Did OpenAI Cheat to Solve a 90-Year-Old Math Problem? & Teens’ Test Scores Are Plummeting

SPEAKER_1 (0:01)

JPMorgan has teams to support you at every stage of your growth. As former founders who have worked in local markets themselves, the bankers at JPMorgan bring deep sector knowledge and lived experience to their clients. JPMorgan can help you navigate complexity with confidence backed by real-world insights and an entrepreneurial perspective. Get what your startup needs now at jpmorgan.com or grow without limits. JPMorgan is the bank of the innovation economy.

Neal Freyman (0:28)

Good morning, Brew Daily Show. I'm Neal Freyman.

Toby Howell (0:31)

And I'm Toby Howell.

Neal Freyman (0:32)

Today, why teen reading scores are falling off a cliff.

Toby Howell (0:36)

Then OpenAI solved a famous math problem, but did they cheat to do so?

It's Wednesday, September 9th. Let's ride.

Neal Freyman (0:51)

Monday might have been Labor Day, but today is another kind of Labor Day. That's because September 9th is the most popular birthday for people in the United States, according to research from FiveThirtyEight. They looked at the registered births in the US from 1994 to 2014, and found that September was the most popular birth month, and the ninth day of that month, today was the most popular birthday.

Doing a little back of the napkin biological math, babies that are born in early September were conceived in late December, which checks out.

Toby Howell (1:25)

So I went through and found some famous people born today, and I gotta say would make for a heck of a dinner party. You got Adam Sandler, you got Hugh Grant, Michael Buble, then you get Leo Tolstoy and my favorite Colonel Sanders. Sounds like a pretty fun group to hang out with. My brother's born in September, all these September birthdays. Too many people being conceived around the holiday period. A lot of people inside, a lot of bad weather. I understand why it's happening. Just September birthdays arise up. And our word from our sponsor, hotels.com.

Neal Freyman (1:59)

Neal, when you stay at a hotel, must have continental breakfast. If I don't start my day with a waffle, it's ruined.

Toby Howell (2:06)

Sounds like you could use business preferences from hotels.com. It helps travelers find the right hotels faster by remembering their go-to work trip requirements.

Neal Freyman (2:16)

Travelers set their preferences once, including fully responsible stays, Wi-Fi, breakfast, gym access, and more. Then future business trip searches automatically surface matching hotels.

Toby Howell (2:25)

That way you can book work trips faster. Just set your preferences once and get results tailored to your business travel needs. Plus, get up to 20% off select hotels instantly with member discounts and earn rewards for future travel.

Neal Freyman (2:38)

Sign up for their free rewards program and start earning on every work stay at hotels.com/morningbrew.

That's hotels.com/morningbrew. Loyalty members 18 plus.

Toby Howell (2:51)

There are seven problems in the world of mathematics that are equivalent to winning the Super Bowl and World Series at the same time. Established in 2000, they are called Millennium Problems and carry with them a $1 million prize if you can solve them. In the last 26 years, only one has been cracked, but OpenAI thinks it just solved another. Called the Navier-Stokes Problem, it seeks to describe how liquids and gases move, which underpins things like weather forecasting, airplane wings, and even blood flow. The main question mathematicians were trying to solve is, does a fluid always remain well behaved or can it get riled up to the point where it does something impossible, like water speeding up forever and even blowing up? OpenAI says, yes, it can. It found a scenario where, quote, an initially smooth fluid at rest can develop a singularity in a finite time. For mathematicians, it's a major finding, a literal once in a millennium breakthrough. For us normies, the answer itself is secondary to how it got there. OpenAI released a swarm of agents on Navier-Stokes, 100 of them initially, but after it was clear progress was being made, the number swelled to 10,000. In total, those agents exchanged 2.7 million messages, the total time it spent working was 88 hours, and the total compute costs in the millions of dollars, all by an unreleased model still in testing.

But the funniest part is the reason OpenAI sick a swarm on the problems is because it had caught wind that Anthropic had cracked two other millennium problems and didn't want to be left behind. We were curious if ours could do it too, Sam Altman said. Neal, this story has it all. Crazy math, a little jealousy between labs, and apparently some plagiarism claims for good measure.

Neal Freyman (4:32)

This has all blown up in OpenAI's face from a PR perspective and devolved into a messy, very intriguing drama involving OpenAI perhaps playing dirty here, and who should actually get credit for solving the problem. I look forward to watching the Nathan Fielder documentary one day because I do not think math gets more spicy than this.