In partnership with
shared.image.missing_image
shared.image.missing_image
Good morning, {{first_name | AI enthusiast}}.
OpenAI didn’t just ship a new model this week — it says its AI helped crack three problems that have stumped humans for decades, including a foundational assumption about computational complexity that’s stood since 1999.
OpenAI brought in outside mathematicians and a Nobel physicist to check the work before saying a word publicly, and it’s promising to release the raw data for anyone else who wants to try. If AI models can really produce new, verifiable results on open problems instead of just remixing what’s already known, how long before “AI-assisted proof” becomes a normal line in a research paper’s methods section?
Today in AI Brief:
OpenAI’s AI disproves a 1999 complexity assumption
Anthropic’s IPO filing reveals its real numbers
Microsoft offers buyouts to 60,000 workers
Stop typing AI prompts. Start talking.
You think 4x faster than you type. So why are you typing prompts? Wispr Flow turns your voice into ready-to-paste text inside any AI tool. Speak naturally, tangents and all, and Flow cleans it up. Available on Mac, Windows, iPhone, and Android.
OpenAI’s AI Cracks Three Decades-Old Problems
In Brief: OpenAI said its AI produced three scientific results this week in computer science, mathematics, and physics, bringing in outside experts — including Nobel laureate David Gross — to review each claim before publishing.
The Details:
OpenAI’s model produced a disproof of the Exponential Time Hypothesis, a 1999 assumption about how hard the hardest satisfiability problems can get, then formalized the proof in the Lean assistant for the hypothesis’s original authors to review.
The math proof cracked the local smoothing conjecture across every dimension at once, a problem open since 1991 that mathematicians had only solved dimension-by-dimension, with the 3D case falling as recently as 2025.
A physics proof established existence for Yang-Mills theory (not the mass-gap portion, so the $1 million prize stays unclaimed) — and the same day, OpenAI shipped GPT-6.1 Terra, a cheaper, faster model available now in ChatGPT Work and Codex.
Take Away:
OpenAI pledged not to claim academic credit and promised to release all the raw data for outside verification, a notable show of restraint for results this dramatic. If the disproof and the conjecture proof hold up, it’s a sign AI models can now contribute original results to open problems, not just summarize what’s already known.
Anthropic’s IPO Filing Shows Triple-Digit Growth
In Brief: Anthropic filed its S-1 with the SEC for a Nasdaq listing under the ticker ANT, revealing $16.3 billion in revenue for the first half of 2026 alone.
The Details:
Revenue hit $16.3 billion in the first half of 2026, after the company brought in just $4.59 billion for all of 2025.
The $80.7 billion net loss sounds alarming, but $72.4 billion of it comes from a non-cash accounting adjustment on convertible notes — the actual operating loss was $693 million, down from roughly $2 billion a year earlier.
Claude wrote more than 90% of Anthropic’s own code in September, and the company says Claude’s internal usage topped its monthly payroll cost by April.
Take Away:
The filing paints Anthropic as a company growing faster than almost any traditional business ever has, even as its accounting shows how distorted “loss” can look once convertible notes are involved. It also doubles as an unusually candid admission: the filing itself warns that some of Claude’s safety-related work could become “difficult for any individual or group of humans to understand and control.”
SPONSORED BY GLADLY.AI
A Single Typo Shouldn’t Cost You a Sale
A shopper types “running shos” and gets nothing back. Another types “red waterprof trail runners under $150” and hits the same wall. Both knew what they wanted.
Traditional site search relies on exact keyword matching, leaving shoppers stranded at “0 results found” screens.
Gladly understands what shoppers mean, typos and all. Instead of a dead end, they get products that fit what they asked for and a conversation that keeps them shopping.
Microsoft Offers Buyouts to 60,000 Workers
In Brief: Microsoft is offering voluntary buyouts to about 60,000 employees — roughly a quarter of its global workforce — as it keeps trimming headcount while pouring money into AI data centers.
The Details:
Chief People Officer Amy Coleman detailed a standard severance package in a Monday memo, open to eligible employees at the vice president level and below in the U.S., Canada, India, and the U.K.
Xbox, Windows, Copilot, and Finance are among the participating divisions, while data center, sales, and marketing staff are excluded from the offer.
Employees have until early December to decide, with those who accept exiting by the end of the year.
Take Away:
Microsoft keeps choosing buyouts over blanket layoffs as it tries to free up budget for the data centers powering its AI push, which it has said is straining its finances. A quarter of the workforce getting an exit offer — even a voluntary one — is a striking number for a company still posting massive profits.
Everything else in AI
AgiBot choreographed 110 of its Expedition A3 humanoids dancing in sync for nearly six minutes in Jinan, setting a Guinness World Record with ultra-wideband positioning holding the formations together.
Authentic Brands Group sued Anthropic for $41 million, claiming Claude was trained on 274 Elvis Presley songs without a license.
Pollen Robotics unveiled Microduck, a $399 robot duck demoed by Nvidia’s Jensen Huang at GTC Berlin that’s already pulled in more than 20,000 preorders ahead of Christmas delivery.
Canva launched Craft, its first in-house image-generation model, which ranks #1 on DesignArena for portraits and landscapes and is available now through Canva and its public API.