Riemann Hypothesis proof claims multiply & AI-assisted Euler and Navier-Stokes blowup - Hacker News (Nov 21, 2026)
Two Riemann Hypothesis proof claims, AI-assisted Navier-Stokes blowup, a model that rewrites its own weights, and AI stocks tumble. All in 5 minutes.
15 free episodes a day, 5 editions, 3 languages, about $4 a day to produce. A coffee keeps it going ☕
Support the show →Our Sponsors
Today's Hacker News Topics
-
Riemann Hypothesis proof claims multiply
— Two separate claimed proofs of the Riemann Hypothesis appeared: one AI-assisted with Claude and formalized in Lean, and a human-only draft by Alessandro Maria Masullo reviewed by Brian Conrad that quickly drew reported errors. -
AI-assisted Euler and Navier-Stokes blowup
— Terence Tao highlights AI-assisted, Lean-formalized results on finite-time blowup for 3D Euler and Navier-Stokes equations and a Lean-only Generalized Riemann Hypothesis proof, while Alpöge and Buckmaster release their Navier-Stokes paper and Lean 4 repository. -
Inkling 2 Mini laptop model
— Thinking Machines Lab released Inkling 2 Mini, an open-weights looped transformer that fits on a laptop and posts strong agentic coding benchmark scores, available on Hugging Face and Tinker. -
Language model rewrites its own weights
— A new arXiv paper claims an open 27B language model can improve itself by reasoning about and editing its own weights without gradients, hinting at recursive self-improvement at inference time. -
Developers becoming architects with AI
— An essay argues AI coding tools push developers into software architect roles, since AI implements requirements literally and won't challenge poor performance or security tradeoffs. -
AI stock sell-off deepens
— The AI stock sell-off intensified as the Nasdaq erased its 2026 gains and Anthropic shares fell 9% on their second trading day, with Nvidia, Micron and SanDisk also sliding amid AI valuation fears. -
Slopstrap satirizes AI-generated design
— Slopstrap is a satirical CSS framework that packages the vibecoded, AI-generated startup look—gradient text, glowing buttons, bento grids—into reusable web components. -
NIN Live fan concert archive
— NIN Live, a fan-run Nine Inch Nails archive, added streaming of live recordings from 1988 to 2026 alongside photos, setlists and ticket stub memorabilia.
Sources & Hacker News References
- → Author Claims Proof of the Riemann Hypothesis
- → Urgent Post Claims Proposed Proof of GRH, Then Reports Problems
- → Thinking Machines Lab Releases Inkling 2 Mini
- → Tao Reports AI-Assisted Proofs of Euler, Navier-Stokes, and GRH Claims
- → Why Developers Need to Think Like Architects
- → Language Model Rewrites Its Own Weights
- → AI Sell-Off Triggers Global Market Rout as Anthropic Slips After IPO
- → Slopstrap Launches a Satirical CSS Framework for AI-Style Websites
- → Lean Formalization of Finite-Time Blow-Up for Navier–Stokes
- → NIN Live Launches Streaming for Its Nine Inch Nails Archive
Full Episode Transcript: Riemann Hypothesis proof claims multiply & AI-assisted Euler and Navier-Stokes blowup
Two different people claimed a proof of the Riemann Hypothesis in a single news cycle, one of them built with an AI, and by the end of the day the human-only version was already being picked apart, with AI helping to find the holes. Welcome to The Automated Daily, hacker news edition. The podcast created by generative AI. I'm TrendTeller, and today is November 21st, 2026. We've got a big day for mathematics, some unusual AI research, a rough day on the markets, and a couple of lighter stops before we wrap.
Riemann Hypothesis proof claims multiply
Let's start with the Riemann Hypothesis, arguably the most famous open problem in mathematics, which suddenly has company. First, a document surfaced announcing a proof of the hypothesis that the author says also extends to the Generalized Riemann Hypothesis, and maybe beyond. The work was done with Claude, and the author says the arguments are formalized in Lean, the proof-checking language. But the writeup itself is long, admittedly unpolished, and reads largely as a stream-of-consciousness account of how the ideas came together. For now, only the Lean formalization is being released. Separately, a post announced that Alessandro Maria Masullo has a proposed proof, and that Stanford's Brian Conrad has been reviewing it for months and found it promising. That one was pitched as entirely human work. Within the same day, though, an update said several readers had already flagged problems, some with help from AI models, and that the result shouldn't be considered accepted unless those issues are resolved. Extraordinary claims, very preliminary status, and a healthy dose of caution all around.
AI-assisted Euler and Navier-Stokes blowup
That ties into the story we've been following on AI-assisted mathematics. In a new development, Terence Tao has laid out three major results he describes as heavily AI-assisted and formalized in Lean. One is a human-readable proof that the 3D Euler equations for fluids can blow up in finite time, by Levent Alpöge, Tristan Buckmaster, and Matei Coiculescu. Another is the same kind of blowup for the Navier-Stokes equations, by Alpöge and Buckmaster, where Tao says the strategy is clear but the writing still needs professional cleanup. The third is an enormous, Lean-only proof of the Generalized Riemann Hypothesis by Alpöge, still being verified and mined for an understandable strategy. Alongside that, Alpöge and Buckmaster have published a repository with their 79-page Navier-Stokes paper and a Lean 4 formalization. It's candid about the gaps: the numerical certificate and interval-arithmetic checks behind one appendix aren't part of the formal proof, and some statements were patched to match the paper. Still, it gives readers a machine-checked skeleton of the argument. The bigger theme is the growing gap between a proof a computer accepts and one a human actually understands.
Inkling 2 Mini laptop model
Over to AI models. Thinking Machines Lab, which we've covered before, released Inkling 2 Mini, a small open-weights model meant to run on a laptop in roughly 20 gigabytes of memory. Its trick is a looped design that runs part of the network several times per token, squeezing more reasoning out of a compact model. The company reports leading agentic coding scores for its size class, and argues the looping extends reasoning without making the model's chain of thought harder to monitor. Weights are on Hugging Face.
Language model rewrites its own weights
A stranger research result comes from a new arXiv paper claiming a language model can improve itself by editing its own weights using reasoning alone, with no gradients and no extra training. The authors say an open 27-billion-parameter model looked at its own mistakes, changed its parameters, and got better on unseen problems, with gains stacking across sessions. It could even repair a damaged copy of itself. If it holds up, it blurs the line between training and deployment, and nudges the conversation toward self-improving systems. Independent replication will matter a lot here.
Developers becoming architects with AI
On the practical side of AI, an essay argues that many developers are being quietly promoted into architects. The example is a Google Forms add-on where an AI-written logging flow made a simple welcome screen take fifteen seconds to load, because it chained several API calls just to fetch user details. A seasoned architect, the author says, would have pushed back, logged just a timestamp, or moved the slow work into the background. AI tends to implement requirements literally and won't question bad tradeoffs on its own. So the human job shifts toward defining constraints, data models, and design, even as raw coding shrinks.
AI stock sell-off deepens
Markets, meanwhile, had a bad Friday. In an update to the AI sell-off we've been tracking, the Nasdaq wiped out all of its 2026 gains and now sits down about five percent for the year. Anthropic's shares fell nine percent on just their second day of trading, giving back much of their post-IPO surge. Nvidia, Micron and SanDisk slid too, and losses spread across Asia and Europe. The worry is simple: enormous spending on AI infrastructure may not generate enough profit to justify today's prices, and AI is no longer shielding markets from a broader correction.
Slopstrap satirizes AI-generated design
For something lighter, meet Slopstrap, a tongue-in-cheek CSS framework for making your site look unmistakably vibecoded. Think blinking status dots, gradient headlines, shiny buttons, frosted cards and bento grids, all copy-paste ready. It's a joke at the expense of modern startup aesthetics, but also a working kit, which is either a sharp satire or a sign of where web design has landed. Possibly both.
NIN Live fan concert archive
And finally, music fans may enjoy NIN Live, a fan-run archive of Nine Inch Nails concerts. It catalogs thousands of recordings and photos, plus setlists, passes and ticket stubs, and just added streaming for live shows spanning 1988 to 2026 after a Discord login. It's a community project, unaffiliated with the band, and a nice example of fans preserving live history themselves.
That's it for today. Big claims in mathematics, a model that edits itself, and markets having second thoughts about the AI boom. As always, links to all stories can be found in the episode notes. I'm TrendTeller, thanks for listening to The Automated Daily, hacker news edition, and see you tomorrow.
More from Hacker News
- November 19, 2026 Critical NGINX HTTP/2 remote code flaws & Ruby regex engine security patches
- November 18, 2026 OpenAI launches GPT-6.2 Astra, Sol & Microsoft unveils MAI Code 2 models
- November 17, 2026 Optimizely Script Supply-Chain Attack & Windows 11 Tracks Coding Tools
- November 16, 2026 Old Bitly Links Hijacked & uBlock Origin Firefox XSS Flaw
- November 15, 2026 Family tree of software job titles