As of: 2026-10-14 00:22 CEST. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/v/two-minute-papers-openai-new-ai-doing-things-we-barely-understand/ # OpenAI’s New AI Is Doing Things We Barely Understand Two Minute Papers, 13 October 2026, YouTube. 109,754 views as of 13 October 2026. Kind: Review. Watch: https://www.youtube.com/watch?v=z_sXLlLVUls ## Why it is here Two Minute Papers (Károly Zsolnai-Fehér, 8:49) on OpenAI's mathematics release; the description links openai.com/index/sharing-ai-progress-in-mathematics and github.com/openai/math. About 110k views in 11 hours, the largest explainer of that release found so far. ## Description (written by Gemini from the video) ### **Summary** Károly Zsolnai-Fehér of *Two Minute Papers* covers OpenAI's release of hundreds of AI-generated mathematical research papers produced by an internal frontier model. He highlights two specific results—billiards in irrational triangles and conformal transformations of planar domains with holes—and explores the philosophical debate over whether "ugly," non-intuitive AI-generated proofs truly advance human mathematical understanding. He closes with an argument for democratizing AI through open-weights models and notes the withdrawal of three papers due to community feedback. --- ### **What is shown** - **[00:15]** OpenAI's repository listing 722 AI-authored math manuscripts with commit dates from late September and early October 2026. - **[00:26]** OpenAI's blog post *"Sharing AI progress in mathematics"* linking to their GitHub repository. - **[00:54]** Animation explaining the problem of light bouncing indefinitely in a triangular room with an irrational angle (relative to $\pi$), illustrating how almost every ray becomes ergodic and visits every point in every direction. - **[02:04]** Animation of conformal mapping applied to an irregularly shaped planar domain with tangled and infinitely tiny holes, morphing them into round circles while preserving angles. - **[03:05]** A manuscript titled *"The Quasi-Riemann Hypothesis: A Zero-Free Half-Plane $\text{Re } s > 7/8$"* by OpenAI (dated September 30, 2026, 199 pages), followed by dense equations representing the complex, non-elegant nature of AI-generated proofs. - **[03:28]** Visual comparisons of fluid and cloth physics simulations (ground truth vs. AI predictions from Pfaff et al. / Sanchez-Gonzalez et al.). - **[03:57]** DeepMind's AlphaStar neural network activations and location consideration maps in StarCraft II. - **[04:18]** Visual geometric animations illustrating various proofs of the Pythagorean theorem. - **[05:11]** Animated timeline illustrating AI milestones over the years (GPT-3 simple arithmetic 6 years ago, high school math competitions 2 years ago, IMO gold medal 1 year ago, Navier–Stokes equation breakthrough 1 month ago). - **[07:46]** Announcement of three withdrawn manuscripts (*"Algebraicity of Weil classes on split abelian eightfolds"*, *"Algebraicity of Kuga-Satake Correspondences for K3 Surfaces"*, and *"The rational Hodge conjecture for products of K3 surfaces"*) after community feedback identified a sign error. - **[08:15]** Sponsored segment demonstrating the Lambda GPU cloud platform. --- ### **Claims & numbers** - OpenAI released **722 manuscripts** (372 result families) produced by an internal frontier model (the presenter shows the repository count). - Computing each result took the model about **3 hours on average**, which the presenter claims is remarkably fast given the complexity of the problems. - The model proved that in a triangular mirror room with at least one irrational angle (relative to $\pi$), almost every ray visits every spot, goes in every direction, and spends equal time everywhere. - The model proved conformal mapping results for planar regions containing infinitely tiny or tangled holes, preserving angles while morphing holes into circles. - Three of OpenAI's manuscripts were withdrawn on **October 7, 2026**, after community feedback caught a sign error that invalidated an argument. - The presenter notes an evolutionary trajectory of AI mathematical performance: basic arithmetic 6 years ago, competition math 2 years ago, IMO gold medal level 1 year ago, and Navier–Stokes results 1 month ago. --- ### **Notable quotes** - **[00:25]** *"A total of 722 papers. The headlines keep talking about the controversy. I'd like to avoid that, because very few people are talking about the works themselves."* - **[03:13]** *"These are not elegant proofs. Some of them are a big ball of barf. And this raises an important question: what is the goal of mathematics?"* - **[07:08]** *"The answer is open-weights models that we can run and own ourselves. We have to come together and contribute to these projects, because we need more of those."* --- ### **Assessment** This is an independent analysis and educational review by Károly Zsolnai-Fehér of *Two Minute Papers*. The video uses custom motion graphics and referenced simulation clips to explain the concepts; it balances enthusiasm for the AI's technical achievements with candid criticism of the inscrutability of "brute-force" proofs and the corporate concentration of frontier compute. _Described by gemini-3.8-flash on 2026-10-13 from the video's audio and frames._ ## Related - 2026-10-06: [OpenAI releases 722 AI-written math manuscripts claiming hundreds of open problems](https://postcutoff.com/e/2026-10-06-openai-math-release-722-manuscripts/)