- Need2Know, by Cheddar
- Posts
- OpenAI Invisibly Watermarks Text in EU
OpenAI Invisibly Watermarks Text in EU
Plus: ChatGPT Cracked 300 More Math Problems
This week’s world-famous-news-haiku-competition™ is about how Americans are spending more money on hobbies despite economic pressure. Send me your entry — to haiku at cheddar dot com — by noon ET Thursday (today!), for consideration by your Cheddar peers. But first? News!
Matt Davis — Need2Know Chedditor
Table of Contents
What’s the Stock Market Up To, Eh?
Companies Mentioned in Today’s Newsletter
$OPEAZZX ( ▼ 0.31% ) $ANTHZZX ( ▼ 0.06% ) $DIS ( ▼ 1.45% ) $CMCSA ( ▲ 0.4% ) $CBS ( 0.0% ) $FOX ( ▼ 2.65% ) $AMZN ( ▼ 2.19% )
OpenAI Invisibly Watermarks Text in EU

(Cheddar.com)
No wonder those French students are rioting. (Er, that isn’t why they’re rioting, Matt — Ed.) To satisfy the EU AI Act’s transparency rules, OpenAI $OPEAZZX ( ▼ 0.31% ) is rolling out an “invisible text watermark” across the EU for its ChatGPT and Codex models. Instead of slapping a visible stamp on your text, the system uses a method called “textGrain” to subtly shape word choices using a secret key to sort next-word predictions. Readers cannot see the pattern, but an AI detector can — and the signal even travels with the text when copied and pasted. L’horreur!
Of course, every digital lock invites a clever workaround. OpenAI’s tests revealed that replacing just 10% of words with synonyms tanked the detection rate from roughly 92% to 66%. Short passages, math answers, and translated text are also notoriously hard to spot. Acknowledging these hurdles, OpenAI said, “These limitations contribute to our decision to provide initial detector access only to approved researchers and expert organizations.”
Thank the Lord.
Furthermore, OpenAI acknowledged that watermarks “can indicate that an OpenAI system generated or processed part of a passage, but not how much human judgment, editing, or creativity went into it.”
Exactly. Plus, it’s an EU-only rollout. Sucks to be French! Ha ha! And it comes two months after Anthropic began watermarking Claude globally. That move triggered backlash from users who argued they had supplied “the instructions, context, decisions” while Claude was merely “the tool.” OpenAI had previously held back on text watermarking over fears users would flee to un-watermarked rivals. But for now, if you want your AI essay to stay under the radar in Europe, just…keep a thesaurus handy!
Quote of the Day
Kalshi’s contracts pose numerous threats. They undermine the integrity of our elections by incentivizing misinformation and other forms of election manipulation for profit.
ChatGPT Cracked 300 More Math Problems
Just a month after OpenAI's internal AI model cracked a notoriously difficult Millennium Prize Problem, the lab has dropped 372 groups of results that included AI solutions to more than 300 math problems across number theory and algebraic geometry.
The academic world is unhappy. Reacting to a proof regarding the well-known “quasi-Riemann hypothesis,” Alex Kontorovich, a distinguished math professor at Rutgers University, wrote on X, “If a human did this, it would be an instant Fields Medal, no questions asked.” Meanwhile, even a rival researcher at Anthropic $ANTHZZX ( ▼ 0.06% ) called the release “obviously the most significant moment in mathematical history.”
Mathematicians now feel like collateral damage in a trillion-dollar corporate arms race. Dozens of Fields Medal winners signed a furious open letter warning that the tech industry's rush “threatened to undermine the very purpose of math,” explicitly pleading, “We ask them to stop testing advanced mathematical problems on proprietary models.”
OpenAI’s response to the academic revolt was essentially a polite corporate brush-off. The lab countered, “It is important to continue to evaluate our internal frontier models on mathematics and other sciences, so we can accelerate developing the tools to advance those fields.”
In other words, mathematicians freaking out is good publicity before the company floats.
The Emmys Move to Prime Video for $15M

(Cheddar.com)
For the first time since 1949, television’s biggest night is cutting the cord. The Television Academy has officially abandoned its 31-year-old “wheel deal” — the annual broadcast rotation among ABC $DIS ( ▼ 1.45% ) , NBC $CMCSA ( ▲ 0.4% ) , CBS $CBS ( 0.0% ) , and Fox — to sign an exclusive six-year global partnership with Amazon’s Prime Video $AMZN ( ▼ 2.19% ) , reportedly worth around $15 million annually.
The move marks a seismic shift for legacy networks, who spent decades passing hosting duties around like a hot potato. TV Academy chairman Cris Abrego defended ditching linear television: “The wheel deal was incredibly successful for the Academy and the Emmys for a long time.… But as we approached the end, I felt strongly that it shouldn’t just default to the way it always had been done,” he told Variety. Abrego also pointed to the annual headache of resetting creative plans and sponsor relationships: “The continuity is difficult when you’re starting new every year.… We wanted a real kind of commitment and investment, to build and to evolve.”
Plus there was that fat Amazon check waiting, of course. And for Amazon, placing Hollywood’s crown jewel in front of its streaming portal is a massive flex. Under the deal, the telecast will stream free without a paywall across more than 240 countries and territories. Mike Hopkins, head of Prime Video and Amazon MGM Studios, called it “a fair deal” and welcomed rival platforms: “We want to have Netflix’s $NFLX ( ▼ 1.25% ) biggest stars, and we want to have Disney’s, and we want to have everybody else’s as well,” he said. And for viewers terrified that unconstrained streaming runtimes might create an endless marathon, Hopkins offered reassuring mercy: “I think a three-hour award show is plenty.” Us too.
Song of the Day: Kofi Stone, ‘Sicily’
"Sicily" is a highly praised, standout track from Birmingham rapper Kofi Stone's fourth studio album, The Last Flower to Bloom, which dropped this moth. The song features a guest appearance by R&B singer Mahalia and has something called “emotional depth,” which sounds a lot like this newsletter. (No, it doesn’t, Matt—Ed.)
Kalshi Touts ‘Transparency’ for the Midterms

(Kalshi’s midterm predictions)
If you want to gamble on congressional gridlock with the same cold intensity you bring to fantasy football, 2026 is your golden era. Prediction market Kalshi $KALSZZX ( ▼ 0.04% ) has officially rolled out a suite of “transparency tools” ahead of the midterm elections, evidently aiming to suggest that, er…betting on elections is, you know, fine.
As political betting surges mainstream, Kalshi is introducing three new features: “historical data benchmarks,” “real-time trade feeds,” and “warning labels for low-volume markets” where a handful of traders can distort the odds. Kalshi CEO Tarek Mansour hyped the launch: “This is a special time for us. Four weeks from now, Americans will be voting in the midterm elections. And this is really the first election after prediction markets have gone mainstream,” he said. Mansour added, “We are excited to showcase the powers of these markets in what feels like the first meaningful prediction market election.… It’s important to put these markets in context.”
The transparency push comes after, let’s say, a bump in the road in August, when Kalshi gave a primary candidate in Wisconsin a commanding 95% win probability — only for them to lose in an upset. Defending the platform's math on social media, Kalshi co-founder Luana Lopes Lara noted, “5% is not 0%.”
But telling traders that a 1-in-20 longshot isn't mathematically impossible doesn't soften the blow when their money vanishes. As democracy gets gamified, I’m sure there are absolutely no possible negative consequences.
Better Markets, meanwhile, issued a scathing critique: "Kalshi’s contracts pose numerous threats. They undermine the integrity of our elections by incentivizing misinformation and other forms of election manipulation for profit."
Free Yourself From Advertising Forever!
Now you can sign up for an optional ad-free version of Need2Know! Subscribe for just $5 a month, or $50 a year, and you can continue to enjoy this reasonably high-quality newsletter uninterrupted. Bonus: The immense satisfaction that comes from supporting journalism*!
*This counts as journalism, right?
ADVERTISEMENT
Built for the unstoppable.
The Dell Pro powered by Intel® Core™ Ultra with Intel vPro® adjusts power usage based on how you work, so your battery life never slows you down.
Starbucks Sued Over ‘Sugar-free’ Drinks

(Cheddar.com)
If you’ve been sipping on Starbucks’ $SBUX ( ▼ 3.23% ) protein lattes thinking you were a model of health and wellness, you might want to pause before taking your next gulp. A proposed class-action lawsuit filed in Seattle federal court accuses the coffee titan of deceptive marketing across eight beverages in its protein line, including hot and iced "Sugar-Free" Vanilla and Caramel Protein Lattes and Matchas.
While advertised as “sugar-free,” the complaint alleges these drinks actually pack between 13 and 21 grams of sugar per serving. That’s equivalent to eating a donut while convincing yourself it’s kale. The culprit? Naturally occurring lactose in the drinks' protein-boosted milk base. Under federal regulations cited in the suit, beverages labeled "sugar-free" cannot exceed 0.5 grams of sugar per 12-ounce serving. The plaintiffs argue that health-conscious consumers would “not reasonably expect beverages with ‘sugar-free’ in their names to contain that much sugar.”
Starbucks, however, is firmly pushing back against the bitter allegations. Firing back, the coffee chain declared, “We believe these claims have no merit.” Starbucks maintains it doesn't add sugar to the drinks, relying strictly on sugar-free syrup for flavor, and points out that full nutritional disclosures — including the natural milk sugar — have always been clearly posted across its menus, website, and app.
The lawsuit seeks monetary compensation for impacted buyers and an injunction forcing Starbucks to drop the "sugar-free" label on any drink containing… you know… sugar.
Should You Check Your 401(k) Today?
👎️
Not if you checked it yesterday, mate. Basically, things are flat since August.
Poll of the Day: Hindsight Odds Are 20-20
Poll of the Day: There’s Hope for You, Yet
We asked: How Did You Get Your Job? You answered:
🟩🟩🟩🟩🟩🟩 I applied for it without knowing a soul and beat out the competition with my talents. (266)
🟨🟨🟨🟨⬜️⬜️ Connections, baby! (178)
444 Votes via @beehiiv polls
![]() | Want more Cheddar? Watch us!Search “Cheddar” on Samsung, YouTube TV, and most other streaming platforms. N2K is the tip of of the cheeseberg for financial news, interviews, and more. |
Need2Know is covered by Cheddar’s Terms of Service
*We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites. This post may also contain other affiliate links, meaning when you click the links and make a purchase, Cheddar receives a commission at no extra cost to you.
P.S. So, you remember the cheese puns that used to open this newsletter? Suffice to say, they were divisive. Now, thanks to a thing called “dynamic content options,” I can offer you the option to see cheese puns again, if you’re one of the thousands who got in touch bemoaning their departure six months ago. All you need to do is answer “true” on this survey, and submit it. If you never want to see cheese puns in this newsletter again, don’t click that link, don’t fill out the survey, don’t submit it. Just keep reading and pretend this conversation never happened. Mmmkay? Thank you.


