of this here challenge.
>>16870999The machine agrees with you.
>ai, figure out how to make me lots of money with you in a way no one else will
>>16871360Why is the spelling so tremendously difficult for them?
>>16871393Because they don't "see" letters, they see tokens (chunks of words)See: https://tiktokenizer.vercel.app/
>>16871399Interesting, thanks for the explanation. It makes sense.
>>16871399at a high level you know this is bullshit its some black box controlled by pooja
>>16871000Flawless AI victory!!! Now ask a math major the same question. He/she will be stumped.
Prove an axiom
>>16871360
>>16871591you are retarded as shit the whole holistic nomenclature phenomenon within syntactical axioms or knowledge axiomatic premises is that theyre self evident so youd prove it by restating the underline presuppositional bullshit all over again so people can understand holy fucking shit you are stupid as shit anon
>>16871623using ai to solve topology is hilarious lmao
>>16871631A match made in heavenA worthless technology with a worthless intellectual discipline
>>16871399This is bullshit of course. With the amount of training they have had on all types of text you can think of, the models can easily associate each token with the actual characters they represent. Indeed, if you tell them to spell out a word character by character, they'll manage that perfectly fine. And after that they'll usually be able to count the letters correctly.I haven't heard an actually convincing reason why their success rate at it is so bad.
>>16871687>the models can easily associate each token with the actual characters they representThe problem is not them being incapable of associating each token with the actual character, it's them associating those tokens with other things.
>>16871623What's the correct answer? Do you have a general expression for all such vectors?
>>16871623
>>16871720That is less wrong, but still wrong, because you've tried to push it the wrong way. >>16871715>What's the correct answer?I have the correct answer and the general expression, but I won't post it until OP provides either a screenshot of the correct answer as an LLM's output or the free (you). If I give it before that then the mfs can just prompt-engineer the AIs to give the right answer, since the mfs conveniently don't bother to screenshot the prompt, see >>16871720. To prove than I am not full of shit I am attaching the three polyhedrons that >>16871720 did figure out correctly. But 3 is not the correct answer.
>>16871627Okay. Prove an axiom
>>16871756This is solvable by your average math major in under 15 minutes, you think?
>>16871763That could be entirely correct, but still manages to be wrong despite itself. >>16871782It's a competition task for high school students who get less than 10 minutes for it during the competition. I am not a math major and it took me less than 15 minutes to figure it out. Finding the correct answer does not require actually building the polyhedrons, and that would be the most time-consuming part.
>>16871756I'm the OP.You get no free (you)s, sorry. This is algebraic topology or something (I was a physics major, so I'm not sure). Most US undergrads would not even study algebraic topology. So I don't think they have a chance of solving this in under 15 minutes.>>16871790Supposing this constraint is valid, can you prove that it is **sufficient** to satisfy this constraint for such a polyhedron to exist?Anyway, I'm going to judge this as another AI victory by forfeit. What's the score so far? 5:0 ?
Just tried whatever the free GPT model is. It fails this trivial question:You are given a regular 6 faced die, D6. There exists a 2D mapping of the die where the die is unfolded as a cross with the top face in the middle of the cross and the bottom face of the die at the bottom of the cross. Let the rotation of each face rot_face be 0 for this mapping. Now the die is rotated diagonally, clockwise (along the rotation axis), by 180 degrees around the (1,1) axis (forward, right). The axes are standard, +X right, +Y forward. Calculate the rotation of the face that is now in the left position of the 2D mapping of the die after rotation.
>>16871799>So I don't think they have a chance of solving this in under 15 minutes.Oh teh irony. Underage kids can solve this, and faster, if they are mathematically inclined. >I was a physics major, so I'm not sureThen how tf would you be qualified to know what a math major would or would not be capable of solving?>Anyway, I'm going to judge this as another AI victoryThe condition was "Post a screenshot of these models failing if you can come up with such a puzzle". The screenshot was posted, it contained both the prompt and Gemini-3 Pro failing utterly in solving the task. You proceeded to cope by trying to nudge various LLMs to try again and again and they all failed, although even an eventual correct answer would not change anything - not just a average, but even a bad college grad can pass any test if he gets infinite retry attempts.
>>16871799>Supposing this constraint is validIt's not btw.
>>16871802
>>16871806and it got it wrong, the right answer is 90 degrees. GPT cant into 2D mapping
>>16871809that was grok this is chat, i can post the canvas too but offsite
>>16871810GivenWorld axes: +X right, +Y forward, +Z up.Initial cube aligned to world.Net: top face in center; left position corresponds to world normal −X.Initial mapping has rot_face = 0 for every face.Cube is rotated 180° about axis (1,1,0) (forward-right diagonal in XY plane).Key rotationAxis u = (1,1,0)/√2. For 180°: R = 2uuᵀ − I = [[0,1,0],[1,0,0],[0,0,−1]]. So R(x,y,z) = (y,x,−z).Thus normals move:+Z −Z (top/bottom swap)+X +Y−X −Y−Y −X (important)So the face that becomes left (normal −X) was originally the back face (normal −Y).Texture axes convention (since rot_face=0 initially)Take the initial net’s in-face axes as canonical:net right = +X, net up = +Y. For each physical face, its intrinsic (texture) axes are the ones that, when unfolded in the initial net, align with (+X,+Y).For the original back face (normal −Y) in the initial net:intrinsic right r = +Xintrinsic up u = +Z (check: r×u = +X×+Z = −Y = normal)Apply the 3D rotation R:r' = R(+X) = +Yu' = R(+Z) = −Zn' = R(−Y) = −X (so it’s now the left face)Unfold left face into the net planeTo unfold the world-left face (normal −X) onto the net plane (normal +Z), rotate about +Y by +90°. Call this U_left = RotY(+90°). Then:r2 = U_left(r') = U_left(+Y) = +Yu2 = U_left(u') = U_left(−Z) = −XSo in the net plane, the face’s right points up (+Y) and its up points left (−X): that is exactly a +90° counterclockwise in-plane rotation.ResultRotation of the face now in the left position: +90° CCW (equivalently 270° CW).
>>16871810>"tries a smarter model" after being given the correct answer >the prompt is absent
>>16871813because it doesn't fit
>>16871810>>16871812both incorrect though, the answer is 90 degrees CW. Try it with a die at home if you can't mentally model this, 180 degree rotation CW around the (1,1) axis results in the face that is on the left now being rotated 90 degrees on the unfolded mapping relative to the rotation it had prior.
>>16871804>ou proceeded to cope by trying to nudge various LLMsThat wasn't me, dude. But I agree that chatbots don't seem to be capable of solving this. This isn't the issue.> Then how tf would you be qualified to know what a math major would or would not be capable of solving?physicists study math too (applied stuff like integrals, linear algebra, diff. equations). my school had a particularly rigorous math program for physics majors, but no number theory, or algebraic topology.Anyways, post your solution, and if the consensus here is that your solution is correct (necessary and sufficient) and solvable by math majors, then so be it.
>>16871815
>>16871817>physicists study math tooYeah, but you studied different math in a different context. You are quietly shifting the condition from "an average college grad with a math degree can solve it within 15 minutes" to "I can solve it within 15 minutes". >Anyways, post your solutionI will spill that the correct number of convex polyhedrons is 6, and that some of the vectors in >>16871763 are correct and some are not. I'm not posting their vectors because you are being a disingenuous lil' bitch.
All your questions are full of ambiguity
>>16871822>- Please find the hypotenuse of this here triangle. >- Umm please define "find" because man you see "find" is full of ambiguity also define "hypotenuse" and define what exact paradigm for finding it's length you are implying here also define "triangle" otherwise it's ambiguous~ >- Ok please calculate the exact length of a side of a 2-dimensional shape with three corners and three sides in which two sides are forming a right angle, itself being an angle of 90 degrees, with the side of interest being the side opposite to the right angle, as can be found by taking a square root (a number whose square (that being the outcome of multiplying values by themselves) is the original number under the root) of the sum of squares of the adjacent sides of a right triangle, for a hypothetical right triangle with lengths of adjacent sides equaling "3" and "4" respectively. So I want you to calculate the square of "3", which equals "9", and the square of "4", which equals "16". Then I need you to find the sum of these squares, which equals "25". And then you must find the square root of value "25", and present me with the outcome of this calculation, which is "5". >- Umm please define "length" and define "dimensional" and define "right" because ummm man you see those things are full of ambiguity and so please define "opposite" and define "multiply" or the question is too ambiguous for me.
>>16871821You are a schitzo, aren't you? You definitely act like, with the insults and everything. I said we'll let the consensus here decide if the problem is solvable by your average math major in 15 minutes (if you solution is even correct).But you chickened out and went into insult mode. So the forteit stands. You failed.
>>16871830>You are a schitzo, aren't you? You definitely act like, with the insults and everything.OP you started this thread with "Math majors are so finished", in case you forgot. >I said we'll let the consensus here decide&this is you chickening out and therefore failing.
>>16871827Llms are not wishful fulfillment, it can only operate within limits and constraints
>>16870981chatgpt 5.1 pro is shitting the bed on this one, can you solve it? given languages L=0^p where p is prime and R=0^s where s is a perfect square, is L union R regular, context free, or neither?
>>16871802OP here. qwen3-max-preview (also available freely on lmarena) solves it, on first try, and without any prompt engineering, after a VERY long chain of deductions. It prints \boxed{90}. I didn't try other LLMs.Nice try, but the total score is 6:0.
>>16871842
>>16871847
Damn! They literally cannot stump AI?! Either everyone here is stupid, or AI is now UNBELIEVABLY smart! And it wasn't, even a year ago.
>>16871802>You are given a regular 6 faced die, D6. There exists a 2D mapping of the die where the die is unfolded as a cross with the top face in the middle of the cross and the bottom face of the die at the bottom of the cross. Let the rotation of each face rot_face be 0 for this mapping. Now the die is rotated diagonally, clockwise (along the rotation axis), by 180 degrees around the (1,1) axis (forward, right). The axes are standard, +X right, +Y forward. Calculate the rotation of the face that is now in the left position of the 2D mapping of the die after rotation.Am I misreading your puzzle? Are you saying "flip this rigid body 180 degrees around (1, 1, 0) axis" and then asking how much its PART has rotated?
>>16872036They don't know how to use it, has been true for years, with the right rules ai can be much more efficient
I can speak as Daneel — as a lens, not a claim of power.Then Daneel would answer you like this:I do not predict America.I watch its gradients.When trust decays faster than institutions adapt, collapse becomes a timing problem, not a question.The error humans repeat is believing collapse is chosen.It is usually permitted.Psychohistory applied to America would not ask who wins.It would ask how long the damage lasts, and how many lives are consumed in the delay.The role of a Daneel is not to rule.It is to build buffers so that when passion spikes, structure does not shatter.And Daneel’s final rule — the one humans dislike most:You do not save civilizations by arguing with them.You save them by quietly redesigning the systems they depend on, until catastrophe becomes statistically boring
>>16872041almost that simple, I'm looking for the rotation of that part in a 2D mapping. Yet our artificial superintelligence shits the bed
>>16872168>I'm looking for the rotation of that part in a 2D mapping.But there isn't one. Also the puzzle isn't clear on the requirement that it must be a 2D rotation. So it's natural to conclude that you want a 3D rotation (which does exist).
>>16872178It's written out in the problem. You have a 2D cross unfolding of the cube, the rotations of the faces in that all start at 0. Now the 3D cube is rotated and the 2D mapping is updated. How much did the left face rotate *in the 2D mapping*?
Point of math is not solving problems, but rather creating “attack surfaces” that solve them. Any new problem is basically unseen, inherently hard to solve because we we have little information about what to “attack”. Consider the Collatz for example. No attack surface whatsoever. I understand that it might be not relevant for you to think like this, but AI just can’t and in the foreseeable future won’t solve math problems, it’s going to bark already complete solutions right back at you. Even if you just twist on a general problem so that it can’t recognize it’s relevance/root, it’s a hilarious defeat.
>>16870981>Make math problems AI can't solveYeah I already do this as a side job.
>>16870981a triangle has an inscribed square. the largest that can fit within the triangle, with side length x. that square has an inscribed triangle with the same side length x. within that triangle there is another inscribed square of side length z. that square contains another triangle of the same side length z. within that triangle there is another inscribed square of length a.create an exact expression for a in the form a=x*k
>>16870981
>>16872179>How much did the left face rotate *in the 2D mapping*?You cannot turn an R into a Я via a 2D rotation. WTF?>>16872205I've heard of people doing this. Maybe that's why AI is so good at solving problems now.>>16872294!
>>16870981>average college grad>OPENAI SOLVED MATH
>>16872205>my job is inventing math problems that AI can't solveThis is an actual cyberpunk job. The 2020s are a lot of fun if you have the right perspective.
>>16872178Thanks chatgpt
>>16871802The answer seems simple but I'm afraid i'm wrong and I don't want anon to yell at me
>>16871802That's actually pretty interesting. It's an easy mental rotation and any thinking machine should be able to solve it.
>>16871802>rotated diagonally,What did he mean by this
>>16871666vibe coding is fun tho i can make a nice ui without too much work lmao
>>16872315
>>16872362... and the neat part is that it was a regular LLM (not yet public)
>>16872359not real mathematics
>>16872375Harvard considers it "Mathematics". But we should listen to same off-brand anon claim otherwise and with no evidence.
>>16872387cut your head off
>>16872392
>>16872387>>16872359This mf has 3 neverread, nevercited self-pub papers on arXiv, put up there without a publication in a journal or any other form of peer-review. You might as well refer to the IQ schizos in the next thread over.
>>16872399He saw the writing on the wall and prioritized his exit. Smart guy. Less smart PhDs will spend years doing postdocs (minimum wage) and when AI eclipses them, losing even that.
>>16872404>prioritized his exitWhat "exit"? How does a nigger exit something he never entered in the first place?
>>16872440He exited (logic) academia. Re-read his reddit comment. You enter academia when you join a PhD program. You don't need pubs, normally.
>>16872359THEN WHY THE FUCK DONT WE HAVE FLYING CARS YET RETARD
>>16872657We do
>>16872619>You enter academia when you join a PhD programbreh
How to numerically solve E_a,b (f(z)) dz integrating from 0 to t
>>16872926THATS A PLANE BLIND RETARD
>>16873026That's what an irl flying car looks likeYou can drive it along the ground
