In partnership with
shared.image.missing_image
shared.image.missing_image
Good morning, {{first_name | AI enthusiast}}.
OpenAI says one of its internal models just cracked a Millennium Prize math problem worth $1 million — a feat mathematicians have chased for decades.
But the celebration didn’t last long. An NYU professor says OpenAI got suspiciously close to his own unpublished work right before rushing to publish first. Is this proof AI can now do original math research, or a warning about how frontier labs treat the people doing the underlying thinking?
Today in AI Brief:
OpenAI’s math breakthrough sparks a credit fight
Meta launches its first personal AI agent
ChatGPT’s new Sketch tool for images
Go from AI overwhelmed to AI savvy professional
AI will eliminate 300 million jobs in the next 5 years.
Yours doesn't have to be one of them.
Here's how to future-proof your career:
Join the Superhuman AI newsletter - read by 1M+ professionals
Learn AI skills in 3 mins a day
Become the AI expert on your team
OpenAI’s Navier-Stokes Proof Ignites a Credit Fight
In Brief: OpenAI says an unreleased internal model produced a full proof of the Navier-Stokes equations, one of math’s seven Millennium Prize Problems, but the win came within days of an NYU mathematician accusing the company of racing to claim work he says was already his.
The Details:
OpenAI’s effort burned 300 billion output tokens — about $22.5 million in compute — running roughly 10,000 agents for 88 hours to produce and verify the proof.
NYU’s Tristan Buckmaster says he told an OpenAI mathematician about his own near-finished proof on September 3, and three days later OpenAI’s Sébastien Bubeck said an internal model had already solved the same narrow case.
Buckmaster alleges Bubeck pushed him to drop his Anthropic co-author’s name from the credit, at one point asking “why would you ruin your career?” — OpenAI says it never saw his unpublished work but can’t rule out that de-identified Codex usage data helped.
Take Away:
Whether or not OpenAI’s model got an assist from the mathematicians it’s now accused of scooping, the episode shows an “AI solves a Millennium Prize problem” headline is now within reach for a frontier lab. That milestone will invite far more scrutiny of how the winning proof actually got made, not less.
Meta Launches Muse, Its First Personal AI Agent
In Brief: Meta introduced Muse, calling it the world’s first personal AI agent — a system that books travel, drafts emails, and works toward bigger goals on its own, even after you close the app.
The Details:
Muse runs inside a dedicated Muse Secure VM that isolates the agent and a user’s data, while a separate “Sentinel” agent watches its actions and asks for approval before anything sensitive.
The agent is rolling out free on iOS, Android, and muse.ai in the US first, with AI glasses integration coming and a premium tier for heavier use.
Users can opt out of having their Muse interactions used for model training, addressing one of the biggest privacy concerns around always-on personal agents.
Take Away:
Muse is Meta’s clearest bet yet that the next AI battleground isn’t chatbots but agents that act on your behalf across email, calendars, and smart home devices. Whether people trust an always-on agent with that much access will decide if Muse becomes a daily habit or a privacy headline.
SPONSORED BY THE DEEP VIEW
Become An AI Expert In Just 5 Minutes
If you’re a decision maker at your company, you need to be on the bleeding edge of, well, everything. But before you go signing up for seminars, conferences, lunch ‘n learns, and all that jazz, just know there’s a far better (and simpler) way: Subscribing to The Deep View.
This daily newsletter condenses everything you need to know about the latest and greatest AI developments into a 5-minute read. Squeeze it into your morning coffee break and before you know it, you’ll be an expert too.
Subscribe right here. It’s totally free, wildly informative, and trusted by 600,000+ readers at Google, Meta, Microsoft, and beyond.
ChatGPT’s New Sketch Tool Turns Doodles Into Art
In Brief: OpenAI rolled out ChatGPT Images 2.5, its newest image model, headlined by a Sketch tool that lets users draw directly inside a chat and use the drawing as a visual reference for the final image.
The Details:
Typing “@Sketch” in a conversation opens the drawing canvas, and new templates for formats like posters and merch give users a faster starting point than a blank prompt.
OpenAI says the model produces more natural lighting and richer textures, and generates images up to 50% faster than Images 2.0.
Two companion API models, Flare and Sunburst, ship alongside it — priced at $30 per million output tokens — for developers who want speed or precision editing.
Take Away:
Sketch turns image generation from a guessing game of prompts into something closer to actual creative direction. For an audience that skews toward builders and creators, that’s the kind of “try it now” feature that tends to outperform pure model-release news.
Everything else in AI
Evan Hubinger warned there’s a “greater than 10%” chance AI causes human extinction within a decade, after a fellow Anthropic researcher resigned over the same fear.
Cognition raised more than $2 billion at a $48 billion valuation, with its Devin coding agent’s run-rate revenue nearly doubling to $900 million since May.
Google DeepMind launched AlphaGenome Atlas, a 1-petabyte map predicting the effect of every possible single-letter change in human DNA.
Bedrock Robotics deployed autonomous excavators with empty cabs on active job sites in Texas and Nevada, matching near-human productivity without an operator.