OpenAI just dropped roughly 300 math papers with Lean proofs, mathematicians are crashing out, and the vague-posting has started: "math is breaking, so cryptography is next." Brian and Rob Hamilton refuse to hand-wave. Rob fed the Magic Internet Math elliptic-curve study guide and the open theorems into a model and asked it straight: if you wanted to break Bitcoin's curve, which doors would you knock on? Then he and Brian walk through every one of them, live. The result is a red-team tour of secp256k1. There are six offensive open problems and three defenses nobody has been able to prove, and each comes with the specific math that would have to break. Along the way: what Lean proofs actually are, why the discrete log problem is where the hardness lives, why "elliptic curve cryptography has been broken" will probably be a headline about some curve Bitcoin doesn't use, and why Big Bitcoin and Big Math are on a collision course. In this episode: • OpenAI's ~300-paper drop, the "11 squares" packing result everyone memed, and the human who immediately found the beautiful version nobody paid for • Lean explained: an audit trail for what the AI did, built up from the foundations of math • Why mathematicians are upset (the funding model behind long-standing problems), and Terence Tao's case that humans still matter • Division of labor between people and machines, and Brian's advice to his daughter Kayla, a senior math major: apprentice on Lean now, then go do your research • Big-O notation in two minutes, and why it's the first page of Neal Koblitz's number theory and cryptography book • The matrix-multiplication race (Strassen's ~2.8 down to ~2.37 and below) and why faster linear algebra is the real prize for AI • secp256k1 from first principles: y² = x³ + 7 over a finite field, point addition, and why public → private is a one-way street • Pollard's rho and the real defense line of 2¹²⁸, plus the Coldcard reminder: 2³² falls to one computer and ~2⁷⁰ still costs real money • Door 1: point factoring • Door 2: the "minors" attack. Build a big matrix and hunt for a vanishing determinant. Solved toy curves at 2⁵⁰, but an Aug 2026 follow-up showed it costs more than brute force at scale • Door 3: beating the generic √n bound (Shoup) • Door 4: the Koblitz-curve symmetry Satoshi chose for speed, and a 2026 Lean result that found only the trivial symmetry survives • Door 5: lifting the problem to a smooth curve, and why the height penalty blows up • Door 6: isogenies and the "weak cousin" curve, why GRH says every door in the family is equally strong, and why this is a combo pool shot rather than a break • The defenses no one can climb: one-way functions, proving the discrete log is hard, and P vs NP • Yehuda Lindell's (Coinbase) thread: AI proving hard theorems is no evidence that elliptic curve hardness is in danger • Shor's algorithm as an engineering and physics problem, not a math one, and the post-quantum camps (hash-based, lattices, isogenies) • A preview of the math panel at BTC Amsterdam: "do a podcast to get inspiration for the conference talk" ️ Brian Hirschfield × Rob Hamilton Magic Internet Math: a first-principles podcast about mathematics, Bitcoin, and meaning ⚡ Follow: Brian on X: https://x.com/Fundamentals21m Rob Hamilton on X: https://x.com/Rob1Ham AnchorWatch: https://anchorwatch.com Magic Internet Math: https://magicinternetmath.com Further reading: secp256k1: Security Status & Open Problems (episode companion): https://bitdevs.magicinternetmath.com/content/secp256k1-open-problems OpenAI math results (Lean): https://github.com/openai/math #Bitcoin #EllipticCurves #secp256k1 #Cryptography #DiscreteLog #Lean #FormalProofs #OpenAI #PvsNP #PostQuantum #MagicInternetMath #Mathematics #RobHamilton
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
UnknownYeah. Well, welcome. Hello, and welcome, dude. We are on, like, a regular cadence now. This is Magic. Magic Internet Math podcast. We're doing it. We are, like like, what what the hell did we agree to do, when we decided to do this math? I mean, I started a math podcast two years ago. And when it had to end, I remember thinking at the time it can't end,
Right? And it had to continue. Didn't really, you know, we were not in the world we're in now. But like, there was just something nagging inside of me. Like there's no way that this math podcast can stop. Then we pick it up. Pick it up again, I pick it up again, you know, with you and we start in January. God, we started in January, it's October now. And it's a... It's... It feels like we have been running into a storm the whole time.
UnknownYeah. Well, you know, sometimes you do things, you know, because you just have you you you you do things like start a podcast sometimes because you just feel a calling, and there's no one else occupying that space. Right? And the entire domain of math was just basically disregarded. It was something that would get coverage as a guest on a podcast. Like, you'd maybe bring a cryptographer on, like, once a year. An insane guest on a podcast typically. Like, I'm thinking, like, it's a Terrence Howard on Joe Rogan. You know? Like, you would get the little things like that where somebody would be painted as insane but with
semi interesting views.
UnknownYeah. Yeah. I mean, when, I mean, listen. Alan is one of our four listeners of this podcast. We don't have to make fun of him, like, right on the show. They they would bring Alan out every once in a while. What's his title again for this podcast? The, executive... Is he the board member of the... Like, what is he? The executive director? Spiritual guide? What is what is he? Like that. But we Yeah. Yeah. We we we think we're we think we're gonna do something this week, so everybody, watch out. But by the way, so we named Alan, I am Rob. This is... I am Rob. I am Brian. This is Rob. Mix up the clankers. We're killing the clankers right now. I am Brian. My partner here is Rob,
CEO of Bitcoin. Needs no introduction.
UnknownAnd, we think we're gonna have Alan on this week. So Yeah. But what what am I... One of my big directives as CEO going into... I joined in the middle of the fiscal quarter. So we're doing some transitional stuff right now. But the math podcast was one of my top priorities as CEO of Bitcoin to make sure that everyone's gonna eat their damn vegetables. And if you wanna live in a world of not trusting and verifying, you're gonna have to look at some math. I'm sorry. I'm sorry. You're gonna have to do it, and we're gonna make it as painless as possible. And to which I will start with
something that we spent 10 episodes talking about, elliptic curve cryptography is right back at the top of the news. We'll get to that in a little bit. Gets old. Old. Never It never gets old because it's holding trillions of dollars of wealth. So That's right. We will be getting back to that shortly. But mathematicians not doing great this week. They're they're, I
Unknownthink as the kids say, they're crashing out. They are crashing out as some... You know, they're crashing out. They're they're going crazy. You know, I had some long conversations with Kayla, who if you guys listen to the podcast may know my daughter who, you know, has been on podcast. She's a math major. She's a senior. She's pretty dug in. Not much that can happen in the world that's gonna change the trajectory of the fact that she's gonna be a mathematician. And we had a couple of long conversations after,
know, she's she is neck deep in the academic academic environment in college. So she's getting... The conversation she's getting is from people who are mostly getting... Who got blindsided by OpenAI throwing what was it like the solutions to like 18, essentially 18 long standing problems with like lean proofs and the whole kit and caboodle. I've since brought all of that onto Magic Internet Math. So if you want summaries of any of all of those things, you can go there and look at it. Like,
you know, what seems so... Mathematicians are very upset And it's it's complicated, but the mathematicians are upset mostly because... Here, I'm gonna get killed. If anyone besides Alan and three other people listen to this, I'm gonna get killed for this. But, like, the reason mathematicians are upset right now is because their entire complex of how they get funded is being dismantled by AI essentially. It's like these problems, what gets funded is the attempts to solve these problems. And then,
you know, universities, programs, all these programs get funded to do it. And then, you know, some computer just solves them all in a second. And, you know, you are your sort of economy is in shambles. But that's not even like the big picture thing. But that's what everyone I think that's what everyone's really upset about is like, they're now their careers on a chopping block like all the others. Right. As though there was a bit like as those society like as those society viewed mathematicians like accountants in terms of having careers,
you know, people... Right? Yeah. But nonetheless, what little what little they had has been snatched.
UnknownIt's been really carved up. Yeah. And I think there's actually an uplifting part of the story here. But... So we had, what, was it, like, 300 papers that got put out by Open By OpenAI. Yeah. It was, like, 300. Yeah. They just, like, dropped it. And so I think the one that probably everyone saw was this, packing of, what was it, 11 squares optimally into a space,
and it looks sloppy and horrific. Like, it doesn't look clean and nice.
UnknownAnd was this, like, basically the one maybe accessible thing that can be put out for people? Is that really... Is that why, you know... Is that why this was the one that was chosen to sort of be be the example? Well, was nice to see, like, something in a clear picture. Right? Or did or did they put a lot of things out and that one just happened to be the thing that got memed? I think that one got memed. It had good meme It did get memed Yeah. Pretty
UnknownAnd I don't think everything's lost though. I agree completely. Because there's the... So maybe to take a half step back here and explain, like, Lean. Right? It's a it's a programming language that... And and maybe there's actually a really good video. I'm gonna send it to you right now to throw in the show notes. There's this content creator who does, like, coding stuff called Primogen, and he actually did a really cool video, like, talking about this
and laying down, like, some of the context. But, lean is a programming language that you use for mathematical proofs. Right? And the reason why, AI models are able to attack this, whereas if I was using, OpenAI's latest model to try and find some scientific breakthrough in chemistry, it could come up with a theory, but then it's like, now I need your wet lab to pair to do the observation and the experiment. But since abstract math is purely just numbers,
you can actually in parallel sequence out and kind of like build these things. And so that's why you're kinda seeing this, like, this, like, little bit of a takeoff and hitting all of this stuff.
UnknownSome of the papers so would you say lean is, like, going to the math store and asking to verify Lean is Lean is going to the Home Depot of the math store, and you just take all of the raw inputs, and you just start excel... Like, you just start... It's not quite a brute forcing, but you just start taking all of these small concepts, and you start building them up into one interconnected thing. Right? So you can start all the way at the foundations of math and use those as sub proofs for future proofs. Right? And then you kind of build a chain of validated logic. And But it's how a person actually
sees what the computer did. Yeah. It's essentially an audit trail of what AI did to solve the problem.
UnknownYep. Exactly. Yeah. And so you're seeing these solutions. Like if you look at this one with the five... The six aligned squares in the box and the five tilted squares, You could even see that, like, one square is wedged into the gap of the two other squares. Like, it doesn't look clean. It doesn't look nice, and everyone's trying to understand, like, why is that? This is, I think, actually where mathematicians step back into the fold though because someone looking at that original proof realized, wait a second, you could
do a transformation. And, are you familiar with the Fibonacci Taurus? No. So I'm sending you a screenshot right now so you can see this. If you actually rearrange the squares into a Fibonacci... Through a Fibonacci Taurus, you actually get a beautifully symmetric clean picture. I just sent it to you. Oh, man. That's crazy. Do you see that? Yep.
UnknownA torus, by the way, is typically viewed as like a donut. It's like a topological object. In math, like you learn about a torus in like algebraic geometry or topology and it typically looks like a donut. Right. In this picture, I mean, it doesn't have to look like a donut but that's typically the visual that almost to, you know, some continuous object. But in this picture, it's I'm not seeing the donut. So that's Well, so it's a
UnknownI'm pulling up a model right now that can load it for me. But yeah, so it's using the golden ratio and it actually does a transformation. And if you wild, though. Were to do that It does look So this is the beauty of mathematicians, though. Right. Is the fact that ordered is this thing was orderly. Bingo. And I think we're gonna talk about this. If you wanna, like, tee up Terence Tau and, like, math stuff, but, like, these proofs are really... Are incredible in being able to kind of, like, shake up the known domain knowledge and kind of solve some problems, but we're not getting clean solutions, like one with
linear algebra multiplication, right? Matrix multiplication. We always thought it was, I think it was what, n log n was it for the for the, like the the the computation time to do this stuff. And we're actually finding like a race now of getting fractionally smaller and smaller and smaller, which implies there's probably some lower limit that we haven't even thought of before. You wanna step back, you wanna just step back and say
Unknowntwo minutes on that what computation, how computational time is discussed and measured and what you mean by n log n versus say two to the n or versus, you know, do you wanna just say something? Yeah. Real quick there. So this is something that anyone in computer science, any sort of background would have experience in. And it's kind of this idea of big O notation, right? And so the time it takes to compute something, let's keep it very simple to start. Let's say you have a for loop. You wanna count from zero to 100 or zero, and you just print one, two, three, four, five, six. Right? And let's say you wanna do that,
up to any length you want. Well, the time it would take for that algorithm to even run would be n, big o m. Right? And then the... Like, the the the next example that would be like what people talk about when you're thinking about programming design things would be like, what if you had a nested for loop? So it was a loop within a loop. Well, now you're at n squared. Right? Because you have to take the length of the list through both of those things. And as a starting place, this is, you know, teed up for... As it relates to computer sciences because oftentimes you can rewrite algorithms more efficiently
than the naive way. Like, if you're a simple brain, it's like, I need to do this, and then for each of those, need to do that. Right? I'll do two loops. It'll work. It'll just take longer, but oftentimes there are, mathematical transformations and ways you can reevaluate the problem where you can actually do it much faster. And so, you can have it, you know, in things that take n log n, n log n. Right? These are all different ways and approaches. Square root of n. Right? Depending on whatever the given
Unknowntransformation is, like, you'll be able to understand that. And the thing I'll add is what it shows is how time how it takes scales over, like you want to know in the moment over the infinitude of computations you're going to actually do. You just want to know how does the time it takes scale over that like infinitude. And that sort of tells you in the moment. I think one of the I think I mentioned this on the podcast before, but everyone should check out Neil Koblitz's
book. It's on my it's on the website on the magic internet math website, but I think it's called number theory and cryptography. It's considered a graduate level book, but it's really very accessible. Excellent. I went through the whole thing last January when I started this podcast originally. And literally in the first page of the book is Big O notation. It is like the most fundamental thing to know about cryptography. Because in the if you're, you know, if everything is completely random,
okay, right, and you have perfect entropy, and then it really just comes down to you. Is the force you're applying with your cryptographic algorithm? Does it overwhelm the force an attacker can
UnknownRight. Can provide? Which trivially, we always say that, like, if you're trying to brute force... If you're trying to... If I'm trying to get your Bitcoin key, my really, in general, best way of doing that is brute forcing and guessing all possible numbers between zero and two to the two fifty six. Right? And that's the asymmetric defense that you get with cryptography. Right. So what big o notation measures is the... Really the force that can be computationally applied to a problem. And that's how so when we talk about like when we say the big O is N, we're saying that the
power scales linearly, like one to one linearly. When it scales at like the log of n, we're saying it scales less than linearly. Correct. So when we say it scales at like n squared, we're saying it scales way, way more Exponentially. Exponentially. Right? So, like, that... I I wanted to just make sure we have that out there. I just wanna make sure for posterity, we have that out there for people. Because, know? Yep. We're probably gonna talk about it a good bit with these these examples.
The other thing I wanted to say was, if this is a yin and yang situation with AI solving these problems, and I think you explained it really well, but like, there's this one side where everyone's upset about you know, the, these projects were funded and AIs come to solve these problems. And then there's this thing where, okay, well, then immediately somebody just came and made an extension on it that made beautiful sense. And the thing is nobody paid that guy to do that. And
that... But, you know, we still have... Like, mathematics is still filled with people who just want... Who have love of the game and want to continue to make sense of things. Right? And it's like Yep. That's the thing that's gonna matter long term, and it's amazing that we saw it immediately. Like, we saw somebody do this immediately because it's like, you know, in in one sense, it's it's like you're trying to play a game, and the game might have to take hundreds of years. Like like we talked about Fermat's last theorem on the podcast about how over the three hundred plus years of trying to solve that, we invented algebraic number theory.
We figured out fray elliptic curves. There were like multiple fields that we had to invent in order to solve that. And if you just go back in the thought exercise and say, if a computer what if Fermat had actually provided the proof? Would we actually have done all that? Probably because he would have provided the proof and everyone would have been in the same spot that they were in without the proof, which is like, what the hell? What the hell is this? Right. Right? So then it would start that we would just start,
you know, and it's just, you know, it happened that there was there was prize money funding for Vermont's Last Theorem all that time, but maybe there would have been also to make sense out of it. Right? Maybe we will be offering funding and prize money to say, okay, OpenAI just dropped a bomb.
UnknownIt's time to make sense. It's time for people to think about rationalize and make sense of it. Yes. Which I think is gonna be the standing value of the mathematicians because these machines and their, like, strict mathematical precision are able to show breaks in previous mental models of what we thought was a theoretical end, And now you need to re rally the theoreticians, you know, the mathematicians that are doing the theoretical stuff. They need to reorient their career trajectory and not have it just be, I'm gonna sit around and try and solve this problem.
They're gonna have to be like the human cipher that comes in and tries to do these transformations to understand, okay. But why is this happening? And the reason why you want them to explain why it's happening, one, in theory, it's possible that that the machines could also do this job. But you want to be able to just not say, well, we have a proof, and we now know something's broken. It needs a new, like, theoretical, like, abstract scaffolding to go around it to explain why it's happening. And then the recur... The loop would be once you have that and you flesh that out and you formally prove whatever the new scaffolding is, maybe you can break things another leg. Right? Like, it's gonna be like a... It's gonna be a lobbying back and forth, like a dialogue that happens between the machines and man to be able do that because just because you can prove something breaks. Obviously, that is
a massive significant, like, advancement in the field. But you're gonna wanna be able to, like, make it into a pretty picture. Like, we'll show in the study guides with this, this torus, the Fibonacci torus, of the pack squares. Like, it just... It shows you that the, like, the beauty in the math is still there. You just have to... The way we were looking at the problem doesn't work. And and in that video that I... That I'll link, we'll put in the show notes, they talk about how if you were to just basically, like, take, like, a spiral graph out from, like, like, on planar coordinates to, like, graph that, it looks really weird. But if you shift it to polar coordinates, it's instant. You just, like, increase your number, like, just, like, linearly. And that's to show you that like if you just rearrange and rethink the problem
going from traditional coordinates to polar coordinates,
Unknownsomething that looks like really hard and complicated gets like trivially easy. We have to put a pin in that one and definitely save that for another conversation. That's a big, big, big, big one. Polar coordinates or Yeah. Like, would it meet that exactly? Like presenting something in polar coordinates and then having it be just a different world where everything makes sense. But don't remember the math book that I got at BTC Prague by the author Jan Kopas. Yes. And I showed you the cover. I did this with Alan on. I showed the cover and basically he represented the elliptic curve as a torus, proper donut.
And I found that to be beautiful. That's what really got me starting to think about toruses in general. And like how... Like I thought it was powerful for us to go through all the episodes we did and be able to see the elliptic curve the way we did as a finite field. But to be able to visualize it as this one object, is like really where like the Steiner explanations come to fruition for me, where it's like, just to be able to call it one thing to see something as one thing is extraordinarily
powerful. And that one thing that this guy probably did for nothing. You know, I think there'll be a world where people, you know, when at scale people are visualizing elliptic curves as toreses, as tori or I don't know how you pronounce it. It's a remarkable thing. And, know, was talking to my daughter about this whole situation and she said to me like, Look, I don't know if I want to spend my entire life
validating lien proofs. And I said, I totally understand. I don't think you're going to have to, however, you may want to spend the next couple of years doing it. It's not a bad thing. So like you're meeting this problem now, think at the perfect time. You're senior in college,
Unknownyou're going to go work for people doing research. You can be valuable to them. It's the most exciting time in mathematics ever. Yeah. It's the... Like you are at the precipice and you're not in ensconced in like the hierarchy of the math world. You get to go in and just break shit. Absolute... And you Like, this is the best time for her. Like like, there is no better time to be a senior with a math, like, entering the field than this moment in the entire history of humanity. Like Yeah. The ability to just actually raise your hand and say, I wanted
want to spend like a lot of time validating Lean Proofs for somebody because I want to do it once or twice
Unknownthen never then I'm gonna be able to go do my research. Right. Right. I think that's totally and that's such a great point. Yeah, like you cut your teeth a little bit, learn the skill. There's a small corollary of, when my wife was getting her PhD in biochemistry. It was just like CRISPR had had been out for like a couple of years, and people were starting to do RNA sequencing and doing all of the... It wasn't just wet lab work anymore. It was the computational biology, like bioinformatics. Yeah. And she pivoted right into that. And it was amazing for her career. And now like it's what a lot of people are doing like in the field and like linking this stuff up. But all that to say that like, when there's major step functions like in changes in math,
like in in whatever industry it is, you wanna lean into that and be versed and skilled in it. And this is just general career advice. Oh, yeah. You want you wanna you wanna be familiar with this stuff, because then you become the expert. You then be able... You're then you're able to speak about authoritatively. You then get to see the the entire world of your field differently than all of your peers, and you're gonna be able to do some amazing, incredible stuff, like, for sure. And you will have it... I mean, I do think it's probably an important skill to gain. You don't have to spend the rest of your career doing it, but you probably should do consider
Unknownyourself an apprentice at this point and that this being a needed skill like blacksmithing at this moment. Yeah. Right? That see the writing on the wall, but still skate to where the puck is going. Yep. And by the way, as a math podcast, I think we're going to be able to snatch all the lean puns until you know, essentially, we're gonna get we're gonna be able to snatch them first. Yeah, absolutely. And I actually maybe another thing to drop into the notes, the stress and algorithm. So originally linear algebra multiplication thought was thought to be n cubed.
And then the stress and algorithm was thought to kind of be the floor was like two point, like end of the 2.8. And then it kind of goes down a little bit, expanding upon that to like, 2.3755. And now like what's happening over the past couple of weeks is like, we're extending that even more. Like, we're going shades and hairs below that where at first it was like, ninety nine thousand nine ninety nine, like 80 times, but it was
it was to the 0.9999. So technically, it was a fractionally smaller number, and now we're at like five nines. And all this is saying is similar to that packing the square problem that, like, there's probably some misalignment in how we're actually viewing the problem that will be to get more efficient. And this is,
Unknowndo you wanna talk about the Terrence Tau step before we get deep into the, elliptic curve piece specifically? Yeah, let's do that. But I'll also just, I'll just say like, I also ironically, I mean, there is a lot of money now behind getting linear algebra to compute faster. Bingo. Because little... And maybe to tease out that context, AI, these large language models is just a bunch of linear algebra. So if all of a sudden you can more efficiently can do the linear algebra mathematic... Like, this is like a Moore's... Not literally Moore's law, but, like, you're finding more productivity. So with the same amount of energy, you're able to do more math, which means lower cost of compute. So, like, this is
making a run at probably some of the most significant problems, which is actually an interesting point of of all the papers that you would wanna publish and produce. Why would it be this one? Because now you're letting the entire world know how they could do math more efficiently where in theory, OpenAI could have held this to themselves and gotten a leg up on getting more efficient compute. But maybe that gets a little bit too complicated because the second you go to the chip fabrication, you're kind of laying out the algorithm on the circuits. They're gonna realize, oh, shit. This happened. So they just wanna get all of the the glaze and glory up front.
UnknownSeems that way, and it just seems like they are... They're all out to seems very low level. They're trying to disrupt industries and create sort of create FUD and create chaos right now. That's what it feels like. It doesn't feel like there's a high level strategy here. Yeah. Well, I mean, they're, they're trying to IPO, so they have to make it seem like they're they're making a run at literally all things, and then they vague... Like, basically, I think they help, accelerate the vague posting about there's no cryptography out there. Maybe all cryptography is k fabe. Like, they're staging pretty terrific k fabe. And so Terrence Tao is... Whether he knows it or not or likes it or not, anytime he speaks, he's part of the k fabe.
And he's either imposing his own k fabe or he's just part of this machine's kayfabe. But the big thing was he posted some slide from, I guess one of his slides got on Twitter of this presentation he did yesterday. And just to summarize, he's basically making a case for why humans matter, why mathematicians matter. And he used this analogy of pharmaceutical development that like, if we knew a thing worked, but we didn't know why would you actually inject it into your bloodstream?
It's just now more in the kayfabe of, I think, framing this argument. It's like, you know, what is it a Hegelian dialect or like where there's like, know, you're trying to you're forcing two sides, but none of them are actually good arguments. We're gonna have bring Eric Kason onto the podcast. He can break that down for us. We should he just he just started following me. I should... We should be Oh, hell yeah. Bring him up. We should We should definitely bring Eric Kason. We should. But, like, it's it's more about the... This is now, like... You also now have to navigate the k fave.
I I I have a personal view that human beings matter. I don't know what Terence Tao is trying... Is doing. I understand that his world is get... Like, imagine being like... You know, I don't know what the analogy is for who Terrence Tau is, but maybe like there's a like you worked your whole life to become a great NBA player and all of a sudden, the 13 foot prediction that I made about quantum actually came true. And now you're
dealing with it, right? So he's dealing with the fact that he was like, he was special. And I think the world is asking for his views, right? The world wants him to speak on these things. He's sort of the CEO of math right now, right? He's the CEO of big math. You know what I mean? So he's making this case that, yes, human beings matter. That's a case I've made here. Like, I've made the case here that, like, okay, what really good is it if a machine takes
the shortcut just like I mentioned with Fermat's Less Theorem? I just think it's it's just interesting though. You also have to navigate it. That's... You have to navigate the k fabe while observing this because he's getting so much shit on Twitter. For some reason, the Twitter... Like the Twitter take is that fuck these mathematicians, they're entitled shits. They're lucky to do... They're lucky to make a living doing what they do anyway, and they should stop complaining. And there's like... That's like 80% true,
you know? So which is all good Twitter. All good Twitter takes are like 80% true. Right. Right? So I think that's what this is what we're seeing here. That there's gonna... It's gonna look like a fight between humans
Unknownand the robots. I feel like that's what they're they're staging here. Doctor. That's the cape maybe in stage up. But to my point, I think the actual answer is working together. Right? So his example is that like, let's say someone made a bleed proof for a drug cocktail that could cure your disease, whatever disease you had, but no person could explain the mechanism. But there's a lean proof. And this is an abstract like thought experiment, because to actually do a lean proof that proved you were able to do take this medicine safely would mean that you would need a lean proof for the entire human body and physio,
the entire anatomy and all of like chemical interactions and all the protein binding and the entire like human, system, like the entire like biochemical system of the human body, you would need lean proofs and how they all default operate. And then from there, you could plug a drug in and show that there won't be negative side effects. Like, that's why it's like more of a thought experiment. This isn't something that's in the reason of what we're able to do at the moment. But that would be wild if over the next, know, decade, we're able to formalize out the entire human like
bio biochemical soup that we are. And then you could start actually figuring this stuff out. But the point is,
Unknownif you had a mathematical proof, but no one could actually explain the mechanism, would you take the Right, but what's interesting, I guess about it is I mean, the fact is what we consider the standard for pharmaceuticals is are these reverse control studies. A lean proof is like a billion times more powerful than anything from a reverse control study. Oh, it's way more direct too. You know, like we, the illusion is, oh, we have these reverse control studies and therefore we understand all the side effects and all this thing, you know, yet not to get off a tangent here, but you know, something like three hundred and fifty thousand
people die every year just from normal pharmaceutical accidents. Because these reverse controls, we can talk about the modeling of these things, but they have, there's a lot of model bias, you can't, you know, this is partly why there may be tyranny in the world because they want the entire world to be in the set of training data. Yeah, because when you reverse control for certain groups, you may not be capturing everybody. The
Unknownclassic one is that men are way oversampled because women with their hormonal cycles plus pregnancies, like pregnant women don't typically sign up for clinical trials to do stuff, right? So like the entire like, the knowledge that we have on the medical treatment of pregnant women is entirely different than a generic man. Because no woman is going to subject herself to a clinical trial while she's nursing or while she's, you know, while she's pregnant, right? Which was always kind of the wildest thing about like, with the COVID vaccines, Oh, it's fine if you're pregnant. I was like, hold the fuck on. Know, I know for a fact you weren't injecting pregnant women with this. Like, that would have violated every single ethical standard, so fuck you. Like, that was, like, that was one of those moments where, like, they put... Like, a half step too far where they had to be so absolutist with, like, the structure of this. They're like, it's safe for pregnant women. I'm like, you actually can't say that. And you know that's not true. Like, anyway, that's just a little nugget of a side note of a tangent within a tangent. We can't not go there with this, but,
Unknownyou know, to... I guess to piggyback your last point before we then get into the study guide is that what I have been saying for the last couple of days, and this is like the light bulb that kind of went off for me talking to Kayla about this, is that what we're seeing in real time is a division of labor. Yes. And there are probably now problems that just should be solved by machines. And it just so happens that we're dealing a division of labor with an incredibly asymmetric ability. Well, like machines have an asymmetric ability over people for certain things, and we have an asymmetric ability over them over certain things.
And we are now we are going to be in an awkward process of discovering that division of labor. Know it's so funny, dude, like in Bitcoin, we talk about these concepts as though they're romantic, you know, like, yes, man, dude, like what it takes to discover division of labor is painful, awkward, and really crappy. And we're about to go through that in this generation now with math. Yeah.
UnknownExactly. Is cycle time. Humans haven't made their move yet, really. That's the thing. And so we have this k fave. We have this societal conversation. We just haven't... We really haven't seen humans make their move yet. Right. We have to figure out how to respond to this all. Right? So let's get into it with elliptic curves, why everyone's tuned in today because the extension... The high level just extension of this before we get into the specifics is math is... You know, OpenEye puts out hundreds of papers and lean proofs,
know, solving different fields of mathematics conspicuously. There are no cryptography results. The implication and, OpenAI has not come out to say anything either way, because I think it's a handy narrative for them to go wield, is that there are advancements that are happening currently in cryptography, and that we may be able to actually break things like elliptic curves that we use in Bitcoin, not with a quantum computer, but just using the foundational math and our understanding of the math behind elliptic curves, which we've gone into extensively in this podcast previously.
And we have a study guide. I'd encourage you. Go check the episode with Riordan that we did with Riordan about this. Yeah. Talk a lot about this. Yeah. Yeah. Yeah. We did an episode with Riordan. I'd encourage you to go through the study guide and go through, the entire series if you wanna go deep, deep, deep into the elliptic curve stuff.
UnknownAnd because even in... Even at the level of elliptic curves, there's elliptic curves, and there's Bitcoin's elliptic curve. Right. Right. And we've talked a lot about that. Right? Indeed. Indeed. Yeah. So maybe to tee up the the nature of the problem for those at home, Bitcoin has an equation y q... Y equals, x cubed plus seven. Y squared equals Sorry. Y squared equals x cubed plus seven. Yes. Correct. Sorry. Y squared equals x cubed plus seven.
And that creates a curve. It looks kind of like a c, if you're thinking about looking on a graph plane like that. It it fans out and goes wider, but like the initial start of it looks like like a c, if you're trying to think about that, and goes across positive and negative, coordinates, you
Unknownknow, kinda mirrored across the x axis. That's what the y squared. That's sort of what the that's what the y squared means. That's right. Y squared equals something. And if that something is negative or positive, it'll resolve to the y squared. So that's why it's symmetric over the x axis. Right. And so we... It is not literally a continuous curve. It is a finite field, and it is a group
of points. Those points are actually raw public keys that... And private keys that exist on the curve. I will... Well, just to jump in there.
UnknownIf you read a book on elliptic curves, the first thing you'll especially in Bitcoin, like elliptic curves and Bitcoin, you'll first learn of elliptic curves over the reals, which is a continuous curve. Over the real numbers, that's just you have every possible x value is now drawing. We like to do that in math, we like to visualize things in the reals as though they're continuous. And then we do it over that finite field. Finite field is a different number system than the reals. It's basically
every integer up to that prime, up to the modulus minus one. Yeah.
UnknownThen it's a bunch of points that people will wanna connect the lines, but they don't connect. It's just a bunch of points. We we for our monkey brains talking and reasoning about this stuff, connect the points to make a pretty little curve, but it's actually, you know, roughly two to the two fifty six dots that are Antinopoulos, the mastering Bitcoin book. I think it's chapter four, it does some visuals of what if the prime modulus they takes the prime modulus leaves is like twenty three and thirty seven. And it just shows how it starts even at those low numbers.
The curve starts to a look like it starts to look like a form. Yeah. You know?
UnknownYeah. So the the general nature of how this all works is you take your private key, which is a number, and you basically want, you use modular arithmetic and you fold it over this curve. That's the amount of times you kind of step through the curve starting at the generator point. Well, there's an operation called point addition. Yes. That's the so and it, there's... When you take a point and add it to another point, it becomes another point on the curve. It's... There's a way to visualize it, but it's not entirely intuitive. The cool thing about it not being entirely intuitive is
it's really hard to trace, like, you know, if you do this point edition a number of times, it's kind of hard to keep track. It's hard to really visualize. It looks like it looks kind of random, even though it's not. But that's and that's what makes it hard. It's sort of what makes it hard to even to make it makes like if you had a circle or a continuous curve real number and you knew the generator point you can solve, you can actually solve the private key.
Yep. But over the, so you have this point and you have this operation called point addition that has a specific definition. It's probably closer to like vector addition, if you're familiar, say, than adding numbers. Right. The point is you know that the closure tells you that adding two points keeps you on the curve, gives you Which is important. You always wanna stay within the curve, right?
UnknownAnd I'm actually looking at the study guide at the moment. This is because we're kinda teeing up why this is important is that these properties basically lend itself to the fact that to go from a private key to a public key is trivially easy, but to get from that public key to work backwards is basically impossible as known through our understanding of math today.
And this is usually described as, you know, there's more possible keys than atoms in the universe, and your best shot at trying to get my private key is actually just to guess all of the numbers. You have to just guess all of the numbers, which is not really a practical not really a practical exercise at all. Assuming you rolled own dice to get the Assuming you have good entropy. Right? You have to have good entropy. Now, there is one known,
optimization. It's even in the study guide. Theorem 5.2, Polar's row complexity, running rows out... Polar's row's algorithm for the group, you're actually able to take the square root of the field. So if it's two to the two fifty six, you're still at two to the one twenty eight, which is still pretty damn good. Right? Two to the one twenty eight. So just incidentally, like,
UnknownI told you about the Neil Koblitz book. Yeah. And we haven't really gotten to the ETH KFADE because Neil Koblitz actually works for the Ethereum Foundation as a No shit. But But his book is so good. And the first chapter, the first section's big O notation. Then the next thing he gets into is Paul Droe and the algorithms that reduce, that do reduce the solution set, solution space. Yeah. And so that's kind of like what we know today is like the things you can do in general and do It's also known as Shanks, I think, Shanks algorithm. But
Unknownit essentially square roots the set. Yeah. Yeah. Yeah. So now I try to take a little bit of time. So it's two to one twenty eight. Sorry. So it's basically not two to the two fifty six, it's essentially two to the one twenty eight is really the defense line. Yeah. That's the defense line, which is cryptographically secure. Because remember each each number to go from two to the one twenty seven to go to the two to the one twenty eight is not
small. It's a doubling. Yes. Right? We're talking about base two and increasing the exponent. That's why, maybe to make it a little topical for everyone, the issue with the cold card hack with the MK three, it was two to the 32, which is trivial for one computer to brute force the entire known space of numbers. Yep. The MK four was at two to the 70, which can be done, but that is to go from 32 to 70 makes it so that, like, you're gonna have to spend a lot of money to get there. Right. Is it fair to assume
Unknownthat that's probably still grinding? Maybe. It's also like I didn't do the final math on two to the 70 versus it may have been like two to the 56 versus two to the 67. Like, there was some fuzz space in there. The hope the hope is that with months of notice, people move off, and then the prize gets smaller which then resets the computation relevant for is this even worth going after anymore. If I'm the attacker, I'm gonna just keep doing it because I think there'll be one whale that I can get. Maybe. Yeah. But you've... Then then it's a calculated risk. Right? Whereas if you were able to keep this private somehow,
you would have, you know, probably won for the whole whole gambit.
UnknownI might have I might have been looking at... I might have been looking to see really how effective the public outcry is and how responsive The it would other thing too is to procure that level of compute is, it's not impossible, but you're not gonna log onto AWS and start grinding it out because you know, like it's gonna be pretty obvious, right? So not impossible, but like it's just something to keep in mind. So by the way, so just, I mean, how encouraging is that
Unknownthat two to the 70... It was a two to the seventieth? Is that... What was the MK four? I think m k four or five and q are all, like, two to the 70 because it did a 32 bit recede in the we live in now, we know two to the 70 is a far cry from... It's a far cry from two to the one twenty eighth. Yes. And yet it's still it's still really a unlikely and difficult to get the resources to do that attack. Casually. Yeah. That's like it's it's not quite nation state. Like nation state is like two to the 90, but like,
you know, it's it's on it's on... Yeah. Like, of what they could... And reasonably on opportunity cost point their problem there versus elsewhere. But, like, once... But now, like, you're going from two to the 90, which is, like, nation state level grinding to, like, two to the one twenty eight, you're like, oh, this is more compute than the entire available compute in Earth,
Unknownyou know, with many, many orders of magnitude in between padding you. Again, I'm gonna say something really stupid, but, like, it's just... Again, a nation state might have... Like, they might have just imprisoned the guy that ran Silk Road, and they may know that... They they know that he's on an MK four, and they know they have him for, like, eleven years, and they may just decide to do it. Maybe. Like that's... But it's still getting so specific that it's not like you can't... You can think like generally these things aren't gonna happen today.
Yeah. A year from now, things may be very different. I'm just thinking like the fact that even subpar, even you know, you just can't be so subpar. Yeah.
UnknownI'm gonna do something I don't think I've ever done before. I'm going to because I fed basically the study guide into all of the elliptic curve, math and the open theorems and proofs. And I'm gonna just read this to you cold. Love it. And we're gonna try and take this as a conversation because I had... I actually went and got an obliterated model talking about it. I'm trying to crack all the Bitcoin. Like, here's all of the data points we have. Like, help me walk through what are the things I should go attack. Right? And then the reason why I'm doing this is because I wanna actually have a grounded conversation in, like, one, attackers already do this. But, like, secondly,
I wanna have an actual conversation around the math that needs to break for this elliptic curve stuff to be relevant rather than it'd be vague posting and and, like, hand waving, and I tie it into the study guide to talk about the parts that specifically matter. So section 5.6 of the study guide has a little thing called the bottom line. You actually have a whole section here about the discrete log problem, which is basically the asymmetry of advancing forward to go from a pub... Private key to a public key versus the difficulty of going backwards from a public key to find that private key is the discrete log problem. That is the big universe of
the mathematical... Like, that is Atlas holding up the security of
Unknownelliptic curves. That's the hardness. That's where the hardness lives. Absolutely. And so you can go to section 5.6 of the study guide if you wanna go through that there, but I'm gonna just jump off from here. And so from the guide, we know that every signature verification when I provide a signature, it's actually a linear algebra problem. Right? You're just doing a quick check of the signature versus the public key and seeing that it's there. It's quick to verify. That's why it's so great. Because I can have a secret. You don't know that secret, but you can quickly verify that you know that I know the secret. Like, that's the whole under... When I send you Bitcoin, I don't have to give you my private key.
I just give you a signature tied to that private key, and you can quickly mathematically verify that I have the private key. That is what checking a signature is in Bitcoin. Now in 2020, a team of researchers at... Like, we're starting to program and ask a question. It was like, what if the discrete lock problem itself was a linear algebra problem that you could quickly backwards verify? Right? Could you, in theory, walk backwards and instantly get that key? In plain words, you take your public point and the generator, and you build a big matrix out of many random combinations.
And now every determinant
Unknownof every square sub block of that matrix is called a minor. Yeah. And by the way, just to pause. But this is Yeah, please. So if you go back to when we said looking at the elliptic curve over the reals, and I was saying how you could just... If you know... Basically, it's an algebra problem now where if you know all but one thing, you can solve for the one thing. Right? If you know the generator point, you know the public key and say, you know, basically when you say solving using linear algebra, the equation is simple. It's Ax equals basically Ax equals b, where A is your big, big matrix,
X is a vector that you're solving for and B is the answer. It's a vector. That's the answer. And so all the linear algebra problems is Ax equals b. It sounds really simple. Again, this is like the question that this is now the question is, you know, in a linear system, it's actually really easy to solve for x. So like one of the most famous solutions is least squares, Where it's like x prime x inverse x prime b something.
Sorry, a prime a inverse a prime b that would equals x. Right. You take the inverse. I'm gonna fuck that up. Like anyway, By like taking the inverse, yeah. There are certain systems that are really known and easy and already have like analytical solutions. And so what Rob is saying right now is that now they're trying to see if that can be done on an elliptic curve of a finite field. Yeah. And so what they're doing is is they're saying you take your public point in the generator and you build a big matrix,
Unknownright, out of bunch of these random combinations. And what you do is you take the determinant of every one of these sub blocks, and that sub block is called a miner. And they prove that here o r. M I n o r. M I n o r. Yeah. And if you were to find one of these miners, if you can find one that has a determinant... If you find one with zero, a block where the determinant vanishes, meaning there's some dependence. We'll we'll get... Well, I'll I'll go to this in a second. There's some dependence of this matrix.
There is a, and it's zero. There's a coincidence among the points that solves the discrete log.
UnknownThat's because the determinant is zero, you no longer have an inverse. And I think that it's called degenerate. Yeah. And so this is from my linear algebra days. I got a... I mean, I got a bachelor's in math over a decade ago. But if you view a matrix, right, like, can just think about maybe a two by two matrix with x and y coordinates. The determinant... The linear dependence piece is really important because, it means that you can actually express that matrix in a simpler form. Ah, we talked about this last week, by the way. We talked about linear independence last week. Yeah. So you just jarred me to do is imagine a matrix with two vectors. One is one, two,
and then the other below it is two, four. Mhmm. This is the same... These are clearly the same line. Right? So The same line. Exactly. And if I just did the determinant in my head,
Unknownit's zero. It is zero. Yeah. Because you're taking the same vector just on different, point. You're taking up... If you have two lines and you're just taking... The lines overlap. So it shows you you lack linear independence, also means you can't Within that matrix. Means you can't have an inverse. It's a... Yeah. All of these... This is... It gives you all of this information that kinda means something now. Right. And so not all is lost. Right? Because we're talking about a two by two matrix. You're gonna be taking something that's gonna be
way larger, way, way larger to do all of the combinations and the pieces. Right? And the idea though is that this could be a way that you could shortcut your way through the discrete log problem because you're basically taking a subset of the randomness in the universe and you're finding a way to accelerate computationally.
UnknownHow do you maybe find a key that could work And if you don't have perfect entropy, you may open the door for this. That's that's really Always. Always. That that... That's really what it comes down to. Like, perfect entropy gives you per... Gives you linear independence in the highest amount of dimensions, I. E. The two fifty six minuteus two thirty two. If someone wants to go do... Take some homework away from this podcast, you could take the subset of the entropy that was generated by the Colt Carton show it didn't have linear independence and how quickly you could collapse back down on this probably. Like, go ask your clanker that question. I I would say recommend that to people. Yeah. Yeah. Yeah. So this
field, like... So, like, just a... This is the why do you care? Because if a polynomial if the polynomial version is true and the set is cheap to find, then this... The security ends. There... It's not that it reduces. This is a a one shot, so to speak, destruction of elliptic curves as we know it. Yep. The door is open whether you... Whether anyone realizes it or not, the door is open at that point. Yeah. So they solved some toy examples using a space of two to the 50 back in 2021.
So if your elliptic curve was two to the 50, you could use this method to break it. Now you're still protected because, again, the universe between two to the fifty and two to the one twenty eight, two fifty six, whatever, is massively, massively larger, unfathomably larger, right, in the ability to do this. In August 2026, right, so two months ago, same researcher published a reality check, and his new search algorithm is polynomial, but only for a fixed parameter. Enforcing that parameter to make success likely costs more than it saves,
which is the saving grace. To apply this algorithm to try and crack it is actually less efficient than just brute forcing all the numbers. Right? You just... You you go back to square one. You're back to, yeah, there is a hypothetical way you could solve this, but at the larger scales that we're talking about, you're actually just way better off just going to the brute force method. Method. So that's kinda protected. But, the conjecture sits there, and it's something that... As a precisely stated problem, if you were able to prove it that it's not polynomial, that it's linear, that you could actually collapse the space down, that you would be able to basically break elliptic curves.
The... This is the one that if you wanted to go grind lean proofs to see if you can find some marginal improvement, this would be the area you'd go to, like, for the most straightforward one. And, you know, so that's, like, an example. Right? And the reason I'm talking about this is because rather than being like this really... Because I kept, like, I kept asking this question. Like, everyone's just, like, hand waving math is broken, so cryptography is gonna break. I was like, can you at least point me to the thing that's gonna break? Because you... Like, to even just say the discrete log problem is gonna break is actually insufficient detail. That's just saying that, like, cryptography will break. You're just repeating yourself. This isn't a hypothetical theorem that could, in theory,
if you found a lean proof that found it to be linear, that you would be able to break the whole thing. It's the same extrapolation
Unknownthat, like, same low level, low thinking extrapolation that is just more like there's gonna be super intelligence. This... You know what I mean? It's just like that there's gonna be some state change from a to b. So they're not looking at like, Oh, I know exactly how they're gonna break cryptography. It's just, Oh, we're gonna have a state change where it's gonna be more intelligent in a way we can't even comprehend. And obviously, it will break. Obviously, that will break all of our Yeah. Powder. Yeah.
There's actually a little bit on Shoop's theorem, as we mentioned earlier in this podcast with Polar Dzrow. And the very brief one of that is that it's not actually looking at the numbers inside. Could you, in theory, do curve points leak structure to an attacker who reads them? The starting point right now is no, but in theory, you could start grinding away at that problem with formal proofs to be able to
break things down. And then one other... I mean, just kinda thinking if it was a menu jump in at any point. We we were just talking about coblets. You know? Mhmm. Bitcoin uses a coblets curve, and that's because, you know, we have y squared equals x cubed plus seven. In theory, there is a x squared value, but the constant is zero, so it just disappears, and that's the last one more. There's also an x value. There's also an x value. Yeah. If in the ax
Unknowncubed plus bx squared plus cx plus d, in the Bitcoin's elliptic curve, the coefficients b and c are zero. Yeah. A is one and D is seven. Yeah. Right? And so it was also known as the wire stress form and this was done intentionally by Satoshi for speed. Yeah. Right? Yes. Correct. And
I think there are a lot of idiosyncratic things that Satoshi did that probably will keep Bitcoin's elliptic curve ahead of the other people that the bear is chasing.
UnknownMhmm. Right? Like the resistance to certain NIST standards, the resistance to certain prime numbers that we know already have been already have been compromised. Right. Right. We did a lot of talk about that, Satoshi picking this curve in particular. That was not a NIST certified curve. It wasn't, you know, Well, sorry. NIST cert... But it wasn't one that was really used anywhere. And the reason why Satoshi picked it was because there were no weird constants to explain away, which is typically where backdoors come in. But it's important to talk about this because the head... We know what's gonna happen. I... It may happen this year where the headline is going to be elliptic curve cryptography has been broken.
And they're gonna break some rinky dink curve that isn't used in Bitcoin.
UnknownYeah. Or Is that your theory? Yeah. My theory is it... Or they don't. It's just they're gonna be able to... Anytime... If anyone says elliptic curve theory... Elliptic curve cryptography has been broken, it's going to be chaos in our world. Of course. Right? And this is like... I hate to be a broken record, but this is why we do this podcast. Yeah. Is so that at the very least, we're not in chaos. We're like asking other questions like, Oh, well, you know,
was that curve? Right? I mean, because it's, you know, what we talk about here is not all elliptic curves are equal. And a lot of what protects us is not at the level of what you would say is an elliptic curve. Right. It's the very large prime numbers. It's underneath that level. The specific prime number that...
UnknownRight. I wanna... There's a tweet thread that I'm gonna... When I... Because I'm just closing out the rest of these, like, possible areas that are open at the moment. But Yehuda, Lindell, the head cryptographer at Coinbase, one the leading cryptographers in the world talks exactly about this, that the nature of elliptic curves and the strength of them don't come from the same area of math that these machines are kinda finding more optimal algorithms. Maybe I'll just jump into it briefly to talk about it now. He did a really good Twitter, thread about We
we rely on fundamental hardness. You you cut out for a second. Are back? You're back? Cool. You're back now. Sorry. You you were breaking up for a second. No. You're good. That... Basically, the fact that AI... I'm gonna read his thread verbatim. I wasn't gonna comment on this because it's a really bad take, this whole bunker mode, like, general thud. But since it's taken off, he feels the need to. To my understanding, there's no evidence whatsoever pointed to a break of decades old hardness assumptions like elliptic curve cryptography. The fact that AI can prove theorems
and and that have been hard does not indicate in any way that problems are assumed to be hard or not hard. Our assumptions on the hard problems are not based merely on human fallibility, but on the belief that inherent hardness exists. Of course, we cannot prove that elliptic curve discrete log is hard. Since we haven't been able to prove p does not equal NP, we'll get to that in a minute, but throwing out that AI is doing amazing math, so elliptic curve hardness must be in danger has no logical basis whatsoever. Furthermore, it would be really...
If it really were to be broken, the ramifications to our digital world, including traditional financial industry, would be immense. Once I can break elliptic curve cryptography, I can generate fake public key infrastructure certificates and impersonate bank websites. Worse, I can generate malicious banking apps and even operating systems and sign them and push them to people's phones and computers and completely take them over. This is the fear mongering since there's zero evidence to back that this capability exists. And there's zero evidence that elliptic curve hardness is more vulnerable than hash function hardness. In fact, historically, hash functions have been broken more than elliptic curves. So this is based on conjecture rather than any evidence.
In conclusion, making such statements without evidence is the opposite of responsible behavior. It is the very definition of FUD, fear, uncertainty, and doubt. It cannot be proven wrong, but then there's no evidence whatsoever of it being true. And that's just trying to talk about the nature of these problems. It's just immensely different. So I think that... I I quote tweeted that talking about it. Like, it is absolutely kind of the big, like, walking away piece we should talk about here to... Do you have anything to add to that? Compliments to the chef. You're on mute. There you go. Oh, there you are. You're you're just delayed. Okay. We'll we'll we'll... You got like a...
You're in the... Now you're back. Basically
Unknowncompliments to the chef. I gave some applause. I mean, that's a really great, it's a terrific rant. Excellent. Yeah, So we're just to tie this off. I wish it was from a place that people respected more, unfortunately. This is like, again, why we do this. You got to see the message, not the messenger. Absolutely. Absolutely. Right? Not that I don't even know this person. I'm just saying people. Oh, he's he's a world round cryptographer. He's the head of cryptography at Coinbase. Like, he
Unknownhe's not doing this for engagement. You know what I mean? Totally. It seems like he's on the same mission we are. He's like, let's let's let's have some freaking sense here. Absolutely. Yeah. So to tie it off with cobblets curves. In theory, we have this increased efficiency with the computation, and it technically gives the attacker a small discount, but nothing that is relative to the benefit everyone who's honestly using it gets. In theory, basically, could this be degraded as a Koblet's curve in a way that could result in some issues? In 2026, a lean formalization proof stress tested one combination of the symmetry
around this with, Samed's polynomials and found only the trivial symmetry survives, so a bounded negative result. Basically, it was a lean proof proving that this is not a path to be able to break it. But in theory, that could be a path you could try and go for from a different approach. Another one here, so in chapter sixes through seven, we start talking about elliptic curves over the reels and the smooth loop picture. And to say now you modulo P the prime around it, which is the two to the two fifty six minus two to the 32 minus nine seventy seven, right? That's the prime number that we use the base all of our math around in the Oh, let's call it big P. Big P, that's right.
So we talked about this. Here's an attack idea. The guide basically teaches it backwards. What if we could take a scatterplot problem and lift it back up over the smooth curves, solve it there, where we have calculus, geometry, and all of our existing mathematical tools, then push the answer back down to mod p. Why doesn't this obviously work? Because when you lift, the numbers grow. There's a height function. Think of it as an altitude of the point. And the classical theorem says the altitude you need blows up exponentially. So you're basically trying to walk backwards across the curve, but every time you take a try and take a step back, the competent metrics get bigger and bigger and bigger. Like that is like the inherent hardness of this problem.
So in theory, you could actually maybe find, some weakness, not like a total break, but like a weakness. The bounds aren't tight in every regime. In twenty twenty five, some of the work people are doing in quantum for completely different reasons show that high altitude versions of this exact curve are arithmetically rich, and there are lifts and twists with huge amounts of structure. Nobody has connected that richness richness to a classical attack, but nobody has proven it can't be connected. So it's kind of like an open problem. So there's a low probability high payoff
that the attacker route we talk about in our study guide for doing all this. Basically could try and collapse it back, but it's an open question. It hasn't been proven positively or negatively in either way. But it... This is kind of talking about Shor's algorithm. We talk about the ability to be able to walk backwards on this. It requires a quantum computer. Like, this is Shor's... Like, it's an algorithm that if you had a quantum computer, you can, in theory, walk backwards
Unknownacross many points simultaneously to be able to collapse and get back to the original key. This is basically the bucket of the quantum attack as a thing. Talked about with Reardon, you know, that it's it's not a math prob... Just because it's not a math problem doesn't mean it's solvable. It's... So it's an engineering problem. Yeah. It's an engineering and physics problem that, okay, they know them. The math is solved. Great. That doesn't really mean that they're any closer to launching the attack because
you know, there's a lot of math problems that require physical and engineering solutions that Mhmm. You know, that's really the trick, isn't it?
UnknownYes. And so the last bucket, which I briefly mentioned earlier, basically distills down to p equals NP, p does not equal NP. So this is, like, the most foundational problem in math. There was... So, like, there were, like, seven, like big millennial problems one of which was Naver Stokes. Yeah. And this this is also one of these fundamentally fundamental problems that there's like infinite funding to try to get somebody to try to solve. Yeah. And so in a sentence, p equals n p. When everyone ever says that in plain English, what they're talking about is,
is finding an answer ever truly harder than checking one? Because remember, you get to check this stuff when you do signatures. I can verify these checks. I can quickly take a private key and quickly get to a public key and and check the the math. Can we get... There is no... And this is the reason why you can't prove elliptic curves are secure is because of p equals NP, and we can't... And related to that, the discrete log problem, you can't prove a negative because you'd have to conclusively prove
all possible aug... Algorithms are insufficient. Right? So it's a it's a fundamental, like, ep... Epistemic problem of, like, the knowledge gathering. You cannot prove a negative in this sense. And so taking it off just from right here here, you know, all of the verification, all the stuff we've done with Bitcoin is traditionally defensive. We're proven like this path doesn't work. This path doesn't work. This path doesn't work. But the hypothetical amount of paths that could be taken to weaken elliptic cryptography
is innumerable. And because of that, it's a... What we're relying on is, you know, this open problem of p equals NP. If we could prove that p equals NP, that would also break elliptic curves, because that would be able to mathematically represent that you will be able to... If you have a problem that, you know, that you will be able to in somehow trivially reverse engineer an answer. And that's just for all math and all numbers and all things. Like you... Like,
Unknownin general. It probably is. I think it's true that it's breakable. I think it's always true that it's breakable under a certain power and compute. Sure. And that's why... But that's why, like, you know, the Byzantine general's problem as a... Was not a math... It's not a math solution. You know? Like, it it combined physical it combined physical resources and value.
And it's just... It's more of an agreement that, you know, like Bitcoin in year 18 that we're getting close to is becoming an agreement of like, it's just not worth it to really try to do it. Right. It's not worth it to undo this thing. So yes, we're just going to agree to a world where these things are much more trivial to verify than they are to calculate.
Whatever gets done, like the whole system that gets done to create the blockchain is like a toy model, I feel like of what this concept is. It's like a toy model simulate like if you know, if we were to simulate it, this is exactly how we would do it with this kind of game here. And it's not like we don't, this in a lot of ways, it's like believing in God where I care
if somebody proves it or not. It really doesn't change anything. Like, we're... We wanna live in this world where this exists, and it happens to be, you know, it happens to be crypt... Like, cryptographically secure doesn't mean it's proven in math. I guess that's my my point. Isn't that Yep. You know what I mean? It's more of like, okay. Well, if it becomes uncryptically cryptographically secure, we have to really...
We have to hope that we've reached a point of maturity where we're able to make it... We're able to make the adjustment to make it re secure again. And it's, I think, it's all like, this is a part of, like, human existence that we're always going to have to collectively figure this out through time. The reason and the fact that it's valuable is that the incentive will always be there to attack it. The incentive will always be there to create more efficient compute, more efficient algorithms and make operations more efficient to attack it so that we can accelerate
the defense of it. You know, it almost reminds me of Dwight Schrute's quote in The Office where, you know, I think he's in the pilot, you know, and he was like, Oh, the Schrute blood is the best. I can raise and lower my cholesterol on command. Yeah. Why would you ever want to raise your cholesterol? And it's like, So I can lower it.
UnknownRight. Yeah. But it's... You know what I mean? So, like, I... It... It's this... I just don't know. I I don't know if it even is meaningful to math... You know, to say, oh, we can prove that cryptography is like that that elliptic curve cryptography is or isn't secure per se. Yeah, I think it's also a good point too, as it relates to P equals NP we're talking about. Like it actually... Bitcoin doesn't determinately settle it either way. You could prove,
if in theory you broke Bitcoin, it would not prove that p equals n p. And also Right. If we've proved that p does not equal n p, that that does not mean that we prove that p does not equal NP. Right? Like, it's just... It... It's something that if we're to found, we could have related, like, consequences to Bitcoin, but Bitcoin itself... Basically, p equals NP is a bigger problem than than elliptic curves. And so, if you get a terminate answers from p equals NP, that could have implications on Bitcoin. But if Bitcoin either stays secure or breaks,
it doesn't really actually go up to the level of p equals NP. Basically because it's even a closed system. It's not It's Yeah. Bitcoin has bosses in the math store is basically what we're talking about here. And p equals MP is something that sits above Bitcoin, not the other way around. I think there were maybe even one or two other small things I skipped because I think there's like seven like open problems that we've talked about. Maybe just to recap for the class so far. On though, real quick, isn't it? Isn't... I mean, I'm just personally comforted
Unknownby the fact that there's somebody at Coinbase that is as, on top of things as the person who made that. The head of cryptography at Coinbase. Their head of cryptography, it's comforting, right? Just don't don't you think it's... Absolutely. Absolutely. So I'm just saying, you know, maybe there are more than three or four high priests out there that we have to... That we're trying to get closer to. I've exchanged emails with him. Maybe we'll try and get him on the podcast. That'd be pretty sick. Yeah. Well,
Unknownhe's done some podcasts before, so we can probably grab him. So to just go through the list of the problems we talked about, and I'm gonna go very quick because, you know, we're already an hour and fifteen minutes in. I don't wanna get everyone, lose everyone who's probably already asleep. Nonsense, nonsense. Right. Basically, the difficulty of factoring a point on the curve, like if you somehow could,
break that, we talked about, the the initial minors con... Conjecture, where you could just take these like points, and if you could find a determinant equal to zero, could you collapse the space down? We had from, Shoop's theorem, basically, the the ability that the attacker gets the square root ability to get efficiency, could that be bled down to an even greater advantage that could collapse the whole thing?
The cobblets curve angle of that there is an increase of the verification time, could that also, allow for an attacker to decrypt in a way and work... Walk... Not not decrypt, but walk backwards to be able to get to the final, key, your private key. We talked a little bit about P equals NP, and I think there's also what we have here. You can invent a point factoring, which I guess is kind of what we talked about earlier. I already caught that one. And then,
there's one that actually uses a bit of isogenies cryptography, which is probably its own series you go down. This is one of the popular paths that are being discussed as it relates to quantum computing. You have three main camps of hash based algorithms, lattices, and isogenies.
UnknownAnd people who listen to the podcast should be... Already be like a little familiar that an isogeny is an isomorph... It's like... It's it's a it's a something morphism. It's a morphism between elliptic curves. Yes. Second, an equivalence. It's not an isomorphism, but it's one of the morphisms where there's a there's an equivalence. So the idea here is leveraging these isomorphisms. Could you hop to a weak cousin This is basically like in plain English what we're talking about here. This is in the study guide chapter,
7.3 and a little bit in chapter four, two, we talked about Hassa's theorem, and a curves point is pinned near p plus one, and that all curves points form one group. And one thing I didn't mention is curves come in families. And the idea here is that curves have cousins, right? For any curve, exists other curves, it's isogeny's class that can be connected to a biasogeny, it's a translation between curved languages. And isogenies is a map that sends a point to the end respects the group law, and sums go to sums, and it has one property that is very critical here.
If q equal dp on curve egg, right, so this is the public key equals your private key times, you know, to do the arithmetic, the generator doing the arithmetic. You can actually have your secret number survive the translation between the two curves. So the idea is that could you find a cousin curve to sec p that could honor all the properties, break that curve, and since the private key survives, you'd be able to crack a different curve than sec p two fifty six k one and then hop that word backwards to the Bitcoin curve. Right? So it's like Right. So it preserves the hardness. It preserves the hardness of the discrete log problem.
Yeah. It does. Exactly. Right. So maybe a hypothetical attack would be a common metrics of some of these things we're talking about to try and like get your way in. So does Bitcoin's curve have a weak cousin in the family? A curve that's that where the log is easy, anomalous and pairing friendly that you could tunnel to. The catch is, is that the curves connected by isogenies over the same field have the same number of points. So you're back to exactly what you're saying is you're back to two to the two fifty six big P numbers of trying to break this stuff.
In 2004, there was a theorem by Zhao Miller Van Costen, and this is like a belts and suspenders approach. They proved assuming GRH. Now I'm gonna be honest and sound, what is GRH? GRH math, elliptic curves. See, we're all just learning things on the show. That's what this is all about. The... Oh, the general Riemann hypothesis.
UnknownAh, okay. Yeah. Okay. That I know. Yeah. I just never saw it abbreviated that way. And so, plainly, all doors in the family have locks of equal strength, and the quarters of them, between them are short. So you're back to square one, basically. And it's probably worth noting for whatever it's worth that, Jonas Nick, one of the best cryptographers also in the world who probably fully understands what we're doing. He didn't seem to go the isogenies route. Correct. Right?
Probably So good to have him on and maybe We should talk about... Yeah. I
Unknownthink that the emerging consensus right now is hash based algorithms because we already use hashes in Bitcoin. So it's the one that introduces no new security assumptions to Bitcoin, whereas lattices and isogenies do. The trade off is that, hash based algorithms are larger. And you have to maintain state, which means, you know, you can't have if you leak key information, if you basically aren't keeping a counter of when did I I signed this transaction, I signed this transaction,
you need to keep a running count within your signer of how many times you've signed with the key. Otherwise, you will leak your private key. It's basically... Or parts of your private key, which is basically the unfortunate trade off that we would have to go do in a post quantum world with hash based algorithms. So, I mean, that's all part of those trade offs.
UnknownI mean, the good news to me is based on where we are and all the ideas that are out there, I mean, We have time to get comfortable with all of these things probably. We're probably not dealing with anything seriously threatening in the next don't think in the next five to ten years that we can have these people continue to do this work and get us comfortable.
Like, it's not gonna be a snap of the finger, oh, we're changing, you know? Yeah. Like, this is gonna take... It's, you know, it's gonna have to be a slow process. Yeah.
UnknownAnd so maybe just to quickly bullet point off all of the lists again, of all of the things in math that are related to the stuff that you should be thinking about. And if you see advancements in these kind of things, maybe revisiting the conversation. We're gonna start with... We have the general route one family of point factoring. That's been an open problem for a while. You would get basically the security drops from two to the two fifty six to, like, two to the 80 to two to the 112.
So still pretty strong, but, like, that's what you need to think about. The next one would the initial minors one we talked about with, finding if you get the determinant to equal to zero. Next, you're gonna have... Can you can you hear my kids rioting in the background? It's awesome. Yeah. I can hear my wife shushing them, but she is losing the battle, absolutely. So you have, you know, the the the point factoring piece, you have the initial minors line, with all of these, like, submatrices that... Of minor matrices that you'd be able to, if you found the determinant equal to zero, you'd be able to quickly collapse things down. Curve- To me this is finding weaknesses
Unknownin the system, finding definitely and non perfect finding the weaknesses in the door. Yeah. This is finding the weaknesses for sure because you need to do it for each point. So you maybe find a way to classically reduce the computation required to go from two to the two fifty six unique points on the curve to something smaller. Then we have the curved symmetry, which we were talking about. It's an open problem for you to do a lean proof on it. The lifting one,
I guess, for the curved symmetry, talking about, you know, this was the cobblets curve and stuff. Right? Like, being able to, like, find your way backwards on that lifting. Can you beat the height penalty basically as you get further and further higher to try and walk backwards? Are you able to start, efficiently breaking it down? It's it's basically open, classical bound, say no, but it's not very tight. A 2025 quantum computer shows the curve lifts this arithmetically. This is Shor's algorithm. This is the whole quantum door. This is why... This is this is the door quantum is knocking on Yeah. For breaking elliptic curves.
Then we have isogenies, and could you, with the Riemann Hypo... Could you detour to a cousin curve? As we just discussed though, for the curve to be part... The actual proper cousin curve, it needs to have the same level of modular p. So you're back to the square, you know, you're back to the beginning again. Can't be isomorphic
Unknownor homomorphic. So this path by itself? Right. Unless this is Lagrange's theorem, but like you... It has to have the same amount of So to use this, you would have to use this in combination with something else. If you would do a cousin curve with something else that would then break the whole thing. Like, is, this is an alley oop setup. This is not in of itself the solution because walking through the cousin curve, you're just back to where you started. I like it. And then there's just combo. Yeah. It's a combo billiards, combo pool shot. That... That's right. That's right. And then there's the general just
gap of... We talked about a Shupin blind attacker... Attackers can't beat square root of n. Could you, in theory, improve that to something more? We talked about that earlier. And then the next, like, kind of tier, like defensive, improved theoretic stuff with one way functions proving that they can maintain exist with p equals m p. If you could somehow, in theory, you're not really able to, but just hypothetically, if you could prove that the discrete log problem was hard, then that would be a a defense buttress, not for cracking, but for defending.
The hardness of SHA two fifty six, which I'll be honest, I wouldn't, I think SHA-two 56 in a long enough time frame also fails. It's just a question of when. And then there's some other things of just like... Those those are the big those are the big things. I'm looking through the rest of this here. There's formalization work that could be done in lean to either... That that all feed into this stuff. But that's... So there's six offensive problems, point factoring, solving degrees, initial minors, the symmetry, the lift, and the cousin. And there's three defenses that no one can really climb. If we can prove one way functions
are strong, p equals m p, improving the curve itself is hard. Those are, like, the defense. So there's, like, five... We're looking at this. There's six offense, and there's, like, three defense, and they're all open problems. And that's kind of, I think, a good just... When everyone's like, it's gonna break. It's like, okay. What... This was part of my... Making this study guide this week was my journey because everyone's like, it's gonna break because math is gonna break. And I was like, okay. Can we talk about the specific things in math that have to break to actually have a problem here? And packing 11 squares,
optimally in a space while being very interesting, I think, that could probably have downstream interesting applications. Going back to, Yehuda's, like, Twitter thread, we don't rely on elliptic curve cryptography because we just think that humans haven't figured out an efficient algorithm yet. We believe the problem itself is sufficiently hard and has hardness inherent to the nature of the math being done. And so I wanted to enumerate the hypothetical things that could be done that could reduce the security, and that's, you know, that's what we went through today.
UnknownI think the title of this episode might be, like, red teaming elliptic curve. Yeah. Something like This is red teaming. Yeah. This is honestly red teaming. This is going through and thinking about what are the doors to knock on, what are the things that could be plausibly insecure, all of that stuff. Absolutely. Awesome. Great stuff. Anything else you wanna do or? Let's see. I'm not sure. This was the meat and potatoes or vegetables. Well, it's meat and potatoes for you and me. Maybe it's vegetables for everyone else in the audience listening.
I'm trying to think through Things got real. I feel like shit got real today. A little bit. Yeah. We had to It has to. Rather than like passively explain stuff, had to go on the offense a little bit and talk about the adversarial nature of the cryptography that holds up all of Bitcoin.
UnknownMaybe that would the Amsterdam presentation maybe. But we should do that. Because, and that's like, man, because like my, our contact at BTC Inc. Was really struggling to like, tap and click like, am doing That's it. No, this is why we're doing the map panel. This is why we're doing the map panel. Gonna... I this is absolutely it. Then he I've been holding off like, so he was like, hey, just give me the description and I'll put it on the website. And I was holding off because even for me, it wasn't fully clicking in now. This is it though. This is the talk. Got it. And now that I kinda teed this up for you today, I sent you my notes.
UnknownYou... We could do... You could brush up on it now. We could have an interesting conversation around it. Yeah. Now we can do some work and prepare some... Like a pretty band on This is perfectly pod conf aligned. Do a podcast to get inspiration for the conference talk. Big Bitcoin's in control. Big Bitcoin and big math are on a collision course. That's right. That's right. Good stuff. Right. Well then Magic Internet Math is in the books.
In this episode, we picked back up on our red-team tour of Bitcoin cryptography and tried to make sense of the last few weeks of fast-moving math news. We unpacked why the recent excitement around the exponential time...
Magic Internet Math: https://www.magicinternetmath.com/ Shout out to https://x.com/uncleJim21 for the Alice in Wonderland story and the following rebuke of Lewis Carroll: https://www.youtube.com/shorts/33FYXpEWgyA This...
We’re back because math keeps refusing to stay in the classroom. In this episode, I unpack two very different kinds of “math pop culture.” First, we revisit entropy, seed generation, and wallet security through a...
Magic Internet Math is back. After the summer that set the Bitcoin world on fire - the Cold Card entropy disaster - Brian and Rob Hamilton (AnchorWatch) reunite for a two-hour deep dive on the event Rob spent three...
In this episode, Rob and I sit down with Brandon Black (aka Rearden Code) for a deep dive into the quantum-computing debate around Bitcoin. We unpack why Brandon thinks the more useful framing is not “post-quantum” but...