OpenAI says it cracked a 90-year-old math problem.

When you need more leads, you want to do more, right?

But what do you do if your leads are on LinkedIn? LinkedIn outreach still depends on the limited number of profiles already on your team.

Aimfox Avatars lets you rent dedicated, customizable LinkedIn profiles that start conversations and hand interested replies to your existing sellers.

Add outreach capacity without hiring another SDR or turning every employee account into a prospecting channel..

Book more calls with Aimfox Avatars

1.  Meta ships a personal agent that runs on its own computer. Meta introduced Muse, an agent app that connects to your email, calendar, shopping and smart home and then does the chores, from booking travel to haggling down a bill. Each user gets a dedicated cloud machine where the agent and their data live, with a separate "Sentinel" process that must approve anything leaving for the internet. It is free for most tasks, US-only for now, and it pays with one-time-use cards from Stripe's Link so your real card number stays hidden. The obvious question is whether people will hand Facebook's parent that much access; TechCrunch asked exactly that.

2.  ChatGPT Images 2.5 arrives with a sketchpad. OpenAI released Images 2.5, promising more natural lighting, edits that change only what you asked for, and generation up to 50% faster than Images 2.0. The feature people are talking about is Sketch: draw a rough layout inside ChatGPT, add a line of description, and the model turns the doodle into a finished image. It is rolling out to every ChatGPT tier, and two API versions (Flare for speed, Sunburst for precision) ship for developers.

3.  An Anthropic researcher quits, and a colleague says he is right. Jacob Coxon, who spent three years on pretraining at OpenAI and then Anthropic, resigned with a post accusing both labs of "racing straight to self-improving superintelligence and gambling with our lives." It has passed 171M views. Then Evan Hubinger, who leads alignment science at Anthropic, replied that Coxon is correct, put his own odds of AI wiping out humanity within a decade above 10%, and said the company does not yet have a plan for aligning superintelligence.

4.  The Sam Altman movie has a trailer, and it is not a puff piece. Neon released the first teaser for Artificial, Luca Guadagnino's film about the five days in 2023 when Altman was fired and rehired, with Andrew Garfield in the lead. Amazon MGM walked away from the project earlier this year, not long after its $50 billion deal with OpenAI, as Variety notes. It premieres at the New York Film Festival on October 5 and opens in theaters on Christmas Day.

The claim. On September 8 OpenAI published what it calls a resolution of the Navier-Stokes existence and smoothness problem, one of the seven Millennium Prize Problems. An unreleased internal model, which OpenAI says is well beyond GPT-6 Astra, argues that a smooth fluid at rest can be pushed into a vortex that stretches like spaghetti until its speed becomes infinite in finite time. The proof comes with a formal verification in Lean, and OpenAI says it will not claim the prize money.

Why it matters. Navier-Stokes describes how fluids move, which is why it sits inside aircraft design, weather forecasting and blood-flow models. Showing the equations can break down tells physicists where the smooth picture of a fluid stops being trustworthy. If referees accept it, this would be the first Millennium problem resolved by a machine. The full paper runs to the technical details.

How it was done. OpenAI describes a swarm of roughly 10,000 coordinating agents working for about 88 hours, exchanging 2.7 million messages and consuming around 130 billion output tokens, before GPT-6 Astra spent another 17 hours formalizing the result. The effort began on September 1, after the company heard a rumor that two Millennium problems had already fallen.

The dispute. That rumor traced back to Tristan Buckmaster of NYU and Levent Alpöge of Anthropic, who had spent a year on a closely related problem using an unusual approach. In a public statement, Buckmaster says he had uploaded drafts of that work into Codex, and that OpenAI, which could see those logs, then produced a proof along the same lines. OpenAI denies it, and has said no user data was accessed.

OpenAI's answer. On September 10 the company updated its announcement with the results of an internal investigation. It says Buckmaster's Codex prompts could not have influenced the model, including through training, that neither its researchers nor its agents saw the pair's work before it was public, and that the two proofs differ. It also credits Alpöge and Buckmaster with priority on a related result for the Euler equations.

The bigger worry. Whether or not the proof holds up, the episode hands academics a concrete version of a fear they have voiced for years: that pasting unpublished work into a lab's tools means the lab can see it. Every researcher who checks a draft with Codex or ChatGPT now has to weigh that. The math will be settled by referees. The trust question will not.

Construction-ready. Former X product chief Nikita Bier spent a weekend feeding land surveys and satellite photos into AI and says the renovation concepts it produced were close to buildable, predicting a future with few middlemen between consumers and the last mile of labor (1.7M views). The architects in his replies were not convinced.

Check the machine's work. OpenAI put the Navier-Stokes proof, the Euler result and both Lean formalizations on GitHub, so anyone with a Lean install can verify the argument line by line rather than take the announcement on faith.

Your 80s headshot. Images 2.5 now lets you share the prompt behind any picture, and OpenAI's own 80s flashback prompt is the one making the rounds. Drop in a photo of yourself and prepare for the windbreaker.

Under the hood of Muse. Skeptical of handing Meta your inbox? The company published a technical explainer on how the Sentinel process, credential vault and audit trail are meant to work, and says an encrypted version that even Meta cannot read is due later this year.

Founder grunt work. OpenAI has bundled a set of ChatGPT plugins for small businesses, covering the invoicing, scheduling and admin tasks that quietly eat a founder's week.

Turn any meeting notes into a project plan

This one works in ChatGPT, Claude or Gemini, and it shines with a reasoning model like GPT-6 Astra or Claude Fable if your plan includes one. Paste a transcript or your scribbled notes and you get back a tracker with owners, deadlines and risks, plus a flag on every task that nobody actually owns. Try it on the messiest meeting from last week and see how much fell through the cracks.

Swap [PROJECT NAME] with the actual initiative (Q4 launch, office move, website redesign) and paste the raw transcript into [PASTE MEETING NOTES]. Advanced tip: ask for the table as a downloadable spreadsheet, then run the same prompt after next week's meeting and ask the model to show what changed between the two trackers.