📈 first Derivative [122]

People bet $30 million on Polymarket trying to predict the next pope and almost everyone got it wrong

(Raja Ravi Varma)

🤖 The WSJ recently covered the question of whether LLMs have world models or not. If you’d read fD, you would’ve read about the same research last summer, when I had my friend and fD reader, Keyon Vafa, contribute a few lines about his machine learning work.

📚 Started reading Team of Rivals. I’ll be posting some highlights as I go along here.

Highlights in this issue: popes & Polymarket, the Houthis, Kurdish militias

Good reading,

—TK

If you were forwarded this email, subscribe here.

---

🍸 Woke up hungover last weekend? I wrote about why that actually happens. Next time, take Last Call to help you metabolize alcohol and feel better the next day. fD readers get 20% off their first order with promo code “FD20”

---

🇻🇦 People bet $30 million on Polymarket trying to predict the next pope and almost everyone got it wrong. The Catholic Church made history and elected its first American, Cardinal Robert Francis Prevost, as Pope Leo XIV. If you read last issue you would’ve seen his name among the trending contenders.

I wrote a bit last time about why I expected the prediction market on the next pope to be less useful than in most cases. Structurally, there was not a lot of information going around and the people who had it weren’t likely to share it or introduce that information to a market themselves.

Nate Silver wrote on why the conditions where prediction markets impress were almost entirely absent in the case of the conclave:

In such situations, the conventional wisdom can just feed back on itself. Maybe it was reasonable a priori to think that Parolin or Tagle were more likely choices than Prevost, for instance. But these probably ought to have been pretty weak priors. Once the media begins to write about them as frontrunners, however, those perceptions can become entrenched. Everyone assumes that everyone else knows something, when in a case like this, they probably don’t.

And prediction markets potentially contribute to this process because of their seeming mathematical precision. That they can be extremely wise in some circumstances doesn’t mean that they will always be.

---

🇾🇪 NYT is reporting Trump ended the military operation against the Houthis after 30 days without substantial results

But the results were not there. The United States had not even established air superiority over the Houthis. Instead, what was emerging after 30 days of a stepped-up campaign against the Yemeni group was another expensive but inconclusive American military engagement in the region…

the cost of the operation was staggering. The Pentagon had deployed two aircraft carriers, additional B-2 bombers and fighter jets, as well as Patriot and THAAD air defenses, to the Middle East, officials acknowledged privately. By the end of the first 30 days of the campaign, the cost had exceeded $1 billion

Honestly, I respect the realistic decision-making here, although the campaign was initially planned for 8-10 months in scope. The Saudis and Emiratis have been bombing the Houthis for years. At the same time, I remain confused as to how the Houthis are so formidable, a deep dive for later. Maybe with modern technology, defending your territory and air space is asymmetrically easier.