🛎️ A 37-Year Math Record

Good Morning, AI Enthusiasts!

First they outlasted the mathematicians. Then Wall Street funded the expansion. Now they’re getting hands.

MATH

Claude Pushes a 37-Year Math Record

👀 What's happening: Anthropic set an unreleased Claude on the Riemann Hypothesis. It didn't prove it, nobody has in 167 years, but it moved a number humans couldn't. The share of zeta zeros provably on the critical line: mathematicians pushed it 0.8% across 37 years. Claude added 25.6% in a day and a half, from 41.6% to 67.2%. And it's not alone, through 2026 these records have fallen almost monthly, from Erdős problems to decades-old conjectures. Two years ago we ranked AI by olympiad medals on problems with known answers. Now it sets records on ones nobody has.

🌍 How this hits reality: This wasn't genius. It was stamina. The pieces sat on the shelf for years, Montgomery in 1973, Goldston and Suriajaya naming the exact missing step. No human fitted it. Claude did, across 650 failed attempts, 60 agents and 31 million tokens, half of them producing nothing. The edge wasn't a sharper idea, it was refusing to stop. Humans quit at the thirtieth dead end. The machine was calm at the six hundred and fiftieth.

🛎️ Key takeaway: This still isn't a proof, and this method may still won't become one. But 37 years versus 36 hours is the whole story. The next field-defining breakthrough won't belong to whoever out-thinks the field, but to whatever outlasts it.

TOGETHER WITH FRAMER

Launch your website faster than ever with Framer

Framer is the pro website builder where AI agents work inside the product to make real, reviewable changes to your site. Agents can generate layouts, edit responsive breakpoints, refactor components, update CMS collections, and ship SEO-safe changes, with every edit visible on the canvas and under human control. Trusted by teams at companies like Miro and Perplexity.

Agents bring speed and flexibility; you bring judgment and control. Framer Agents:

• Plug into your existing stack: connect Claude Code, Cursor, Codex, or Gemini CLI and run agents directly on the canvas.
• Workflows that act on your site: agents operate directly on your site model (canvas, components, CMS, and SEO), so output stays editable and production-ready.
• Manage and review: run agents in a branch, diff changes, review with teammates, and merge only what you approve.

COMPUTE

Nvidia’s $500B Debt Flywheel

👀 What's happening: Jensen Huang unveiled a plan with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to raise over $500B from outside investors to build AI data centers. The pitch: a GPU isn't a depreciating server anymore, it's an "AI factory," an investable asset that earns revenue and holds resale value like a power plant. The rental data is real, H100 prices rose through 2026, and so is the Wall Street backing. Huang insists it isn't circular financing: the capital is independent, the deals underwritten by the banks, NVIDIA just provides the platform.

🌍 How this hits reality: Except NVIDIA will backstop up to 25% of a project's residual value. A truly independent asset doesn't need its maker guaranteeing it's worth something. That 25% is the thread tying NVIDIA back into the loop it claims to stand outside of. And the structure rhymes: in the dot-com bust, Lucent and Nortel used vendor financing, lending customers the money to buy their gear, and the flywheel looked flawless until real demand fell short of the demand financing had propped up.

🛎️ Key takeaway: Huang describes a one-way flywheel: more compute, better AI, more revenue, more compute. But flywheels spin both ways.

AGENTS

Meta Gave Local AI Hands

👀 What's happening: Meta Superintelligence Labs released Muse Glimmer, a 30B open-weight agent model under a permissive Apache 2.0 license, runnable on a Mac or a single consumer GPU with no cloud, no per-step API calls, working even offline. On its own benchmarks it nearly matches GPT-5.5 at tool use and agent tasks, while scoring roughly half of frontier models on hard general reasoning. Meta shipped the weights, though not the training data, and llama.cpp, Ollama and MLX support lands within days.

🌍 How this hits reality: Don't mistake it for a local ChatGPT. Ask it to reason through something open-ended and it stumbles. But that isn't the job. Meta didn't shrink a genius; it cut the deep thinking it couldn't fit and maxed out the one axis that matters for grunt work, calling tools and executing steps. That makes it a fit for fixed, repeatable SOPs: read the files, call the tool, run the script, retry on failure. Not an employee who thinks, a machine that follows the procedure, a few thousand times a day, on your device.

🛎️ Key takeaway: Meta built hands, not a brain, dumb at reasoning and sharp at execution, which is exactly half of the work local personal AI has always been missing.

CLAUDE

Anthropic Marked the World for Europe's Rule

👀 What's happening: Any Claude model launched on or after August 2 now ships with an invisible text watermark, woven into the output's statistical patterns so it survives copy-paste and some editing. Older models have until December 2 to comply and are being retrofitted with no timeline. The trigger is EU AI Act Article 50, which only governs AI sold into Europe, yet Anthropic switched the marking on worldwide, across all five product lines. Paid users can't turn it off, it won't publish how the watermark works, and the detection tool isn't out yet.

🌍 How this hits reality: A rule that covers one market got shipped to every user on the planet, with no regional pilot and no opt-out. That isn't transparency, it's overreach, and it hands rivals a gift. OpenAI faces the same Article 50 and still hasn't shipped a matching policy. It doesn't need to win users now, it just needs to stay less aggressive than Anthropic and let them walk over. A watermark you can't switch off just turned every calmer competitor into a migration pitch.

🛎️ Key takeaway: Anthropic stamped every user on Earth to please a regulator that only asked about Europe, and handed every rival that stayed calm a reason to poach them.

DAILY TL;DR

• Bernie Sanders urged Meta, OpenAI, and Anthropic to pause frontier AI development, warning Congress may step in if labs keep scaling risky systems.
• OpenAI expanded Daybreak with GPT-5.6-Cyber, giving vetted defenders limited access to frontier cyber models.
• Microsoft is reportedly preparing to unveil Maia 300 this fall, pushing deeper into custom AI chips to reduce Nvidia dependence.
• ByteDance and other Chinese platforms are shutting down AI companion services as Beijing tightens rules on emotionally manipulative chatbots.
• OpenAI reportedly completed a $7 billion employee tender offer at an $852 billion valuation.
• Linux is seeing larger kernel update waves as AI-assisted review helps developers push more small fixes into the pipeline.
• OpenAI acquired NextSlide, adding prompt-to-presentation tooling to its broader ChatGPT productivity push.

READ MORE

Let the Future Come to Your Inbox

Stay ahead without drowning in information. We turn the most important signals across AI, tech, marketing, and future products into 5-minute reads you can actually finish.

• AI Secret uncovers what really matters in AI
• Bay Area Letters decodes tech and business shifts from Silicon Valley
• Robotics Herald tracks how robots move from labs into daily life
• Marketing Secret breaks down real growth and go-to-market playbooks
• The Hardwire explores hardware, consumer tech, and what’s coming next

TOGETHER WITH US

AI Secret Media Group is the world’s #1 AI & Tech Newsletter Group, reaching over 2 million leaders across the global innovation ecosystem, from OpenAI, Anthropic, Google, and Microsoft to top AI labs, VCs, and fast-growing startups.

We've helped promote over 500 Tech Brands. Will yours be the next?

Email our co-founder Mark directly at mark@aisecret.us if the button fails.