The Download

Illustration: Morning Brew Inc., Photos: Adobe Stock

TL;DR: Right now, the math world is in an uproar. The breakneck pace of AI-made proofs and other discoveries has left human mathematicians worried—not just about machines taking their jobs, but that something gets lost as math is automated. And it puts the field squarely in the middle of an existential anxiety other disciplines have already faced—or could soon.

What happened: Earlier this week, OpenAI posted 722 papers showing progress that an unreleased model had made on open math problems, grouped into 372 “families.” (Raise your hand if you’re also getting number fatigue already.)

But according to some mathematicians, this kind of mass unloading of findings does more harm than good and isn’t all that helpful to the field. It’s part of an escalating debate over the cost of AI’s rapid push into math. Last month, 28 Fields medalists (math’s closest thing to a Nobel Prize) signed a statement saying AI companies’ goals were “severely misaligned” with mathematicians’.

AI’s math speedrun: OpenAI’s dump this week is just the latest in a string of dizzying AI math progress, which has come with a fair share of controversy—including over the issue of proper credit and how much AI discoveries lean on human mathematicians’ prior work.

But the worry over machine-made math goes beyond a credit fight: Some mathematicians say AI’s “skip to the end” approach, as Quanta Magazine puts it, cuts out a crucial middle stretch of doing math. What’s often a long slog of working toward a proof can turn up new math tools and ideas along the way. And so far, critics argue, AI companies have put little effort into helping humans understand their models’ solutions, which leaves the field with less “fertile ground” to build on from each breakthrough, as one mathematician noted.

The revenge of the humans: Plenty of mathematicians are anxious and angry, and there’s even a perception that AI companies are engaging in “mobster behavior,” one math professor told Wired. Some have stopped including open conjectures in their papers so AI scrapers can’t grab them.

Bottom line: Software engineers have already had to rethink their place in a field AI is reshaping. Now it’s mathematicians’ turn to weigh how much human involvement is worth when AI can potentially get from Point A to B faster. And more industries—including your own—could start doing the same. —WK

      

Sponsored By BILL

If your expense process involves a shoebox of receipts, a spreadsheet that makes you groan, or a lucky rabbit’s foot that doesn’t actually feel lucky, listen up.

BILL Spend & Expense has built AI-driven receipt matching and automated categorization where expenses get captured, sorted, and reported fast. Here's what that looks like:

• receipt matching that happens automatically, no shoebox digging required
• AI categorization that cuts manual entry down to almost nothing
• real-time spend visibility across every team and department
• budget controls that flag overspending before it becomes a problem
• one sync to your accounting software so close is actually close

Take a 30-minute demo and see what expense tracking can look like when it works, and BILL will send you a WHOOP wearable as a thank-you for your time.*

Google’s Fitbit Airs on the side of less is more

Disclosure: Companies may send us products to test, but they never pay for our opinions. Our recommendations are unbiased and unfiltered, and Tech Brew may earn a commission if you buy through our links.

Not every coach comes with a mustache and a box of biscuits. The one you can use with your Google Fitbit Air lives in an app and has the same Ted Lasso energy: all encouragement, no yelling. (And honestly, the only kind of coach I listen to.)

I’ve used the Whoop (see our review here) and the Apple Watch, and both seem built for people who want to optimize every heartbeat. I’m more of a big-picture person: I want to know I slept OK and moved enough, then get on with my day. At $99.99, the Fitbit Air is a screenless band that tracks your sleep, heart rate, steps, and workouts without making a whole thing of it.

How we tested it: I wore it nonstop for about a month, only taking it off to shower (although it’s water-resistant up to 50 meters). The band pairs with Google Health (formerly the Fitbit app), where all your stats are logged. Otherwise, it mostly minds its own business, apart from a daily nudge to get your workout or steps in.

I asked its Gemini-powered Google Health Coach (free for three months with the band, then $9.99 a month) to build me a workout plan. After a gym-style intake questionnaire, it made one tailored to my goals. We went back and forth before I confessed I simply wanted to keep doing what I was doing, and it rebuilt the plan around my existing routine.

The Good: The price is a fraction of what you’d pay for a Whoop (where memberships start at $199 a year) or an Apple Watch (starting at $399 for the Series 12). The band is also lighter and thinner than both, so it’s much less clunky to sleep in. The battery lasts a week as promised, which is great for a band this light. (My Apple Watch 7 needs to charge every other day at best—with the 12 having up to 24 hours of battery life, too.)

The Bad: The band could saddle you with yet another subscription to keep track of. The basics (sleep, heart rate, steps) still work without Google Health Premium, but you’ll lose the AI coach, personalized exercise plans, guided workouts, and deeper sleep insights (unless you pay for Gemini).

Verdict: ✅ Signal —LC

Sponsored By Outreach

AI the amplifier. If your systems are fragmented, AI agents will simply scale the fragmentation. That’s why Outreach created the AI Maturity Model. It evaluates your org across five dimensions: workflow standardization, data trust, cross-team alignment, inspection and accountability, and AI readiness. See how it works in their on-demand webinar.

The Zeitbyte

The Sam and Dario chronicles

Illustration: Morning Brew Inc.. Photos: Getty Images, Adobe Stock

Daniela Amodei’s stuffed animal role-play apparently isn’t the only unusual anecdote from journalist Kevin Roose’s new book, which follows how the AI industry has been racing to achieve artificial general intelligence (AGI) in the last few decades (you know, the version you see in Ex Machina)—including details from within OpenAI and Anthropic.

Some of the standout moments, per Business Insider:

• During a company retreat, several OpenAI execs performed a “ritual” that involved burning a demonic-looking effigy—which was supposed to represent “misaligned AGI that had slipped free of human control.” (Cut to: everything that’s happened with misbehaving AI models in the last year.)
• Another vivid moment shows OpenAI demoing GPT-2 to future investor Microsoft back in 2019. In response to a fictional scenario on the first thing Satya Nadella does after becoming CEO, the AI model replied, “…fire all the employees and pivot the company to Azure.”
• Dario Amodei apparently guarded an AGI essay he’d written at OpenAI like it contained nuclear codes, hiding copies of it within hundreds of pages of unrelated philosophical papers.
• An early version of Claude took Anthropic staffers’ request for a “killer guacamole recipe” too literally, accusing them of trying to murder people.

We might not have reached AGI yet, but we do have a growing number of bizarre AI world incidents—and a whole lot of AI-generated videos of famous frenemies Dario and Sam doing everything from dancing to “Gangnam Style” (it’s just as ridiculous as you may think) to starring in Succession parodies. —WK

Chaos Brewing Meter: ☕☕☕/5

Sponsored By Slate Flosser

That’s some hot floss. To quote your dentist, you probably aren’t flossing enough. You don’t need more time, just better tools. Slate electric flossers help make flossing easier and more effective. Their patented three-in-one flosshead system provides a deeper clean that even dentists will notice. Check them out.

• ChatGPT will soon look a whole lot less like a chatbot.
• The Curiosity rover captured this stunning image of dawn over some cliffs on Mars.
• If it’s a day that ends in Y, a company is releasing its own AI agent. This time, it’s Google’s turn—but its agents are specifically for the workplace and come with some very useful tools.
• If you’ve ever wondered what it’s like to spend $1 trillion, this game from Wired is made for you.
• Nostalgic gadgets are so back: Even this discontinued Apple product is getting cloned.
• Scared a smart lock will leave you locked out? Ring has a surprisingly old-school solution on the way.

^{*A message from our sponsor.}

This time last week...

Readers’ most-clicked story was about the newest “most advanced” Gemini model being released. And, judging by its name, it’s colorless, odorless, and tasteless—it’s also designed for four very specific careers.