Anthropic/Alibaba, Instagram betting, FIFA
Stealing profitably, Betting on gender reveals, Creating fixable problems
Margin Points:
→ Alibaba stole Anthropic’s crown jewels and somehow got poorer as a result.
→ Instagram is going to enable betting on gender reveal videos.
→ FIFA is creating problems and conveniently has the perfect solution handy.
Stealing profitably
Is someone richer after stealing the crown jewels?
Apparently not, according to today’s coverage of the Anthropic/Alibaba spat. Anthropic is accusing Alibaba of using 25,000 accounts to access Claude to glean information to help Alibaba improve their own AI model.
Alibaba’s stock fell a few percent in trading in Hong Kong. The Hang Seng index was down a few percent as well and the Bank of China fell even more. So, hardly a big movement at all.
And would we not have expected Alibaba’s stock price to have gone up? I mean, they are basically getting state-of-the-art R&D (valued in the billions of dollars or yuan) for free. Shouldn’t the stock move up?
Anthropic alleges that the 25,000 accounts exchanged 29 million messages with Claude in order to help train Alibaba models.1 Let’s assume for a second that you were monitoring Claude’s outputs in real-time and started to notice this type of attack happening. One thing you could do would be start feeding Alibaba bad advice. Not ridiculously bad, but time wasting, token burning, booby trapped stuff. You are smart, right, so you can do that. Make a maze and have them burn metal trying to scamper through it.
But instead you give them real good answers? This now reflects poorly on Anthropic. You’ve been railing for years about this being a threat and you weren’t able to monitor it and react dynamically. At least route them to a cheaper model and keep charging for the more expensive model and pocket the difference. That would help the P&L for the upcoming IPO.
This thinking could explain the slight stock drop. OK, we got some free R&D out of this, but we should price in that eventually Anthropic will get smart enough to interfere with the answers it gives and then the gravy train will end. That they know about the 25,000 accounts means they are on it—slowly—and unguarded flank is being fixed. According to the WSJ:
It’s not the first time Anthropic has said Chinese AI labs are using its technology to train their own AI models.
In a blog post in February, Anthropic said it had identified industrial-scale campaigns by three AI laboratories to extract Claude’s capabilities.
So Anthropic has had an eternity to fix this problem. Lengthening the lead over competitors and open source models surely falls somewhere on their priority list after growing revenue and going public.
Conveniently, Alibaba’s business model is not reliant on getting big enterprise contracts from US businesses that they could be blocked from selling to. The stock rises if this was free advertising to European and Asian businesses that the Alibaba models are pretty good now. The stock falls if there were concerns on further tariffs, hurting their ecommerce business, brought on by their Anthropic escapade .
Looking at it from the exact wrong angle—here’s a take on Alibaba’s stock:
“Given this is not the first distillation allegation targeting Chinese firms, I expect the reputational damage … to be small,” said Laila Khawaja, a research director at Gavekal Technologies.
Chinese company steals Western technology is hardly new. Chinese company is successful in stealing from world leader in AI is a net positive on Alibaba’s stock.
This reputational damage is entirely on Anthropic. They are leaking state-of-the-art information that they can’t get control of. Claude now looks smart enough that it’s worth using to make Chinese models better, but dumb enough that it isn’t able to identify what is happening and just gives up the goods.
Stealing the crown jewels is profitable, but cat and mouse usually isn’t.
Betting on gender reveals
Meta is working on a prediction market called Arena. I can see it incorporated into Instagram and Facebook, coming soon. The name connotes sports, and I’m sure they’ll do that too, but why not really leverage the advantages that come with all the content and the relationships with creators they have.
On Instagram, they can take a gender reveal video and slap pink and baby-blue buttons under it and wham, you’ve got one of the more popular pieces of content on Instagram now built for social betting.
Instagram can grow the in-app gambling through creators already on the platform. It can offer creators a share of exchange fees on markets they create, promote, or pump. Instagram might not need to pay anything—creators might decide the algorithm is smiling kindly enough on engaging with Arena that they just choose to promote it to expand their own reach.
Going live on Instagram from your kid’s Little League game? Well—now your followers can speculate on whether little Jimmy can get a hit.2
Much ink has been spilled on Polymarket and Kalshi working with influencers to pump markets. Meta has a built-in platform for mega-influencers to share and potentially monetize markets around their lives and interests. Which dirty soda flavor will I be slurping today? Will I be divorced by the end of the year?
Meta has huge distribution and better targeting than anyone (except maybe TikTok). Instagram already knows your favorite sports and teams. They have your attention. They have a real social element. Arena could be additive to their other engagement metrics. More time in app. More commenting. High retention.
Kalshi and Polymarket have both leaned into social elements of the trading, but it would obviously be mogged by the social graph that Meta brings to the table. Arena wouldn’t even need to be that successful to have an impact on Kalshi and Polymarket and Robinhood. Presumably, all the influencers that are paid to promote other prediction markets will see their reach throttled on Instagram. There’s always TikTok for them to pretend to win money on bets they didn’t actually make.
Arena might help the other prediction markets by continuing to move the Overton window on sports betting. Kalshi/Polymarket/Robinhood might also benefit if Meta decides to move quickly and buy Novig (up-and-coming sports-only competitor) which would knock out a well-funded competitor with traction.
Meta could also buy Fanatics for a parlay bet. You’d get a fully regulated U.S. exchange3 to take real money wagers and you’d get the merchandise business as well for deeper sports activations. That’s a sub-$20B acquisition for Meta. It also comes with a World Cup sponsorship.4
Don’t read anything into the fact that Meta is starting with an ‘internet points’ version which doesn’t involve betting real money.5 This allows them to move faster and experiment gleefully.
They’ll figure out the best way to ring the register with real money soon enough. Don’t believe that? Of course, you can bet on it…
Creating fixable problems
As the U.S. gets ready for its final group stage match today, we realize how complicated the formula is for deciding which teams advance to the next round. With 48 teams in the field, we cut to 32 in the knockout round. What we have is confusion, gratuitous complexity, and waiting on strange unrelated results—like if the groundhog saw his shadow—to learn who is still in and who they’ll face.
FIFA introduced this problem, knowing that they have a great solution for it.
Today’s WSJ profiled the Scottish national team now having to wait three days to figure out whether it will advance in the tournament. They lost last night to Brazil, which may have been their last game. In the past, you knew right away if you had made the knockout round.
Working out which third-place teams advance is taxing. Only 8 of the 12 teams make it.6 Which 4 don’t make it? Well, there is a formula for that. You need either a math degree or an LLM to make sense of it.7 Everyone seems to be resigned to just being told when it happens. There is something lost in that. We used to have the group of death. Now we have the group of mild headache.
It’s not just the fans checking Reddit threads when they should be watching. We have what is being preemptively called the “Disgrace of Kansas City” coming up on Saturday. FIFA’s fugazi 48-team tournament has created an own-goal:
This week, we’re about to see that the format has one gigantic problem.
The source of the trouble is Group J. ... This is the rare group where it might be worse to come second than third. And now, Austria vs. Algeria is turning into the World Cup game no one wants to win.
Something similar happened for different reasons8 back in 1982, when the “Disgrace of Gijón” occurred. A game where no one wants to win is called a dead rubber match and it feels deeply un-American happening here. Gijón led to rule and format changes.
FIFA knows the end state is 64 teams in the tournament.9 With 64, the math is simple. 48 was just a stepping stone as everyone can see the quality and fun of watching teams like Curacao and Cape Verde. The tournament has been very enjoyable. You can then play that out to imagine what Antigua & Barbuda and Wallis & Futuna will look like when they sneak into the tournament as the 63rd and 64th teams.10
Certainly, we must move immediately to 64 teams to solve this problem that FIFA themselves created. Was that so hard?
Caught my eye:
An Apple designer made a cool golf cart // LVMH and Hermes’ heir are fighting // SK Hynix files to raise $29 billion on the Nasdaq // Wendy’s is a meme stock now, up 26% on a “we need to save Wendy’s” Reddit thread. Maybe GameStop should have bought them before the post? // Micron surges 18% // Slate opens preorders on its $24,950 bare-bones EV pickup—crank windows, optional speakers—and the real money is in 175 accessory options and $500 wraps // Cerebras drops on its first earnings as it forecasts margins shrinking to 37% from 47% // Caleb Williams’ trademark application for Iceman denied because of “likelihood of confusion” with a 1988 filing by LaCrosse Footwear for insulated boots
Subscribe and forward it to your friends. Have feedback, thoughts to share or comments? Comment or reply.
Stealing seems like too strong a word for what is actually going on. It’s possible Alibaba’s engineers were asking Claude for feedback, which is a terms of service violation. As far as the accusations go, there doesn’t seem to be any hacking in to steal model weights.
The CFTC, who regulates prediction markets, has weighed in to block betting on Little League games. In theory there are markets to be had on whether little (or not so little) Frank’s birthday is what it says on the birth certificate he produced and his dad notarized or if that is a fake.
Via the Crypto.com partnership Fanatics has in place
Maybe still enough time to get that deal done before the final?
Basically like Manifold Markets
There are 12 groups of 4 teams to begin with in the first round. Each team plays the other three teams in their group.
I tried, briefly. Not easy, even with an LLM.
Final matches in the group stage weren’t started simultaneously as they are now.
NCAA Football playoffs seem to also be boiling the frog to get from the current 12 teams to 16 or 24 teams in the playoffs. It started with 4 teams in the playoff back in 2014.
They might not dominate the competition or inspire large nations to combine for the sake of World Cup supremacy. Italy & Uruguay combining across geographies would be interesting though.