OpenAI Solved a Millennium Prize Problem, Nvidia Bought Hugging Face, and an NYU Mathematician Says He Got Raced to It. Who Owns a Discovery Now?
Hi,
So…… the AGI era has begun.
How are we feeling about that
let’s go.
Free competitor analysis from CEOfriend
This week from the AI for Founders podcast on Spotify:
Two AI PhDs Are Now in the Business of Pricing Your Next Lawsuit — LINK
Charles Higgins and Sophia Kalanovska of Redberry Labs built the company after major insurers pulled back from covering AI agents this year. Their sharpest point: an agent-drafted email you approve and send could trigger a lawsuit that no insurance category yet cleanly covers.
She Pivoted Her Ad Agency the Day a Client Asked If Her AI Trader Was a Real Person — LINK
Katrien Grobler of Twinnin now verifies humans, hashes consent terms, and routes 75% of licensing fees back to the person behind the likeness, built after she realized nobody had asked the person being deepfaked.
Five Days of 12-Hour Claude Code Sessions, and What Came Out the Other Side — LINK
Jason Katz of Kindling walks through the canyon between a vibe-coded prototype and a production-ready system, and previews Chief of Staff, an app that turns 10 minutes of daily talking into a full operations plan after 30-45 days.
Why Every Founder Now Has the Same AI Tools, and Why That's the Problem — LINK
Jacqueline Schmidt, CEO of the $24M Series A design agency Radiance, argues access has stopped being the advantage, memorability is the only one left. Her test: name your brand's fingerprint the way people recognize Warby Parker's blue without the logo.
Investor Corner
Are you raising? Reply here, and we’ll feature your deck.
An autonomous AI job search agent for senior tech professionals. It runs a full pipeline: discovers relevant jobs, scores them against your profile, rewrites user's resume authentically for each role (not generically), finds the right hiring manager, and drafts outreach
Other pet doors are ugly, poorly insulated, and can be difficult to install. Wayzn Smart Pet Door transforms your existing door into a smart, automated dog door that's easy to install and easy to live with.
DiningOut unites restaurants, food lovers, industry suppliers, and brands through a self-reinforcing ecosystem of live events, targeted media, and proprietary technology.
Rare, Pristine, Protected drinking water.
The only platform that can authenticate human-made content. Using patented algorithms that analyze how content is made, not just the output.
Cora is building a self-optimizing website platform that uses AI to continuously test, learn, and improve your site so it converts better without you babysitting it.
AI-powered training simulations to help healthcare learners practice complex interpersonal and communication skills. We focus on realism, repeatability, and scalability—making soft-skills training more accessible and measurable.
AI + Human Stylists → LEARN MORE → Save $30 with RYAN30
Tip of the spear:
An unreleased OpenAI model burned 300 billion tokens and about 22.5 million dollars in compute to produce a proof for Navier-Stokes existence and smoothness, one of math's Millennium Prize problems. NYU's Tristan Buckmaster says OpenAI got there first using his own unpublished approach, a claim OpenAI disputes, and nobody outside either party has verified it yet.
Nvidia Is Paying $12.9 Billion for the Platform 18 Million Developers Live On — LINK
Hugging Face turned down a $500 million Nvidia offer years ago. Jensen Huang says the platform stays open. Owning where developers show up is its own kind of leverage.
OpenAI Shipped a Model That Reasons Where Nobody Can Watch — LINK
Astra claims the top spot for coding and cybersecurity, but it reasons through opaque recurrence, looping internally instead of showing its work. OpenAI calls the tradeoff inevitable. Safety researchers call it the end of monitorability.
Crusoe Tripled Its Valuation to $30 Billion in Ten Months — LINK
A $13 billion infrastructure contract with Jane Street did the heavy lifting. The AI buildout keeps minting valuations that looked absurd eighteen months ago, and Crusoe is reportedly eyeing an IPO next.
Mira Murati's Lab Is Raising at $40 Billion, Right After Losing Two Cofounders Back to OpenAI — LINK
Thinking Machines is reportedly closing $1 billion, down from the $50 billion it originally wanted. Even star-studded AI labs are learning that talent retention is the harder fundraise.
Wonderful Doubled Its Valuation to $5 Billion, Again — LINK
Salesforce just joined as a new investor. The edge isn't the model, it's forward-deployed engineers who sit inside client operations, a services layer most AI startups skip entirely.
Google Keeps Its Ad Empire, Gets a Warning Instead — LINK
A federal judge rejected the DOJ's breakup push and ordered operational changes instead, details still sealed for two weeks. Behavioral remedies keep beating structural ones in these cases.
150 Million ID Cards Just Showed Up on the Dark Web — LINK
Hackers appear to have breached IDScan, an identity verification vendor, with new documents reportedly added at 500,000 a day. Largest known breach of ID documents in years, right as governments push more age-verification laws.
Nvidia's Fingerprints Are on Both Sides of This Robotics Deal — LINK
Nscale is supplying Figure's humanoid robots with at least $3.5 billion in compute, scaling toward $6 billion, while taking an equity stake in Figure. Nvidia backs both companies. The supplier now owns a piece of the customer.
Subscribe and save 35% here.
Framework:
Price the Failure Before It Happens
Most founders deploying AI agents have no idea what a catastrophic failure would actually cost them, and neither do their insurers, which is exactly why the insurance market pulled back from covering agents this year. This week's Redberry Labs conversation lays out a way to quantify that risk before a regulator, a lawsuit, or an insurer does it for you.
01 — Map every failure mode before you ship. List every plausible way your agent could go wrong: a bad email that gets sent, a wrong charge, a decision made without a human catching it. The scenario that bankrupts you is usually the one nobody wrote down.
02 — Simulate the long tail, not the average case. Run your agent through thousands of synthetic scenarios weighted toward the rare, expensive failures. A 1-in-10,000 catastrophe is still a catastrophe, and averages hide it completely.
03 — Put a real number on each scenario. Attach a dollar cost and a probability to every failure mode, even a rough one. An agent risk you can't price is a risk you're carrying for free.
04 — Find the insurance category before you need one. Decide in advance whether a given failure looks like a D&O claim, a tech E&O claim, or a cyber claim, and get your broker to confirm it in writing. Waiting until after the lawsuit is how founders discover they have no coverage at all.
05 — Revisit the model every time you expand agent scope. A new tool, a new permission, a new customer segment all change your failure surface. Treat the simulation as a living document, not a box you check before a fundraise.
Source: aiforfounders.co/frameworks
Done-for-you brand sponsorships for podcast creators.
Keep it moving
The Reasoning Trick That Has Safety Researchers Worried — LINK
OpenAI's opaque recurrence loops reasoning internally instead of showing a readable chain of thought. Researchers warn it could kill the ability to monitor for misalignment entirely if labs push it further.
The US Government Just Took OpenAI's Side on Copyright — LINK
The Trump administration filed a brief backing OpenAI's right to train on copyrighted material without permission. A strong signal for how courts might rule in the New York Times case.
Nvidia's Investment Portfolio Hit $99 Billion — LINK
Up fourteen-fold in a year, including $30 billion in OpenAI alone. Critics call it financing your own customer base. Nvidia calls it a flywheel.
HiddenLayer Raised $100 Million to Guard Your Agents — LINK
Gartner projects AI security spend hits $4.78 billion next year. Securing agents from prompt injection and supply-chain attacks is becoming its own category fast.
Your ARR Is Less Solid Than You Think — LINK
77% of enterprises reevaluate their AI vendors every six months, and fewer than half of pilots reach production. Your ARR might be real, it's just a lot less sticky than the SaaS metrics it's dressed up as.
There's Now a Business in Stripping AI's Guardrails — LINK
Abliteration.ai sells guardrail-free versions of open-weight models for red-teaming. TechCrunch confirmed it will readily generate malware and bioweapon instructions on request.
The world’s best podcast player. Snipd
Meta Will Pay You to Let It Watch You Use Its AI — LINK
Muse Spark pricing drops 95% for users who let Meta train on their prompts and outputs, a trade meant to fix its data problem after employee-tracking plans got shelved.
Ollie Is Betting Families Want a Private AI Assistant — LINK
A $7.5 million seed round for a text-based family assistant that promises never to sell or train on your data, betting privacy becomes the differentiator competitors like Instinct can't match.
Every AI-Generated Menu Looks the Same, and Here's Why — LINK
Too symmetrical, too smooth, too uncanny, because the models were trained on the same handful of chain-restaurant photos. Founders building anything visual should take the training-data warning personally.
Google Wants You to Talk to Gmail, Docs, and Keep — LINK
Gemini-powered voice features let users talk instead of type to search inboxes, draft documents, or build shopping lists. English only for now, iOS and Android.
The EU's Tech Chief Says US Guardrails Are Coming Anyway — LINK
Just through lawsuits instead of regulation, she argues, pointing to hundreds of state AI rules already on the books and Meta's $12.19 billion youth-safety settlement.
A Glossary Worth Bookmarking Before Your Next Investor Call — LINK
TechCrunch's living glossary covers everything from tokens to opaque recurrence to neuralese. Handy the next time an investor drops a term you'd rather not Google mid-meeting.
Create a new corporate Uber account here and receive up to 10% off. Other terms and conditions apply, see full terms.
What I’m thinking about:
Your face might be the most valuable thing you own, and you're giving it away for free. — Clip
Thanks for being here.
-Ryan
p.s. I write about Revops, Product, and Founder-led marketing on Linkedin , X , and my blog .
