A large study of Khan Academy’s math tutor program found no extra benefits from AI. “Seeking help with one’s own confusion remained a choice, and most students declined it most of the time.”
#aitutor
hechingerreport.org/proof-points...
Yeah, AI is obviously never going to be help rogue states make bioweapons, it's not going to be solving another Millennium Prize Problem, nor is it going to ever get better. EA has a track record of crying wolf and have been trying to push pandemic preparedness since like 2010 as a vanity project.
A new metric for 'interesting' math theorems cuts overlap with Mathlib from 91.9% to 30.6% Researchers define a theorem's intrinsic interestingness as the ratio of its proof length to its statement length, then use a 27B model built on that metric to generate genuinely novel theorems instead of…
Yeah among other things they absolutely stole those mathematicians proof of Navier-Stokes, regardless of if they were deliberately trying to. OpenAI has zero control of what happens to data users give the chat bot.
SciAm displays hilarious double standards.
Now that AI solved it, the forced option Buckmaster was working on, and Córdoba and Martínez-Zoroa were praised for, became the "wrong" problem.
Also, for the Euler problem, OpenAI solved the "right" problem and Buckmaster the "wrong" one.
Did anyone see that " math taunt" from the Iranians aimed at the Orange Guy? The math equation was posted by Mohammad Bagher Ghalibaf, the Speaker of the Iranian Parliament and a chief negotiator of Tehran. Ghalibaf fired what international media dubbed a "maths missile" on social media. Ghalibaf posted a modified version of the Taylor Rule, a famous 1993 formula created by economist John Taylor t…
If a computer solves a millennium prize problem but there are no humans around too see then did it really solve a millennium prize problem?
The Villanova study supposedly proved that AI content is indistinguishable from human writing. I'm saying it did nothing of the sort. litmagnews.substack.com/p/lets-study...
Ok, we agree there needs to be regulation.
We disagree on the capabilities. It's not true that OpenAI simply "yoinked" the results of a mathematician. From Buckminster's own statement, the majority of the work he was doing was to make the verified proof by another LLM suitable for publication.
If you drop a sloppy AI-generated proof of some theorem onto the internet hoping to be hailed as a hero, think twice. Mathematicians are sick of it. https://arxiv.org/abs/2609.24779
Forecasters were mostly wrong about AI.
On capabilities
If you release a crappy AI-generated proof of some theorem, don't think you're going to become a hero. People will make fun of you.
3 researchers offered this proof that OpenAI's proof doesn't solve all cases (if I understand right, their "solution" doesn't have any of the real world applications a real solution would).
n.b. the researchers used AI to "decipher the OpenAI paper" 🤨
arxiv.org/html/2609.20...
AI.MATH With the AI.MATH initiative, we put forward draft positions on the future of mathematical research and mathematics degree programmes in the age of AI, alongside questions and possible answers. This is an initiative from Claudia Alfes , Thomas Nikolaus , and Andreas Thom. https://ai.math.ms/en/ also see: https://proofsandprompts.com/2026/09/23/ai-math-announcement/ #research #math #AI
One difference is that referees for journals are anonymous. If one of us rushes to build on one of OpenAI's papers before anyone else gets the chance, it will be very public.
More generally, I think, and indeed hope, the whole idea of being admired for "winning races" is coming to an end soon.
The scenario you describe would indeed be extremely unfair, and I don't think I'm stepping out of line by saying that.
For the benefit of anyone who might think this refers to the field of mathematics (which is in a state of extraordinary upheaval), let me make clear that it is in fact about a literal field. When I last checked, I was delighted to see that some willows had gone from waist high to taller than me.
Ask arXiv to add a toggle to state whether chatbots can train on your work: math.mit.edu/~elmos/petit...
We also plan to be as open as possible about the recommendations we end up making, including explaining why we made them.
We are reading all of them, and while it is impossible to agree with all of them (since, unsurprisingly, a wide range of views are expressed), my sense is that the views of the committee are pretty representative, even if the people are not.
I'd like to add that we are very aware that we are not representative of the mathematical community, and for that reason we greatly value the approximately 500 views we have received via a form linked to here.
agmai.org
If you want a better idea of how the newly formed Advisory Group on Mathematics and AI sees its work, and in particular its immediate task of advising OpenAI on how best to release their results, I recommend a recent blog post of Martin Hairer.
proofsandprompts.com/2026/09/22/w...
So it really depends. For the Navier Stokes solution, yes, the AI proof provides no mathematical value, but it’s a strong proof of concept of how proofs could be developed by AI. There are tons of areas in economics, engineering, and physics where you often need a proof just for your result.
OpenAI claims to have proved 100 more big theorems. I urge all of you to fill out the form at https:// agmai.org/ and politely ask their board of nine "elite mathematicians" to initiate a more democratic approach to dealing with this potential flood of AI-proved results. https://www. theverge.com/ai-artificial-int elligence/999167/openai-elite-mathematicians-panel
So your position is that the thing solving millennium prize mathematics is not thinking? Sometimes old traditions don’t take into account new possibilities, and it’s a time of such possibilities
So the claim about 'AI@ solving the hardest maths problems remains no more than internet bollocks. No verified claims and no prizes awarded. www.explainx.ai/blog/millenn...
Here's something AI can't do for you: enjoy On Proof and Progress in Mathematics by Bill Thurston.
I went to the first discussion at the Fields Institute about AI in # math . Some good comments but overall, not super inspiring. A lot of people on the panel seem to aim for some middle ground. Not enough focus on the ethics of it all for my taste. You can probably still sign up if you're interested in the remainder of the series (in-person and remote participation is possible): https://www. field…
And sure: a single proof to a single math problem can be explained away. Could be coincidence. But the pattern is creating multiple proofs of lang standing well known math problems is pretty striking and is evidence for (some!) intelligence.
I am by no means an expert, but it seems overly confident to me to assume that sustained human effort directed at a machine generated proof can do nothing to uncover novel–and transferable–insights about a problem.
I understand the fear around change, but change also opens up new opportunities
So # OpenAI wants to be friends again? https:// openai.com/index/advisory-grou p-on-mathematics-and-ai/ It's hard to understand what this means. Everything is so slick and a bit meaningless. Maybe they used AI to write their statement? "Importantly, the group will not be responsible for advising us on how to pace our internal progress on mathematics." 🤔 Sounds to me like: "OK, we will solve your…
New math + AI developments in fluid mechanics. ~ Javier Gomez-Serrano. https:// youtu.be/TcEefrWrddA # AI4Math
Interestingly I just listened to a talk with a mathematician where he complained about how poorly written the (AI drafted) paper surrounding the Navier Stokes proof is
Just explained to a class of first years how academic publishing works and they were appropriately incredulous. "HOW much money do you make from publishing an article? Wait, you PAY money to publish an article??"
From the mouths of babes.
El miedo al lado oscuro de la IA en la comunidad matemática. ~ Anabel Forte Deltell. https:// theconversation.com/el-miedo-a l-lado-oscuro-de-la-ia-en-la-comunidad-matematica-292151 # AI4Math
“Furthermore, last Thursday, three mathematicians posted a proof of their own that showed that OpenAI’s method can never be extended to solve the full problem. In other words, the loophole will never be closed, barring some completely new idea.” # Math # NavierStokes # ai https://www. scientificamerican.com/article /did-openai-solve-the-wrong-navier-stokes-problem/
Tracking AI impact in math research: https:// mathurls.com/?q=AI%20%20 https:// epoch.ai/frontiermath https:// epoch.ai/data/arxiv?view=graph https:// epoch.ai/benchmarks/proofbench ?view=graph&tab=release-date https:// matharena.ai https:// benchlm.ai/benchmarks#category -math https:// lean-lang.org/eval/ https:// trishullab.github.io/PutnamBen ch/leaderboard https:// sorrydb.org https:// lea…
Stop trying to send the signal that you've “solved math”: this is impossible anyway, and you've only succeeding in making problem-solving as completely worthless as producing endless hours of AI slop videos.
This is impossible to argue against, for the same reason you can't construct a logical proof in math without a definition of terms and a couple of axioms.
Fine, so you don't have to have read all of the possible definitions but you have to have some conceptual understanding of what you mean