Blackbirdphys
Region: US
Monday 20 July 2026 20:58:39 GMT
136616
7747
229
473
Music
Download
Comments
mathscribbles :
I had a counter example ages ago but was too lazy to talk about it
2026-07-20 23:26:02
1379
bubbz777777 :
Ah yes, the Jacobian conjecture, one of my favorites.
2026-07-21 04:09:10
8
the underdog hedgehog :
“claude find a non trivial zero for the riemann hypothesis (make no mistakes)”
2026-07-21 03:06:01
538
j_cholman :
"why cant normal humans prompt these proofs" well for one, the average human doesn't know what a Jacobian is and two, obviously people who work at the ai companies have access to more powerful models that haven't released to the public for testing purposes.
2026-07-20 23:47:25
362
I'mHereForTheComments :
its because thats likely the 10,000th attempt. the Ai is really good at gaslighting you that it's right and you need to have a really deep understanding of the subject matter to know when it's lying
2026-07-21 01:58:53
155
James :
I’ll spoil it: they have unlimited tokens, so they do a (brute force) shotgun approach. Open 50 sessions on Fable max each exploring some of the harder problems out there. If you solve just 1, it’s a huge win. If you solve none, you mark down what you tried on every problem, and keep spending. Solving these problems is huge PR for them, so it’s worth it to put some of their math folks on this task. They aren’t better at math, they are making more attempts.
2026-07-21 02:10:51
110
zeta :
I'm biased but so many of these comments are dismissing how important the intuition of a mathematician is. decades of research and study provide the knowledge to be able to steer a model. right now provides a unique time to test it. we have the counterexample now so we know where to look. try providing a prompt to your favorite model and see how close you can get it to the answer while having Internet search disabled so the model can't cheat and lookup the answer. it won't be in its training data, so getting it to the answer will depend on your prompting skills and how good you can steer the model.
2026-07-21 03:01:56
50
George_Lincoln_Rockwell :
EASY! According to MY CALCULATIONS 🤓 134=raised to 15 x + 79 - x² = x + 7 So x = 1 and x = -2 V = π112 (9-x2) ² - (x + 7) 2dx V = π12 (x4-18x² + 81) (x² + 14x + 49) if V = pi / 12 (x4 - 19x2 - 14x + 32) dx19 V = x [-x-12 y = x + 79 - x² = x + 7 So x = 1 and x <562) = -2 V = π112 (9-x2) ² - (x + 7) 2dx V = π12 (x4-18x² + 81) - (x² + 14x + 49) dx V = pi / 12 (x4 - 19x2 - 14x + 32) dx 19 V = x [-x-12x3-7 y = x + 79 - x² = x + 7 So x = 1 and x = -2 V = π112 (9-x2) ² - (x + 7) 2dx V = π12 (x4-18x² + 84) - (x² + 14x + 49) dx V = pi / 12 (x4 - 19x2 - 14x + 32 9 - x² y = x + 79 - x² = x + 7 So x = 1 and x = -2 V = π112 (9-x2) ² - (x + 7) 2dx V = π12 (x4-18x² + 81) - (x² + 14x + 49) dx V = pi / 12 (X4 - 19x2 - 14x + 32) dx19 V = x [-x-12 y = x + 79 - x² = x + 7 So x = 1 and x = -2 V = π112 (9-x2) ² - (x + 7) 2dx V = π12 (x4-18x² + 81) - (x² + 14x + 49) dx V = pi / 12 (x4 - 19x2 - 14x + 32) dx 19 V = x [-x-12x3-7 y = x + 79 - x² = x + 7 So x = 1 and x = -2 V = π112 (9-x2) ² - (x + 7) 2dx V = π12 (x4-18x² + 81) - (x² + 14x + 49) dx V = pi / 12 (x4 - 19x2 - 14x + 32) as you know 9 - x² produces y = x + 79 - x² = x + 7 So x = 1 and x = -2 V = π112 (9-x2) ² - (x + 7) 2dx V = π12 (x4-18x² + 81) - (x² + 14x + 49) dx V = pi / 12 (x4 - 19x2 - 14x + 32) dx19 V = x [-x-12 y = x + 79 - x² = x + 7 So x = 1 and x = -2 V = π112 (9-x2) ² - (x + 7) 2dx V = π12 (x4-18x² + 81) - (x² + 14x + 49) dx V = pi / 12 (x4 - 19x2 - 14x + 32) dx 19 V = x [-x-12x3-7 y = x + 79 - x² = x + 7 So x = 1 and x = -2 V = π112 (9-x2) ² - (x + 7) 2dx V = π12 (x4-18x² + 81) - (x² + 14x + 49) dx V = pi / 12 (x4 - 19x2 - 14x + y = x + 79 - x² = x + 7 So x = 1 and x = -2 V = π112 (9-x2) ² - (x + 7) 2dx V = π12 (x4-18x² + 81) - (x² + 14x + 49) dx V = pi / 12 (x4 - 19x2 - 14X + 32) dx19 V = x [-x-12 y = x + 79 - x² = x + 7 So x = 1 and x = -2 V = π112 (9-x2) ² - (x + 7)= IS 134=raised to 15 x + 79 - x² = x + 7 So x = 1 and x = -2 V = π112 (9-x2) ² - (x + 7) 2dx V = π12 (x4-18x² + 81) - (x² + 14x + 49) if V = pi / 12 (x4 - 19x2 - 14x + 32) dx19 V = x [-x-12 y = x + 79 - x² = x + 7 So x = 1 and x <562) = -2 V = π112 (9-x2) ² - (x + 7) 2dx V = π12 (x4 answer is: I DONT KNOW
2026-07-21 01:31:53
8
This Kiss :
"why can't I prompt it to do this?" I've probably never interacted with a non-quantized Claude model. probably a heavily quantized one at times. the staff mathematician at anthropoic? yeah, he's using fable unquantized with a million dollars worth of compute to throw at it.
2026-07-21 01:05:22
30
Adam :
What's a conjecture?
2026-07-21 04:01:33
19
ThiccYoda :
Awe damn it man, I was close to getting that polynomial too. Common sense if you think about it
2026-07-20 22:56:24
99
Rassy :
is it too late to change degree 😭😭😭
2026-07-20 21:07:46
10
transient :
let's just casually solve a hundred year unsolved problem so we can falsely give ai the credit as good marketing
2026-07-21 07:13:53
7
Luke :
the average person doesnt even know what a conjecture is, why the hell do they think theyd be able to understand what a proof even looks like
2026-07-21 14:26:56
8
DrKSmith94 :
Agree not surprising. Why does the rocket scientist know more about making rockets fly higher and faster? Why does the surgeon know more about anatomical variation? The LLM is a tool that needs an expert when being used in this capacity.
2026-07-21 02:27:19
8
llllllllIIIIIIIIlllll :
The guy who proved the first twin prime gap worked his entire PhD on this conjecture.
2026-07-21 00:31:17
23
fmi :
Well sometimes even simple prompts can actually solve stuff
2026-07-21 00:04:15
9
domoardrugs :
i had a counter example years ago but mine wouldn't fit on the margins of my tweet!
2026-07-21 01:28:40
8
Sasha :
Why do you need AI to find a counterexample though? Wouldn’t brute force be more efficient? I can’t imagine there’s any real shortcuts AI would be taking here unless there’s something special about this counterexample
2026-07-26 00:10:19
0
Cris :
all these theorical math problems solved and yet we 1) depend on fossil fuel energy for most planet 2) we haven't travelled outside our solar system 3) we don't have even an efficient energy mechanism with high efficiency and minimal energy loss 4) we don't have understanding of dark matter and how it makes most of our universe and how it originated,let me go with useless Anthropics useless marketing for their AI since Fable flopped because again the model they promised was again a marketing stunt
2026-07-21 12:04:00
0
Nate :
I’m not a talented mathematician but I’ve had opus and fable do lean verified theorems regarding Tao’s prop 1.17 it’s not a huge proof but I taught myself Fourier transformations and eigenvalues through this. The entire repo is public.
2026-07-21 15:35:07
4
El Chonco :
it's also important to know that Fable 5 is opus 5.0 it's not Mythos...
2026-07-21 22:57:27
0
Dan :
Simple: The AI does not do all the heavy lifting is a co-worker on the results and obviously that guy used to check, go through and verify... the problem. No, you cant ask for start to finish to any AI to do something as complex like this on a single or multiple tries
2026-07-21 08:39:18
4
grassrootscajun :
how is this useful in real life?
2026-07-21 20:14:35
0
ConorCon :
This guy is a world class math mathematician and likely has an IQ of at LEAST 160 maybe 170 and we’re wondering how he can prompt the LLM whereas the average person cannot? Holy shit we’re cooked
2026-07-21 10:40:47
4
To see more videos from user @blackbirdphys, please go to the Tikwm
homepage.
