←
Paata Ivanisvili441 posts
Paata Ivanisvili avatar

Paata Ivanisvili

@PI010101

Professor of Mathematics @ UC Irvine.

◷ Joined October 2019▧ 92 Photos and videos
Paata Ivanisvili avatar
Paata Ivanisvili@PI010101·3h

Post attachment

Okay, here is a problem for testing AI. You do not need to know any abstract mathematics to understand the statement, only how to add and multiply 1’s and -1’s. I am offering a monetary prize for a correct solution, paid via X Money on X. I will also check the proof myself and retweet it. If the solution truly impresses me, I am happy to increase the prize. The question arose almost accidentally while I was working on something else. I remember spending around five hours thinking about it without finding a solution. It is also one of my least-upvoted questions on: mathoverflow.net/questions/4… It appeared before AI became good at mathematics, so this is a fair challenge: can AI beat roughly five hours of human work? The rule is simple: the solution (share the full AI transcript) must be purely AI-generated, with no substantial mathematical hints from a human.

Post attachment
Paata Ivanisvili avatar
Paata Ivanisvili@PI010101·18h

Journals should judge papers by the results they contain, not by whether AI was used.

Jensen Huang avatar
⟳ Paata Ivanisvili retweeted
Jensen Huang@JensenHuang·Nov 17

Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me. I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA. I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation. Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO. Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours. Jensen

Paata Ivanisvili avatar
Paata Ivanisvili@PI010101·Nov 17

Grok built a short counterexample to Problem 7.4 in Figalli–Guerra–Kim–Shahgholian arxiv.org/abs/2506.23608 I want to explain it a little, because the example is cute and it was a bit surprising. A harmonic function on a domain that vanishes on an open subset must vanish everywhere. Geometrically: if it maps an open set to a single point, then it maps the entire domain to that point. A variational way to think about it is by minimizing the Dirichlet energy: the minimizer of the integral of the square of its gradient, with fixed boundary values, is a harmonic function. For vector-valued harmonic maps, the Dirichlet energy is the integral of the sum of the squares of the gradients of each component. For these vector-valued harmonic maps, there is a stronger geometric statement: if an open set is mapped into a smooth curve, then the entire image lies in a straight line. In particular, rank at most one (of the matrix of first derivatives) on an open set implies rank at most one everywhere. Now introduce an "obstacle": require the map to stay outside a given body, allowing contact with its boundary, and minimize the energy subject to this constraint. Problem 7.4 asks whether rank at most one still propagates when the body is convex. The authors had already constructed a counterexample for a non-convex body. Grok shows that the answer is negative even when the body is a ball grok.com/share/c2hhcmQtNA_f9…

Elon Musk avatar
⟳ Paata Ivanisvili retweeted
Elon Musk@elonmusk·Nov 14

Inspecting ship 42 right now

0:15
Paata Ivanisvili avatar
Paata Ivanisvili@PI010101·Nov 14

Meanwhile, the authors

Post attachment

The authors take full responsibility for all contents of this manuscript, the same way they have fully read and agreed to the terms of service of every app they’ve ever installed.

Paata Ivanisvili avatar
Paata Ivanisvili@PI010101·Nov 12

Jensen’s inequality is famous, though less famous is Gaussian Jensen’s inequality.

Post attachment
Paata Ivanisvili avatar
Paata Ivanisvili@PI010101·Nov 10

Even sharing a paper with a partial result is risky -- someone with more compute and access to internal models can finish it and post the full solution before you do, and your contribution disappears. You either release a sharp result or you post nothing.

陶哲轩担心的事,真的发生了。😕 一位数学家在 YouTube 上讲了自己还没解开的问题。 一周后,一个毫无相关背景的陌生人发邮件给他: “我用 AI 帮你解出来了。” 他是 Heriot-Watt 大学的 Francesco Fournier-Facio。他的决定是,以后不再公开讲没做完的研究。 “这很可悲,但对青年学者来说,这就是现实。”

Cursor avatar
⟳ Paata Ivanisvili retweeted
Cursor@cursor_ai·Nov 9

Claude Fable 5.2 is now available in Cursor! At max effort, it's the top model on CursorBench at 60%. Fable 5.2 scores 56.9% on CursorBench at default effort, up from 49.2% for Fable 5.1.

Post attachment
Paata Ivanisvili avatar
Paata Ivanisvili@PI010101·Nov 6

Congratulations -- this would be a major result if it holds. I’ve spent a fair amount of time on this problem myself. A positive resolution essentially means quantum algorithms can be classically simulated in polynomial time on most inputs on big class of problems. PS. I recently obtained a partial (quasipolynomial-time) result with AI and was polishing the draft; since the problem is now claimed to be solved, I’ll just release the raw AI draft here: paatai.github.io/ai-proofs/a…

Today Physical Superintelligence PBC is announcing a proof of the Aaronson-Ambainis conjecture, one of the central open problems in quantum complexity, with major implications for quantum computing speedup. psi.inc/research/aaronson-am… Thread 👇

Paata Ivanisvili avatar
Paata Ivanisvili@PI010101·Nov 4

Proudly saying “No AI was used in this” is not going to age well. Your agent count is the size of your intelligence. Scale or get left behind.

Paata Ivanisvili avatar
Paata Ivanisvili@PI010101·Nov 3

A random variable X is said to be a symmetric Rademacher if it takes two values +1 and −1, each with probability 1/2. A random variable X is called a biased Rademacher (with bias q) if it takes two values with probabilities q and 1−q such that 𝔼X = 0 and Var(X) = 1. The biased reverse Khintchine inequality states that for any q ∈ (0,1) other than 1/2, one can improve the Cauchy–Schwarz inequality ‖Y‖₁ ≤ R(q) ‖Y‖₂ for some R(q) < 1, where Y is any linear combination of i.i.d. biased-q Rademachers. For some time I was curious about the sharp value of R(q), as it has some interesting applications. At some point (before the AI era) I was even suggesting this problem to my students. Recently AI agents worked overnight and produced a 192-page PDF identifying the sharp constant R(q). I scrolled through the proof very briefly, and my first impression was that this is one of those computationally hard monsters. I absolutely do not have that much energy and time to read and verify everything line by line, so I eventually ended up formalizing the proof in Lean. In the process of formalization, AI simplified the proof to 72 pages. It is about 534,000 lines of Lean in 823 files. Here is a graph of the sharp R(q), and the full details (PDF + Lean code) are available under item #12: paatai.github.io/ai-proofs

Post attachment
Paata Ivanisvili avatar
Paata Ivanisvili@PI010101·Nov 1

My bots have been running nonstop for 48 hours and found 166 possible new impact sites on Mars: paatai.github.io/vibe-projec… I’m also running a parallel project looking for anomalies in galaxy images — about 12 candidates so far: paatai.github.io/vibe-projec…

Post attachment

Grok Bot screened years of public NASA Mars images and flagged a few dark spots that weren't there before, possibly fresh impacts that haven't been publicly recorded. I'm tagging @NASA and @HiRISE in case they find it useful. More details paatai.github.io/vibe-projec…

Post attachmentPost attachmentPost attachment