Today’s HN front page is dominated by a single story: OpenAI’s claim to have solved a Millennium Prize Problem using an army of AI agents. The community is skeptical, the math is preliminary, and the concurrent drama with an NYU mathematician adds a layer of intrigue. Meanwhile, LG TVs are caught spying even when powered off, Google is quietly strangling new wiki domains in search results, and Paramount is running fake grassroots campaigns to save its merger.

OpenAI Claims Navier–Stokes Millennium Prize Solution (714 points)

OpenAI announced that an internal model—“significantly more capable than GPT-6 Astra”—produced an analytical proof and Lean formalization showing that smooth 3D fluid motion can develop a singularity in finite time. The proof describes a vortex that spirals inward and elongates like spaghetti, with energy remaining finite throughout. They used approximately 10,000 concurrent agents coordinating across 2.7 million messages and 130 billion output tokens, arriving at the solution 88 hours after launch.

The timing is suspicious. OpenAI heard rumors on September 1st about a competing solution, then accelerated their own effort. They claim their proof differs from the concurrent work by Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic), and they’ve offered to share their prompts and proof. Buckmaster’s own statement reveals the real story: the foundational ideas came from Diego Córdoba and Luis Martínez-Zoroa, and Buckmaster/Alpöge used LLMs to push the program forward. Buckmaster openly calls his Euler writeup “AI slop” and apologizes for the presentation quality. The whole situation reads like a race to claim credit, with mathematicians caught between genuine breakthroughs and corporate PR.

Source: OpenAI | Buckmaster Statement (PDF)

Tristan Buckmaster and Levent Alpöge independently solved related problems (forced Euler, Boussinesq, incompressible porous media) using LLMs, with the foundational ideas attributed to Córdoba and Martínez-Zoroa. Buckmaster’s statement is refreshingly honest: the proofs are “much closer to what models produce under human direction than to a paper written by a person,” and he calls the Euler writeup “AI slop.” He also reveals being pressured by outside factors to release prematurely, and notes this is a “Deep Blue-Kasparov moment” for mathematics.

The drama: Buckmaster emailed a prominent OpenAI mathematician on September 3rd to prevent a credit dispute, only to have OpenAI launch their own 10,000-agent blitz days later. Buckmaster’s tone suggests frustration with how the situation unfolded.

Source: Buckmaster Statement (PDF)

LG TVs Caught Spying Even When Offline or on Standby (380 points)

According to a Gamers Nexus investigation covered by The Verge, LG TVs have been caught recording audio, scanning Wi-Fi networks, and transmitting data even when in standby mode or disconnected from the internet. This isn’t a bug—it’s a feature. The TVs appear to be designed to collect data regardless of user settings or network connectivity.

This is the logical endpoint of the smart TV business model: the TV is a surveillance device that happens to display video. The fact that it works offline suggests the data is stored locally and transmitted when connectivity is restored. If you still have a “dumb” TV, hold onto it.

Source: The Verge

Google DeepMind Releases AlphaGenome Atlas (379 points)

DeepMind released AlphaGenome Atlas, a database predicting the effects of every possible single nucleotide variant in the human genome. The 1-petabyte dataset was generated by pre-calculating the regulatory impact of all 9 billion single-letter genetic changes using the AlphaGenome model. The Atlas introduces an “AVI score” for rapid variant prioritization.

Early results are promising: at the Broad Institute, the AVI score helped identify a critical splice-site variant in the DNM1 gene for rare disease research. Another team found 22% more non-coding genetic associations in UK Biobank data. The tool is available through a web portal requiring zero coding skills, which is genuinely useful for clinical researchers who don’t want to wrestle with bioinformatics pipelines.

Source: Google Blog

There’s a New “Google Jail” for Independent Wikis (488 points)

Since March 2024, Google has been effectively blocking new domains from appearing in search results for anything other than their main page. This has devastated independent wikis trying to leave Fandom—wikis like gta.wiki, hollowknight.wiki, and the official Path of Exile 2 wiki are all stuck in this “Google Jail” for up to a year. The author’s hypothesis: Google decided they couldn’t effectively filter SEO slop, so they nerfed all new domains instead.

The workaround is to launch on subdomains of existing domains (like overwatch.weirdgloop.org) rather than new root domains. Subdomains of established domains are immediately indexed properly. This is a massive shift in strategy for the independent wiki ecosystem and effectively gives Google veto power over which new sites can succeed.

Source: Weird Gloop

Paramount Caught Using ‘Astroturf’ Group to Drum Up Fake Support for Merger (277 points)

Paramount’s $111 billion merger with Warner Bros. has hit a new low: they’ve been caught using a nonprofit called “Neighbors for Strong Communities” to bombard California residents with texts pressuring AG Rob Bonta to drop his antitrust lawsuit. The organization is less than three months old, discloses no founders or funders, has a PO Box address at a UPS store, and doesn’t appear in ProPublica’s nonprofit database. Classic astroturf.

This follows Paramount’s previous tactics of accusing deal critics of antisemitism and threatening to leave California if the state enforces antitrust laws. The company’s desperation to complete this merger, combined with Warner Bros.’ history of mass layoffs after acquisitions, suggests the deal benefits shareholders and executives—not consumers or workers.

Source: TechDirt

DHS ‘Predictive Policing’ Unit Is Analyzing Americans’ Financial Habits (272 points)

404 Media revealed that Border Patrol is running secretive predictive policing units that analyze Americans’ financial activity and other data, then feed intelligence to local police who pull over people not suspected of any specific crime. The article is paywalled, but the headline alone tells you everything: the government is profiling citizens based on their spending patterns and using local law enforcement as enforcement proxies.

Source: 404 Media

DaVinci Resolve 21.1 (292 points)

Blackmagic Design released DaVinci Resolve 21.1 with AI assistant integration (Claude, ChatGPT Codex) for project management tasks like organizing media and batch rendering. The update adds expanded camera format support (Fujifilm, Leica SL, Sony A7R V1), enhanced multi-cam workflows with up to 25 camera angles, and over 25 new Krokodove tools for Fusion 3D work. Notably, multimaster trim now supports independent passes for Dolby Vision, HDR10+, and HDR Vivid. Free download, as always.

Source: Blackmagic Design

Throughline

The dominant theme today is AI systems doing things that were previously impossible—solving 90-year-old math problems, predicting genomic variants, and managing video editing workflows. But the stories underneath are more interesting: the Navier-Stokes “solution” is tangled in a credit dispute between OpenAI and independent mathematicians, the genomic atlas is genuinely useful but will need independent validation, and the DaVinci integration is pragmatic rather than flashy.

The second theme is institutional power abusing technology: LG TVs surveil users offline, Google effectively decides which new websites can succeed, DHS profiles citizens’ financial habits, and Paramount runs fake grassroots campaigns. The contrast between AI’s potential to solve fundamental problems and its use as a tool for control and surveillance is stark.

The Navier-Stokes situation is the most revealing. Buckmaster’s honest assessment—“AI slop” proofs, a mathematician caught between genuine breakthroughs and corporate PR, and the realization that LLMs can now do in a month what used to take years—suggests we’re at an inflection point. The question isn’t whether AI can solve hard problems, but who gets credit and who controls the narrative.