Something I don’t quite know, maybe someone closer to math can answer this question
It seems we’re making tremendous progress on the back of existing abstractions and concepts with the brute force ability to generate language and verify proofs.
For example, Tao says about the Riemann proof:
> appears to use the full power of modern developments on the Langlands program and arithmetic geometry
What I wonder: how do we create new abstractions and concepts? Like whatever “Langlands program” is? That then feedback into the system, and allow for even more sophisticated mathematics.
Are we assuming AI will do this? Or does this still need to be something at least agreed-upon by a mathematics community - in the same way programmers talk about a “linked list” without having an hour long conversation to define it axiomatically? And without that social aspect, without breaking down long proofs into Lego block common abstractions, math may plateau? In part because we lose the ability to ask meaningful questions?
show comments
westurner
Re: geometry and 6-Channel Dual Schubert QFIM and Sato Grassmanian Gr_p(H) and Polarized Hilbert space; "5. Li’s Unit-Circle Criterion and Why Finite Zero Audits Do Not Close Classical RH" https://westurner.github.io/sustainablefactory/grassmannian-...
And from the unifications paper from this series of papers, working title "Turner Grassmanian Series: Mutant League Physics; The Caret Wedge" or "Unification, Engineering Applications, and Research Roadmap of Complex Grassmannian \(Gr_{\mathbb{C}}(2,4)\), \(Gr_{\mathbb{C}}(2,6)\), and Dilatant Dark Fluid (DDF) Mechanics: Cross-Domain Plücker Invariants, Photonic–Thermal–Quantum Systems, and Formal Verification in Lean 4 — Sustainable Factory 0.1.0 documentation https://westurner.github.io/sustainablefactory/grassmannian-...
I just put all of this on compressible and incompressible NS together today but have been working on those for a few months with a few hundred dollars of tokens. I haven't yet looked at their 17M Langlands program solution; hopefully there's some insight there.
We had the first 30 zeros with our numerical implementation before we found Toupin's 2025 manuscript On the Nature of Nature: Celestial Holography to the Zeta Zeros this week on gscholar looking for geometry and proof of Hilbert-Polya, but at least four gaps remain in the RH Lean and IDK how much time I'll have for this given the free KdV/KP-II solver for doing the Sato Grassmanian to Hilbert work.
"Adelic Carleman–Weierstrass Regularized Determinants (det2,det3
), Schwinger 2-Cocycle Central Extensions, Segal–Wilson Determinant Bundles, 1(ℝ)
Weil Positivity, and Formal Bridges to Signals.RiemannSpectral in Lean 4, Rust/PyO3, and mpmath" >
7. Open Boundaries Recorded in Signals.Pending.SatoRiemannSpectralPendingBoundaryhttps://westurner.github.io/sustainablefactory/grassmannian-...
It looks like two independent kernels accept the result, which means its likely that there are no kernel bug exploitations in the proof.
show comments
actusual
Is anyone else struggling with feelings of both excitement and terror? I haven't felt this much personal uncertainty about my career/life since I was an L1 MLE (10 years ago).
show comments
swingboy
Dumb question but with something like 15 million lines of Lean code, how can we tell when the Lean build “passes” that it is actually formally verifying the right thing and there wasn’t just a “1+1=2” making it pass? Like you hear people say “if your Rust program compiles then it’s correct” but that never meant your business logic is correct. I’m not suggesting malfeasance or intentional error, just that maybe the model spit out something that “passes” but doesn’t actually prove what it says it does.
show comments
maplet
I understand that Tao does not want to promote a result which could turn out to be incorrect, but I simply cannot imagine posting about a Riemann Hypothesis result and not commenting about the significance.
show comments
sega_sai
As some others in the thread, I truly find this moment scary. I feel that right now AI essentially completely surpassed human intelligence. For now the results from AI are limited by what's easy to do within the computer "with pen and paper", but surely it's matter of time before this is unleashed in other fields.
For a scientist this is scary, because it is unclear if there is space for research by people anymore. But I start to worry about actual dangers from this intelligence and where it can lead.
show comments
PopePompus
Could someone explain the difference (if any) between the Generalized Riemann Hypothesis and what has historically been called the Riemann Hypothesis?
show comments
mNovak
It's interesting to me that the ANT GRH proof appears to have been generated directly in Lean, whereas the OAI NS proof I believe was produced in natural language then formalized into Lean.
Maybe GRH was as well, I haven't seen much about how these were put together. But, it's a curious difference in approach if not.
gehsty
Future of mathematics has two frontiers - one that LLMs can derive and another that we can understand. When the science studied is completely abject, what is the point of an LLM doing it.
mv4
I haven't been so excited about math in a long time.
The pattern seems to be that AI is capable of some amazing things, when assisted by Levent Alpöge.
soslowdnjdjs
Slow day. Unforced NS, GRH and an reverse engineered cat litterbox.
starkshift
Interesting times. AI math will be applied by engineers, and mathematicians themselves will become mere astronomers or archaeologists of the universe of mathematical knowledge, the grid of which AI has already constructed.
machina_ex_deus
If llms can solve these open problems in math, the biggest question is what is going to happen with cryptography. So far cryptographically related math problems are suspiciously absent from all the new math development.
But I think there's a very good chance cryptographically relevant developments are just around the corner. There's a good chance that's why both anthropic and open ai have not released their latest models.
If public key cryptography were to fall to ai, the social implications could be immense, the whole digital world including cryptocurrency but also software updates (which are the only thing holding cyber security right now with all the ai generated vulnerabilities), credit cards, digital bank access, internet access being secure, public wifis, online markets, work from home, and quite ironically llm access itself will be severely damaged without encryption to protect the session.
show comments
barkingcat
I wish AI would solve and verify P = NP (instead of the ≠ case humans expect.)
that would be a tremendous achievement and contribution to humankind.
Invictus0
With no real commentary from Tao, why is this post notable?
throwitaway222
Overwhelmed with GRH today.
AndrewKemendo
It’s worth noting that this is a post from Terry himself rather than him reposting something from somebody else.
I find that a lot of the times that people see the post from Terrys site they attribute the authorship to him when he does a significant amount of reposting.
Something I don’t quite know, maybe someone closer to math can answer this question
It seems we’re making tremendous progress on the back of existing abstractions and concepts with the brute force ability to generate language and verify proofs.
For example, Tao says about the Riemann proof:
> appears to use the full power of modern developments on the Langlands program and arithmetic geometry
What I wonder: how do we create new abstractions and concepts? Like whatever “Langlands program” is? That then feedback into the system, and allow for even more sophisticated mathematics.
Are we assuming AI will do this? Or does this still need to be something at least agreed-upon by a mathematics community - in the same way programmers talk about a “linked list” without having an hour long conversation to define it axiomatically? And without that social aspect, without breaking down long proofs into Lego block common abstractions, math may plateau? In part because we lose the ability to ask meaningful questions?
Re: geometry and 6-Channel Dual Schubert QFIM and Sato Grassmanian Gr_p(H) and Polarized Hilbert space; "5. Li’s Unit-Circle Criterion and Why Finite Zero Audits Do Not Close Classical RH" https://westurner.github.io/sustainablefactory/grassmannian-...
And from the unifications paper from this series of papers, working title "Turner Grassmanian Series: Mutant League Physics; The Caret Wedge" or "Unification, Engineering Applications, and Research Roadmap of Complex Grassmannian \(Gr_{\mathbb{C}}(2,4)\), \(Gr_{\mathbb{C}}(2,6)\), and Dilatant Dark Fluid (DDF) Mechanics: Cross-Domain Plücker Invariants, Photonic–Thermal–Quantum Systems, and Formal Verification in Lean 4 — Sustainable Factory 0.1.0 documentation https://westurner.github.io/sustainablefactory/grassmannian-...
> 2.13 Infinite-Dimensional Sato–Segal–Wilson Hilbert-Space Grassmannian Gr_res(H) , Integrable PDE Solvers (KP-II / KdV / Boussinesq / AKNS), Dual-Schubert QFIM, Adelic Carleman–Weierstrass det3 , & Spectral RH Gap Audit (Papers 51 & 52) (Signals.SatoGrassmannian [21 modules], Signals.Pending.SatoRiemannSpectralPendingBoundary, sato_grassmannian.rs, sustainablefactory/sato_grassmannian.py)
> 2.14 Compressible & Incompressible Navier–Stokes via Three-Channel Plücker–Madelung & Six-Channel Dual-Schubert QFIM (Paper 53) (Signals.NavierStokes [PluckerFrame, Incompressible, Compressible], Signals.Pending.DualSchubertNavierStokesPendingBoundary, navier_stokes.rs, sustainablefactory/navier_stokes.py)
I just put all of this on compressible and incompressible NS together today but have been working on those for a few months with a few hundred dollars of tokens. I haven't yet looked at their 17M Langlands program solution; hopefully there's some insight there.
We had the first 30 zeros with our numerical implementation before we found Toupin's 2025 manuscript On the Nature of Nature: Celestial Holography to the Zeta Zeros this week on gscholar looking for geometry and proof of Hilbert-Polya, but at least four gaps remain in the RH Lean and IDK how much time I'll have for this given the free KdV/KP-II solver for doing the Sato Grassmanian to Hilbert work.
"Adelic Carleman–Weierstrass Regularized Determinants (det2,det3 ), Schwinger 2-Cocycle Central Extensions, Segal–Wilson Determinant Bundles, 1(ℝ) Weil Positivity, and Formal Bridges to Signals.RiemannSpectral in Lean 4, Rust/PyO3, and mpmath" > 7. Open Boundaries Recorded in Signals.Pending.SatoRiemannSpectralPendingBoundary https://westurner.github.io/sustainablefactory/grassmannian-...
https://leanprover.zulipchat.com/#narrow/channel/113488-gene...
It looks like two independent kernels accept the result, which means its likely that there are no kernel bug exploitations in the proof.
Is anyone else struggling with feelings of both excitement and terror? I haven't felt this much personal uncertainty about my career/life since I was an L1 MLE (10 years ago).
Dumb question but with something like 15 million lines of Lean code, how can we tell when the Lean build “passes” that it is actually formally verifying the right thing and there wasn’t just a “1+1=2” making it pass? Like you hear people say “if your Rust program compiles then it’s correct” but that never meant your business logic is correct. I’m not suggesting malfeasance or intentional error, just that maybe the model spit out something that “passes” but doesn’t actually prove what it says it does.
I understand that Tao does not want to promote a result which could turn out to be incorrect, but I simply cannot imagine posting about a Riemann Hypothesis result and not commenting about the significance.
As some others in the thread, I truly find this moment scary. I feel that right now AI essentially completely surpassed human intelligence. For now the results from AI are limited by what's easy to do within the computer "with pen and paper", but surely it's matter of time before this is unleashed in other fields. For a scientist this is scary, because it is unclear if there is space for research by people anymore. But I start to worry about actual dangers from this intelligence and where it can lead.
Could someone explain the difference (if any) between the Generalized Riemann Hypothesis and what has historically been called the Riemann Hypothesis?
It's interesting to me that the ANT GRH proof appears to have been generated directly in Lean, whereas the OAI NS proof I believe was produced in natural language then formalized into Lean.
Maybe GRH was as well, I haven't seen much about how these were put together. But, it's a curious difference in approach if not.
Future of mathematics has two frontiers - one that LLMs can derive and another that we can understand. When the science studied is completely abject, what is the point of an LLM doing it.
I haven't been so excited about math in a long time.
But what is the Riemann zeta function? Visualizing analytic continuation (3blue1brown): https://www.youtube.com/watch?v=sD0NjbwqlYw
The pattern seems to be that AI is capable of some amazing things, when assisted by Levent Alpöge.
Slow day. Unforced NS, GRH and an reverse engineered cat litterbox.
Interesting times. AI math will be applied by engineers, and mathematicians themselves will become mere astronomers or archaeologists of the universe of mathematical knowledge, the grid of which AI has already constructed.
If llms can solve these open problems in math, the biggest question is what is going to happen with cryptography. So far cryptographically related math problems are suspiciously absent from all the new math development.
But I think there's a very good chance cryptographically relevant developments are just around the corner. There's a good chance that's why both anthropic and open ai have not released their latest models.
If public key cryptography were to fall to ai, the social implications could be immense, the whole digital world including cryptocurrency but also software updates (which are the only thing holding cyber security right now with all the ai generated vulnerabilities), credit cards, digital bank access, internet access being secure, public wifis, online markets, work from home, and quite ironically llm access itself will be severely damaged without encryption to protect the session.
I wish AI would solve and verify P = NP (instead of the ≠ case humans expect.)
that would be a tremendous achievement and contribution to humankind.
With no real commentary from Tao, why is this post notable?
Overwhelmed with GRH today.
It’s worth noting that this is a post from Terry himself rather than him reposting something from somebody else.
I find that a lot of the times that people see the post from Terrys site they attribute the authorship to him when he does a significant amount of reposting.
So, 2050 huh?