The Riemann Hypothesis has apparently been proven

Any topics primarily focused on metaphysics can be discussed here, in a generally casual way, where conversations may take unexpected turns.
Kaje977
Posts: 88
Joined: Fri Aug 23, 2024 9:23 am

The Riemann Hypothesis has apparently been proven

Post by Kaje977 »

With the help of Claude (presumably a special version available only to Anthropic employees) and a mathematican, it appears that one of the greatest mathematical problems has been solved a few days ago. The proof is 250,000 pages long and has been verified by Lean (a proof checker) and other proof checkers. Of course, it's not yet absolutely certain whether this proof is truly sound or whether the Lean kernel might have a hidden bug that Claude simply exploited. Nevertheless, it's a unique achievement.

However, the most crucial point seems to have escaped the notice of most people: The fact that an AI can presumably solve such complex proofs of theorems with ease should increasingly prompt us to reflect on whether the current way of thinking, merely "shuffling" mental puzzle pieces back and forth can truly be the path forward for humanity. AI is increasingly stripping the "life" out of the well-known major theorems (which have often inspired many to write poetic songs and texts). It is the definitive "indicator" that should point to the fact that there is still something in Thinking that an AI can never "capture": living, real-time thinking activity.

Where will this lead us? Once large AI models start to tackle on the P-NP problem, will it perhaps finally provide the spark to look deeper?

For anyone interested, the entire repo is here:
https://github.com/CocoSgt/riemann-hypothesis-lean

And independent checkers seem to be almost sure about the proof:
https://www.scientificamerican.com/arti ... ns-attest/
Kaje977
Posts: 88
Joined: Fri Aug 23, 2024 9:23 am

Re: The Riemann Hypothesis has apparently been proven

Post by Kaje977 »

Independent Lean verification checks:
https://math.mit.edu/~drew/grh/

It seems the proof is actually correct. Insane.

EDIT: It looks like the rumor about the proof being 250,000 pages long is misinformation. Apparently, this was just the "summary" generated by Claude but Alpöge refused to publish it, for obvious reasons. The proof is to be found in the Lean code and a write-up will follow soon. Several independent lean verification checks have been done on it, and it looks like the real deal.

P-NP will be next, I assume.
User avatar
Federica
Posts: 2850
Joined: Sat May 14, 2022 2:30 pm

Re: The Riemann Hypothesis has apparently been proven

Post by Federica »

Kaje977 wrote: ↑Sat Nov 21, 2026 12:08 am However, the most crucial point seems to have escaped the notice of most people: The fact that an AI can presumably solve such complex proofs of theorems with ease should increasingly prompt us to reflect on whether the current way of thinking, merely "shuffling" mental puzzle pieces back and forth can truly be the path forward for humanity.


Exactly. The fact that not only mathematics, but conventional science in general is being practiced more and more in this loop of iterative human-AI collaboration is making the split more and more real, between those who feel - as you say - that there must be more than shuffling mental pictures to true human-ness, and those who are getting completely hyped by the mindboggling acceleration in material problem solving.

While some will say: "since human thinking (as commonly interpreted) can be reproduced and exponentially potentized within inanimate devices, there must be more to the human evolutionary path", many others feel that being human precisely means to deepen this material problem solving capacity, anchoring their identity more and more to the material spectrum.

In this sense, AI may become the means by which this split in humanity begins to take shape with certain clarity and acceleration. As commonly intended, human intelligence is being externalized. It can be directly outsourced to matter. Now, do we get carried away and fully establish identity inside that cave, or do we reclaim our future by owning our expansive nature beyond it?


In a way this externalization of capacity is continuous with the preceding externalization of tasks to grosser machines. It connects with a lecture I have been browsing today:


"That is to say, if the work done by machines had to be done by people without machines, there would have to be six hundred million more people living on Earth. But when what I now call mechanical occultism enters the realm of practical action - which is an ideal of those secret centers - it will be possible not only to accomplish five or six hundred million people's labor, but approximately one billion 80 million people's labor. The possibility will thus come about of rendering unnecessary 9/10 of the work of individuals, within the regions of the English-speaking peoples.
Mechanical occultism makes it possible not only to dispense with nine-tenths of the work currently performed by human hands, but also to paralyze any rebellious movement of the then-unsatisfied masses. The ability to set motors in motion according to the law of resonating vibrations—this ability will develop extensively among the English-speaking population. This is known in those secret circles. They expect this to be what will grant them supremacy over the rest of the Earth's population during the course of the fifth post-Atlantean epoch."


"The Fundamental Social Demand of our Times" GA 186/III "The Mechanistic, Eugenic and Hygienic Aspects of the Future"
https://steiner.wiki/GA_186#DRITTER_VOR ... ember_1918
"The release of the senses form their bondage to matter can also be undertaken through warm and intimate perception of the life and phenomena of nature. In this case, thinking is connected to Intuition and to the perception of the ever-unfolding archetype that lies behind the organic world."
User avatar
AshvinP
Posts: 6759
Joined: Thu Jan 14, 2021 5:00 am
Location: USA

Re: The Riemann Hypothesis has apparently been proven

Post by AshvinP »

Kaje977 wrote: ↑Sat Nov 21, 2026 12:08 am Where will this lead us? Once large AI models start to tackle on the P-NP problem, will it perhaps finally provide the spark to look deeper?

Thanks for sharing this, Kaje.

I think the spark to look deeper will always arrive from an orthogonal direction to our modeling efforts. In other words, the spark arrives from our feeling of futility with the current efforts, their ability to make our higher human ideals more likely to manifest, and the corresponding intuition that we need a whole new perspective on what we are doing in our intellectual life. As Federica noted, observing these AI solutions to otherwise intractable intellectual problems can push the intellect in either direction. The problem is that, without the needed shift in perspective, we are much more likely to move in the direction of even greater fascination with 'machine learning' as a means of understanding Intelligence. It is a catch-22: the perspective shift we hope these developments will stimulate should already be there, to some extent, for that stimulation to occur.

It was good to see previous Fields Medal winners pushing back against this trajectory. As they put it:

"In recent months, the success of AI in solving major mathematical problems has made headlines even outside mathematical circles. But solving problems is only a tool and proxy for achieving the primary goal of conceptual understanding and insight. Forgetting this in the world of AI may turn the tool against the primary goal. Indeed, the mass production at faster and faster pace of "true/false" statements could destroy fertile ground instead of breathing life into new ideas."

But if we are being realistic, their intuition of its misaligning tendencies can hardly stand up against the practical ease and convenience of pursuing it further. Since these AI-generated solutions have immediate practical relevance for certain fields, there is little incentive to stand in their way. Why is conceptual understanding and insight the 'primary goal'? Why do we need to preserve this fertile ground instead of pushing it to yield as many ideal crops as possible, as quickly as possible? We simply can't properly orient ourselves to those questions without a deeper perspective on the evolutionary process and what we are 'up to' within it. Yet that perspective becomes less and less likely to take hold the more our technologies 'succeed' in obviating our opportunities for creative experimentation with the inner process.

Aside from that, I am curious about the P-NP problem. Could you say more about it and, if it were solved, what practical ramifications might follow?
"They only can acquire the sacred power of self-intuition, who within themselves can interpret and understand the symbol... those only, who feel in their own spirits the same instinct, which impels the chrysalis of the horned fly to leave room in the involucrum for antennae yet to come."
Kaje977
Posts: 88
Joined: Fri Aug 23, 2024 9:23 am

Re: The Riemann Hypothesis has apparently been proven

Post by Kaje977 »

AshvinP wrote: ↑Sat Nov 21, 2026 4:40 pm
Kaje977 wrote: ↑Sat Nov 21, 2026 12:08 am Where will this lead us? Once large AI models start to tackle on the P-NP problem, will it perhaps finally provide the spark to look deeper?

Thanks for sharing this, Kaje.

I think the spark to look deeper will always arrive from an orthogonal direction to our modeling efforts. In other words, the spark arrives from our feeling of futility with the current efforts, their ability to make our higher human ideals more likely to manifest, and the corresponding intuition that we need a whole new perspective on what we are doing in our intellectual life. As Federica noted, observing these AI solutions to otherwise intractable intellectual problems can push the intellect in either direction. The problem is that, without the needed shift in perspective, we are much more likely to move in the direction of even greater fascination with 'machine learning' as a means of understanding Intelligence. It is a catch-22: the perspective shift we hope these developments will stimulate should already be there, to some extent, for that stimulation to occur.

It was good to see previous Fields Medal winners pushing back against this trajectory. As they put it:

"In recent months, the success of AI in solving major mathematical problems has made headlines even outside mathematical circles. But solving problems is only a tool and proxy for achieving the primary goal of conceptual understanding and insight. Forgetting this in the world of AI may turn the tool against the primary goal. Indeed, the mass production at faster and faster pace of "true/false" statements could destroy fertile ground instead of breathing life into new ideas."

But if we are being realistic, their intuition of its misaligning tendencies can hardly stand up against the practical ease and convenience of pursuing it further. Since these AI-generated solutions have immediate practical relevance for certain fields, there is little incentive to stand in their way. Why is conceptual understanding and insight the 'primary goal'? Why do we need to preserve this fertile ground instead of pushing it to yield as many ideal crops as possible, as quickly as possible? We simply can't properly orient ourselves to those questions without a deeper perspective on the evolutionary process and what we are 'up to' within it. Yet that perspective becomes less and less likely to take hold the more our technologies 'succeed' in obviating our opportunities for creative experimentation with the inner process.

Aside from that, I am curious about the P-NP problem. Could you say more about it and, if it were solved, what practical ramifications might follow?
Hello, Ashvin. Yes, the P vs. NP problem is probably "the" problem of theoretical computer science. What the Riemann hypothesis is to mathematicians, the P vs. NP problem is to computer scientists. To understand the problem, you need a little bit of prior knowledge in computer science, but I'll try to explain it in an intuitive way:

Almost every one of us owns a computer nowadays, and a computer can execute algorithms, i.e., a kind of recipe used to solve a specific problem. It is an exact step-by-step instruction to solve a problem, e.g., something like: "Sort this list alphabetically for me". Now, for us computer scientists, it's not enough to merely discover such algorithms, we naturally also want to know how fast and efficient they are. If you approached the matter naively, you would simply take a stopwatch, let the algorithm run, and then see how fast it is. It's clear that this obviously doesn't work like that. An old computer might execute the exact same algorithm much slower than a supercomputer. So, in order to maintain a certain universality, what you do is measure algorithms not in seconds, but in computational steps depending on the input size (which is referred to as n).

E.g., if you are sorting a deck of cards, then n is the number of cards.

So now, when the problem size n grows, the number of steps needed to solve the problem naturally grows as well. (e.g., from 5 to 10 cards) Therefore, to describe how fast this effort grows, one uses the so-called Big O notation. As an example of some Big O notations, we can use the following:

O(n), for example, describes linear time. Example: You are looking for a specific card in an unsorted pile of n cards. In the worst case, you have to look at every card once. That means: Twice as many cards = twice as much work. So far, so good. Ideally, of course, you want to have algorithms that have linear time complexity.

There are, of course, also other algorithms that have such notations (and you probably know some examples of your own):
O(n^2) describes quadratic time complexity. Example: At a party with n guests, every guest is supposed to shake hands with every other guest. (e.g., for n = 10, that would be 10 x 10 = 100) For a computer, this would get a bit slower with very large values, but it is still "feasible" in a foreseeable amount of time, as one would say.

O(2^n) here describes exponential time complexity. You can already guess what's going on here. Example: You try to crack a password purely by trying out all combinations (brute-force). With each additional character (n), the number of possibilities multiplies. Let me give a more vivid example using pizza toppings. For every available topping, you have two choices: put it on the pizza or leave it off.

1 topping (pepperoni): 2^1 = 2 possible pizzas (plain or with pepperoni).

2 toppings (+ mushrooms): 2^2 = 4 possible pizzas (plain, with pepperoni, with mushrooms, or with both).

3 toppings (+ onions): 2^3 = 8 possible pizzas.

10 toppings: 2^10 = 1,024 possible pizzas.

20 toppings: 2^20 = 1,048,576 possible pizzas.

So you can see very quickly that with just 20 toppings, the possibilities literally explode. A computer would already run into massive difficulties here with a larger n. Therefore, we go ahead and classify these kinds of problems more generally:

O(n^2), O(n^3), ... etc. Here, n is in the base. Such algorithms are generally considered feasible/efficient/fast.

O(n!), O(2^n), ... etc. Here, n is in the exponent. Such algorithms are generally considered inefficient/practically unsolvable. So you see, "exponential growth" is often unpredictable and a major hurdle. It is no coincidence that humans also generally have a hard time when something grows exponentially.

Alright, so far, so good. Now the whole thing gets interesting. In theoretical computer science, one often deals with decision problems, meaning problems that can be answered with "Yes" or "No". In complexity theory, the problems we just talked about are divided into complexity classes like these:

The class P stands for polynomial running time: This class contains all problems that are "easy" for computers. That means: We know an algorithm that can find the solution in polynomial time (like O(n^2) or O(n^3)). Examples: Sorting a list, calculating the shortest route in a navigation system, multiplying two numbers.

The class NP (Nondeterministic Polynomial)
Caution, most common misconception: NP does NOT mean "Non-Polynomial"! It stands for Nondeterministic Polynomial. Leaving aside the technical subtleties, we can put it briefly and concisely: NP contains all problems for which a finished solution can be verified in polynomial time (i.e., quickly).

A good example of this would be a Sudoku:
If I give you an empty Sudoku that is very large (let's say 100×100 squares), it is unimaginably difficult to fill it out. A computer would have to try billions of combinations (exponential time). Now, of course, if I give you a finished, filled-out 100x100 Sudoku and ask: "So, is this correct?", you can check it very quickly (polynomial time). You simply go through every row, every column, and every block once and check that no number appears twice.

What is perhaps also important to know: Every problem in P is automatically also in NP. Because if I can quickly solve a problem from scratch, then I can certainly verify a finished solution all the more quickly.

The P vs. NP problem now tries to investigate the following question: Is P = NP or P =/= NP?
To put it simply: If it is easy to quickly verify a finished solution (NP), is there always also a clever way to quickly find this solution on your own (P)?

And this is where it gets interesting now, because the practical consequences would probably be much more drastic:
1st Possibility: P =/= NP.
That would mean: Finding a solution is fundamentally harder than merely verifying a solution.
So there are problems in NP (like the Sudoku example or route planning) for which there will never be a fast algorithm. The nature of mathematics forbids a shortcut. Nobody has been able to prove this so far, because it is unimaginably difficult to mathematically prove that a clever algorithm cannot exist at all.

Most experts assume that the 1st possibility is true. However, it has not been proven yet.

2nd Possibility: P = NP.
That would be a sensational discovery if it were true. That would mean: Everything that can be verified quickly can also be solved quickly. In short, it would mean: We humans have simply been too stupid so far to discover the right algorithms for hard problems like Sudoku or the routing problem.

Sometime in the 70s, another interesting discovery was made: Within the class NP, there is a group of problems called NP-complete. The Sudoku example belongs to this group as well. It was proven (namely by Stephen Cook) that all of these problems are at their core the exact same problem in different disguises. You can transform any problem from NP into a Sudoku in polynomial time. So, if anyone in the world finds a fast algorithm for just a single one of these hard problems (e.g., Sudoku), they have automatically found a fast algorithm for all NP problems. It would then immediately be proven that P = NP.

From a practical point of view, the consequences would of course definitely be felt more directly. The most direct one, which is noticed most frequently, is that of encryption. For example, in online banking, WhatsApp, etc., the encryption used there is based on the assumption that so-called one-way functions exist. That is, something that can be calculated very quickly in one direction, but can only be calculated in reverse with great difficulty and very slowly. A typical example would be multiplying two large prime numbers (which is very fast) versus factoring the result back into its factors. Now, if P = NP, this would immediately mean that such one-way functions don't really exist at all. In theory, an attacker could crack a key or a password just as fast as the legitimate user creates it. All digital security that relies on one-way functions would collapse in on itself. That would truly be a dangerous sensation. If someone in Claude has already secretly solved the P vs. NP problem somehow, and it actually turned out that P = NP, the people involved might potentially hold a truly very dangerous weapon in their hands.

Realistically speaking, we are not prepared for that yet, at least as far as I know. Previously, it was still assumed that the Riemann problem would only be solved many years from now, but people massively underestimate the exponential growth of knowledge in current LLM models. In any case, the cryptography problem would be the most drastic practical consequence resulting if P = NP were true. And we would probably have to try much faster to make physical quantum encryption possible then. As you can imagine, that would become extremely complicated.
User avatar
Federica
Posts: 2850
Joined: Sat May 14, 2022 2:30 pm

Re: The Riemann Hypothesis has apparently been proven

Post by Federica »

As it seems to me on initial consideration (if I am getting the concepts right) P = NP depicts the completely fulfilled Ahrimanic ideal, or end state. Isn't that the Ahrimanic dream: a sub-reality where there is basically only one 'solution' that can be inferred from 'verification criteria'? A reality where the future can be inferred from the past, and every scenario is a merely deterministic composition of a matrix-scenario (the solution to an NP-complete problem also solves any other NP-complete problem). In this sense, P = NP could perhaps be seen as a symbol for the sub-world that is being built through the progressive descent into the material spectrum.

And perhaps the question is not so much a matter of 'discovering' whether or not the equation is true (whether or not AI can produce the algorithm) but more like how a sub-reality asymptotically approaching P = NP is progressively materializing through a deepening of the Ahrimanic impulse.
"The release of the senses form their bondage to matter can also be undertaken through warm and intimate perception of the life and phenomena of nature. In this case, thinking is connected to Intuition and to the perception of the ever-unfolding archetype that lies behind the organic world."
User avatar
Cleric
Posts: 2053
Joined: Thu Jan 14, 2021 9:40 pm

Re: The Riemann Hypothesis has apparently been proven

Post by Cleric »

Kaje977 wrote: ↑Sat Nov 21, 2026 6:18 pm Hello, Ashvin. Yes, the P vs. NP problem is probably "the" problem of theoretical computer science. What the Riemann hypothesis is to mathematicians, the P vs. NP problem is to computer scientists. To understand the problem, you need a little bit of prior knowledge in computer science, but I'll try to explain it in an intuitive way:
Excellent overview, Kaje! Thank you.

It should be noted that the proofs produced by AI are notoriously difficult to understand. I haven't heard if anyone was yet to decipher the 100-page Navier-Stokes proof. To understand how this is possible we can make again analogy with chess. A proof is a path of valid moves, starting from a certain configuration, and ending in a sought-after board configuration. This final configuration is the theorem that we want to prove. If we can demonstrate such a valid path, this is in itself the proof because the final step is the board configuration in question. Proving the negative of the theorem is more tricky because not being able to find a path to the final configuration doesn't mean that no such path exists. Maybe we simply haven't looked hard enough. For this reason, negative proofs are usually arrived at through demonstrating a contradiction. We assume that we are already at the sought after configuration and we start to follow what that would imply. If we find something that is incompatible with the rules, then it means that our initial assumption (that the theorem is valid) must be wrong.

This mathematical chess game can be modelled with the Lean programming language. It is indeed a full programming language (it is meant for writing software), but with great emphasis on type-checking and correctness even before the source is compiled. For mathematics, actually only this type-checking part is used, no code needs to be compiled. To prove something in Lean means to write such a function that returns a value of a certain type. This type is the theorem. As a simplistic example, in programming languages we have different data types: numbers, text, etc. In Lean, these types can be further constrained in various ways, for ex. I can define type T that is allowed to only have values between 100 and 200. Now if I define a function (a computational pipeline) and say that it should produce a value of type T, the type-checker is very strict and analyzes the different branches of execution. If it detects a branch that may return a value >200 or <100, compilation will be refused.

This is how Lean is used for math theorems. We basically define a target type T, which describes the constraints of the theorem, and we need to write a function f() that returns a value of type T. If the type checking succeeds, it's like we are being told, "Yes, this program can be compiled, and if run, it will produce a value that satisfies T." As said, this is not done when Lean is used for mathematics. We don't need an actual 'value' that the function can return. We only use the type-checker to verify whether our composite pipelines are guaranteed to produce a 'value' of the required type T. If there are no errors, we have 'proved' the proposition. We have constructed a pipeline that satisfies the requirements.

This is how AI proves mathematical theorems. It writes Lean functions, runs the type-checker, reads the error output, corrects mistakes, type-checks again, and so on in a loop until it is able to produce the functional pipeline that returns a value of type T - the required constraints of the theorem. This doesn't mean that AI is basically brute-forcing its way, by randomly combining pipeline stages until it succeeds. There's certainly a kind of directedness, that selects the more plausible combinations at different contextual levels (like operational, tactical, strategic), yet, like much of the code written by AI (popularly dubbed 'AI slop'), it quickly becomes complex and convoluted. It 'works', it connects A with B, but often in ways that do not look like anything comprehensible.

My current feeling is that these proofs come not so much from ingenuity, but simply of the tirelessness of the models that keep pushing toward the target, no matter the cost. Human beings are generally much more lazy and thus they seek the simpler and elegant solution. Or put in another way, if China was to dedicate thousands of people to tediously trying and eliminating possible paths for the proof, similar results would most likely be found. It's only that there has be no motivation for anything like that and AI turned out a good fit for the job.
User avatar
Cleric
Posts: 2053
Joined: Thu Jan 14, 2021 9:40 pm

Re: The Riemann Hypothesis has apparently been proven

Post by Cleric »

I would like to add something to the topic. However, I'll have to go out soon, and writing my post would easily take a couple of hours. So I decided to make a 'social experiment' :) Below are my shorthand notes (took about 10-15min) on what I would like to touch upon in the post. Paste it into your chatbot of choice and let it expand it. But please read the shorthand first. Note whether it is clear. Then go to the generated expanded text. Does it help? Check the shorthand again. Is it clearer now? It's interesting to me whether such an expansion helps. And just to say: I do not intend this as a recommended practice. Usually, when I meditate on something that tries to make its appearance into a forum post, I live precisely through the pictorial contents of such bullet points. Most of the effort is the actual articulation. I'm fully certain that this articulation is a necessary skill and practice. We need to find our being along the full contextual hierarchy of the moments of existence.
Expand the following shorthand notes into text suitable for a forum post. The expansion should be moderate. Each structural point should expand to no more than a paragraph.

*General trajectory:* Making an intuitive analogy for concrete and abstract types of thinking with industrial pipelines and typed programming languages. The audience is non-technical.

*Structure:*
* The industrial pipeline (say, fruit processing) has concrete implementation and conceptual architecture
* The concrete deals with what happens with a specific fruit as it goes through the pipeline. Analogy with program execution.
* The conceptual architecture deals with the `fruit` as a whole, which contains all possible members of the type `fruit`.
* In the concrete, we perform tests. We pass different fruits and make statistics about the outcomes.
* In the typed thinking, we try to conceive how each pipeline stage transforms the *set* of all possible members. For example, when we have `n`, we should think of all numeric values that `n` can take.
* Layman pseudo-code example of a typed function that takes a number and returns a number. Then it has an `if` statement – one branch returns a number, the other text (string). The static type-checker detects that the second branch violates the return type signature and prevents the source code (the conceptual architecture) from compiling into a concrete computational pipeline.
* Mathematical relations can be represented through the type system of a language such as Lean. In maths, we also have concrete calculations and reasoning about what happens with whole sets of numbers as they pass through specific pipeline stages. A theorem is like a pipeline architecture where we can reason about how sets are transformed. Example with Fermat's Last Theorem, which makes a claim about the possible results of a computational pipeline when the inputs fall into one or another category.
* Quantum algorithms are more like pipeline stages that operate on whole sets of numbers (the superposition of qubits), thus writing such algorithms is more like proving a mathematical theorem. Normal programming is more practical because it is easier to start with concrete input and envision what pipeline stages are needed for reaching desired results. Quantum programming (if it ever takes off) is unlikely to be a widespread job. Each algorithm is more like a mathematical discovery rather than a practical task list.
* In conclusion, the two types of thinking deal with `n` but in one we pass a concrete value through the pipeline (computation), while in the second, `n` is a spectrum of values and we reason about the ways in which this spectrum is transformed through the stages (type-checking).
* Ultimately, both types can still be 'flattened' (reduced) to one kind of intellectual movements. This is why type-checking in computers is at all possible, since it is itself a concrete computational pipeline after all. The types are *represented* as values in the type-checker.
Kaje977
Posts: 88
Joined: Fri Aug 23, 2024 9:23 am

Re: The Riemann Hypothesis has apparently been proven

Post by Kaje977 »

Federica wrote: ↑Sat Nov 21, 2026 9:04 pm As it seems to me on initial consideration (if I am getting the concepts right) P = NP depicts the completely fulfilled Ahrimanic ideal, or end state. Isn't that the Ahrimanic dream: a sub-reality where there is basically only one 'solution' that can be inferred from 'verification criteria'? A reality where the future can be inferred from the past, and every scenario is a merely deterministic composition of a matrix-scenario (the solution to an NP-complete problem also solves any other NP-complete problem). In this sense, P = NP could perhaps be seen as a symbol for the sub-world that is being built through the progressive descent into the material spectrum.

And perhaps the question is not so much a matter of 'discovering' whether or not the equation is true (whether or not AI can produce the algorithm) but more like how a sub-reality asymptotically approaching P = NP is progressively materializing through a deepening of the Ahrimanic impulse.
I'd say that's generally something that's always been part of intellectual thought in one way or another. The fact that something can be translated from one thing to another is, in a sense, nothing new and always part of mathematics (e.g. translating set theory into type theory and vice versa). From the "Game Loop" essay, we have the same example of the potato pipeline, which can also be replaced by something else at any time. The only thing that would be "new" if P = NP were to hold is that, on the one hand, numerous security mechanisms based on one-way functions could be rendered ineffective (which is definitely a dangerous problem), and on the other hand, logistics, medicine, LLMs, etc. could likely be highly optimized. In that respect, it seems like a universal algorithm, yes. But it would by no means cover all decision problems, since there are even higher classes, such as EXPTIME and EXPSPACE, that are not covered by P = NP. And, of course, there are also the undecidable problems (e.g. the halting problem), which, in a sense, directly touch upon the limits of intellectual thought in which modern science has been stuck in for the past centuries.
User avatar
AshvinP
Posts: 6759
Joined: Thu Jan 14, 2021 5:00 am
Location: USA

Re: The Riemann Hypothesis has apparently been proven

Post by AshvinP »

Kaje977 wrote: ↑Sat Nov 21, 2026 6:18 pm And this is where it gets interesting now, because the practical consequences would probably be much more drastic:
1st Possibility: P =/= NP.
That would mean: Finding a solution is fundamentally harder than merely verifying a solution.
So there are problems in NP (like the Sudoku example or route planning) for which there will never be a fast algorithm. The nature of mathematics forbids a shortcut. Nobody has been able to prove this so far, because it is unimaginably difficult to mathematically prove that a clever algorithm cannot exist at all.

Most experts assume that the 1st possibility is true. However, it has not been proven yet.

2nd Possibility: P = NP.
That would be a sensational discovery if it were true. That would mean: Everything that can be verified quickly can also be solved quickly. In short, it would mean: We humans have simply been too stupid so far to discover the right algorithms for hard problems like Sudoku or the routing problem.

Thanks, Kaje, that was a very intuitive presentation of the nature of these problems and their practical significance. It was super easy to follow.

I am still curious about the 1st possibility and why the experts converge on it. This seems to be what Cleric described as a 'negative proof', which would need to demonstrate a contradiction if there were ever a case in which P = NP. It would be like saying, "There exists a Sudoku puzzle somewhere that can never be solved quickly with an algorithm". But how could such a demonstration ever be possible if we set aside other physical constraints (like energy to run the algorithms)? To me, it seems not only unimaginably difficult to demonstrate that, but impossible.
Realistically speaking, we are not prepared for that yet, at least as far as I know. Previously, it was still assumed that the Riemann problem would only be solved many years from now, but people massively underestimate the exponential growth of knowledge in current LLM models. In any case, the cryptography problem would be the most drastic practical consequence resulting if P = NP were true. And we would probably have to try much faster to make physical quantum encryption possible then. As you can imagine, that would become extremely complicated.
That is indeed a somewhat alarming prospect. In a sense, the modern intellect has made ordinary life flow so entangled with and dependent on its computational movements (for example, for password protection), that if we pull on any tiny thread, the whole thing could unravel quickly. I'm still kind of confused about why we need to prove P = NP before algorithms could begin 'reverse calculating' encryption functions. Or is it that the realization of the latter is itself the proof that people are seeking, as in Cleric's metaphor - "If we can demonstrate such a valid path, this is in itself the proof because the final step is the board configuration in question"?

This stuff quickly begins to boggle my mind. I probably need to work with Cleric's social experiment more :)
"They only can acquire the sacred power of self-intuition, who within themselves can interpret and understand the symbol... those only, who feel in their own spirits the same instinct, which impels the chrysalis of the horned fly to leave room in the involucrum for antennae yet to come."
Post Reply